EPIDEMIOLOGY

Epidemiology

Published by Quantum Scientific Publishing a division of Sentient Enterprises,

Inc. Pittsburgh, PA. Copyright © 2017 by Sentient Enterprises, Inc. All rights
reserved.

Permission in writing must be obtained from the publisher before any part
of this work may be reproduced or transmitted in any form or by any means,
electronic or mechanical, including photocopying and recording, or by
any information storage or retrieval system. All trademarks, service marks,
registered trademarks, and registered service marks are the property of their
respective owners and are used herein for identification purposes only.

QUANTUM SCIENTIFIC
PUBLISHING
 table of
CONTENTS
Chapter 1 − Introduction to Epidemiology 1
Chapter 2 − John Snow’s Story of Cholera 5
Chapter 3 − Infectious Epidemiology 9
Chapter 4 − Environmental Epidemiology 15
Chapter 5 − Psychosocial Epidemiology 21
Chapter 6 − Forensic Epidemiology 25
Chapter 7 − Ecological Study Design 31
Chapter 8 − Cross-sectional Study Design 37
Chapter 9 − Case-control Study Design 43
Chapter 10 − Cohort Study Design 47
Chapter 11 − Randomized Controlled Trial (RCT) Study Design 51
Chapter 12 − Surveillance and Mortality Data 55
Chapter 13 − Screening 61
Chapter 14 − Prevalence and Incidence 67
Chapter 15 − Ethics 71
Chapter 16 − Risk Factors 77
Chapter 17 − Sample, Population, and Sample Size 81
Chapter 18 − Sources of Data 85
Chapter 19 − Types of Surveys and Variables 89
Chapter 20 − Constructing Survey Questions 93
Chapter 21 − Answer Choices and Layout in Survey Construction 99
Chapter 22 − Survey Layout 103
Chapter 23 − Use of Incentives and Participation Proportion 107
Chapter 24 − Biomarker Matrices and Collection 111
Chapter 25 − Making a Map – MAUP 119
Chapter 26 − Chloropleth Maps 123
Chapter 27 − Making a Map – Elements 129
Chapter 28 − Use of Tables in Reporting Results 133
Chapter 29 − Graphs and Charts 137
Chapter 30 − Person, Time, and Place 143
Chapter 31 − Causality 147
Chapter 32 − Sampling Error and Chance 157
Chapter 33 − Sampling Frame 161
Chapter 34 − Confounding and Effect Modifier 165
Chapter 35 − Types of Bias 173
Chapter 36 − Generalizability 177
Chapter 37 − Publication Bias 181
Chapter 38 − Crude Rate, Numerator and Denominator 185
Chapter 39 − Direct Standardization 191
Chapter 40 − Indirect Standardization 195
Chapter 41 − Relative Risk 199
Chapter 42 − Odds Ratio 203
Chapter 43 − Confidence Interval 207
Chapter 44 − Evidence-Based Research 211
Chapter 45 − Public Policy 215
Chapter 1 - Introduction to Epidemiology
Chapter Objectives
• Deine epidemiology

• Describe what epidemiologists do

• List where epidemiologists work

What is Epidemiology?
The most common guess is that it is a skin disease. You probably remember from biology
that the outermost layer of the skin is called the epidermis. From Greek and Latin roots,
“epi” means “outside” and “derm” means skin. Therefore, epidermis means outside layer
of skin. However, epidemiology is not a skin disease.

Three Greek words make up the term, epidemiology: “epi” means on or upon; “demos”
means people; and “logos” means the study of. Therefore, epi demi ology means the study
of what is upon the people.

The formal deinition of epidemiology is the study of the distribution and determinants of
health related states in speciic populations, and the application of this controlled study of
health problems.

In more basic terms, it is the study of diseases in populations.

Epidemiology looks at epidemics (sudden increase) and endemic (always present) dis-
eases. One of the underlying beliefs of epidemiologists is that disease does not occur at
random. In other words, epidemiologists believe that there is some pattern to the disease. It
may be related to the persons, place or time.

What do Epidemiologists do?

Epidemiologists are disease detectives. They work with others to igure out the reason for
health problems in a community, region, country or countries.

They use special tools to look for clues. They count the number of people who have a dis-
ease or injury in a community, region, country, or countries. Using math, they igure out if
there is a pattern of disease or if a problem exists.

Just like police detectives, they gather information in a systematic way. Systematic means
getting things done using clearly deined steps. As in other scientiic disciplines, epidemi-
ology uses the scientiic method to deine questions, design and perform the experiment,
and analyze and present the data.

1
When epidemiologists solve the problem or have a good idea about why the problem oc-
curred, they let the public know. This can happen in many ways, such as making announce-
ments through the media, suggesting a change in a law, and educating the public.

Epidemiologists also use their understanding to prevent further illness or disease. They
do this by making suggestions about what factors may play a role in the pattern of disease
or illness. An example is health inspectors for restaurants making sure the pork is cooked
enough because we know pork that is not thoroughly cooked may cause illness.

Where do Epidemiologists Work?

Epidemiologists work in many different places. Some epidemiologists work in the local
health department. In the health department, one job of an epidemiologist may be to ig-
ure out the reason for an outbreak of food poisoning or whether a neighborhood has more
children in that neighborhood diagnosed with childhood leukemia than in other neigh-
borhoods. Some epidemiologists work in hospitals, non-proit organizations, or in large
governmental agencies, such as Centers for Disease Control and Prevention (CDC) or the
World Health Organization (WHO). Epidemiologists in these larger governmental agen-
cies often work on health issues at the national or international level. For example, CDC
and WHO would be involved in understanding the spread of avian inluenza or any cases
of poliomyelitis (short name is polio). Another group of epidemiologist works in universi-
ties. These epidemiologists either work on describing the whos, whats, whens, and wheres
of a disease or the whys and hows of a disease. In other words, they do the research that
uncovers relationships between different factors and disease. This information is used to
help improve the health of populations by establishing health programs and policies.

What have Epidemiologists Taught Us?

Many truths about disease and illness were discovered by epidemiologists. A hundred and
ifty years ago (mid 1850s), an epidemiologist showed that cholera was spread through
contaminated water. Remember, this was before we knew about bacteria. About 50 years
ago, epidemiologists (the Framingham Study, started in 1948) were able to tell us about
risk factors for heart attacks and strokes. Before then, we had little knowledge about who
was at risk for a heart attack or stroke or why some people were more likely to have a heart
attack or stroke and others were not. Epidemiologists were the irst to show that cigarette
smoking was linked to lung cancer (Drs. Doll and Hill; study in the mid 1950s). Epidemi-
ologists were also important is showing that a drug given to pregnant women, diethylstil-
bestrol (DES), harmed their babies. However, this harm did not occur until the children
reached adolescence or young adulthood. One of the harms was that daughters of mothers
who took DES could develop a rare cancer of the vagina or cervix. Epidemiologists have
taught us a lot about the health of people.

2
Summary
Epidemiology is the study of health and disease in populations. Epidemiologists are inter-
ested in both epidemic and endemic diseases. Epidemics are sudden increases in the inci-
dence of a disease. Endemic diseases are those that exist at a stable level in a population.
Epidemiologists work to igure out the causes of diseases in speciic areas or populations.
They use the knowledge they gain to prevent future disease. Many of the laws about restau-
rants and sanitation are based on epidemiological research. Public health inspectors have
the job of enforcing those laws to help prevent outbreaks of disease. Epidemiologists have
also helped us understand the way disease spreads through a population.

Concept Reinforcement
1. Explain epidemiology and discuss what epidemiologists do.

2. Explain why epidemiologists work in so many different types of organizations.

3. Discuss the role of epidemiologists played in understanding the effects of DES.

3
4
 Chapter 2 - John Snow’s Story of Cholera
Chapter Objectives
• Describe John Snow’s cholera story

• List the important inding of Snow’s study

• Describe the two competing beliefs about health at that time

Around two hundred years ago, death spread from India, China, Russia, Western Europe
and then to North America. The reason was cholera. It is an infectious disease caused by
a bacteria. Millions of people died of cholera. It was one of the most rapidly fatal diseases
known. A person could die within 3 hours of falling sick. At that time, it was one of the
most feared diseases because it seemed to make people look less human because they
seemed to shrink due to loss of water from diarrhea. Although there are still people who get
cholera in the world today, it is no longer a disease that kills millions.

Beliefs About Cholera in the Mid-1800s

Bad Air Belief

In London, there were two main schools of thought about the cause of cholera. One group
of people believed that cholera was caused by miasma. Miasma means bad air. People try
to protect themselves by wearing sweet smelling herbs. Men smoked large cigars. Because
of the miasma belief, London oficials searched for bad smells. Below is a cartoon showing
the search for bad smells.

Cartoon showing search for bad air by London oficials

5
Bad Water Belief
When John Snow was 18 years old and in his 4th year as an apprentice to Dr. Hardcastle,
a cholera epidemic hit London. Dr. Hardcastle had so many sick patients that he sent John
Snow to treat sick coal miners alone. He could do little to help them and many died. He did
a lot of thinking about cholera. In his early 30s, he was a successful doctor in London. He
thought that cholera was spread through contaminated water.

Two water companies supplied water to London. One company, the Southwark and Vaux-
hall Water company, got its water from an area of the Thames River known to be polluted
with sewage. The other water company, the Lambeth Water Company, pulled its water
north of the city where the water was less polluted.

John Snow showed that people who drank water from the Southwark and Vauxhall Water
Company were much more likely to get cholera compared to people who drank water from
the Lambeth Water Company. In one month, 286 people who drank water supplied from
the Southwark and Vauxhall Water Company died of cholera compared to 14 people who
drank water supplied by the Lambeth Water Company.

He wrote and spoke about his indings. Unfortunately, he was not believed. He was one of
the irst to use a survey of people and look at where the disease was occurring to igure out
the cause of the disease.

In August of 1853, anther epidemic hit London in an area close to John Snow’s home,
which was in the west London district of Soho. John Snow mapped out the places where
people died of cholera in one area of London. This area, the Soho area of London, had
many, many deaths. The people in the Soho area had one thing in common. They used the
Broad Street pump’s water.

As further evidence of the culprit, John Snow noted that the work house, which is like a
prison, had only 5 deaths. The work house was a smelly place. If the miasma idea was
correct, then most of the 55 inmates should have died. In reality, the workhouse was a
safe place to live. The workhouse had its own deep well and did not use the Broad Street
pump’s water. Another local place, the brewery, did not have any deaths. Again, it was a
very smelly place. Based on the miasma idea, many people who worked in the brewery
should have died of cholera, but no one did. Why? The brewery workers were given free
beer to drink, so they did not drink the water.

Map made by John Snow showing places that people died in the Soho area of London

6
After much detective work, John Snow was able to show that the water from one pump, the
Broad Street pump, was the source of the disease. However, his idea was not accepted and
he died a couple of years later, in 1858, at the age of 45. Finally, in 1884, the bacteria that
caused cholera was re-discovered. John Snow’s idea was at last believed.

This example of detection work in iguring out the reasons that people get sick or die still
continues today. It may be hard to imagine that people had no idea about microscopic or-
ganisms and their role in causing sickness and death. Even when John Snow identiied the
source of cholera as drinking bad water, he did not know what was actually causing the ill-
ness. This also is common today in epidemiology—epidemiologists identify the risk factor
but not the cause of the disease or death.

Summary
John Snow is known as the father of ield epidemiology because he went into the ield to
gather data. He interviewed people and he mapped the location of cholera deaths to igure
out the common factor. In the irst stage of John Snow’s research into the cause of cholera,
he igured out that people who drank the water from Southwark and Vauxhall Water Com-
pany were likely to get sick. Next, he went door to door and igured out how many people
died of cholera in a small area of London. He also asked them where they got their water.
In this area, people usually got their water from the nearest pump, the Broad Street pump.
He also found out which water company serviced that pump. Through this detective work,
he felt sure that drinking bad water was the reason people got sick and died of cholera. John
Snow’s efforts to igure out the risk factor for cholera—drinking bad water—, his survey-
ing people to get information about risk factors, and his mapping of the area of cholera
deaths,is a classic example of epidemiology.

Concept Reinforcement
1. What factors may have created the situation in which people thought it was bad air
that caused the cholera epidemic?(Hint: think about where refuse was deposited in
the mid-1800s and rich and poor people’s living conditions and nutritional status.)

2. What are two means used in epidemiologic detective work to discover risk factors
associated with disease or death?

3. Why would a densely populated city be more susceptible to the cholera epidemic
than the rural countryside?

7
8
 Chapter 3 - Infectious Epidemiology
Chapter Objectives
• State the 3 major factors in infectious disease transmission

• Describe the procedures of investigating an outbreak

• Deine herd immunity

Controlling infectious disease is important. Although the United States has fewer deaths
these days than 100 years ago from some types of infectious diseases, such as cholera or
tuberculosis, some types of infectious diseases still kill many people. For example, pneu-
monia-inluenza was the 8th leading cause of death in 2005 in the Unites States.

Epidemiology Triangle
The epidemiologic triangle is one way to show the major factors necessary for an outbreak.

Epidemiology triangle

9
he Agent
The biological agent in infectious disease is studied in microbiology. In epidemiology, we
use the microbiologists’ discoveries. For an infection to occur, the agent must be present.
There are six types of microbial agents: bacteria, viruses and rickettsia, fungi, protozoa,
helminths, and arthropods. Each type of agent causes certain diseases. See the table below
for a list.

Agent Diseases Comment

Leading killers in the 19th century.
TB (Tuberculosis), salmo-
Antibiotics are used to treat this type of
Bacteria nellosis, streptococcal infec-
infection. Concern about bacteria resis-
tions (strep throat)
tant to antibiotics.
Virus means poison in Latin. Some
Hepatitis A, Rocky Moun-
Viruses and people can be carriers of viruses, mean-
tain Spotted Fever, chicken
Rickettsia ing they do not have any illness but can
pox, inluenza, AIDS
infect others.

Fungi means mushroom in Latin. Some

Fungi Athlete’s foot, ringworm fungi are used for good, such as yeast to
make bread. Others can cause disease.

Single cell organisms that cannot be

Malaria, giardiasis, amebia-
Protozoa seen with the naked eye. They live in
sis
soil, wet sand, and fresh and salt water.

Found in moist tropical regions. Also

Roundworms, tape worms, known as parasitic worms.
Helminths
snail fever
They live inside the person.
Malaria, Rocky Mountain Living things carry the disease agent:
Arthropods
spotted fever, encephalitis mosquitoes, ticks, lies, mites.

he Environment
The environment means the place where the agent lives. The environment is made up of
many factors:

Physical environment

• Humidity

• Weather

• Temperature

• Type of landscape

10
Social environment of the people

• Culture

• Behavior

• Personality

• Attitudes

For example, in the mid-1800s, water was drawn from the Thames River in London. One
water supply company drew its water from the river that was near sewage dumping area.
In the warm months of the year, the cholera bacteria could live in this water. Fecal matter Typhoid Mary.
(also known as stool or bowel movement) from someone sick with cholera contaminated
Mary Mallon
the water. People who drank this water became sick with cholera. The physical environ-
(9/23/1869 –
ment consisted of a warm river climate. The social environment consisted of the habit of
11/11/1938) was a
pouring raw sewage into the local river and drawing drinking water from this river, too.
famous carrier of
typhoid. She was a
he Host cook for the wealthy
in New York City.
An infectious disease agent needs a favorable environment to live. The agent must also Health oficials
have the ability to infect a person. Once infected, a person may or may not show signs of discovered that she
sickness. If the infected person does not get sick, he or she is called a carrier. One very had infected the
famous carrier was Mary Mallon. She never believed that she infecting people. She felt she families where she
was healthy. However, she was a typhoid carrier. was a cook. She was
arrested and held
People have ways of blocking infections. Most agents cannot enter our body because our in isolation for 3
skin protects us. Tears and saliva can wash away infectious agents. Our stomach juice is years at a hospital on
very acidic and therefore kills many agents. Finally, our immune system can ight infec- North Brother Island.
tious agents. However, people still can get an infectious disease. However, Mary Mal-
lon did not feel sick
If a person gets sick, there are four outcomes: and did not believe
that she was a car-
1. C
omplete recovery. In other words, the person goes back to feeling the same as rier. After 3 years on
before the illness. the island, she agreed
2. P
ermanent disability. In this case, the person is not the same as before. For ex- to never work as a
ample, if a person gets polio, that person may recover but have a limp. Because the cook and was freed.
virus destroyed some motor neutrons in the leg, muscle weakness in the leg leads However, she started
to limping. working as a cook
again at New York’s
3. D
isigurement. A bad break, for example of the pinkie inger, can end up with the Sloane Hospital for
pinkie inger not being able to be straightened. This is a disigurement of the hand. Women. She infect-
ed 25 women there.
4. Death. Illness can lead to death. Mallon was arrested
again and sent to the
island. She never
left the island again.

11
Herd Immunity
Herd immunity is a term that means that a population, group, or community may not be-
come infected with a disease because most of the people have been vaccinated. In other
words, not everyone in a community has to get a vaccine for the whole community to be
protected. To become sick with an infectious disease requires contact with another person
who has the infection. The likelihood of coming in contact with a infected person is very
small when most of a community is vaccinated. For example, if only three-quarters of the
population is vaccinated, polio can spread throughout the community. On the other hand,
if eighty-ive percent of the community is vaccinated, the chance of polio spreading in the
community is very low. The exact percent of vaccinated people needed to protect the entire
community depends on the disease. For polio, it is approximately 85%. This is called herd
immunity. Very few people get polio today in the United States and we do not have epidem-
ics of polio because more than 85% of the children are vaccinated. However, if more and
more parents choose to not vaccinate their children against polio, we may see an epidemic
because the percent of vaccinated children falls below 85%. So, even if you do not get a
vaccine or a lu shot, you can thank those who do get one. They are keeping the community
healthy.

Typhoid Mary

So Who is More Likely to Get an Infectious Disease?

As we grow older, our ability to ight infection decreases. People who do not have balanced
diets cannot easily ight infectious diseases. Some people are born with the ability to ight
infectious diseases less well than others. It is in their genes. Older people, malnourished
people and people born with certain types of genes are more likely to get infectious dis-
eases.

12
Detecting an Outbreak
An epidemiologist follows ive steps in detecting an infectious disease outbreak.

Step 1: Verifying the outbreak

The irst step is to make sure there is an outbreak. For example, if a customer in a restau-
rant said that he became sick because of the food he ate, the epidemiologist would have to
make sure he did have food poisoning and not an upset stomach. If there is an outbreak,
the emergency room is usually looded with lots of people with the same complaint. Doc-
tors are very good at making a diagnosis: in this case, whether or not the person had food
poisoning.

Step 2: Identiication of All People with the Illness

The second step is inding all the people with the disease or illness. Once found, the people
describe their symptoms. A laboratory test may also be done at the hospital to help igure
out the illness or disease. Mapping where the people were when they began to feel sick
also occurs during step 2. Information on the date and time when people felt sick is also
gathered. At this point, the epidemiologist may be able to igure out what infectious agent
is responsible for the outbreak.

Step 3: Make a Tentative Explanation for the Outbreak

The third step involves making a statement that indicates the possible source of the infec-
tion, the likely agent, and the likely method of spread.

Step 4: Compile Evidence

The fourth step involves inding evidence to support or not support the statement made in
Step 3. During this time, the search continues for additional infected people.

Step 5: State the Cause of the Outbreak

The ifth and inal step uses the results of the investigation to make conclusions. These
conclusions may cause a restaurant to close or a beach to shut down. The purpose of the
action is to prevent another outbreak.

13
Summary
No country is immune from infectious diseases, though the important infectious diseases in
speciic countries vary. One way to ight against some infectious diseases is through vacci-
nation. If most people in a community are vaccinated against an infectious disease, such as
polio, it is unlikely that an epidemic will occur. Some diseases cannot be prevented through
vaccination and there are outbreaks. If this happens, epidemiologists act as detectives to
igure out who was infected and how they were infected. Once this is known, protective
action often occurs so others in the community will not become infected.

Concept Reinforcement
1. What do you think about Typhoid Mary? She never felt sick and never believed she
was a carrier of Typhoid. Do you agree with her punishment to make her live most
of her life on an island?

2. Do you think everyone should be required to get vaccines or is it OK to rely on

herd immunity?

3. How are these ive steps in investigating an infectious disease outbreak similar to
the methods used by police detectives in solving a crime?

14
 Chapter 4 - Environmental Epidemiology
Chapter Objectives
• Deine environmental epidemiology

• List the three sources of pollution

• List three factors to consider in exposure to a toxin

In the 21st century, attention has turned to environmental exposures and health outcomes.
Some experts believe that about one quarter of the diseases and deaths worldwide are
linked to environmental factors. These environmental exposures are outside the control
of the exposed individual. These can be from exposure in the workplace, toxic chemicals,
indoor and outdoor air pollution.

Albert Einstein, a famous scientist, said “The environment is about everything that is not
me.” Environmental epidemiologists look at environmental exposures that are related to
health. The deinition of environmental epidemiology is the study of diseases and illnesses
linked to environmental factors. There are many kinds of health problems, such as cancer,
birth defects, or infectious diseases. Each one of these may be linked to environmental
exposures.

There are three types of sources of pollution:

• Point sources

• Line sources

• Area sources

The type of pollution source plays a role in solving the problem of where the exposure is
coming from. If the source can be identiied, solutions can be proposed to reduce or elimi-
nate this exposure.

Example of point source pollution

15
Point source means that the pollution comes from one place. That place can be identiied.
Factories are point sources. The smokestack or the discharge pipe are both examples of
point sources in factories. Containers full of toxins are also a point source. You can point
to the place where the exposure begins.

Example of line source pollution: Electrical transmission towers and high-voltage

power lines near Albuquerque, New Mexico. The arrow indicates place where
crackling sound is heard.

The best example of a line source of pollution is power lines. Another common line source
of pollution comes from vehicular exhaust along a highway. Like point sources, line sourc-
es can be easily seen. The difference between the two is that point sources are in one place,
whereas line sources follow a line of exposure—along the line of high-voltage power lines
or along the line of a highway.

Example of area source pollution: Air pollution in Beijing China

Areas sources of pollution come from afar. The origin may be a line or point source, but
the affected area faces pollution of its entire area. Air contaminants from sources far away
can create local air pollution problems. Local water can be contaminated by toxins that are
carried from point or line sources and fall into the local water. This can contaminate the
local water.

There are many examples of mass illness, disease or death linked to environmental expo-
sure over time.

16
Examples of Environment Exposures Linked to Health

he Great Fog of London

A great fog fell on London in 1952. At irst, Londoners did not realize that this fog was a
sign of disaster. They were used to fog. However, many things came together to make this
fog deadly. When it inally lifted, about 12,000 people died, mostly from breathing prob-
lems. This was one of the irst disasters that made people realize that the quality of the air
is strongly linked to health. This is an example of area source of exposure, the air pollution.

These days, it is common to hear about ozone alert days in the summer. These are days in
which ozone, an air pollutant, is high and people who have trouble breathing need to be
careful.

Great fog of London in 1952

Love Canal
Hooker Chemical Company made a canal and illed it with hazardous chemicals, topped
the canal with dirt, and sold it to the city of Love Canal for $1. In the late 1950s, about
100 homes and a school were built on this land. Many of the homeowners did not realize
they were living on top of a toxic waste dump. Through natural weathering, the buried
toxic waste containers migrated to the surface and leaked their contents. Children playing
returned home with burns on their hands and faces from these chemicals.

One of the health effects from exposure to these toxic chemicals was birth defects and
childhood cancer. In 1978, President Carter approved emergency inancial aid to help deal
with Love Canal. The State of New York bought the homes of 221 residents so they could
move somewhere else. This is an example of point source of exposure–the containers illed
with toxins.

Love Canal, New York

17
Times Beach
A dusty small town outside of St. Louis Missouri hired Russell Bliss, a waste hauler, to oil
the roads in the town. He sprayed the roads with his oil mixture from 1972-1976. Unknown
to the residents of Times Beach, the oil was not just used oil but it had large amount of
other toxins, including dioxin. Dioxin is thought to be the most toxic chemical made by
man. After looding in 1982, the Environmental Protection Agency (EPA) found dangerous
levels of dioxin in Times Beach’s soil. The federal government bought the homes of Times
Beach residents and evacuated the town in 1983. The former residents of Times Beach
were shunned by people in their new communities. People were afraid that exposure to di-
oxin was contagious (which it is not). Over 265,000 tons of soil were collected and burned
from March 1996 to June 1997 to get rid of the dioxin. The cleanup cost was $110 million
dollars. This is an example of an exposure beginning at a line source–the oiled roads.

Times Beach, Missouri

Gulf War Syndrome

In the previous examples, the link between environmental exposure and health outcomes
has been determined. In case of Gulf War syndrome, much is unknown.

By the late 1990s about 15% of the 700,000 US veterans of the Gulf War complained of
health problems. The combat vets say that their health problems are linked to something
that happened to them while serving in the war. Some of the wide-ranging symptoms are
chronic fatigue, headaches, dizziness, loss of balance, memory problems, joint pain, indi-
gestion, muscle weakness, skin problems, and shortness of breath. This is called Gulf War
Syndrome.

There have been a lot of disagreements about whether or not the Gulf War Syndrome is a
physical, medical condition. Many believe it to be psychological; the result of serving in
combat. Others believe it is the result of exposure to some toxin. Although many ideas ex-
ist, the detective work is ongoing.

Oil well ires during the irst Gulf War

18
Factors About Environmental Exposures
There are many things found in the environment that may not be good for you. However,
our body protects us from most of these toxins. Harm from a toxin depends on several fac-
tors:

• Amount of toxin—how much toxin is the person exposed to?

• Toxicity—how dangerous is the toxin?

• Exposure duration—how long a person is exposed?

• Mode of entry—how does the toxin get inside the person?

• Timing—when does the exposure occur (e.g., during pregnancy, childhood,

adulthood)?

• Health of the person—the state of the person’s health at the time of exposure.

• Genetic makeup—people have differing abilities to deal with an exposure based

on their own genetic makeup.

All these factors must be considered at the same time. The impact on people’s health var-
ies by type of toxin. Some toxins are deadly in very small amounts, such as botulinum.
Foodborne botulism is caused by a bacteria that creates a deadly toxin. This toxin causes
paralysis. Because store-bought canned goods are required to be cooked at 121 °C (250 °F)
for 3 minutes, it is rare for someone to get botulism from eating store bought food. Home-
canned foods with low acid content, such as carrot juice, asparagus, green beans, beets,
and corn are more likely to be contaminated with botulin toxin The effects of other toxins
depends on the timing of exposure, such as exposure to methylmercury, a form of mercury.
The source of methylmercury is eating ish. Mercury seems to accumulate in babies during
their time growing inside their mother. Exposure in utero compared to adulthood has much
more known serious health effect . For an environmental epidemiologist, iguring out the
source of exposure is one step in solving the reason for a disease. Another important step is
thinking about other factors that play a role in the occurrence of the disease.

19
Summary
The relationship between environmental hazards and the health of human communities is
of growing concern in the 21st century. The type of pollution source varies from point, area
or line source. Whether these sources affect a person’s health depends on many factors
such as timing of exposure and exposure duration.

Concept Reinforcement
1. In your community, can you give an example of the three sources of pollution?

2. What kind of land was your school built upon? Is there anyway that this land could
be contaminated from its past use?

3. What does environmental epidemiology mean to you?

4. Does your community have ozone alert days? How does it affect your life?

20
 Chapter 5 - Psychosocial Epidemiology
Chapter Objectives
• Deine psychosocial epidemiology

• Discuss the stress concept as it relates to disease

• Describe examples of the sociocultural inluences on health

Psychosocial epidemiology focuses on stress and health. Psychosocial epidemiology also

uses surveys to gather data. These surveys ask about knowledge, abilities, attitudes, and
personality traits as they are linked to health. Epidemiologists are learning more and more
about the importance of non-biological factors and health.

There are many psychosocial factors, such as gender (male or female), age, socioeconomic
status (e.g., how rich you are), health risk behaviors (e.g., whether you smoke, drink al-
cohol or use illegal drugs), life events (e.g., accidents, deaths, moving parental divorce),
personality and temperament(inborn personality traits), as well as factors related to the
place of work.

Personality, lifestyle and environment can all affect whether a person gets sick or not. A
balance between an individual’s coping skills and his or her stress level can also be the tip-
ping point of whether one gets sick or not. An individual’s belief system can inluence the
course of a major illness. For example, women with early breast cancer who scored high on
helpless measures were more likely to relapse or die within ive years of being diagnosed.
Psychosocial epidemiologists look for these links.

Examples of Psychosocial Factors and Health

Breast Cancer
Women who have mastectomies with an attitude of “I’m going to beat this” do better than
women who have mastectomies and are less positive. Those with a positive attitude live
longer and are less likely to have a recurrence of breast cancer. Size of the tumor, the age
of the women, or the severity of the illness does not matter.

Cardiovascular Disease (CVD)

Cardiovascular diseases, such as a heart attack, are the leading cause of death and disabil-
ity in the United States. Type 1 personality—those who are very hard-driving, ambitious,
competitive, time-urgent, and unusually quick-tempered and tightly wound, are at an in-
creased risk. Type 1 personality is the irst psychosocial factor to be accepted in the medical
community as a CVD risk factor.

21
Cystic Fibrosis (CF)
Cystic Fibrosis is the most common inherited life shortening disease among Caucasians.
People with CF live to around 40 years of age. Males live longer than females. One neces-
sary treatment for CF is eating high-fat, high calorie diets. Adolescent girls are less likely
to eat this kind of diet. This may be due to societal pressures adolescent feel to stay thin.
This may be one partial explanation of the difference in male and female length of survival.

How do Psychosocial Processes Inluence Health?

This idea is very complex. The cause and effect path is not straight. For example, smoking
cigarettes for decades will signiicantly increase the chance of getting lung cancer. The path
of smoking to lung cancer is straightforward. However, when looking at psychosocial fac-
tors, the path is not so straightforward. It may be that some of these psychosocial factors
play a role into the length of survival.

Often, psychosocial epidemiology involves diseases that have many risk factors. The inlu-
ence of each risk factor plays a role. However, we do not yet understand to what degree
each factor is important. For example, hypertension, blood lipids, smoking, diet, and per-
sonality traits are all known risk factors for heart disease.

Stress
It seems like common sense that stress affects health status. In the late 1960s, scientists
described the 10 leading life change events that cause stress. These are:

1. Death of spouse
2. Divorce
3. Marital separation
4. Jail term
5. Death of a close family member
6. Personal injury or illness
7. Marriage
8. Bring ired from a job
9. Marital reconciliation
10. Retirement

The more stressful life events happen, the greater chance a person has of getting sick. Of
course, this does not mean that each person will get sick with a life change. It does mean
that someone is more likely to get sick. For example, in college, inal exam week is very
stressful. Some students become sick either during the exam week or immediately after the
tests are over. We do not understand why some students get sick and some do not but we
do know that stressful events increase the likelihood of getting sick.

22
Laughter
In contrast to stress, laughter has been shown to be positive. Using humor decreases stress,
lessens pain, and improve the quality of a person’s life. Some studies have even shown that
humor improves a person’s ability to ight infection. Dr. Patch Adams believed that laugh-
ter, fun and play increased the recovery rate of patients. A movie was made about him.

Summary
Psychosocial epidemiology focuses on the study of stress and health. The key tool used to
collect data for psychosocial epidemiology studies is the survey. These surveys collect in-
formation about gender, age, socioeconomic status, health risk behaviors, employment, life
events, and personality and temperament. Personality, lifestyle, environment, coping skills
and stress level all affect whether a person becomes ill. Stress has been shown to contribute
to illness, while laughter and positive attitude has been shown to be positive.

Concept Reinforcement
1. What makes iguring out the psychosocial factors associated with disease dificult?

2. What do you think has more impact on one’s health, being stressed or being happy?
Why?

3. How does one’s culture and environment relate to health? Think about the cultural
difference between the Western vs non-Western country.

23
24
 Chapter 6 - Forensic Epidemiology
Chapter Objectives
• Deine forensic epidemiology

• Describe who works with epidemiologists

• Provide an example of the use of forensic epidemiology

Forensic epidemiology is a new ield. Since the mid-1970s public health and law enforce-
ment oficials have worked together on health problems that may have a criminal intent.
The term forensic epidemiology was irst used in 1999. It describes an expert witness who
was an epidemiologist. After the 9/11 event in the fall of 2001, the need for a ield called
forensic epidemiology was obvious.

The basic deinition of forensic is “used in legal proceedings” and the term epidemiology
is the study of the distribution and determinants of health related states in speciic popula-
tions, and the application of this controlled study of health problems. By combining both
terms, forensic epidemiology means the use of epidemiologic methods as part of an ongo-
ing investigation or a health problem that may have a criminal intent.

Forensic epidemiology is about epidemiologists and law enforcement working effectively

in the event of a threat or attack involving biological or chemical weapons. The primary
impact of forensic epidemiology has been in the courtroom. Epidemiologists testify as ex-
pert witnesses. More recently, epidemiologists have worked with law enforcement oficials
in the ield, such as during the 2001 anthrax investigation. Anthrax is a serious disease that
is caused by bacteria. In the 2001 anthrax investigation, the bacteria was included in enve-
lopes and people who opened the envelope got sick.

In the Courtroom
Epidemiologists are often experts when a case involves environmental exposure and dis-
ease or death. Epidemiologists can help explain scientiic principles about exposure or ex-
plain statistical principles used to igure out the chance of the disease (or death) occurring.

However, science and law are different in basic ways. For example, science is forever
changing. The “truths” today may be “falsehoods” tomorrow. Law is inal. Epidemiology
is the study of populations and not individuals. In court cases, individuals are involved. The
court decision is about alleged harm to an individual(s). Law requires a causal link between
the exposure and the disease (or death). In epidemiology, it is often impossible to make a
100% link between the exposure and the disease. Even with these differences, epidemiolo-
gists are increasingly being used in the courtroom to help with cases.

25
 In the Field
Several legal issues arise when law enforcement oficers and epidemiologists work to-
gether in the ield. Both are interested in protecting the public. However, the rules for each
may be different. Areas that may cause problems are:

1. gathering of evidence during public health investigation that is admissible in court

2. access to premises

3. establishing and maintaining a chain of custody of evidence

4. disclosure of conidential health information by public health to law enforcement

5. restricting a person’s freedom of movement following exposure to communicable

diseases

Let’s look at 3 of these legal issues more closely.

Disclosure of Conidential Health Information by Public Health and

Law Enforcement
One important part of epidemiology is the protection of human rights. This can be a prob-
lem in the case of possible criminal intent to hurt another.
Tuskegee Syphilis
Study South Dakota Situation with HIV Infection
In 1932, the Public Epidemiologic research was the leader in determining the link between sexual contact and
Health Service and the spread of HIV. The transmission path is now widely known. Many public health of-
the Tuskegee Insti- ices provide free HIV testing. However, there are legal issues about this practice. In a
tute began a project widely published case in rural New York State in mid-1997, six young women with newly-
on the natural course diagnosed HIV infection reported having had sexual contact with the same infected man.
of syphilis. They Using voluntary partner notiication and contact tracing, 17 cases of HIV were identiied.
enrolled poor, illiter- Publicity surrounding this situation led to legislative attempts to adopt new criminal expo-
ate black men from sure laws. Although this new law was not adopted in New York, a similar one was adopted
rural Alabama (399 in South Dakota.
with syphilis and 201
without syphilis). In South Dakota in 2000, the new law made intentional (knowing) exposure of others to
Those with syphilis HIV a felony, for which a person could get up to a 15 year prison sentence. One of the irst
did NOT receive persons found guilty under this law was a college freshman who had sexual intercourse
proper treatment with his girlfriend after he found out he was HIV positive.
when penicillin
became the accepted However, during the course of this investigation, a major problem arose. The South Dakota
treatment. This is a health authorities were required to not tell anyone besides the person with HIV about the
classic example of test results even if they believed that this person was not telling his partners. As a result of
violation of human this issue, speciic language was introduced to allow health oficials to disclose this kind of
rights in a research information to law enforcement oficers.
setting.

26
Gathering Admissible Evidence in Public Health Investigators
Even when the situation is very clear, such as the threat in the September 2001 anthrax-
laced letter to Senator Tom Daschle. In gathering evidence, law enforcement must comply
with the Fourth Amendment prohibition against unreasonable searches and seizures and
the Fifth Amendment protection against self-incrimination. Self-incrimination means tell-
ing law enforcement information that points the inger at oneself for committing a crime.
If a public health oficial is leading the investigation and does not obtain a search warrant,
then it is likely that evidence obtained cannot be used in the court.

Image of the envelope in which the letter containing Anthrax was sent to
Senator Tom Daschle during the 2001 anthrax attacks.

The return address given in the top left says: “4th Grade, Greendale School, Franklin
Park, New Jersey, 08852.” There is no Greendale School at that address, though there
is a Greenbrook School in the locality.

Tests conducted at USAMRIID conirmed the presence of ine, “energetic”, powdered

anthrax within this prestamped 34 cent transmittal envelope. Also present was a one page
handwritten letter, clues from which enabled the FBI to create a proile of the sender.

The letter was postmarked at the Hamilton Township postal facility at 5:45 p.m. on Oc-
tober 9 2001. By October 11, it had reached the Brentwood postal facility in Washington
which processes US Government mail. Two postal workers, Joseph Curseen Jr.
and Thomas Morris Jr., died after contracting inhalational anthrax at the Brentwood
facility. The 14,000 square foot facility was decontaminated on December 14, 2002
using chlorine dioxide gas. When it was reopened 26 months after the incident the facility
was renamed the “Joseph Curseen Junior and Thomas Morris Junior Processing
Distribution Center.

Image courtesy of the FBI

27
Restricting Freedom of Movement in Response to Public
Health Emergencies
In 2003, the SARS outbreak triggered the widest use of quarantine globally since the early
1900s. SARS stands for severe acute respiratory syndrome. In early 2003, SARS spread
from China to almost 40 countries around the world. In Canada, almost 20,000 people were
kept in home quarantine for 10 days. Only a handful of Canadians refused to stay home.
During this time, all their needs, such as food, medicine and work excuses were met by the
public oficials. In this case, SARS was not purposely given to others. However, forensic
epidemiologist and law enforcement worry about other deadly diseases being given to a
few people intentionally and then being spread, like SARS spread.

Map of SARS cases from November 1, 2002 to April 8, 2003

28
Summary
Forensic epidemiology is a new ield of study. One characteristic of this ield is epidemi-
ologists and law enforcement work together on cases. Both ields involve detective work.
In forensic epidemiology, the health outcome has a criminal intent. It did not happen by
chance but rather was intentional.

Concept Reinforcement
1. Do you think it is OK to release a name to law enforcement of someone with HIV?

2. Would you support the quarantine of anyone upon lying back home who had trav-
eled to a country with an infectious disease outbreak?

3. Do you think it is reasonable to apply the term “reasonable doubt” to dismiss a case
when there will always be less than 100% conidence is an epidemiological inding
linking an exposure to a health outcome that was not immediate, such as leukemia?

29
30
 Chapter 7 - Ecological Study Design
Chapter Objectives
• Deine the basic difference between observational and experimental epidemiology

• Describe an ecological study design

• List 2 advantages of an ecological study

Many study designs are available to the epidemiologist. Choosing the appropriate study de-
sign depends on many things. In general, the amount of information already known drives
the choice of the study design.

Two characteristics of a study will help the epidemiologist pick an observational, quasi-
experimental, or an experimental study. Quasi-experimental design means that it looks
like an experimental design but one key ingredient is missing—the random assignment of
participants. These characteristics are whether the exposure of interest is controlled by the
individual and whether the study participants can be randomized.

Exposure of Interest Not Controlled by Individual

Epidemiologists link exposures to health outcomes. For example, epidemiologists showed
that smoking is linked with lung cancer. The exposure of interest was smoking. This ex-
posure is controlled by the individual because the individual chooses whether or not to
smoke. It would be wrong to have the researcher or government make people smoke in
order to study the risks. In a different situation, exposure to air pollution is not controlled
by the individual. By living in a city, a person has little control over the air s/he breathes.

Randomization of Participants
Randomization refers to the chance of being assigned to one exposure or not. For example,
a participant can be put into a smoking cessation program or not based on the lip of a coin.
If the coin lands on heads, the person goes into a new type of smoking cessation program
and if the coin lands on tails, the person does not go into this new type of smoking cessation
program but rather gets the regular treatment.

Exposure of
Randomization of
Study Type interest controlled
participants
by individual
Experimental No Yes
Quasi-Experimental No No
Observational Yes No

31
Study Design

Experimental
A common experimental study design is the randomized controlled trial (RCT). This type
of study design is used primarily in research and teaching hospitals. One purpose of an
RCT is to test a new drug, therapy or surgical procedure on patients with the disease of
interest. Experimental study designs that take place outside of a hospital setting are known
as randomized community trials. Community trials are usually about education and behav-
ior change at the population level. Common community trials include smoking cessation
programs, physical activity and weight loss programs.

Quasi-Experimental
The second type of study design is quasi-experimental. In this case, the exposure is con-
trolled by the researcher, governmental agency, or even nature. However, unlike RCT,
chance does not dictate who is exposed or not. A classic quasi-experimental study involves
seat belt use and death from collisions. In this situation, the government controls the expo-
sure—wearing a seat belt or not. If you live in New Hampshire, you do not have to wear a
seat belt. However if you live in any other of the 50 states, you are required to wear a seat
belt. If we wanted to compare death or injury rate between two states, a person in a state
with laws requiring the use of seat belts could not be randomly assigned to wear a seat belt
or not. Researchers cannot ask people to break the law. The quasi-experimental design
might be comparing deaths from collisions in New Hampshire and Vermont. Even though
there may be some people in Vermont who choose to break the law and not wear a seat belt
and some people in New Hampshire who choose to wear a seat belt (even though it is not
mandated by law), researchers could still look at the statistics comparing the states death
rates by motor vehicular injury.

Observational
The third type of study is the observational study. In many situations, an experiment would
be wrong. For example, it would be unethical to ask some participants to drink alcohol and
drive in town and others to not any drink alcohol and drive in town so that the researcher
could study the effects of alcohol consumption on driving ability. Therefore, a lot of epi-
demiologic research is performed using observational studies. In this situation, patterns of
exposure and disease in populations are carefully measured. Two types of observational
studies exist: descriptive studies and analytical studies. Descriptive studies look at the per-
son, place, and time to estimate disease frequency and time trends. Analytical studies are
used to igure out the reason for a disease. One type of analytical study design is an eco-
logical study.

32
Ecological Study
The group, not the individual, is the subject of an ecological study. For example, in one
study, data on death from breast cancer and the average amount of fat intake per person in
39 countries were collected. Countries with high fat consumption (Netherlands, Denmark,
New Zealand, United Kingdom, Canada, United States, etc) also had an increased level of
deaths from breast cancer. On the other hand, countries with low fat consumption, such
as Thailand, Japan, El Salvador, Ceylon, Philippines, Mexico, Columbia, etc) had many
fewer deaths from breast cancer.

Breast cancer deaths and average fat intake in 39 countries

Once this was shown, the next step was to ind out about individual fat consumption and
breast cancer risk. In this case, we have not been successful in answering the question. We
still do not know if fat consumption is an important risk factor for breast cancer, even after
dozens of studies.

Advantages
An ecological study has several beneits:

1. It is generally less expensive than other types of epidemiological studies.

2. It is easy to do since ecological studies often use routinely collected data.

3. I t provides ideas about the cause of a disease or condition that can then be further
studied using individual level data.

4. I t can show researchers new potential risk factors that might not have been thought
about before.

33
Disadvantages
An ecological study has several limitations.

1. Since the ecological study uses routinely collected data, the quality of the data may
not be as high as we would like. Poorer quality data can lead to incorrect conclu-
sions

2. Sometimes it is dificult to know if the data really relect the number of people with
the disease, since the criteria for being included in the data may vary.

3. Using the data from an ecological study to say something about individuals. This is
known as ecological fallacy. In other words, the misunderstanding occurs when the
individual members of the group are thought to have the average characteristics of
the group.

Ecological Fallacies
One of the main disadvantages is the problem with ecological fallacy. Stereotypes are one
form of ecological fallacy. For example if Hawaiians are considered to be good at math, it
is an error to think that every Hawaiian is good at math. For any one person in the group
of Hawaiians, we do not know their math ability. We only know the average math ability
of the group.

Ichiro Suzuki of the Seattle Mariners

To use a sports example, the batting average of the team will not tell us anything about
one player’s batting average. For example, the Seattle Mariners’ team batting average in
October 2008 was 0.265. What was Ichiro Suzuki’s batting average? It was much different
than 0.265. It was 0.331. If we assumed that Ichiro Suzuki’s batting average was the same
as the team average, we would have made an ecological fallacy.

34
Summary
Two key characteristics describe three types of studies. The irst key characteristic is
whether or not the individual has any control over the exposure. The second characteristic
is whether a participant can be randomly assigned to the exposed or unexposed group. Of
the analytical study designs, the ecological study designs looks at the group to determine
risk factors. One mistake is to apply the indings from an ecological study to an individual.
This is known as an ecological fallacy.

Concept Reinforcement
1. Do you think the laws against driving under the inluence of alcohol were based on
quasi experimental data? If yes, suggest some data.

2. Imagine that your community had an air pollution problem and you are interested
in studying asthma. Choose one of the study designs described in this unit and ex-
plain how you would study the issue.

3. Are the characteristics that describe your school an ecological fallacy?

Explain

35
36
 Chapter 8 - Cross-sectional Study Design
Chapter Objectives
• Describe a cross-sectional study design

• State the pros and cons of cross-sectional study design

• Explain the concept of correlation

Many study designs are available to the epidemiologist. Choosing the appropriate study de-
sign depends on many things. In general, the amount of information already known drives
the choice of the study design.

One type of study is observational study. Careful measurement of patterns of exposure and
disease in populations are made. Two types of observational studies exist: descriptive stud-
ies and analytical studies. Descriptive studies look at the person, place, and time to estimate
disease frequency and time trends. Analytical studies are used to igure out the reason for a
disease. One type of descriptive study design is cross-sectional study.

Cross-sectional studies observe part of the population at a single point or brief period of
time. They often use individual-level data, for example, household interview surveys. A
phrase to describe cross-sectional studies is “snap shots”. It is a snap-shot since a cross-
sectional study captures one picture of the community at one point in time, just like a pic-
ture taken with a camera. Typically the cross-sectional studies describe the frequency of
health outcomes.

Although cross-sectional studies are limited in their ability to test hypotheses, they can
provide insight about the extent of a public health problem. This takes the form of person,
place and time.

Cross-sectional studies are useful. One thing that can be igured out from cross-sectional
studies is the magnitude of a problem. For example, by gathering data from a survey of col-
lege students about drinking more than 5 alcoholic drinks on a single occasion , the severity
of the problem can be determined.

Information from cross-sectional studies can be used to focus more research into iguring
out why a problem exists. Once a problem is identiied, then much more research follows
to igure out why. For example, cross-sectionals studies have shown very clearly that the
number of children who are obese is increasing every year. A lot of research is being done
to igure out why.

Cross-sectional studies can also be used for planning health programs. For example, if a
binge problem was discovered in the survey of college students, then universities’ health
services might start programs to address this problem.

37
Several national surveys are a cross-sectional study design. In these national surveys, they
are repeated. From these data, trends in disease or risk factors can be shown. For example
data on adolescents, aged 12-19 years from the National Health Interview Surveys con-
ducted from 1974 – 1991 showed that overall smoking levels went down among that group
of adolescents.

Advantages of Cross-sectional Study

A cross-sectional study has several advantages. One advantage is that one study can look
at several health issues. In a survey, people can be asked to tell the researcher about their
smoking, alcohol drinking, medical conditions, physical exercise, etc. Every one of these
topics can be used to show the number of people who smoke, drink alcohol, etc.

Another advantage of cross-sectional study design is the cost. Since the study gathers data
over a short period of time, the costs are a lot lower than other study designs.

In cross-sectional studies, participants are contacted once for an exam or interview. This is
an advantage since it limits the burden on people; they have to spend less time and energy
helping with the research.

Disadvantages of Cross-sectional Study

One disadvantage of a cross-sectional study is that the researcher cannot determine which
happened irst, the risk factor or exposure versus the disease or illness. Both pieces of in-
formation are collected at the same time and it is tough to know which can irst.

Cross-sectional studies are not able to look at diseases that are uncommon. When a sur-
vey is sent to a community, few respondents are likely to have a rare disease or illness.
Therefore, this type of study design is not useful for iguring out trends or frequency of
rare diseases or illnesses. Similar, cross-sectional studies are not very help for studying
outbreaks or epidemics. By deinition the outbreak is a sudden increase in a disease. When
cross-sectional survey is mailed to participants, they answer questions about their health
and lifestyle choices. Since an outbreak is sudden and unexpected the questionnaire would
not ask people whether or not they had that disease.

The third limitation is that cross sectional studies cannot be used to igure out why someone
gets a disease, illness or injury. For people in the study that are survivors of a disease, it is
dificult to igure out which lifestyle factors are linked to the disease and which are linked
to the survival. Cross-sectional studies only describe what exists at the time of contact—at
the time of an exam or interview. Hence the name snap-shot.

Finally, cross-sectional studies cannot predict future health events. In repeated cross-sec-
tional surveys, they can show trends but cross-sectional studies cannot be used to igure out
why these trends happen. Therefore cross-sectional studies cannot predict.

38
Analysis of Cross-sectional Survey Data
One way to describe the relation between a risk factor and disease outcome is to use a sta-
tistical technique called correlation. This term means there is an association between two
variables where the luctuation in the values of each of the variables shows a pattern that
is unlikely to occur by chance.

A few graphs might make this clear.

Weak negative correlation: Adult literacy rates and healthy life expectancy
at birth in years for males

This graph shows that men in some countries who are slightly less educated are likely to
have a slightly longer healthy life. When approximately 96% of the male population is liter-
ate, they live fewer years than when approximately 92% of the male population is literate.

Strong positive correlation: Infant mortality rate and fertility rate in eight countries

This shows that a higher infant death rate is linked with having more children.

39
 Strong negative correlation: Number of hours watching TV and grade point average

This shows that watching TV less is linked to better grades.

Weak positive correlation: percent smoking and per capita income in selected
countries from 2004 WHO data

This shows that a higher proportion of smokers, age 15 and older, are slightly more likely
to have a higher income in selected countries.

Just as a reminder, a correlation does not mean that one of the two “causes” a change in the
other. In addition, these data are about the population. They do not tell us anything about
an individual.

40
Summary
Of the study designs available to the epidemiologist, one of the most common is cross-
sectional design. Cross-sectional study design is an observational study. It characterizes
the population at a single point in time. Correlations are often used to show the relation
between two factors in a cross-sectional study.

Concept Reinforcement
1. What would be interesting factors to gather about your school to be able to create a
snapshot about athletic injury.

2. What kind of correlation do you think exists between high school students playing
sports and weight?

3. What kind of correlation do you think exists between amounts of time a babies
cries and the amount of time a baby is held?

41
42
 Chapter 9 - Case-control Study Design Diethylstilboes-
trol (DES)
Chapter Objectives Illustration
• Describe a case-control study design A classic case-control
study was launched
• Deine a case and a control participant with only 8 young
women who had been
• Deine recall bias diagnosed with an
extremely rare vaginal
One of the underlying principles of all study designs is that disease does not occur ran- cancer in one hospital.
domly, but rather there is an underlying causal pathway to the disease. Of the many types In 1971, Herbst and
of study designs, the case-control study design irst identiies two groups—those with the Scully reported that 8
disease and those without the disease. The case-control study seeks to identify possible cases of an extremely
causes of the disease by determining how these two groups differ with regard to exposures. rare vaginal cancer
and 32 matched
Case-control studies are considered retrospective. Retrospective means that in the study,
controls were inter-
the disease or health outcome is identiied irst and then the epidemiologist works backward
viewed about possible
to try and igure out what is associated with the disease or health outcome. This backward
risk factors. They
looking work is accomplished by asking participants about lifestyle habits and exposures
found that all cases’
at speciic time points in the past. For example, “When you were a child, did your parents
mothers had taken
or others who lived with you smoke in the house?” Another retrospective question might
the synthetic estrogen
be “How old were you when you started smoking cigarettes regularly?”
called diethylstil-
boestrol (DES). DES
had been prescribed
Recall Bias
to pregnant women
One concern about case-control studies is recall bias. Recall bias means that people in the since 1947 under the
different groups remember things differently. Intuitively, this makes sense. When people mistaken belief that
are diagnosed with a disease, it is natural for them to try to igure out why. They will often DES would prevent
think very hard about their past. They will try to ind some reason why they got the disease spontaneous abortions
and not someone else. In addition, they may also research about their disease and note the (miscarriages). Seven
risk factors. Consequently, their recall of past events may be different from someone who months after this pub-
has never had the disease. lication, the US Food
and Drug Administra-
For example, let’s say that a disease is due to being around clothes during adolescence that tion withdrew ap-
were dry cleaned. Dry cleaned clothes emit cleaning luid odors for a few days. The case- proval for use of DES
control study is iguring out if exposure to fumes from dry cleaning chemical are linked by pregnant women.
to the disease. Study participants are asked in a survey about being around dry cleaning in Research continues
the home during adolescents (“when you were a teenager…”). Because there were some on daughters and sons
articles about the possibility of this chemical’s link to the disease, some of the cases had of mothers who took
thought about this possibility. These cases asked their parents and investigated their expo- DES. These adult
sure before they ever received a questionnaire. In contrast, for the controls, the irst time children have several
they were asked to think about this possibility was when they read the question in the health problems that
survey. They answered the question on the spot without investigating the possibility. For are linked to DES
example, they did not call up their parents and ask about this. In this example, even though exposure.
the same number of cases and controls were really exposed to fumes that come off the dry

43
cleaned clothes, the cases were more likely to remember the situation. They more often
reported an exposure than a control. This led to the researchers reporting a link between
the chemical emitted from dry cleaned clothes and the disease. But this link was not really
there. It was due to the recall bias.

To review, in case-control studies, cases have the disease. Controls do not have the disease.
Information about the people with the disease (cases) is compared to people without the
disease (controls). Case-control study design is the appropriate choice for rare diseases.
Remember, the disease is irst selected and data for many characteristics and exposures of
the participants are then collected. Consequently, many possible risk factors for the disease
can be studied since this information is collected in the survey.

Enrollment Process – Selection of Cases and Controls

For the selection of cases, the researcher irst deines a case. Once this has been deter-
mined, the researcher needs to igure out how to enroll these cases.

Misclassiication
The reason these factors are important is the issue of misclassiication. Misclassiication
means that someone is thought to be a case (has the disease) but is really not a case. On the
other hand, misclassiication can also mean someone is a control (does not have the dis-
ease) and is not really a control. For example, we want to study the risk factors associated
with drinking alcohol in pregnant women. Since most women do not realize they are preg-
nant until about 4 – 6 weeks after conception, some women may say they are not pregnant
at enrollment when they really are pregnant. This is an example of misclassiication. Some
controls are really cases. They were classiied incorrectly.

Another way that misclassiication can occur is if the deinition of a case is too broad, there
may be some people assigned to the case population who do not have the disease. On the
other hand, if the deinition of a case is too restrictive, there may not be enough cases to
conduct the study. Because of the importance of minimizing misclassiication, when given
a choice, the researcher would prefer the case deinition be more restrictive than less.

Finding Cases and Controls

Case ascertainment (identiication of cases) rests on the concept that all cases have an equal
chance of entering the study. However, there are few lists that contain all cases, such as a
cancer registry. Another source of cases is patients who go to a hospital for treatment. One
of the concerns about using hospitals is that it is likely the most severe cases are seen at
these places and the distribution of risk factors may be atypical. Consequently, results from
this type of study may not provide results that can be applied to any situation.

Selection of controls is also very important. One fundamental characteristic of the control
population is that the controls have the same characteristics of the cases (except for the
exposure of interest). In other words, the controls are the same as the cases in all respects

44
other than the disease and the risk factors of interest. The underlying concept in a case-
control study is that the cases are presumed to have the disease because of an excess or
deiciency of an exposure, compared to the controls. Usually, proportions of cases and
controls in 5-year age groups, males and female, different races, educational levels are
characteristics that researchers want to be similar for both populations and only the risk
factor of interest different.

Determining the number of controls in a case-control is part of a power calculation,

which is beyond the scope of this chapter. Sufice to say, many case-control studies
use equal numbers of cases and controls. However, the maximum statistical power would
be to use four controls for every case. Determination of the number of controls for a
study is often inluenced by cost and available funding. Sometimes case ascertainment is
very expensive whereas control identiication is much less expensive. Therefore, a
researcher may chose to enroll more controls than cases.

Identifying controls that come from the same population at risk for the disease or condi-
tion is a guiding principal. Several sources of controls are available with each source have
inherent strengths and weaknesses. The researcher needs to carefully weigh the strengths
and limitations of each source.

Advantages and Disadvantages of Case-control Study

As with all study designs, there are several strengths and limitation of the case-control
study design.

Advantages of Case-control Study

One reason to use the case-control study design is that a researcher can focus on a rare
disease. The cases are those with a rare disease. The controls are those without the rare
disease. Cancer is considered a rare disease. Studies about infectious disease choose case-
control study design.

Another advantage of case-control study design is that it takes a short period of time to col-
lect the data. This is helpful in moving forward in understanding the risk factors associated
with a disease. This type of study is also relatively easy to collect data. A survey is sent to
participants or participants are called and asked about risk factors. In some situations, the
participant is asked to give the researcher a biospecimen such as a sample of blood, urine,
ingernail or toenail, or saliva. These biospecimens are measures for a speciic reason that
is linked to a risk factor.

The size of the study is relatively small. The number of participants is usually hundreds
to thousands versus thousands to tens of thousands. The number of participants is closely
linked to the cost of the study. Consequently, the relatively low cost of this study design is
also a strength.

45
Disadvantages of Case-control Study
One of the major limitations of a case-control study is the retrospective design. Remember
this means that participants are asked about exposures and lifestyle habits that happened to
them in the past.

Another disadvantage of a case-control study is the uncertainty of the timing of the ex-
posure and the disease. At what time in the life of the individual was the exposure impor-
tant—-during growth in the womb, during adolescence, during young adulthood, during
menopausal changes for women, during later life, etc. The timing of the exposure is almost
never known and participants are almost never asked about exposure at all the important
periods of life.

A inal limitation of the study design is having controls that really like the cases except for
the disease of interest. On paper it seems really easy to selecting the cases and controls for
the study, but in reality, this is much more dificult. Real differences beyond the risk factors
of interest between these groups can limit the usefulness of the indings.

Summary
One common type of observational study design is the case-control study. The cases are
identiied and their risk factors are compared to the controls. Cases are people with the dis-
ease or illness and controls are people without the disease or illness. For the case-control
study design, irst the disease or illness is known. Second the researcher asks participants
about their lifestyle choices, exposures and behaviors before the disease. Because partici-
pants have to remember information that happened in the past, there is a possibility that the
recall will be different between cases and controls. This is known as recall bias.

Concept Reinforcement
1. What is an example of recall bias?

2. Where would you go in your community to enroll cancer cases?

3. If you asked your aunt or uncle or grandparent about their exposure to DDT, an in-
secticide commonly used in the 1970s, do you think they would remember it? Why
don’t you try.

46
 Chapter 10 - Cohort Study Design
Chapter Objectives
• Describe a cohort study design

• List 2 questions that could be asked using a cohort study design

• Describe the differences between a prospective and retrospective study

A cohort is a group of people that have a common characteristic that is followed over a pe-
riod of time. An occupational cohort is made up of people who share an exposure in a spe-
ciic setting. The other type of cohort is one in which the participants share a non-speciic
exposure and have a general commonality, such as a cohort made up of nurses. A cohort
study design is an observational study design. In a cohort study design, data is collected
from the cohort at least two times.

Types of Cohort Studies

Cohorts can be studied either forward in time (prospectively) or backward in time (retro-
spectively). In the prospective cohort study, the variable of interest is measured before the
health outcome has occurred. In contrast, for a retrospective cohort study, the historical
cohort is reconstructed with exposure status determined from past records and data on the
outcome of interest is determined at the start of the study. In both the prospective and retro-
spective cohort study design, the researcher is comparing exposed and unexposed groups.
The key difference between the two study designs is the calendar time. In the prospective
cohort study design, the exposure and non-exposure are determined as they occur during
the study years. Before the exposure status is igured out, the health outcome of interest is
identiied in the cohort. In the retrospective cohort study design, the exposure is determined
from historic records and the disease status is typically igured out at the start of the study.

Timeline for retrospective and prospective cohort study designs

47
Prospective Enrollment Process
cohort study: The researcher identiies a population (cohort) of interest. For example, in the Nurses’
Framingham Health Study, the researchers selected female registered nurses as the cohort. In 1976,
Heart Study. 121,700 nurses ages 30-55 completed the irst survey.
Cardiovascular disease
At the time of enrollment, the researcher igured out the exposure status and other charac-
(CVD) is the leading
teristics of the cohort. For the Nurses’ Health Study, researchers were interested in the rela-
cause of death and
serious illness in the tion of hormonal factors, a variety of nutrients, diabetes, exercise, and brand of cigarettes
United States, yet little smoked with subsequent risk of coronary heart disease, pulmonary embolism, and stroke.
was known about the A pulmonary embolism is a clot in the lungs.
risk factors associated
with CVD when the The researchers follow the cohort over time with data collection every couple of years.
Framingham Heart They were able to identify women who had the health outcome of interest in the exposed
Study was launched in and unexposed group. In the Nurses’ Health Study cohort, exposure information was col-
1948. The objective of lected every two years. Very few women dropped out of the study. As of 2008, the nurses
the study was to iden- are still being followed. One hallmark of a cohort study is a lengthy follow up period.
tify common factors or
characteristics associ-
ated with CVD by fol- Selecting the Cohort
lowing 5,209 men and
women ages 30-61who A basic requirement in selecting the cohort for a particular health outcome rests on the
had not developed overt concept that everyone in the cohort is at risk for that particular health outcome.
symptoms of CVD,
suffered a heart attack Another is no one in the cohort at the start of the study has that particular health outcome.
or stroke from the town Consequently, iguring out who is part of the cohort depends on the particular health out-
of Framingham, Mas-
come the researcher is interested in studying.
sachusetts. In 1971,
a second-generation
Obviously, the concept of population at risk for the health outcome implies that only people
(offspring) cohort of
who are alive at the time of cohort enrollment are eligible to be part of the cohort. This is
5,124 of the original
participants’ adult chil- the third requirement.
dren and their spouses
were enrolled. Cur- These three criteria for selecting a cohort: (1) at risk for the health outcome, (2) free of
rently, a third generation the health outcome of interest, and (3) alive. However, other important criteria for select-
of participants (children ing a cohort might not make it so straightforward. For example, if a researcher wants to
of the Offspring Cohort) study polio, who should be in the cohort? Would an individual who received the vaccine
is being enrolled. The be included or excluded? If the polio vaccine gave 100% prevention, then by deinition this
Generation III Cohort’s individual would not be at risk of getting polio. However, it is unlikely that anyone could
goal is to enroll 3,500 reasonably argue that having a polio vaccine gives complete and total protection from this
grandchildren of the disease. So the eligibility criteria may not be as simple as irst thought.
original cohort. Over
the years the Framing-
ham Heart Study has led Population-based Study Participants
to the identiication of In a population-based cohort study, the cohort involves either the entire population or a
major CVD risk factors representative sample of the population. One example of a population-based cohort is the
including high blood
Framingham Heart Study. People aged 30-61 and residents of Framingham, Massachu-
pressure, high blood
setts joined the study. Of the town’s population of 28,000, the study had enrolled 6,500
cholesterol, smoking,
obesity, diabetes, and randomly selected residents. In this type of study, the participants were not selected based
physical inactivity. on an exposure or risk factor for a disease, nor did they have differing levels of exposure.

48
Exposure-based Study Participants
Rare exposures cannot be studied using cohort study. For example, it would take a cohort
of at least 1,000,000 children to study childhood cancer using cohort study design because
childhood cancer is a rare disease.

Using a population that we think has enough exposure that is linked to a disease or illness
is another way to select participants. An occupational cohort is a common source of study
participants. Exposure on the job often means that workers are exposed to much higher
levels than the general public. For example, people working in a lead mine, are likely to
have much higher levels of lead exposure than the general public.

Advantages and Disadvantages of a Prospective Cohort Study

Several advantages are associated with a cohort study. One of the most prominent is the
establishment of a sequential order of events. In other words, irst the exposure is identiied
and then, after some passage of time, the health outcome is determined. This type of study
design eliminates recall bias since exposure data is obtained in real time and is not some-
thing that the study participants have to recall. Another advantage of a cohort study is that
several health outcomes can be explored at the same time and over the course of the study.
In the Framingham Heart study, blood pressure and cholesterol levels were two of several
risk factors of cardiovascular disease that were evaluated.

Because of the lengthy follow up required, usually years or decades, several dificulties can
arise. One dificulty is continued funding of the study. The National Institutes of Health
typically award grants for a maximum of 5 years. Consequently, for the Framingham Study
that has been actively collecting data for over 60 years, a dozen grant applications have had
to be awarded. With each grant application, a chance of rejection exists since the amount of
money available to all researchers is limited. Further, it is possible that the cohort will out-
live the principal investigator (PI). For example in the National Children’s Study, a cohort
of 100,000 children will be followed to age 21. The enrollment of the cohort will take over
10 years. Consequently, if a PI is age 50 at the launch of the study which will be launched
in 2009, it is unlikely that the PI will still be in active research and may not even be alive
30 years later, when the data from the completed study are ready for analyses. Of course,
for this study, there will be numerous results along the way for the current researchers.

Another concern with cohort studies is the dificulty of retaining study participants. Sev-
eral scenarios exist in which the number of participant who begins the study is less than
the number of participants who inish the study. This is known as loss to follow up. One
scenario is that study participants can not be located at the next contact period. Participants
can opt out of continuing with the study and participants can die during the study period.
As a general rule of thumb, the validity of the study requires that losses to follow up, for
whatever reason, not exceed 20%. During the enrollment period, not enrolling those who
are deemed likely to not remain in the study, providing incentives for each contact, having
periodic contact not related to data collection activities are all strategies to minimize loss
to follow up.

49
A cohort usually requires a very large sample size (Framingham Heart Study, which is
considered a relatively small cohort study, initially enrolled 6,500 participants and the Na-
tional Children’s Study plans to enroll 100,000 children). The large study size translates
into signiicant study expenses. In addition, detecting rare diseases is not a common goal
of cohort studies.

Summary
Cohort study design is comprised of participants who share a common characteristic. This
observational study design collects data from participants at lease two times over a period
of time. Many cohorts lasts years and a few have been going on for decades. One of the
main advantages of a cohort study is the sequence of exposure and then disease is clear.
Cost of the study and maintaining contact with all participants over the course of the study
are two limitations of cohort studies.

Concept Reinforcement
1. What is the key difference between a retrospective and prospective cohort study
design?

2. What could be the characteristics of a community that would lead to loss to follow
up due to “unable to locate?”

3. What would be an appropriate cohort to study use of electronic communication on

career choices?

50
 Chapter 11 - Randomized Controlled Trial (RCT)
Study Design
Chapter Objectives
• Describe a randomized controlled trial study design

• Describe the reason for using a placebo

• Describe the reason for blinding in a RCT

Of the epidemiological study designs, experimental designs allow researchers to under-

stand the cause of diseases. In the experimental design, the researcher attempts to change
only one thing between the two groups. One type of experimental study design is called
randomized controlled trial.

A randomized controlled trial is called a randomized clinical trial when the setting is a
medical facility. This type of RCT typically investigates a pharmaceutical medication in-
tervention (a new drug), some type of treatment intervention (a new treatment), or some
type of screen (a test used to detect a disease in individuals without signs or symptoms of
the disease) for a disease in a clinical setting. The unit of analysis is the individual. For ran-
domized community trial, one group of people or one community receives an intervention
and the other group or community does not. The unit of analysis is the group or community.

he Enrollment Process
There are three steps to launching a RCT. First, the reference population is determined. In
other words, who is supposed to beneit from the intervention? This is the population to
which the researcher hopes to generalize his/her indings.. The next step involves deining
the experimental population. This is a practical representation of the population. For ran-
domized clinical trials, patients seen by the appropriate doctors associated with hospitals
involved in clinical trials are often the experimental population. For randomized commu-
nity trials, residents of the community are the experimental group. Volunteers are enrolled
from the experimental population who meet the eligibility criteria. The last step in launch-
ing a RCT is random allocation of the volunteers to the treatment group or the control
group. The treatment group will receive the intervention whereas the control group will not
receive this intervention. One ethical aspect of receiving treatment is that the control group
will receive the gold standard of currently accepted standard patient care, if appropriate.

51
Randomization
Randomization means that chance alone determines whether a participant is assigned to the
treatment group or the control group. Why is randomization such an important character-
istic of RCT? A couple of beneits are gained from randomization. One beneit is that sub-
jective bias of the investigators, either overt or covert, is not introduced into study design.
This means that the researchers do not inluence the results of the study, either knowingly
or by some action that they are not necessarily aware they are doing. For example, a study
was carried out in New York City to test a tuberculosis vaccine in the mid 20th century.
Physicians were instructed to divide their eligible pediatric patients into two groups, inter-
vention and control groups. The intervention group received the TB vaccine. At the end of
the study, it was discovered that the physicians selected children for treatment based on the
parent’s intelligence and cooperative nature. Controls were comprised of children of less
intelligent and non-cooperative parents. This non-random allocation of eligible patients
may have inluenced the study results since one could argue that cooperative parents may
be more health conscious, seek medical advice, and therefore have their children at a lower
risk for TB in general. Another beneit of randomization is that the investigator hopes that
the two groups are comparable with regard to characteristics such as average age, sex,
race, severity of the disease, etc. However, randomization is a chance event. Consequently,
randomization does not guarantee comparability because the groups will not be matched
by speciic characteristics, though with larger sample size the likelihood of comparability
increases. This likelihood is a statistical truth. In randomization from one population, the
larger the sample size, the more likely the groups will look alike. For example, imagine
randomly selecting 10 students from a large high school of 1500 students. The average age
of these 10 students are less likely to match the average age of the student body than if 500
students were randomly selected.

Ethical Considerations
Because a RCT enrolls people and, depending on the intervention, a RCT enrolls sick
people, several ethical considerations need to be addressed.

1. The treatment options should be at least as good as other treatment options based
on previous randomized controlled trials. If a good standard of treatment ex-
ists, this should be given to the controls. One rule in medicine is that the patient’s
welfare is the most important consideration. Consequently, it would be unethical to
subject patients to an inferior treatment or no treatment, if one exists.
2. The hypothesis under study in a RCT should be of clinical signiicance and be use-
ful for future patients.
3. Informed consent is vital for a RCT to be ethical. There are numerous examples of
unethical studies because the patients were not informed. A component of informed
consent is the ability to opt out of the study. The medical experiments performed on
prisoners of war during WW II are examples of unethical research.
4. The entire group of participants should be recruited in a timely manner. Having
patients assigned to the intervention group means they are not receiving the current
gold standard of treatment during the course of the study. It is unethical to put these
patients on the waiting list until a suficient number of patients are enrolled if the
waiting time is unreasonable.

52
Blinding
Once the study population has been identiied and assigned to the intervention or control
group, measurements are taken of both groups to determine the inluence of the treatment.
The monitoring of patients for 1) occurrence of events that the intervention is suppose to
prevent, 2) occurrence of unexpected results or 3) end of the illness. Speciic protocols are
used and followed meticulously so that bias is not introduced in the study.

In a single-blind study, the participants do not know whether they are given the treatment
or not. In a double-blind study, not only are the participants kept in the dark about the kind
of treatment they are receiving, but the study staff do not even know who is in the treat-
ment and control groups. Blinding is used to minimize the placebo effect. A placebo effect
is any beneit from treatment with an inert substance (inert means not chemically active),
such as a sugar pill instead of a medication. For the study staff, this placebo effect might
be seen in how the study staff treats the patients based on group status. In some situations,
it is unethical to have patients participate in blinded studies, such as a surgical intervention
because it is unethical to have the control undergo a fake surgery.

Noncompliance
Once the RCT has been designed, participants enrolled, and intervention blinded, if pos-
sible, the next issue that may greatly inluence the study results is noncompliance. Non-
compliance can take two forms: dropouts and drop-ins. Those who do not fully comply
with the treatment protocol are known as noncompliant. In some cases, patients decided
to stop participating in the trial during the course. This is also known as noncompliance.
The other form of noncompliance happens when the control group inadvertently takes the
agent under study. This happened in a heart attack study in which aspirin was used for
the treatment group. Many over-the-counter preparations contain aspirin and some of the
controls took these over-the-counter preparations. Most RCTs have noncompliance issues.
However, statistical tests are used to deal with this issue so that even with poor compliance,
the results of the RCT can be published.

Advantages and Disadvantages of a RCT

The biggest advantage of a RCT is that the study results provide strong evidence since de-
tailed information can be collected at the start of the study and throughout the study. Part
of that advantage is the ability of RCTs to control for other factors that might inluence the
study results by carefully selecting participants. Another advantage is that dose levels can
be predetermined by the researcher.

The major disadvantage of a RCT is the cost of the study. For pharmaceutical medicine, it
can cost millions of dollars to bring a new drug to the market. These pharmaceutical drugs
have to go through 4 phases of clinical trials.

1. P
hase 1 clinical trial begins testing the new drug on a small number of people. Usu-
ally the people are healthy. The purpose of phase 1 trial is to look at the safety of
the drug in people and start to igure out the proper dose.

53
 2. Phase 2 clinical trial takes the drug testing to the next phase. Up to 200 people are
used for Phase 2 trials. The safe and effective dose is igured out. Side effects are
also determined.

3. Phase 3 clinical trials are the irst trial with randomization. Hundreds to thousands
of people are randomized to those in the intervention group or the control group.
The intervention group is compared to the normal treatment.

4. Phase 4 clinical trials move the drug into the general population. Physicians can
prescribe the drug to patients. Safety surveillance continues during phase 4.

RCTs are generally labor intensive, which translates into a high cost per patient. Conse-
quently, most the sample size of most RCTs is constrained. This means that large effects
can be observed, but not modest or small effects. Another disadvantage of a RCT is the
issue of compliance. If there is a problem with compliance, the study indings may be
weakened. Often RCTs take a long period of time to reach a conclusion.

Randomized controlled trials are considered the gold standard in epidemiology since they
are robust experimental designs. The strength of the indings from a RCT far surpasses
those of other designs. Several components of a RCT must be adhered to for it to be useful.
Whether the indings in a RCT can be applied to the general population and ethical con-
cerns are major considerations in RCT.

Summary
The experimental design of RCT allows researchers to control over who receives the expo-
sure as well as the level of exposure. The researchers can more conidently state a cause-
effect relationship with a RCT than observational study designs. Although RCTs are costly
and take considerable time to complete, the indings provide strong evidence. No medica-
tion can be put on the market without going through randomized clinical trials.

Concept Reinforcement
1. Although both randomized clinical trials and randomized community trials are both
experimental in design, what are differences between the two?

2. Why is randomization important?

3. If the study was not blinded, do you think the compliance issues would be differ-
ent? Explain.

54
 Chapter 12 - Surveillance and Mortality Data
Chapter Objectives
• Deine surveillance

• Explain the use of ICD

• State the pros and cons of death certiicates variables

• State the pros and cons of the underlying cause of death from death certiicates

Surveillance is the continuous collection of descriptive information about the population. It

is done is a step-by-step way. These data are analyzed, interpreted, and disseminated. The
fundamental purpose of surveillance is to monitor health. Often surveillance data are used
to shape public policy.

A Brief History
One of the very irst collection of health statistics was in England, by John Graunt in a pub-
lication called Natural and Political Observations Made Upon the Bills of Mortality. This
publication summarized causes of death in London in 1662. One of his contributions was
discovering regularities in medical and social phenomena.

Since early settlers in the New World were mostly English, they were used to registering
christenings, marriages and burials. Beginning in the mid 1600s, all English parishes were
required to collect these data. When the settlers came to America, they continued the prac-
tice of recording births, deaths and marriages at the local churches. However, these data
were not complete or often used. The fear of cholera played an important role in encourag-
ing state legislative action to put in place a system for accurate registration of births and
deaths (registries) in some major American cities in the middle of the 19th century. The
U.S. Census, under the directive of President Theodore Roosevelt, began an annual col-
lection of vital statistics from these registries and by the later half of the 20th century, birth
and death registries throughout the United States were providing accurate data. However,
marriage and divorce data have yet to be sent to the State registrar in all 50 states (approxi-
mately ¾ of the states send marriage license data to State registrars and approximately ½
of the states send in divorce decrees to the State registrars.)

55
Birth Certiicates
The birth certiicate data collection process was automated to a standard set of data in 2002.
Most states have both electronic and paper birth certiicates. The plan is for all states have
electronic birth certiicates by 2012.

In reality, an employee of the hospital where a baby is born typically ills out the birth
certiicate. Since these people are usually not researchers, they do not always understand
the importance of gathering accurate data. It is well known that the smoking and alcohol
answers are not reliable data because mothers may be embarrassed to admit the truth.
Sometimes the employee will ask the mother, “You didn’t smoke during pregnancy, did
you?” Can you see how the phrasing of this question introduces bias and would encourage
the mother to agree with the data collector even if the truth was different? Race is another
data point that is not very reliable on the birth certiicate since the data collector may ill in
whatever s/he thinks is the baby’s race and ethnicity based on their own observation.

Death Certiicates
One of the irst accounts of a systematic analysis of health was John Graunt’s work on the
causes of deaths in London in 1662. In hindsight it is easy to understand why mortality data
would be an excellent tool for understanding population health in the 17th century. It was
surveillance data in which no one could be misclassiied; death is an obvious state. Some
of the data in the records would have given accurate details, such as seasonal variations in
deaths, infant mortality rates, and excess of male deaths compared to females. Other data
would have been less accurate, particularly results using the underlying cause of death
(UCOD) variable since speciic diagnoses of diseases and illness were far from accurate.

Underlying Cause of Death

Underlying cause of death means the cause of death is the disease or complications that
caused the death. All death certiicates have data ields called “cause of death” and “under-
lying causes of death” (UCOD). It is not the mode of dying, such as a cardiac arrest. The
underlying causes of death are deined as the disease or injury which started the train of
events leading directly or indirectly to death or the circumstances of the accident or vio-
lence which produced the fatal injury.

These days the law requires reporting of deaths to the National Center for Health Statistics.
Doctors, coroners, medical examiners and funeral home directors ill out death certiicates.
The National Center for Health Statistics is working on improving the quality of the data,
particularly the underlying cause of death. As you can imagine, the true cause of death may
not always be recorded, such as in the case of a suicide. In other situations, iguring out the
underlying cause of death may not be straightforward, particularly when an individual has
several medical conditions.

56
 Death certiicate form from Michigan

Classiication of Death
First introduced in 1893 by Jacques Bertillon, a classiication system for cause of death
has been revised every ten years by the World Health Organization (WHO) and the coding
has been published in the “Manual of International Statistical Classiication of Diseases,
Injuries and Causes of Death (ICD). It is in its 10th version. The ICD has become the most
widely used classiication system in the world.

The ICD has been revised approximately every 10 years since 1900. These revisions re-
lect advances in the medical ield, changes in our understanding of disease mechanisms
and terminology, and are designed to maximize the amount of information and lexibility
a code can provide.

Changes in disease deinition will not only impact the ICD coding but also the UCOD on
the death certiicate. For example on January 1, 1993, the acquired immunodeiciency syn-
drome (AIDS) deinition was expanded from the original deinition published in 1987 to
one that included all human immunodeiciency virus (HIV)-infected persons with certain
diseases. With this change in deinition, there was a 204% increase in number of reported
cases in the irst 3-months of 1993 compared to the same time period in the previous year.
A larger fraction of people with AIDS who die will have this diagnosis subsequently listed
as the UCOD.

57
Data Quality
In any type of surveillance system, the quality of the data is important. Accuracy is impor-
tant. Is the underlying cause of death correct? How about the race of the baby on the birth
certiicate? Besides accuracy, two other key aspects of data quality are completeness and
timeliness. Completeness means that all the cases and their respective information are rep-
resented in the dataset. Timeliness means the cases diagnosed are represented in the dataset
soon after diagnosis (e.g., within a couple of months). If the data quality is untrustworthy,
then some analyses cannot be done. For example, it is well known that the population esti-
mate of number of Hispanic births is under-reported due to the misclassiication mentioned
above. Therefore, analyzing data by race may not give accurate results.

Functionality of Mortality Data Analysis

Mortality data can serve many functions.

1. One function of a mortality study is to evaluate the effectiveness of a new screen-

ing technique. For example, mammography are a screening test for breast cancer
The key point of information the researcher searches for is the following: “Did
more, the same, or less people die of the disease in which the screening test was
used compared to those who did not have the screening test?”

2. Mortality data can also be a useful indicator of severity of the disease from both the
clinical and public health perspective.

3. Mortality data can also be used to indicate disease risk, if the number of people
who die from the disease is high and survival time from diagnosis to death is short.
Pancreatic cancer is an example in which the mortality data also relects the dis-
ease risk. The prognosis for someone diagnosed with pancreatic cancer is about 3
months.

58
Summary
One function of epidemiology for hundreds of years has been the collection of health data.
From these data, the health of the population can be described. Birth and death certiicates
are collected today. The quality of the elements in the birth and death certiicate is improv-
ing. With high quality records, much can be learned about the health of a community.

Concept Reinforcement
1. If one of the three characteristics of quality data was poor, what would be the con-
sequence?

2. Currently, only births and deaths are collected by all states. What do you think the
advantage would be of collecting records of divorces.

3. If you had the money, what surveillance data would you collect in every state?

59
60
 Chapter 13 - Screening
Chapter Objectives
• Explain the difference between diagnostic and screening tests

• Deine the characteristics of a good screening test

• List the criteria to consider before the implementation of a national screening

program

A fundamental belief in public health is the best approach to health care is prevention. This
is known as primary prevention. However if all cases of a disease cannot be prevented,
the next best approach is called secondary prevention. Secondary prevention is to identify
those who do not have any symptoms of the disease or are in an early stage of the disease
through screening. The purpose of screening is to reduce morbidity and death by catching
the disease early. Early detection of diseases is often linked with longer life. The inal stage
of prevention is called tertiary prevention. Tertiary prevention is limiting the progression
of the disease. An example of tertiary prevention is giving antiretroviral therapy for people
diagnosed with HIV. The antiretroviral therapy does not cure anyone; but it does slow the
progression of HIV to full blown AIDs.

In this chapter we are going to explore the characteristics of secondary prevention in

the form of screening practices.

As stated above, the purpose of screening is to catch the disease or illness in an early stage.
There are many kinds of screening tests used to determine health status. For instance, all
newborns have a mandatory newborn screening test that evaluates 29-50 conditions. Blood
pressure is another example of a screening test that all adults get during their physical ex-
amination.

There is often confusion between screening and diagnosis. Screening usually occurs in
people who do not have any symptoms and are not suspected of having the disease. Di-
agnostic tests are used to conirm whether someone has a disease or not. Sometimes the
screening test and the diagnostic test are the same. For example, blood glucose test can be
used as a screening test if it is given to someone who does not have any symptoms of diabe-
tes. If the same test is given to someone with symptoms of diabetes to conirm a diagnosis
of diabetes, it would be considered a diagnostic test.

61
he Natural History of Disease Progression
Below is a classic diagram of the course of a disease in an individual over time. At each
point on the continuum, the degree to which intervention can be effective is dependent
upon the type of disease. Similarly, the speed of progression from healthy to the inal stage
varies a lot by disease as well as an individual’s characteristics.

Natural history of disease progression

Screening occurs in two places on the continuum. The best time to screen is during the
stage immediately prior to disease onset. For example, colonoscopy can detect polyps in
the colon. Polyps are not cancer and if removed prevent cancer at that site. The second
place screening occurs is when either the precursor of the disease is detectable or when
symptoms develop. For example, a blood pressure screen can determine hypertension, a
potential precursor to stoke. Another example is breast cancer screening to catches the dis-
ease is the earliest stage. Diagnosis in early stage dramatically improves survival. Although
everyone agrees that reduction in morbidity and mortality is an important public health
goal, there are several factors that need to be taken into consideration when deciding about
implementing a screening program.

Population Screening Criteria

Four criteria for population based screening were proposed by Wilson and Jungner in a
report to the World Health Organization more than 30 years ago. These are still considered
relevant today.

1. Knowledge of the disease.

The prevalence of the disease must be balanced with the severity of the disease and the
likelihood of effectiveness of treatment.

PKU Example
In 1934 a Norwegian doctor, Asbjorn Folling, discovered PKU (phenylketonuria) as the
cause of severe mental retardation in 2 young children. PKU is called an error of metabo-
lism because children cannot break down phenylalanine, an amino acid making up approx-
imately 5% of all protein foods. Dr. Folling showed that PKU was an inherited disease. At
that time, the prevalence of PKU in institutions for severely mentally retarded individuals
in western counties was 1-2%. Blood tests were developed in the early 1960s to test for this
genetic abnormality since, if caught early–within the irst 2 weeks of life–mental retarda-
tion can often be prevented. In the United States, the newborn screening test for PKU is
mandated by law in all states except Maryland and Wyoming.

This newborn screening test is worth the societal burden since the consequences of the
disease are quite severe and early treatment is beneicial.

62
 2. Feasibility of screening procedures.

Not only does the test need to be appropriate for disease identiication, but patients have
to agree to be tested. A classic example of a test that does not have wide acceptance is the
fecal occult blood test for colorectal cancer.

3. Diagnosis and treatment.

A screening test can identify those at risk of the disease, in an early stage or precursor stage
of the disease. This kind of screening, as opposed to diagnostic screen, has limited value if
no treatment is available.

For example, someone diagnosed with pancreatic cancer has a very poor prognosis, typi-
cally 3 months to live. The early diagnosis will not change the prognosis of the disease
since the cause of pancreatic cancer is unknown. In other words, precursors to pancreatic
cancer are also unknown. Consequently, instituting a population-based screening program
for this rare disease would not make sense.

4. Cost considerations

One misunderstanding is that screening programs always save money. However, this is
not always true. If the disease is rare and the cost for the screening is high, there may be
considerable health care costs to implement a national screening program. The cost consid-
eration is one of the arguments used by some to resist implementing a national program for
colonoscopy screening for colorectal cancer.

In the United States, cervical cancer, breast cancer, prostate cancer, colon cancer, diabetes
and hypertension are diseases in which screening has been recommended with various
degrees of compliance.

Case Studies

Prostate Cancer: PSA (prostate speciic antigen) Screening Controversy

Prostate cells put out a small amount of protein, called prostatic speciic antigen (PSA) that
may be detected using a blood test. Since the amount of prostate tissue increases with age,
PSA levels normally increase with age, so the normal range gradually increases from under
2.5 ng/ml between the ages of 40 and 49 to as high as 6.5 ng/ml by the age of 79.

It is common knowledge in the medical world that if you are male and live long enough,
you are likely to have cancerous tissue in your prostate. It is also true that males are more
likely to die with prostate cancer than to die from it. Therein lies the controversy. Accord-
ing to the Centers for Disease Control and Prevention (CDC), the current prostate speciic
antigen (PSA) test approved in 1986 by the Food and Drug Administration to screen for
prostate cancer can detect the disease in its early stages. However, epidemiologic evidence
is mixed and inconclusive about whether early detection actually saves lives because re-
searchers do not know how quickly prostate cancer cells grow. Almost all older men have
some prostate cancer cells at the time of their death. This may suggest that prostate cancer
cells grow slowly. It is not clear whether the beneits of screening outweigh the risks of
follow-up testing and cancer treatments.

63
People who support routine PSA screening for men over 45 years of age point out that it
improves early diagnosis of a common, curable cancer at an earlier stage than can be de-
tected by physical exam. Those who oppose routine screening state that it is costly and will
result in unnecessary, invasive biopsies and treatment of people whose prostate problems
would not have resulted in signiicant disease. Opponents to the PSA test also say the re-
sults often prompt men to undergo surgery or other treatments that leave them impotent or
incontinent (impotent means inability to sustain an erection suficient for sexual activity
and incontinent means not having control over urination), even when there is little chance
prostate cancer will kill them.

Breast Cancer: Mammography Screening Controversy

Mammograms are heavily promoted as a breast cancer prevention tool. Of all types of
breast cancer, one type, called in situ, has increased from 2-3% in the early ‘80s to around
16% in the early ’00 of all breast cancer diagnosis. Typically, the treatment is a lumpecto-
my (removal of the lump). In northern Europe, women diagnosed with in situ breast cancer
who have it surgically removed are considered cured. Increased mammography use is the
reason why the percent of in situ breast cancer has increased over time.

The controversy arises regarding the age at which women should be screened for breast
cancer using a mammogram. Mainstream organizations, such as the American Cancer So-
ciety, the American College of Radiology and the National Cancer Advisory Board, part of
the National Cancer Institute, recommend that women begin screening at age 40. However,
epidemiology studies have not found consistent beneit of screening mammography before
the age of 50 years.

Advocates for screening beginning at age 40 feel that this screening tool saves lives since
many doctors believe that mammograms can detect a tumor 80% of the time. Those op-
posed to mammogram use before the age of 50 believe that the screen itself, an X-ray of the
breast, puts women at an increased risk due to exposure to ionizing radiation, even though
it is at a low dose. (Ionizing radiation is considered a risk factor for breast cancer.) Those
who support mammogram use, argue that the dose is so low as to be of little concern. An-
other argument by those who support early use of mammograms is the type of breast cancer
that is more common in younger women—a fast growing, aggressive cancer would be de-
tected earlier through annual mammogram screening. Finally, critics of mammograms for
younger women have emphasized the greater dificulty in detecting small tumors, which
is the major reason for getting a mammogram, due to the denser breast tissue of young
women.

As noted in the two case studies above screening is not without controversy. When a screen-
ing program in instituted, several bias need to be considered when evaluating the eficacy
of the program.

Screening is the principle tool for secondary prevention. Implementing a screening pro-
gram is not necessarily straightforward since the costs or burdens to the population need
to weigh against the beneits. Even when a screening test has been developed, the quality
of the test results needs to be considered. Usually changing patterns of screening directly
inluence the respective disease trends. The development of better screening tests will mo-
tivate some researchers in the quest to improve health for the populace.

64
Summary
Screening allows for the early detection of a disease. In most, but not all, cases, early detec-
tion leads to better health outcome. People diagnosed with an earlier stage of disease, often
live longer. Screening is different from diagnosis, although the same test might be used. A
diagnostic test conirms the disease.

Concept Reinforcement
1. What factors do you think are most important in deciding whether or not there
should be a recommended screening test that is paid for by insurance? For exam-
ple, do you think cost should be considered when deciding?

2. Why is there confusion between screening and diagnosis?

3. Do you think PSA should be given to every adult male over 45 years old? Why or
why not.

65
66
 Chapter 14 - Prevalence and Incidence
Chapter Objectives
• Deine prevalence

• Deine incidence

• Compare prevalence, incidence, and frequency

From birth to death, people experience many health states with varying frequency. Fre-
quency means how often something happens. We all know stories of a seemingly perfectly
healthy young person who suddenly and unexpectedly died one day. Although this often
gets the most press, it is a very rare occurrence. The majority of people luctuate from
healthy, to sick, to be cured of that sickness—experiencing some disability—maintaining
at a diminished level of health, to deceased. These states of health are dynamic in a popu-
lation. At any one time, some of the population will be experiencing each of these states.

The most common measure is to count the number of cases of disease or health outcome.
However, for a count to be descriptive of a group, it is expressed as a proportion. Propor-
tions are often expressed as percentages by using the multiplier of 100. One of the central
tasks in epidemiology is to determine the appropriate denominator to meaningfully de-
scribe and compare groups. The denominator must be the total number of people at risk for
the disease. For highly contagious and deadly cases, such as anthrax, a single case will be
front page news.

Another very common measure is rate. The difference between proportion and rate is that a
rate involves a measure of time. The numerator is the frequency of the disease over a spe-
ciic period of time and the denominator is a population, often at midpoint in the speciic
period. Another characteristic of a rate is that the denominator must be drawn from the case
population. For example if we report the annual rate of birth defects, the numerator is the
number of birth defects in a year and the denominator is the number of births in the same
year.

In determining the frequency of disease, different denominators are chosen depending on

the research goal and availability of data. For example, if a researcher is studying a disease
frequency within a general population, the denominator will be different than if a research
is studying the frequency of a disease (stroke, for example) in a population of people with
hypertension. Paying attention to the denominator is a skill that an epidemiologist im-
proves with practice.

67
Prevalence
Prevalence, sometimes known as point prevalence, means the number of people in the
population of interest (e.g., school district, city, county, state) who have the particular dis-
ease or illness at a point in time, usually when the survey is administered. For example, if
there are 5 people with the disease outcome of interest in a population of 200, prevalence
is 2.5%. Prevalence measures are proportions. Although prevalence is often described as a
rate, it is not. Remember that a rate has a time dimension and prevalence does not.

Prevalence tells us nothing about how long someone has had the disease—it could be a day,
a year or many years. Some people like to think of prevalence as a snap shot in time that
tells us the number of people with the disease.

Prevalence is useful to measure the burden of disease in the community. For example, how
many people with uncontrolled asthma live in the community? This prevalence number
will help determine the number of clinics, the need for emergency medical services, the
number and type of health care providers, etc. Prevalence numbers are very useful in plan-
ning health services.

Case Studies of Prevalence Data

The following are case studies the shows the usefulness of prevalence data.

Prevalence of Swallowing Disorders

Researchers were interested in determining whether swallowing disorders, such as dificul-
ty eating due to some problem swallowing, was an issue for hospital patients. The research-
ers counted the number of patients with swallowing disorders over a 3-week period in two
hospitals and reported that 13% of the patients had swallowing disorders. This prevalence
was much higher than they had anticipated and they ended their article by calling for re-
sources to be used to deal with this problem.

Celiac Disease and Patients with Type 1 Diabetes

Celiac disease: An autoimmune disease which disrupts absorption of nutrients in the small
intestine. Gluten has been identiied as a trigger.

Type 1 Diabetes, also known as childhood diabetes, is disease in which the pancreas does
not make insulin. Insulin is a hormone that helps glucose get into cells to give a person
energy.

If a physician is aware of a high prevalence of celiac disease in patients with type 1 dia-
betes, s/he might be more likely to ask the patient about speciic symptoms and/or order a
test for celiac disease.

68
Incidence Rate
Now we are going to turn our attention to incidence rate. Incidence data is a fundamental
measure used for research on the causes of disease. Disease frequency can be measured
using prevalence or incidence. Incidence refers to new cases in a population at risk during
a speciic time period whereas prevalence is a snapshot of all individuals with the disease
without a time dimension. For prevalence, some of these individuals may have had the dis-
ease for years and some may have been diagnosed in the previous month. In contrast, for
incidence, the numerator is the frequency of new cases which means people with a history
of the disease are not included in the numerator.

The denominator is the number of persons at risk for developing the disease during that pe-
riod of time. Unlike mortality data, in which everyone has some probability of dying, some
individuals do not have a chance of diagnosis for some diseases/illnesses. For example, to
calculate the incidence of ovarian cancer, women who have had an oophorectomy (both
ovaries removed) would be excluded from the denominator as well as men, obviously.
For infectious diseases, those who have lifetime immunity against recurrence would be
excluded from the denominator.

For example, 75 adults in a town of 25,000 adults have been diagnosed with celiac disease
in 2009. The incidence is 75 divided by 25,000 equals 0.003. It we want to express the
incidence rate per 1000 adults, then we multiply by 1000. The incidence rate is 3 per 1000
adults. Reporting the data as 3 per 1000 persons is more understandable to the general
population than stating the incidence as 0.003.

Connection Between Prevalence and Incidence

The interrelationship between prevalence and incidence is as follows: prevalence of a dis-
ease is proportional to the incidence rate multiplied by (average) duration of the disease.
Consequently, for diseases of short duration and high incidence, the prevalence is similar to
the incidence. Why is that? For a disease of short duration, either the patients recover their
health or die. Therefore, over time the number of people with a history of the disease in the
population is virtually zero because those who had the disease eventually die from some
other cause. Remember, incidence does not take into consideration those with a history of
a disease since the numerator only contains new cases. Some infectious diseases meet these
criteria. In contrast, many chronic diseases have low incidence and long duration. With this
scenario, as time passes, the number of people living with the disease increases so that at
any one time period the prevalence would be higher than the incidence.

Example of persons A, B, C, D, E and their respective duration of disease in 2008

69
In the above igure, we are going to ignore the denominator. For 2008, there were 2 inci-
dences of the disease, person B and C. For the other persons, their disease began before
2008 so there was only 2 new cases of the disease. In June of 2008, the prevalence of the
disease was 4 cases. At that point in time, there was 4 people with the disease. Only person
A did not have the disease at that time.

Summary
Two measures in epidemiology is prevalence and incidence. Prevalence measures the pro-
portion of cases in the population at risk. In contrast, incidence measures the number of
new cases in the population at risk over a speciic period of time. For both of these mea-
sures, it is important to pay close attention to determining the population at risk for the
denominator.

Concept Reinforcement
1. Why is prevalence an important measure for population health studies?

2. What would be the formula to calculate the prevalence of females in the class right
now? (Hint: what is the numerator and what is the denominator?)

3. What does incidence tell us that is different from prevalence?

70
 Chapter 15 - Ethics
Chapter Objectives
• Describe the Tuskegee study and Nazi human experimentations

• Describe the 3 criteria for ethical research

• List the elements of consent

Nazi Experiments During WWII

During WWII, medical experiments were conducted on large number of concentration
camp prisoners. The prisoners were forced to participate in these experiments. They did
not volunteer. They did not consent to be part of the experiments. For example, in one ex-
periment prisoners were forced to loat in a tank of ice water for up to 5 hours. The Nazis
were using prisoners to igure out different ways of rewarming survivors. At the end of
the war, doctors captured by Allied forces were put on trial. One of the results of the trial
was the writing of the Nuremberg Code in 1946. This code calls for voluntary consent of
participants, avoidance of unnecessary pain and suffering, and a belief that the experiment
will not end in death or disability.

Death by hanging is pronounced by a U.S. War Crimes Tribunal at Nuremberg upon

Adolf Hitler’s personal physician, 43-year old Karl Brandt. Brandt, who was also Hitler’s
Reich Commissar for Health and Sanitation, was indicted by U.S. prosecutors, along with
22 other Nazi doctors and SS oficers, on war crimes charges. This was the irst case of
alleged criminals being tried after the judgment of the International Military Tribunal.
The Tribunal found him guilty on all four counts, charging him with conspiracy in ag-
gressive wars, war crimes, crimes against humanity, and membership in the criminal SS
organization. Among those criminal acts, was his participation in and consent to using
concentration camp inmates as guinea pigs in horrible medical experiments, supposedly
for the beneit of the armed forces.

71
 He was hanged at Landsberg prison on June 2, 1948 because of his execution of thou-
sands of helpless political, racial, and religious persecutees. This occurred after U.S. Mili-
tary Commander Gen. Lucius D. Clay and the U.S. Supreme court upheld the Nuremberg
Tribunal’s judgment. In a long-winded speech that was inally mufled when the black
hood was thrown over his head, Brandt shouted arrogantly, “It is no shame to stand on
this scaffold; I have served my country as have others before me.”

Hitler was also imprisoned in this fortress-like building near Munich in 1923
following his unsuccessful Munich putsch. During his coninement, he wrote what
was later to become the Nazi bible, “Mein Kampf.”

Image courtesy of the US Army. Caption modiied from the Telford Taylor Papers,
held by Columbia University Law School.

he Tuskegee Syphilis Study (1932-1972)

A study of untreated syphilis in rural impoverished African American males was conducted
in Macon County, Alabama from 1932 to 1972. Volunteers from the community responded
to an announcement of the study and 400 men with syphilis and 200 men without syphi-
lis were enrolled. One important aim of the study was to determine the prevalence and
severity of syphilitic disease process. The men were told the doctors were studying “bad
blood.” Various methods were used to maintain and stimulate study participants’ interest.
Free medicines, burial assistance or insurance, free hot meals on the days of examination,
transportation to and from the hospital, and an opportunity to stop in town on the return trip
to shop or visit with their friends on the streets all helped. Through the tireless efforts of
the project nurse, a very high number of participants remained in the study (87%). At the
start of the study, there was no proven treatment for syphilis. However, even after penicillin
became a standard cure for the disease in 1947, the medicine was withheld from the men.
The Tuskegee scientists wanted to continue to study how the disease spreads and kills. The
experiment lasted four decades, until public health workers leaked the story to the media.
This medical study parallels the medical experiments on WW II prisoners of war in its level
of unethical behavior.

Examination, Tuskegee Study

Image courtesy of the National Archives and Records Administration

72
Willowbrook Hepatitis Study (1963-1966)
Research studies took place at a New York state supported institution called Willowbrook
State School. This was an institution in which mentally disabled children were admitted.
The research study was designed to gain an understanding of the natural history of infec-
tious hepatitis. The children at the institution were infected with the hepatitis virus. During
the 3 years of the study, Willowbrook did not allow any new admissions. However, the
study was able to admit new patients to the institution since it had its own space. Thus,
in some cases, parents could admit their child to Willowbrook only if they agreed to have
their child participant in this study. Many felt that the children and parents had little choice
in participating in the study.

Belmont Report
A meeting in the late 70s was held to address problems arising from researchers acting in
questionable ways. The Nazi experiments, the Tuskegee Syphilis Study, the Willowbrook
Hepatitis Study and the Tearoom Trade study are examples of some questionable ethics.
The Belmont report is an important document. It was written in 1979. The name of the
report comes from the Belmont Conference Center where the report was drafted. The Bel-
mont report explains the ethical principles underlying the medical research in a clear and
concise way.

There are three fundamental ethical principles for using human subjects for research.

1. Respect for persons. This means the researchers need to treat participants with
respect and courtesy. Participation in the study occurs with informed consent by
the participants. The wishes and decision-making ability of the participants must be
respected.

2. Beneicence. This means that the research must maximize beneits for the research
project while minimizing risks to the research subjects

3. Justice. This means that the research is fairly administered. The costs and beneits
to the potential research participants are fairly distributed. The researchers do not
exploit potential participants. The fairness principle goes both ways—fairness in
terms of who is included and excluded in a research study and fairness in terms of
who beneits from the research and who bears its burdens.

The Belmont report serves as a historical document. It provides the moral framework for
using humans in research. It is also the basis of the federal regulations regarding the use of
humans in research.

73
Informed Consent
One important element of participating in a research study is informed consent. Informed
consent means that the person has a clear understanding of the risks and beneits of par-
ticipating in the study. Consenting can occur in two distinct ways. A person can give active
consent. This generally means a signed consent form. A person’s action can imply consent.
If a person ills out a questionnaire, this is implied consent. For a participant who did not
want to participate in the study, s/he just would not ill out the questionnaire. The same
concept applies if a researcher calls you on the telephone and explains a study. This person
will ask you if you would be willing to answer some question. If you say yes and answer
some questions, this is an example of implied consent.

Elements of Consent
There are many required elements of consent. An 8th grade reading level is the target for
this letter. The language of each of the points below has to be at the 8th grade level and ap-
propriate for the people who are potential participants.

Below is a list of eight common elements that are included in a consent letter.

1. A statement that the study involves research. An explanation of the purposes of the
research and expected duration.

2. A description of the procedures to be followed, and identiication of any procedures

which are experimental.

3. A description of any reasonably foreseeable risks or discomforts to the subject.

4. A description of any beneits to the subject or others which may be reasonably

expected from the research

5. A disclosure of appropriate alternative procedures or courses of treatment, if any,

that might be advantageous to the subject

6. A statement describing the extent, if any, to which conidentiality of records iden-

tifying the subject will be maintained. For research involving more than minimal
risk, an explanation as to whether any compensation and an explanation as to
whether any medical treatments are available if injury occurs

7. An explanation of whom to contact for answers

8. A statement that participation is voluntary

It is easy to see why the consent letter is long. All of these elements must be in the consent
letter.

74
Summary
Historically, epidemiologist did not have any oversight. They could do whatever research
they wanted to do. In a few situations, this lead to a clear violation of human rights. In the
largest violation that took place during WWII, some of the Nazi researchers were put on
trial for war crimes. In other situations, no criminal action was taken against the researcher.
Due to some of these questionable practices, studies by researchers in the United States are
examined carefully by peers and institutions review boards to make sure that human rights
are maintained.

Concept Reinforcement
1. What was unethical about the Tuskegee Syphilis Study?

2. What was unethical about the Willowbrook Hepatitis Study?

3. Consent letters are typically long and detailed since elements of consent are re-
quired. Do you think any of the elements of consent could be deleted from the
letter?

75
76
 Chapter 16 - Risk Factors
Chapter Objectives
• Describe age and sex as risk factors

• List the issues regarding the use of race

• Describe the components necessary for occupational coding

Although people differ from one another in an ininite number of ways, there are common
characteristics that are important to consider in epidemiology. These characteristics are
described below and include age, sex, race, ethnicity, income, education and occupation.
Many of these characteristics can be found on the census. Every 10 years the US popula-
tion is counted. The sole purpose of the censuses is to obtain general information about the
population. It began in 1790 and continues today.

Age
The single most important personal characteristic in health research is age. The irst step
in an epidemiologic study is to look at the ages of the population under study. This alone
might be the determining factor that describes the population’s health risk. Most chronic
diseases are age-related meaning that the older a person gets the higher the risk of being
diagnosed with a chronic disease.

Fortunately, age is one of the easiest pieces of information to obtain in the Unites States.
The U.S census provided age-related population data at various levels of geographic units.
In addition, age is a commonly collected item on many surveys.

Sex
Similar to age, sex is generally collected. Knowing if a participant is male or female some-
times is important. In general, women tend to have more diseases and illnesses and men
tend to die younger. The reason for this general trend is not clear. It may be due to biologi-
cal differences between the sexes or it may be due to differences in life experiences. In any
case, it is often useful to collect this information.

Race and Ethnicity

Race relects a social deinition. It is unique for each country. For example, in the 2000 U.S.
census, individual could select multiple categories when asked to self-identify their own
race. A few diseases are more common in some racial groups, such as sickle cell anemia
in Blacks and breast cancer in Ashkenazi Jewish populations from Northern Europe. How-

77
ever, similar to sex, striking differences can be noted in disease incidence, severity and
death rate among the races. The reason for these differences is unknown and may relect
biological differences or differences in life experiences

Ethnicity is based on culture. Ethnicity is different from race. Ethnicity can be divided into
2 categories: “Hispanic or Latino/Latina” versus “Not Hispanic or Latino/Latina.” His-
panic or Latino/Latina is deined by the U.S. Census as a person of Cuban, Mexican, Puerto
Rican, South or Central American or other Spanish culture or origin regardless of race.

However, using race and ethnicity data can be problematic. Studies ask about race and eth-
nicity differently. When this happens, making comparison between studies is very dificult.
In addition, race and ethnicity deinition have changed over time.

United States Census Race Categories in Selected Years from 1860-2000

1860 1870 1900 1970 2000
White White White White White1
Black (of Black or Black, African
Black Black
Negro descent) Negro American, or Negro2
Quadroon* Quadroon*
Chinese Chinese Chinese Chinese
Indian American Indian or
Indian Indian
(American) Native American3
Japanese Japanese Japanese Japanese
Filipino Filipino
Asian Indian4
Korean Korean
Hawaiian Native Hawaiian5
Vietnamese
Guamanian or
Chamorro
Samoan
Other Asian
Other Paciic Islander
Some other race

Source: 200 Years of U.S. Census Taking: Population and Housing Questions 1790-1990.
U.S. Department of Commerce. U.S. Bureau of the Census

*In 1890, mulatto was deined as a person who was 3/8th to 5/8th black. A quadroon
was 1/4th black and an octoroon was 1/8th black.

78
 1
Persons having origins in any of the original peoples of Europe, the Middle East
or North Africa
2
A person having origins in any of the black racial groups of Africa. Terms such as
“Haitian” or “Negro” can be used in addition to “Black or African American”
3
A person have origins in any of the original people of North or South America (including
Central America) and who maintain tribal afiliation or community attachment
4
A person having origins in any of the original people of the Far East, Southeast Asia
of the Indian subcontinent including for example, Cambodia, China, India, Japan,
Korea, Malaysia, Pakistan, the Philippine Islands, Thailand and Vietnam
5
A person having origins in any of the original peoples of Hawaii, Guam, Samoa,
or other Paciic Islands

Socioeconomic Status
Socioeconomic status (SES) is a characteristic made up of many pieces. However, in the
majority of studies, education, income and/or occupation are used for SES status. Partici-
pants usually do not have a problem answering the question about their education level. In
contrast, income is less clear cut. Sometimes people are uncomfortable answering a ques-
tion about their income. In addition, when asking about income, a person could respond
with his or her own income or the income of the family (aka: household income) or some
combination. Further, individual may not think about all sources of income but rather just
report on their weekly or monthly paycheck amount.

Occupations
People change careers about 3 times in their lifetime. Career is different than occupation or
job. A career is a combination and series of occupational positions held during the course
of a lifetime. That means that the series of jobs has something in common. In contrast, a
job is a position held with speciic duties with an employer. When asking about occupation,
some researchers ask for the irst job ever held that was at least 30 hours a week for at least
6 months. Some ask for current job. However, most ask for the “most representative job”
held over one’s lifetime. Unlike asking about education or income, to use the data collected
on occupation, it has to be coded.

Coding Data
To analyze data and produce study results all information collected has to be turned into
numbers. Some data are given as numbers, such as age. Other data are not given as num-
bers and these data needed to be coded. In other words, these data need to have numbers
assigned. As a simple illustration, if someone checked ‘yes’ to the question about having a
tetanus vaccine, this ‘√’ could not be analyzed. Rather, the answer to this question will be
coded as “yes” = 1 and “no” = 2 and for those who didn’t answer this question = 3. There-
fore, all participants would be coded as 1, 2, or 3 for the answer to this question. These
numbers can then analyzed.

79
Occupational Elements
For occupational jobs to be coded three pieces of information are needed: industry, job title
and duties. Identifying the industry and job title is straightforward. This very simple ques-
tion about duties, however, is very important for coding purposes. Although the question
may seem unnecessary in some cases, the response often permits more accurate coding
of the occupation. Obtaining detailed information on the job duties will help with coding.
For example, in the telecommunication industry, telephone company serviceman is the job
title. The duties could be installs telephone in the home (this would be given one code) ver-
sus repairs telephone transmission lines (this would be given an entirely different code).)
Once the data have been collected, the coding can be done using an online reference, such
as One*Net.

Summary
It is important to collect all the necessary information about the participants. Most studies
collect age, sex, race, education and income. Depending on the study, occupational infor-
mation will also be collected. In all cases, the data collected needs to be coded in such a
way that the information can be analyzed.

Concept Reinforcement
1. What would be other information that participants in a study may be reluctant to
provide?

2. Census data is used in numerous way including epidemiological studies. Although

people are “required” to ill out a census form, some people do not. Should people
who do not ill out a census survey be ined? Explain.

3. If you had to create an indicator variable that identiied people of different socio-
economic status, what information would you want?

80
 Chapter 17 - Sample, Population, and Sample Size
Chapter Objectives
• Deine a sample

• Deine a population

• Compare sample vs. population

Most health research collects data from individuals. Paradoxically, researchers often want
Paradox
to know intimate details about individuals but researchers are not interested in actually
knowing the individual. Individual data are stripped of identifying information and com- something that
piled into a dataset. These data are used to tell a story about the group under study. seems to contradict
itself, even though it
Two terms in statistics are important to know the difference: population and sample. is true.

A population means ALL individuals. A sample is a subset of the population. The analogy
is a whole pie is the population and a piece of the pie is the sample. The sample is taken
from the population of interest. To continue the analogy, if we are interested I iguring out
the average number of cherries in a cherry pie, we do not want to sample any kind of pie.
We want to sample cherry pies. The other piece of information to realize is that popula-
tion means ALL members. For cherry pies, it would be an impossible task to collect every
single cherry pie in the world. Therefore, we would have to gather a sample of cherry pies.

Cherry pie

Piece of cherry pie

To illustrate the difference between sample and population, let’s look at driving as our
topic. The researcher is interested in iguring out the predictors of motor vehicle crashes in
California in 2010 data. The researcher deines a crash as having a police accident report
iled. The population would be all motor vehicle crash reports in 2010. If there are elec-
tronic records of these reports AND the electronic records are sent to a central location, the
researcher could use the population of vehicle crashes in 2010 to determine the predictors

81
of a crash. However, if the police reports were not enter electronically or were not sent to
a central location, it would be virtually impossible for the researcher to compile all the po-
lice reports. In the second case, the researcher would use a sample of motor vehicle crash
reports for the study. There are standard ways to take a sample from a population. For the
purposes of this chapter, the important point is to understand the difference between
sample and population.

Sample Size

Scale balances sample size and study cost

How big does a sample need to be? The bigger the sample size, the smaller the random er-
ror and the higher likelihood of inding an effect. However, the bigger the sample size, the
more expensive (and possibly more time consuming) the study. Researchers weigh these
two competing interests: reduction of random error though increased number of partici-
pants and cost of the project.

If a sample is too small, in other words, too few people are enrolled in the project, the re-
searcher will run into one of two problems. S/he may fail to ind any meaningful result and/
or the researcher will not be able to generalize his/her research indings.

If the samples size is too big, there are three issues. The irst issue is a waste of resources
since the researcher need not have gathered all that data. Remember for most cases, gather-
ing data costs money. A second issue is a waste of respondents’ time since some of their
data were not needed. The study results would have been the same with a fewer number of
participants. The third problem is that a very large study has the power to ind signiicant
results for very small, even clinically inconsequential differences.

For public opinion polls, such as the Gallup Poll that is done during election years to give
the public an idea of how the presidential candidates are doing, the rule of thumb is 1500
randomly selected people is enough to provide the estimate of the whole population’s opin-
ion on a subject. More than 1500, will not improve the estimate.

When iguring out how many people to survey, the researcher realizes that as the popula-
tion size increases, the percentage of people needed to achieve a high level of accuracy
decrease rapidly. In other words, to obtain the same level of conidence in the study results,
for larger population the researcher would need a smaller percentage of people surveyed
and for a smaller population the researcher would need a larger percentage of people sur-
veyed.

82
Summary
Everyone makes generalizations about the world through sampling. For example, we might
select a book to purchase by reading a few pages. Often in the fruit section of the grocery
story, we either visually inspect the stand of fruit or pinch a few fruit to select the best one.
However, researchers have clear rules to decide on whether they can takes use a population
or need to take a sample to answer their questions. If they need to take a sample, they pay
special attention to iguring out how many samples are needed to answer their questions.

Concept Reinforcement
1. What would be an example of a question that the population could be used to an-
swer?

2. What would be an example of a slightly different question that only a sample could
be used to answer?

3. What are two examples of sampling that you do in your daily life? A generalization
will have been made through this sampling.

83
84
 Chapter 18 - Sources of Data
Chapter Objectives
• List the key characteristics of any dataset

• List vital statistics mandated for collection

• Discuss the pros and cons of datasets

No matter what kind of study the researcher does, it is only as good as the quality of the
data. Epidemiology deals with populations. One reason the study of populations is impor-
tant is that the indings may be able to be generalized. Secondly, data on populations are
needed to make estimates on the frequency of diseases, illnesses, injuries and deaths. To
help researchers, federal agencies compile a great deal of information about the population.
Much of this information is helpful to researches.

Dataset Characteristics
Several characteristics are any dataset must be described for the researcher to understand
how it can be used to answer his/her research question(s).

Data Source
The irst piece of information is inding out the source of the data. Are the data from physi-
cian’s records, from survey of the general population, from vital statistics (birth or death
records), from registry such as a statewide cancer registry? The source of the data is very
important to know.

Availability of the Data

Some data are publicly available to anyone who signs a conidentiality agreement. Other
data are less easily obtained. In other cases, some data will only be released to the re-
searcher when any elements that could identify an individual are stripped from the records.

Collection Date
When the data was collected will matter greatly. The data collected before or after a public
policy can have dramatic inluence on the answers, the participants give to particular ques-
tions. For example, participants would answer the question about seat belt use much differ-
ent before the mandatory seat belt in a state than after the law was enacted.

85
Sampling Plan
The plan used to collect the data is very important. This will tell the researcher about the
usefulness of these data. If the data were collected from a very special population, it may
not be very useful to the researcher. For example if a survey collected data about Hmong’s
dietary habits, this information would not be very useful if the researcher was studying eat-
ing habits of all people.

Data Dictionary
A data dictionary contains at least three elements to be useful: the variable, the question on
the survey, the type of variable. Speciically, the exact questions that were asked of partici-
pants, the variable names associated to the questions and the variable type (numeric, char-
acter, or date) must be available for the researcher to be able to use these data. For example,
in the dataset a variable is called “dob.” However, the question could have been “What is
your date of birth?” or “What is your irst born child’s date of birth?” Without having the
questions, it would be impossible to know the meaning of the variable “dob.”

Identiiable and De-identiiable Data

Data can be divided into two groups: data that contain personal identiiers and data that do
not contain personal identiiers. Personal identiiers are variables in the dataset that can be
used to identify a person. This is particularly important in epidemiology since protecting
the conidentiality of a person is very important. Several variables can identify an indi-
vidual alone or in combination. Some of these are name, social security number, address,
any date (except year), telephone number, fax number, email address, and medical record
number. These must be stripped from the record before the researcher has access to de-
identiiable data. De-identifying data allows the participants identity to be protected.

For researchers who use identiiable data, they would NEVER publish any results so that
anyone reading the article could identify a person. Often the federal and state governmental
agencies use the rule of ive. It means that if ive or less people are grouped together, then
that number is not published. For example, if there were less than ive women diagnosed
with breast cancer in 2008 in a rural county in Missouri, then on the table listing number of
breast cancer cases by county would not show any data for this county. In some situations,
a rule of 20 or 50 is used. This means that at least 20 or 50 people have to be in a group for
the number to be shown.

Advantages and Disadvantages of Pre-existing Data

Probably the clearest advantage of using pre-existing data is that it is already collected.
It is much cheaper for the researcher to use pre-existing data than to collect his/her own
data. The biggest disadvantage of using pre-existing data is that it rarely contains all the
information that the researcher wishes was obtained. Sometimes the data were collected
for some other reason besides a health query. In this case, many of the standard items in a
survey such as age and sex of the respondent might not have been asked.

86
The data in any study has its own particular strengths and weaknesses. For example, a birth
registry that contains information from the birth certiicates is pretty complete. This means
that almost all births would be recorded in the registry. In contrast, some of the elements on
the birth certiicate are not trustworthy, such as the baby’s race. In this case, the baby’s race
may be recorded based on the mother’s skin color or the mother’s last name.

Vital Statistics
The government mandates the collection of certain vital statistics on everyone in the coun-
try. These statistics include birth, death, marriage and divorce information. These records
are maintained for every person in the country. Every child who is born receives a birth
certiicate, which details the names of the parents, the date of birth, and the place of birth.
Likewise, a death certiicate is issued for every person who dies. This information is essen-
tial for maintaining population records, as well as dealing with the legal issues surrounding
birth and death. Marriage status is also considered a vital statistics. Both marriages and
divorces are recorded as part of the vital statistics of a population.

Summary
Primary data is data that the researcher collects from participants whereas secondary data is
data that has already been collected by someone else. In both cases, the quality of the data
is important. High quality data translates into the ability to have more conidence in the
study results. Sometimes secondary is available but in other cases the researcher needs to
collect his/her own data to answer a speciic question. Vital statistics include birth, death,
marriage and divorce information.

Concept reinforcement
1. What is particularly important about data collected by federal agencies, such as the
census?

2. Why is it important hide the number of people who have a particular disease if it is
less than 5?

3. What kind of data would you prefer to use, primary or secondary?

87
88
 Chapter 19 - Types of Surveys and Variables
Chapter Objectives
• Describe the four types of survey data collection

• Deine validity

• Deine reliability

A survey is one way of gathering information from study participants. The survey must
have an objective. An objective is a question that can be answered from the responses to
the survey. The best question to ask yourself when you are designing a survey is “What do
I really want to know?” This is also a good question to ask when you review each question
in the survey. A health survey, which is what epidemiologists do, asks people about health
issues. A health survey can be used in all kinds of study designs. There are four common
ways to collect data: through the mail, by telephone, in person, and using the internet. In
designing a questionnaire, there is always a tradeoff between asking too many question that
are not necessary to answer the question and too few questions.

Mailed Surveys
For success, researchers need an accurate mailing address. If participants are not happy
with the design or the topic, they will be less likely to complete the survey.

Telephone Interviews
Telephone interviewers often enter the data into the computer as it is being obtained from
the respondent. Of course, the researcher needs to have a telephone number to call the par-
ticipants. Finding a person’s telephone number is becoming more dificult with the popu-
larity of cell phones. Similar to the mailed survey, if the participants are not immediately
engaged, it is likely they will hang up or not even pick up the telephone.

In-person Data Collection

The interviewer has a laptop computer, asks questions to the participant, and enters the
data as the respondent answers the questions. This is called CAPI, computer assistant per-
sonal interviewing. Another method for in-person interviewing is to video or tape record
the interview. In both of these situations, the material will have to be transcribed. The inal
situation is where the study personnel hands out a survey and waits while the participant
ills it out and returns it to the study personnel.

89
Transcribe Web-based Data Collection
To transfer informa- Although appealing, the number of people who respond is typically low. Web based sur-
tion from an audio or veys can only be used for those who have computers or who have access to computer.
video record to paper
records.
Variables
A variable is the name used to analyze the data. In an analytical program, a question is
turned into a variable name. The answer options are the values that populate the variable.
For example, “How old are you?” become an “age” variable. The values will be from 1-110
or whatever the age range of the participants. There are three types of variables: number,
text, and date. All questions can be reduced to these three types of variables.

Number
A number variable can be either a whole number (e.g., 1, 2, 3) or an integer (e.g., 1.2, 3.8).
Whether the participant responds to a question by using the whole number or integer de-
pends on how the question was asked.

A question such as “What is your GPA?” requires an integer response. In contrast, “How
many children do you have?” requires a whole number response.

In analytical programs, if any part of an answer is not a number, then the variable becomes
a text variable. For example, if a respondents telephone number is given as 573-493-2332,
then the variable becomes a text variable. In contrast, if the telephone number is entered as
5734932332, then the variable can be a number.

Date
A data variable provides day, month and year. Sometimes only month and year are provid-
ed. In analytical programs, the data variable is converted to some number by the program.
Typically, the statistician does not know the formula used in the calculation and it really
does not matter. The only important point is that the date be entered in the same way. This
means that it can be entered as numbers (e.g., 4/4/10) or as text and numbers (e.g., April
4, 2010). As long as the dates are entered in the same way for any questions, there will be
no problems.

Character
The third type of variable is a character. Just as the name implies, a character variable con-
tains at least one symbol or letter of the alphabet. The concern with character ields is that
the spacing, spelling and capitalization have to be identical in all entries for the answers
to be counted properly. For example, if the answer to participant’s city is St. Louis, the
entry has to be St. Louis for everyone who lives there. Otherwise, counting the number
of participants who live in St. Louis will give data on each spelling variation. St Louis is
different from St. Louis. Similarly, St. louis is different from St. Louis. As you can see,
there are numerous variations in this data entry. Consequently, when possible, researchers
usually recode text ields to numbers. An answer to “Are you a) male or b) female?” will
be entered as a 1 or 2 in the database. The number 1 means a male response and a number
2 means a female response.

90
Data Validity and Reliability
Data validity and reliability are two ways researchers assess the quality of the research
conducted. Data validity is generally considered the correctness and reasonableness of the
data. The question to ask here is whether the data collected during the survey actual relect
what is happening. Data reliability is a measure of reproducibility of the data. For example,
if you ask the same person a question that has only one correct answer, such as birth date,
on ive different occasions and you get the same answer ive times, the data are reliable.
However, if the person gives different answers, the data are probably not reliable. It is pos-
sible for data to be highly reliable, but invalid. In other words, it is possible to do some-
thing the same way many times and get the same result. However, if the tool you are using
introduces bias to the data, all of the data could be incorrect even though they are reliable.

Summary
Using surveys is one of the key ways that epidemiologists collect information in their
quest to understand risk factors associated with health outcomes. Each question in a survey
is turned into a variable. How old are you? becomes an age variable. The three common
variables are numbers, characters or dates. By thoughtful collection of information, epide-
miologists can answer questions about risk factors and health.

Concept Reinforcement
1. What are the beneits of asking as many questions as a researcher as possible? In
other words, pages and pages of questions?

2. What are the disadvantages of asking as many questions as a researcher as pos-

sible? In other words, pages and pages of questions?

3. Do you think people would respond differently if asked “What is you’re age versus
What is your birthdate?” How about “How much do you make?” versus “Select the
category that best matches your income?” Explain.

91
92
 Chapter 20 - Constructing Survey Questions
Chapter Objectives
• Deine the main properties of a good survey

• Identify features of a problematic question

• List six characteristics of a good question

• Describe the limitations of the three types of questions

A survey is one way of gathering information from study participants. The survey must
have an objective. An objective is a question that can be answered from the responses to
the survey. The best question to ask yourself when you are designing a survey is “What do
I really want to know?” This is also a good question to ask when you review each question
in the survey. A health survey, which is what epidemiologists do, asks people about health
issues. A health survey can be used in all kinds of study designs.

One of the biggest misconceptions is that writing good survey questions is easy. Since we
speak the language and we ask questions every day, we assume that we are all skilled at
making up survey questions. In fact, few people can easily write good survey questions. In
class, students have heated arguments with their teacher about a test because the question
or the answer options were not clear. The beneit of classroom testing is that students have
the opportunity to challenge the question one-on-one with the teacher. In surveys, this op-
portunity is not possible. Consequently, the questions need to be carefully written.

Important Points in Writing Survey Questions

Everyone needs to understand the answer options to multiple choice questions in exactly
the same way. Answers to multiple choice questions must be stated clearly. It is very im-
portant to remember the participants to whom the questions are being asked. It is possible
for different participants to understand the same choices differently if they are not carefully
phrased.

Double Barreled Questions

Only one question is asked and answered per question—no “double barreled” questions. If
a question seems too complex, it is important to make it simpler or split the complex ques-
tion into two questions.

For example, Do you drive or take the bus everyday to school? Yes ( ) No ( )

A respondent can answer “yes” to “Do you drive?” and “ No” to the part of the question
that asks about taking the bus.

93
Mutually Exclusive Questions
The categories (answer options) must be mutually exclusive. Answers to multiple choice
questions must not overlap in any way. This will lead to confusion among the respondents
and likely bias the results of the data analysis.

For example, “What is your annual household income in 2009?”

a. Less than $20,000

b. $20,000–$50,000

c. $50,000–$75,000

d. $75,000–$100,000

e. More than $100,000

If a respondent’s household income is $50,000, which category does the respondent select,
b or c? These answer choices are not mutually exclusive.

Continuous Data
The categories (answer options) that use continuous data include all possibilities within the
range. Consider the nature of the question being asked and the standards used in the target
population. For example, homes can have inside and outside bathrooms. They can also
have full, half and quarter baths. If one is asking about bathrooms, it is important to offer
all of the options so respondents can answer accurately.

For example “How many bathrooms are inside your home?”

a. 1

b. 2

c. 3

d. more than 3

What answer choice would the participant select for 1.5 bathrooms?

94
Unbias Questions
The questions must be unbiased, meaning the question or answer choices are stated objec-
tively. They do not encourage the participants to answer one way or the other.

For example, “Which one of the following do you feel is most responsible for the increase
in obesity in children?”

a. fewer safe places to play outside

b. bad food choices

c. children watching too much television

d. lack of physical exercise during the school day

The term “bad” in option b has a value connotation. A negative connotation as part of the
answer option is not part the other choices. Option c also has a value connotation because
the words “too much” are used.

Negative Questions
Avoid negative questions since respondents will often skip over the negative and answer
the question as if it was worded in the positive. The human brain tends to ignore the word
“don’t.” For example, if you ask a person about the last time they did not do something,
they will likely respond with information on the last time they did that speciic action.

Time Reference
Provide a time reference when appropriate. For example, it is important when asking about
annual income to include a time frame so that data can be compared across all the respon-
dents. One person may think of income in term of the calendar year (Jan-Dec) and others
may think of it in terms of a iscal year (some other 12 month period).

“How many international conferences did you attend?”

a. Did not attend any conference

b. 1 conference

c. 2 conferences

d. 3 conferences

There is no time reference so the participants might answer for one year or during their
career.

95
hree General Mistakes in Writing Questions
Poor survey question development is the biggest mistake made in epidemiologic research;
and it is even committed by experienced researchers.

Another mistake is that researchers want to ask many more questions than are necessary.
As the survey gets longer, less respondents will take the time to complete the survey. Col-
lecting too much information wastes the respondents’ time since this extra information will
rarely be used.

The third mistake that researchers make when constructing a survey is not doing their
homework. This means that they do not read the literature thoroughly to ind out what is
known and what should be asked.

Types of Questions
Question types include open ended, multiple choice, and yes-no (also true-false)

Open-ended
Turning responses to an open ended question into data is very time-consuming. There is
software to help with this but irst the response has to be put into electronic format. After
that specialize data analysis is performed to have qualitative information. Except for re-
searchers that are interested in this type of data, open-ended questions are usually avoided
in survey design.

Multiple Choice
Multiple choice questions provide several answer options that give the respondent a choice
but, if not properly put together the quality of the survey. Many of these issues are de-
scribed above.

Yes-No
Limited information is gained by asking yes-no questions or true-false questions. They are
often used to select a population with a very speciic characteristic for a section of a survey.

96
Summary
Gathering data from surveys is one of the fundamental tools of epidemiology. Being able
to design a survey that effectively and eficiently gathers data to answer a focused question
is a skill that develops with practice. Many researchers have experienced the dismay of
either asking a poor question (and therefore not being able to use the data gathered for that
questions) or not asking a vital question. Understanding the fundamentals of survey design
can arm the researcher with the tools to design a good survey full of useful information.

Concept Reinforcement
1. Of the listed mistakes made in survey question design, which one(s) have you
experienced?

2. Of the three types of questions, open-ended, multiple choice, and yes-no, which
type would take the most time in preparing a dataset for analysis? Explain.

3. In what situation would an open-ended question be better than a multiple choice

question?

97
98
 Chapter 21 - Answer Choices and Layout in
Survey Construction
Chapter Objectives
• Describe the issue with categorical options

• Describe the concept of mutually exclusive

The construction of a survey is a critical part of the survey process. In order to build a
good survey, the researchers must have clearly deined the goals of the survey and the
types of data he wants to collect from the subjects. Since a survey consists of questions, the
researchers must think carefully about each question. The survey respondents will all be
asked the same questions, but may interpret them differently. For example, if you are asked
the question: “Do you have a child in school?,” you will interpret the question differently
depending on whether you have children and the ages of your children. If you do not have
children, you might answer “no,” meaning that you do not have any children. If you have
children who are too young or too old to be in school, you might answer “no.” The response
is the same, but the meaning is very different.

Categorical Options
One method researchers use to present clear questions to survey respondents is providing
categorical options. The categories allow the researchers to limit the number of answer op-
tions, while still providing more than two choices to the respondent.

Categories may include ranges of values, such as income ranges or age ranges. Categories
may also include groupings of speciic items, descriptions, levels of agreement or disagree-
ment, etc. The categories are established based upon the goals of the research. Some com-
mon categories used in epidemiology surveys include: age, occupation, education level,
gender, income, race, and ethnicity. Time is also often used as a categorical option. Epide-
miologic surveys will often ask about diseases, exposures or behaviors from the past.

Common Categorical Options Example Category 1 Example Category 2

1-5 years < 30 days
Time 6-10 years 30-45 days
11-15 years >45 days
Infant 0-5 years old
Child 6-10 years old
Age
Teenager 11-15 years old
Adult 16 years and older
High School 12 years
Some College 14 years
Education level College Graduate 16 years
Graduate School 18 years
20 years or more.

99
Mutually Exclusive
Most survey questions ask respondents to give speciic answers. If these answers do not
overlap, they are called mutually exclusive. When developing questions for an epidemi-
ology survey, it is important that the responses to questions are mutually exclusive. For
example, a yes/no question will get a yes or no answer. If the answer is yes, the answer can-
not also be no. This is called mutually exclusive. Another area where it is very important
to ensure that answer options are mutually exclusive is when asking respondents to select
a range of values that best represents their speciic situation. This could be income, age,
weekly exercise, or any other variable. If the values in the answer options overlap, respon-
dents will not answer consistently: some will choose one option and others will choose the
second option. This will make it dificult to make any conclusions about the study results
because the data were not collected in a consistent manner. Notice the sample in the im-
age below. The irst question is not mutually exclusive. Values are repeated in more than
one category. For example, the value $15,000 appears as an option in both answer “a” and
answer “b.” The second version of the question solves this problem by providing mutually
exclusive answer categories.

Mutually
Exclusive

!"#$%&'()*+,('&'-."+"/001(#230"$*4#(/'$)#5$

/0at
o23o4
best
6esc7ibes
t0e
a44ual
i4co8e
o9
you7
0ouse0ol6:


A.  $0‐$15,000

B.  $15,000
‐
$30,000

C.  $30,000
‐
$45,000

D.  $45,000
‐
$60,000

E.  Above
$60,000


!"#$%&'()*+,(."+"/001(#230"$*4#(/'$)#5$(
/0at
o23o4
best
6esc7ibes
t0e
a44ual
i4co8e
o9
you7
0ouse0ol6:


A.  $0‐$15,000

B.  $15,001
‐
$30,000

C.  $30,001
‐
$45,000

D.  $45,001
‐
$60,000

E.  Above
$60,000


100
Summary
One of the most important parts of designing and implementing a survey to collect epide-
miologic data is ensuring that the questions asked will elicit the correct response from the
respondent. The questions must be clearly stated. If the questions provide answer options,
the categories must be clearly described and mutually exclusive. If the question is unclear,
respondents will interpret it differently, resulting in inconsistent answers that compromise
the quality of the data being collected. If the answer categories overlap, respondents may
provide the “same” answer by using different categories, both of which are correct.

Concept Reinforcement
1. Explain why researchers use categorical options when collecting data from respon-
dents.

2. Discuss a potential problem with using categorical options in surveys.

3. Explain mutually exclusive and discuss why it is important in question and survey
development.

101
102
 Chapter 22 - Survey Layout
Chapter Objectives
• List the key characteristics of layout

• Describe the use of space and length

For questionnaires that the participants ill out themselves, they need to be constructed for
ease of understanding. Several things need to be thought about in the design stage. These
include formatting options, question order, and appearance of individual pages.

Design of the Questionnaire

Two objectives should be accomplished through the design of the questionnaire. One is to
increase the number of responses. The second objective is to reduce the number of ques-
tions skipped or answered incorrectly.

One misconception is that a paper-and-pencil questionnaire allows the participant to care-

fully read the questions and answer them at their own speed. However, people really do
not read the entire questionnaire in a thoughtful way. They use clues from the layout about
what can be read and what can be ignored. It is not uncommon for participants to skip
words. The best strategy is to keep the wording and visuals simple.

Grouping of Questions
A low of questions in a survey is similar to the low of a conversation. One rule of thumb
is to group questions by topic. For example, put all the physical activity questions together.

One of the most important ways to get people to answer a survey is to have them feel it is
relevant to them. In other words, if they feel the survey is asking about something that is
important to them or interesting to answer, they are more likely to ill it out. It is impos-
sible to have the entire questionnaire illed with terribly interesting questions—-some will
be interesting to the participant and others less so. Therefore, the order to the question must
also be carefully taken into consideration.

103
 Rules of humb for Ordering of Questions
In the cover letter that accompanies a survey, the purpose of the questionnaire is described.
Consequently, the irst questions should directly speak to this purpose. For example, if
the survey is about physical activity, the irst questions should be about physical activity.
Later questions can be about other important items such as age, sex, education attainment,
and income level. Questions that ask about opinions or attitudinal scales should go in the
middle of the survey. Concrete questions that just report facts, such as age and income
should go at the end of the survey.

Besides these rules of thumb for placement of questions, questions should be order in a
McCarthyism manner that seems logical to the participant. Generally, it helps to ask irst details about an
During the late event before asking the participant to evaluate the event. For example, it makes sense to
1940s through the irst ask about a medical condition, say a broken leg. First, ask about when, how and type
1950s many thou- of break. It would make sense to follow this with questions evaluating the medical care
sands of Americans received.
were accused of
being Communists
or communist sym- Sources of Order Efects
pathizers. Investiga-
tions into people’s We have lots of information on how the order of the questions can inluence answers. There
lives were con- are 5 issues that a researcher should be aware of when designing a questionnaire. Depend-
ducted. The people ing on the ordering of the questions, big differences in responses can be seen.
singled out were of-
ten government em- Evenhandedness
ployees, entertainers,
This is a value-based response. People usually want to be fair. A survey done in 1948 tested
teachers and union
question order. Some participants were asked irst whether it is OK for an American re-
activist. The proof
porter to write about a visit to the Soviet Union (now known as the USSR, Union of Soviet
of communism or
Socialist Republics). Some participants were asked whether it is OK for a communist re-
communist sympathy
porter to write about a visit to the United States. When the irst question was the communist
was often weak or
reporter question, the percent of positive responses to the American reporter question was
almost non-existent.
37%. In contrast, when the irst question was the American reporter question, the positive
Many people lost
response to the American reporter was 73%. The order changed the responses by almost
their jobs, had their
ifty percent. This is an example of even-handedness. Participants want to be even-handed
careers destroyed,
in their responses so when put in a position of noticing this, for example how they treated
especially those
a communist and American reporter, they may alter their answers.
in the entertain-
ment business, and Anchoring
sometimes wnet to
prison. Almost all of This is a cognitive-based question. The irst question inluences the answer to the second
the prison sentences question. In 1993, a survey was given to students with two questions about cheating in
were overturned university. One question was asked about cheating in their university and the other asked
later. However, the about cheating in universities throughout the U.S. The order to the question changed the
loss of a career could responses. If students were asked about cheating in U.S. universities irst, they were much
not be ixed later. more likely to say that cheating was a problem at their university. This shows that the an-
swer to the irst question served as a reference point or anchor to the second question.

104
Carryover, subtraction, and summary item effect

The order of general questions followed immediately before or after a speciic question in-
luences the response. These can go in either direction and depend on the type of question.
Sometimes, people maintain the same feeling from one question to another (e.g., question
about one’s marriage and then marriage in general). Sometimes, survey responders consid-
er their response(s) to previous questions to adjust the answer to the more general question.

he First Question
No other question is more important than the irst question since this one question can play
a role in whether the respondent completes the questionnaire or throws it away. There are
a few rules of thumb regarding this question. First, the question should apply to everyone
who gets the survey. It should be linked to the purpose of the survey, as stated in the invita-
tion letter and be interesting. It should be an easy question. For example, listing all people
in your household is a lot harder than indicating the number of people in your household.

Page Design
According to Dillman, a leading researcher on survey design, many things can be done to
improve a survey. Below are several pointers.

• Increasing font size helps attract attention

• Bolding helps attract attention

• Use spacing to group items

• Use similar font type to group items

• Begin asking questions in the upper right quadrant of the page

• Avoid use of the lower left quadrant

• Number the questions consistently

• Place more space between questions than between the question and answer options

• Use darker font for the question and lighter font for answer choices

• Vertical placement of answer choices is preferred

• If use horizontal placement of answer choices, all choices should be evenly spaced on
one line

105
Cover Pages
The front cover page will be seen irst. A well-designed cover can motivate someone to
answer the questions. Simple graphics are better than detailed ones since a detailed one
may contain errors or be annoying to the respondents. The title and name and address of
investigator are included for easy reference.

The back cover is often white space. This is a place to invite comments, give a thank you.
The back cover should not compete with the front cover.

Summary
Many pieces need to be considered for a survey to look professional. Additionally, the
design can also inluence participants’ response. A poorly designed survey is more likely
to be put aside and not completed compared to a well designed survey. Besides the design,
understanding how people ill out surveys can help researchers design a survey to maxi-
mize participation.

Concept Reinforcement
1. In a survey asking about risk factors associated with quitting high school, what
should be the irst question?

2. Of all the things to think about in survey design, what do you think is the most
important? Explain.

3. What is a possible example of a question in whichh even-handedness might play a

role?

106
 Chapter 23 - Use of Incentives and
Participation Proportion
Chapter Objectives
• Describe the types of incentives

• List the pros and cons of using incentives

• Deine participation proportion

• Explain the importance of participation proportion

Participation Proportion
All epidemiology studies published in peer reviewed journals spend some time informing
the reader of the selection criteria for participation. Included in this description is the par-
ticipation proportion (it is also commonly known as response rate; as a side note this term
is technically incorrect since the term rate is a special kind of proportion often associated
with frequency.)

Participation proportion is typically reported as a percent. It is the percent of people who

agree to be in the study. Generally, a participation proportion over 70% is considered good
in epidemiology studies. Twenty years ago, participation proportion of 90% was not un-
common. However, these days, less people are willing to participate in studies, it seems.

It is always important to igure out if the people who agree to be in the study are differ-
ent in some important way from those who refuse to be in a study. Even if the sampling
method is random, the results may not be generalizable if the characteristics of those who
participate differ from those who do not participate. Some general characteristics of the
non-responders may be known, and therefore compared with participants such as age and
sex. The more characteristics that can be compared the better. For example, if a researcher
is interested in describing quit attempts by smokers and no heavy smokers participated in
the study, than even with a very good research plan, the results could not be generalized
to all smokers since one subgroup—the long-term heavy smokers—did not provided data.
Unless the researcher was able go gather this information, somehow, a mistake might be
made in reporting the study results by thinking the results apply to all kinds of smokers.

107
Incentives
There are many studies that have looked at the ways to improve participation proportion in
mailed surveys. Some of the indings make sense whereas others are less intuitive.

Money
One of the most powerful ways to improve participation proportion is to give respondents
cash. It is not known how much cash makes a difference. Having $1 included in the mailed
survey may be just as good as having $5. However, $20 would deinitely increase the num-
ber of people who responded.

Besides just giving money, it is better to give people money as opposed to giving people
money, if they ill out the survey. This is known as conditional incentive. In other words,
the respondents will get the money, if they do something irst.

Delivery
How the survey is delivered makes a difference. Special delivery can increase participation
proportion 2-fold. Including a stamped return envelop as opposed to a preprinted business
reply envelop increases response.

Using a window envelop or a larger envelop makes no difference. Similarly sending the
survey to the work address versus the home address does not seem to make a difference.
Using commemorative stamps also does not seem to change the percent of people who
respond to the survey.

Sender
If the sender is from a university compared to the government or business, the participa-
tion increases. Surveys from physicians versus a research group shows no difference. The
ethnicity of the sender has no inluence on participation proportion. If the introductory
letter is signed by senior or more well know person does not change the number of respon-
dents. However, the introductory letter signed by hand versus computer generated signa-
ture seems to slightly increase the participation proportion.

Reminders
Sending a postcard or letter to let people know a survey is coming, called prenotiication,
improves participation proportion. The means of prenotiication do not matter—either by
telephone or by mail have the same participation proportion. Sending a second question-
naire helps improve participation proportion. Mentioning in the introductory letter that the
respondents will be contacted again, if they do not ill out the questionnaire has no inlu-
ence.

108
Content
It makes sense that a survey that is more interesting to the respondent is likely to be com-
pleted. Naturally, shorter questionnaires are more likely to be illed out than longer ques-
tionnaires. Questionnaires starting with relevant questions, easier, and more general ques-
tions at the beginning are more likely to be illed out.

Surveys with open ended questions where the respondent has to write the answer is less
likely to be completed compared to closed questions (e.g., choosing from option list). Of
course asking sensitive questions are less likely to be illed out. It does not matter whether
the answer options are listed in increasing order or whether there are cirecles or check
boxes.

Summary
Researchers strive to have high participation proportion so that they can state their indings
with conidence. Physicians are well known for being very reluctant to ill out surveys. It
is common to have 20% participation proportion for a physician survey. In cases where
the participation proportion is low, the researcher always worries about how closely the
respondents look like the non-respondents. Are the respondents different in some important
way? Some research has gone into iguring out ways to maximize participation. Playing
attention to all aspects of the data collection can make a different.

Concept Reinforcement
1. Why do you think people are less willing to participate in studies these days?

2. What would you consider sensitive question that your peer group would be reluc-
tant to answer? How about your parents peer group?

3. Are there any sensitive questions that transcend generations?

4. What are the trade-offs between illing out the survey and not responding?

109
110
 Chapter 24 - Biomarker Matrices and Collection
Chapter Objectives
• Explain matrices and general collection protocols

• List ive advantages of using biomarkers

• List disadvantages of using biomarkers

Epidemiology uses mathematical formulas to show the relation between exposure and
dis-ease outcome. In this chapter we are going to explore the use of biomarkers,
which are measures of exposure in biological samples used to quantify exposure.

Exposure
In the early examples of epidemiology, researchers did not know exactly what factors or
agents were associated with a disease. For example, Dr. Snow did not know what caused
people to get sick with cholera. He used epidemiology tools, such as maps and interviews,
to help him compile evidence that the agent was something in the water. He actually died
before learning about the discovery of the bacillus that caused cholera. The cholera bacillus
was discovered in 1854 by an Italian anatomist, Fillip Pacini. Unfortunately, Pacini did not
prove that the bacillus that he discovered caused the disease, so his work remained obscure,
and his reports were not translated into English.

In epidemiology, most of the early cases showing a relation between exposure and health
status occurred in occupational settings. Often the exposure to the agent is much higher
in occupational settings compared to the general population. In addition, the agent is also
more readily identiied in the occupational setting than in the general population.

Below are some examples of diseases associated with exposure. Both magnitude of expo-
sure to the agent and identity of the agent are more readily obvious in occupational settings.

• Black lung disease associated with coal mining (inhaling coal dust).

• Mad hatters disease (insanity) associated with working in the millinery (hat) trade
where mercury nitrate was used to treat the felt lining of hats.

• Chimney sweeps and testicular cancer (coal tar).

• Clock dial painters and bone cancer (paint was laced with radium 226 so that the di-
als glowed).

Even though the association of exposure to disease was eventually clear to the general
public, the cause of the disease often remained unknown for some time.

111
Exposure has four components.

• A target, such as a person.

• An agent, such as plasmodium (parasite that causes malaria).

• The vector: The means of getting the agent to the target. A classic example of a
vector is the mosquito. In some cases, in place of the vector is the environment. For
example, air is the “vector” for transmitting ozone.

• Contact between the target and the agent. In the malaria case, the contact is a bite
from the infected mosquito.

Exposure has also been strongly associated with dose. A common phrase in toxicology, or
some would say a basic rule, is “the dose makes the poison.”

A 16th century Swiss chemist, Paracelsus, is credited with this rule. The concept is that
practically every substance on earth (including water and vitamin C) can kill a person if it
is concentrated enough in a person’s stomach or blood stream.

Biomarkers
Biomarkers are cellular, biochemical, or molecular alterations that are measurable in bio-
logical media, e.g., human tissue, cells, or luids.

Three types of measurement are used for biomarkers:

• The level of the substance itself in biological media

• The level of products of bio-transformation of the substances in the same media.

For example DDT is the substance that people are exposed to but once in the body it
is transformed into DDE.

• The biological effect that result from contact of the external agent with the human
body

Advantages to Using Biomarkers in Epidemiology Studies

1. Improve study validity and reduce bias in data

This means that the researcher is much more conident that the exposure really happened.
Remember back to Chapter 4 on self reported data and recall bias. Determining the level
of the agent in the biological sample is much stronger evidence of exposure than
when someone reports that s/he was exposed

2. More nuanced measurements of the exposure

In today’s scientiic world, the level of detection is very low for many agents, in the parts
per million, or parts per billion.

112
In general, the health effects of any toxic substance are related to the amount of exposure,
also known as the dose. Generally, the greater the dose the person is exposed to, the more
severe the effects from that exposure. Some chemicals can cause toxicity at very low doses
and so it is important to be able to understand how these very small amounts are measured.

Concept Extension – Dose Measurement Systems

Parts per million (ppm) and parts per billion (ppb) are the most commonly used terms to
describe very small amounts of contaminants (or agents) in biospecimens. They are mea-
sures of concentration, the amount of one material in a larger amount of another material.
They are expressed as concentrations rather than total amounts. For example, researchers
do not have to measure the total amount of lead in a person’s entire blood supply to under-
stand the level of exposure. They merely measure the concentration in a small sample to
represent the total amount. In doing so, they can compare the concentration levels among
the participants.

Sometimes, instead of using the part per million or billion terminology, concentrations are
reported in weight units; such as the weight of the contaminant compared to the weight of
the total. The metric system is often used. For example, a milligram is a thousandth of a
gram and a gram is a thousandth of a kilogram. Therefore, a milligram is a thousandth of a
thousandth, or a millionth of a kilogram. A milligram is one part per million of a kilogram.
One part per million (ppm) is the same as one milligram per kilogram, mg/kg. In summary,
one ppm equals one mg/kg.

In the metric weight system, a microgram is a thousandth of a milligram. Since a milligram

is a millionth of a kilogram, and the microgram is a thousand times smaller, it is equivalent
to a billionth of a kilogram. Microgram is abbreviated ug. Therefore, a part per billion is
equal to a microgram per kilogram, ug/kg.

In the metric liquid measurement system, the unit of liquid volume most commonly used is
the liter. A liter of water weighs one kilogram. If the contaminant is a solid, it is measured
in milligrams. Thus, one part per million of a solid in a liquid can be written as a milligram
per liter and abbreviated mg/l. Similarly, a part per billion of a solid in a liquid is equal to
a ug/l.

113
Example: PCB Illustration of Dose Measurement
Polychlorinated biphenyls (PCBs) have been implicated in developmental deiciencies and
neurologic problems in children, disruption of reproductive functions, liver disease, effects
on the thyroid and immune system, and increased cancer risks. PCBs are synthetic organic
compounds comprising 209 individual chlorinated biphenyl compounds (also known as
congeners). Although the manufacture of PCBs was banned in the United States in the
1970s, people are still exposed to them. They are persistent compounds that degrade very
slowly. It is important to understand the primary source of exposure. Below is a case study
of PCB exposure from diet versus water.

In determining the important source of exposure, we will compare the relative importance
of PCBs in Great Lakes ish consumption versus PCBs in Great Lakes drinking water con-
sumption. The maximum level of PCBs legally allowed in ish sold in interstate commerce
is 2 parts per million (ppm). (This does not apply to ish sold locally, however.) In contrast,
the average PCB level of the Great Lakes drinking water is 4 parts per trillion (ppt). (No
legal limit exists for drinking water level.)

Remember that a part per million is a million times more than a part per trillion. The ish
levels are measured in ppm whereas the water levels are measured in ppt. So at irst, one
would assume that the ish consumption is a more important source of PCBs than the water
consumption. However, it is not only the concentration level but also the dose—how much
is consumed that is also important.

We consume a lot more water than ish on a daily basis. For example, a few people might
eat as much as a pound of ish a day or as little as one pound every 6 months (180 days).
On the other hand, people generally drink at least 2-4 liters (4-8 pounds) of water a day.

For those who drink little (4 lb/day) and eat lots of ish (1lb/day), it is clear that ish is the
largest source of PCBs. Exposure from great lakes water consumption is 4 parts per trillion
versus the exposure from eating ish, which is 1 part per billion. The same holds true for
heavy water drinkers (8 pounds/day) and light ish eaters (1/180 lb per day). For this group,
the PCB exposure from great lakes water is 8 parts per trillion and from ish it is 0.006 (1
÷ 180) parts per billion. So ish consumption, even for people who rarely eat ish is most
likely the main source of exposure. Consequently, if a PCB advisory was posted, the advi-
sory would target ish consumption and not drinking water consumption.

114
3. Identify individual susceptibility, e.g., genetic biomarkers

Example: PKU Illustration

In 1934, a Norwegian doctor, Asbjorn Folling, discovered the cause of severe mental re-
tardation in 2 young children, PKU (phenylketonuria), also known as Folling’s Disease in
Norway after the discoverer. PKU is called an error of metabolism because children can-
not break down phenylalanine, an amino acid making up approximately 5% of all protein
foods. Dr. Folling showed that PKU was an inherited disease. At that time, the prevalence
of PKU in institutions for severely mentally retarded individuals in western counties was
1-2%. Blood tests were developed in the early 1960s to test for this genetic abnormality
since, if caught early–within the irst 2 weeks of life–mental retardation can often be pre-
vented. In the United States, the newborn screening test for PKU is mandated by law in all
states except Maryland and Wyoming.

Example: BCRA 1 and 2

A history of breast cancer in irst-degree relatives is an important breast cancer risk factor,
particularly if the cancer occurred at an early age. Some breast cancers that cluster in fami-
lies are associated with inherited mutations in particular genes, such as BRCA1 or BRCA2.
BRCA stands for BReast CAncer one and two.

BRCA1 and BRCA2 are major genes related to hereditary breast cancer; they were discov-
ered in 1994 and 1995 respectively. BRCA1/2-affected women carry a high risk of devel-
oping breast cancer, ovarian cancer, melanoma, and pancreatic cancer. BRCA1/2-affected
men have an increased risk of also developing breast cancer, pancreatic cancer, melanoma,
and prostate cancer.

Commercial genetic testing to identify people who have the mutations in the BRCA1 and
BRCA2 genes is now readily available. However, unlike PKU, there is no treatment for
BRCA1 and BRCA2 to decrease the cancer risks.

4. Allows for early detection of disease or illness

Example: HDL cholesterol and atherosclerosis.

Atherosclerosis is known as hardening of the arteries. Complications of atherosclerosis

are the leading cause of death in the United States. Atherosclerosis can be considered to be
a form of chronic inlammation. Michael S. Brown and Joseph L. Goldstein received the
Nobel Prize in Physiology or Medicine in 1985 for their discoveries about the regulation of
cholesterol metabolism and treatment of diseases caused by elevated levels of cholesterol
in the blood.

Commercial cholesterol testing is available to identify people with HDL, “good” choles-
terol and LDL, “bad” cholesterol. Statins are the current drug of choice to lower the LDL
levels.

115
5. Provides a clue as to the mechanism of disease occurrence

Example: Hepatitis C

Hepatitis C is the leading cause of liver transplants. It is a blood-borne infectious viral dis-
ease. It is usually spread through contact with infected blood. Most people with hepatitis C
do not have any symptoms. There is no cure for hepatitis C.

Biomarkers are used to predict a hepatitis C liver transplant patient’s chance of developing
ibrosis, a formation of scar-like tissue that can lead to cirrhosis. Changes in certain types
of liver cells are useful in determining those who are at the greatest risk for developing
ibrosis.

6. Provides data on effectiveness of intervention trials

Example: Isocyanate Exposure

Automobile painters can be exposed to hardeners that are part of the topcoat of automo-
bile paints. These hardeners contain isocyanate resins. Workers exposed to isocyanate may
develop a serious or fatal respiratory disease. An intervention to decrease isocyanate expo-
sure to workers in automobile body shops used urinary biological monitoring (urine test-
ing) of exposure. Workers provided a urine specimen after work each day. In shops that had
ventilation fans turned on all the time, workers were found to have lower isocyanate level.

Disadvantages of Biomarkers
1. There can be considerable variation in levels biomarkers within a person depend-
ing on time of day or month.

Example: melatonin

Melatonin, a neurohormone, is produced from the amino acid tryptophan, in the brain by
the pineal gland. Melatonin’s synthesis and release increases in the absence of light, i.e.,
at night or in darkness. Based on this characterization, melatonin is involved in circadian
rhythm and regulations of diverse body functions.

116
2. There can be considerable variation in the levels of biomarkers between people that
is not related to the disease outcome.

Example: menstrual blood loss for moderate or less severe periods and anemia

Menstrual length has signiicant variation among women with an average of 3-7 days and
the amount of blood loss can be qualitatively described as light, moderately light, moder-
ate, moderately heavy, and heavy. In studies, the intensity of a woman’s period measured
in blood loss was not associated with anemia for women in all categories, except heavy
menstrual periods (>120mL).

3. Considerable intra-individual variation (variation from one sample to another in

the same person).

Example: total mercury in blood

Developmental deicits in children associated with mercury exposure are well documented.
Pregnant women and young children are advised to limit ish consumption due to this
health effect. Total mercury in plasma is associated with both inorganic mercury and or-
ganic mercury with large inter-individual variations in the distribution between red blood
cells and plasma. The inter-individual variation in the blood to hair ratio is also very large.

4. Many exposures have a short half-life.

Example: Atrazine, a common corn herbicide

Atrazine exposure can occur during corn planting season. Two common ways that atrazine
exposure can happen is by inhalation of suspended particles (drift) or ingested from touch-
ing food with hands contaminated with the pesticide and then eating the food or hand (con-
taminated with the pesticide) to mouth behavior. Atrazine’s half life in urine is 24 hours. In
48 hours, all of the atrazine from the previous 48 hours will have been eliminated.

5. Gathering tissues or luids to analyze them for contaminants can be intrusive.

Fear of needles, known as belonephobia, is a condition in which an individual is extremely

afraid of facing needles or injections. It has been reported that up to 10% of the population
has a state of apprehension when facing the possibility of a needle stick. Consequently,
gathering blood is considered signiicantly more intrusive than most other forms of bio-
specimen collections such as toenail collection, urine collection, or saliva collection.

6. Both gathering biospecimens and analyzing them for contaminants is costly.

Current instrumentation allows for detection of minute levels of biomarkers in specimens.

However, with the more precise detection a cost is associated with these analyses.

117
Summary
Perhaps one of the most fundamental problems in epidemiology is measuring the exposure.
For example, cell phone use has been linked to increased risk of brain cancer in some but
not all epidemiology studies. Typically, the exposure is ascertained by asking study par-
ticipants their usage of cell phones, duration and frequency. These indirect measures leave
critical questions unanswered such as internal dose of the hazardous exposure. The use of
biomarkers can, in some cases, provide reasonable data on exposure to the hazardous sub-
stance and thereby strengthen the study indings.

Concept Reinforcement
Explain how occupational exposure helps epidemiologists understand the effects of expo-
sure to speciic factor or agents in the population.

Discuss the advantages to using biomarkers to study population health.Discuss the disad-
vantages of using biomarkers in epidemiologic studies.

118
 Chapter 25 - Making a Map – MAUP
Chapter Objectives
• Describe the modiiable area unit problem

• Describe the advantages of using map for data presentation

• Discuss the limitations of maps for representation of data

Everyone knows the adage, “A picture is worth a thousand words.” The new adage is “a
map is worth a thousand pictures.” The imagery shown on a map is immediate and compel-
ling. With the advent of user friendly desktop software to create maps and the electronic
iles provided by the US census bureau as well as other organizations, researchers can now
display data using maps. When a map is created, it gives the impression of complete infor-
mation. However, like any other type of presentation the quality of the information shown
on a map is only as good as the completeness and accuracy of the data used to make the
map.

One signiicant beneit of using a map is the ability of a map to portray complex informa-
tion in an understandable way. Further, mapping can visually show the data table for ease
of comprehension. Another beneit of maps are their universal appeal. A speciic language
is not needed but rather an understanding of universal symbols. Maps are also visually ap-
pealing and can be used to communicate results while also giving a reader a visual break
from dense text. Maps can also reinforce information by visually showing results that are
described in the text.

Geographic variation of breast cancer mortality by county;

http://www3.cancer.gov/atlasplus/

119
 Map Limitations
There are several limitations to displaying data via maps. Areal bias is one limitation that
can occur. The visual impact of a map is related to both shading and surface area. Surface
area is the area represented by the areal unit, for example some states are much larger than
others in square miles, but may represent a smaller population compared to a state with a
smaller area, but denser population. Most of the New England states are geographically
small but densely populated as compared to the many of the Mid-western states.

Look at the map below that shows the mortality rate by state for brain and nervous system
cancer in white males, 1970 to 1994, all ages.

Without thinking, which states draw your attention? Again, looking quickly at the map, list
the states with the highest mortality rates.

Chloropleth
Brain and nervous system cancer mortality rates in white males, 1970 to 1994
Maps that show
information based on If you are like most people, your eye will be drawn to the dark color. You will probably
area (areal) units and notice Idaho, Montana, and Colorado irst as having high rates. These states are large in
use color to show area and they are colored in black. This demonstrates the areal bias. New Jersey and New
data ranges. Hampshire also has very high brain cancer rates but these states are not noticed as quickly
due to their geographic size.

A second limitation of maps is the modiiable area unit problem (MAUP). This means the
results of a study can appear to change tremendously when showing these data on chorop-
leth maps using different areal units.

120
 Percent of households with income < $10,000 shown by various areal units
in the Bootheel region of Missouri

The third limitation of maps is the inability to show how conident the researcher is in the
results. In tables, researchers have ways to present the data that tells the reader the coni-
dence of the indings. However, there has not been any system developed to do the same
thing with maps.

121
Summary
Although a map must by its very nature distort reality since the 3-dimensional world is
portrayed on a 2-dimensional surface, the researcher needs to be attentive to the construc-
tion of the map. With the availability of user-friendly mapping software, cartography is no
longer left to those with a background in this science. Although there are limitations in the
use of maps, there are also advantages. Mapping is becoming more and more popular in the
visual display of data with the goal of universal understanding.

Concept Reinforcement
1. Why do you think a map seems to be more believable than written results.

2. What educational experiences support the use of maps to convey knowledge.

3. For the most part, the data shown on a map is taken as being 100% true; however,
this is rarely the case. In data analysis, there is always so degree of uncertainty.
How would you portray uncertainty on a map?

122
 Chapter 26 - Chloropleth Maps
Chapter Objectives
• Deine the features of a chloropleth map

• Deine the ways to categorize data

• Explain the assumptions made on color

Maps are powerful tools used to communicate information. There are many types of maps
that an epidemiologist can use. One of the most common types of map is, choropleth maps.
Choropleth maps are based on predeined areal units, such as states, counties, census tracts,
etc. These areas are illed with color that symbolizes ranges in values from the data.

Obesity trends by state in 1990, 1998 and 2007

The obesity map is a choropleth map. It uses states as the areal unit. BMI means body mass
index. It is the calculation that uses height and weight to classify people as underweight,
normal, overweight and obese.

Body mass index calculation

123
Constructing a Choropleth Map
Three fundamental decisions need to be made when constructing a map. These are deciding
about areal units, color selection and data classing. Data classing means how to organize
the data into categories.

Areal Unit Selection

Depending on the data, the areal unit may be predetermined. For example, if the researcher
has data on county level smoking rates, then these data can be displayed at the county
level or shown at a regional or state level. The data will determine the scale used to map
these data. Data can always be aggregated up but not the other way. For example, if the
researcher has county level data, these data cannot be displayed at the town level because
all the cities and towns in the county will have the same value.

Color Selection
For sequential values, the general rule of thumb is to order the color of the data from light to
dark to parallel the ordering of the low to high values. The other type of data is bifurcated.
This is also known as diverging data. It may have an obvious structure such as positive and
negative numbers with zero as the middle value. It is common for high values to be shown
in red and low values to be shown in blue. A less common color scheme in epidemiology
is qualitative. Qualitative colors use a rainbow scheme that purposely has no pattern. The
range of qualitative colors is useful for categories in which the data are descriptive. For
example, qualitative colors could be used to show means of getting to school by neighbor-
hoods in a city (e.g., walk/bike, automobile, public bus, commuter rail/train). Dr. Cynthia
Brewer developed color schemes for sequential, diverging and qualitative categories

Dr. Brewer has provided color schemes for sequential, diverging and qualitative categories.
These are available on her web site.

Color Brewer site, www.colorbrewer.org/

124
When choosing colors for use in maps, it is very important to consider the impact of the
colors on the reader of the map. Some key considerations are whether certain intensities
cause the reader to place higher or lower importance on the data represented by that color,
as well as maintaining an awareness that color blind people’s limitations.

In what media the map will be viewed must also be taken into consideration when choos-
ing colors. If it is anticipated that the map will be copied using black and white print, some
color scheme maintain their differentiation in values while others do not. Similarly, atten-
tion needs to be paid to color choice when the map is projected (LCD use), color printed,
or displayed on a laptop.

Color preferences vary by culture and other characteristics. However, little is known about
the effects of preference on map interpretation. Even without support from research stud-
ies, map designers can use color in an obvious or clever way to reinforce the map message.
For example, adding a gold outline to a dollar sign can reinforce the idea of wealth.

Categories Selection
Another basic decision when constructing a map is how to break up the data into catego-
ries. Categorizing the data differently can tell very different stories. There is no right way
to categorize the data. However, a perceptive reader will pay close attention to this detail
when looking at the map.

Common ways to class data are as follows:

Quantiles: This is putting equal number of areal units into each class. For example, there
are 72 counties in Wisconsin. If six categories are chosen, the quantiles will each have 12
counties.

Equal interval: This breaks up the data ranges into equal segments. In other words, the
entire range of data values from minimum to maximum value is divided equally into how-
ever many categories have been chosen. In this situation, the number of counties could
vary tremendously between classes since the data values determine in which category the
county will be placed.

In a hypothetical map of percent land in mixed zoning, the quantile and equal interval
breaks distribute the data quite differently. Mixed zoning is neighborhoods with two or
more of the following types of land use: industry, commerce, residential, resource pro-
tected, and municipality.

Quantiles Equal Interval

Towns Ranges Towns Ranges
Class 1 A-M 0-13 A-F 1-2
Class 2 N-U 14-28 G-M 3-10
Class 3 V 29-42 N-S 15-25
Class 4 W-Z 43-56 T-Z 25-56

125
 Percent mixed zoning in 26 towns

60

50

40

Percent mixed zoning

30

20

10

0
ry

a
e

Ed id

Q r

y
t
e
y

on

a
d

G k

W r
an
a

en
Li g

g
hn

am
n
s

m
er
te

o
ar
Bo

ra
or
rli

an
ar
am

rg

Id

br
un
ol
n
av

ar

To
Sa

ct
ni
ce

Pe

ob
ue
Jo

Ki
ha

X-
N
nc
w

eo

illi

Ze
Fr

Vi

Yo
D

U
Ad

R
C

Place names

26 hypothetical towns showing percent of land with mixed zoning

Natural breaks (also known as Jenks breaks) minimize the variation within the classes and
maximize the variation between classes. This system is close to the quantile system but
adds some mathematical calculation to improve the it.

Below are three maps of exactly the same data using the same color scheme: mortality rate
by state for brain and nervous system cancer in white males, 1970 to 1994, all ages. Notice
how the story changes for Idaho with the different classing system.

Mortality rate by state for brain and nervous system cancer in white males,
1970 to 1994, all ages

126
Summary
Chloropleth maps are a speciic type of map used by epidemiologists to display ranges of
values using color within speciied areal units. Chloropleth maps contain standard map
components, as well as deinitions of the areal units and the color schemes used to repre-
sent the ranges of values. The ranges of values are classed using different schemes, each of
which is likely to produce a different looking map. In general, color is used from light to
dark based on lowest to highest values for the data. It is important to be careful, however,
to make sure the intensity of the color will communicate well to the reader the importance
of the parts of the map containing that color. It is also important to keep in mind the limita-
tions some people have in seeing color, such as those who are color blind.

Concept Reinforcement
1. What does the color red mean to you? In other words, do you think it makes more
sense to show high values or low values using the color red? How about blue?

2. Does it worry you at all to know that the way the categories are divided can actu-
ally change the look of a map? What could an epidemiologist do to present the data
in the unbiased way?

3. What is it about mapping that makes it so appealing to use in displaying results?

127
128
 Chapter 27 - Making a Map – Elements
Chapter Objectives
• List three features of a map

• Describe the difference in vector versus pixel data

Epidemiologists are not cartographers, people who specialize in map making. However,
map making is becoming more and more common in the world of epidemiology. Of course,
John Snow’s cholera map is shown in practically every introductory epidemiology book.

Soho area of London showing cholera deaths in 1854

Another series of maps that has received wide appeal are the obesity maps produced by
Centers of Disease Control and Prevention (CDC).

Obesity trends by state in 1990, 1998 and 2007

129
All maps are made of three components: points, lines and polygons. Mixing these com-
ponents in sensible ways forms the bases of creating a well designed map. The purpose of
a well designed map is to show the reader what is the most important piece of informa-
tion. The epidemiologist thinks about the elements of the map to be noticed irst and what
elements to be remembered. Map makers think about this visual hierarchy or ordering of
importance. Maps are designed so that some parts appear to be background material and
some parts are prominent, in the foreground.

Components of a Map
The title is one of the most prominent elements in a map. It should stand alone in describing
the map meaning. In other words, a person should be able to read the title and understand
what the map is showing without having to read any additional descriptions in the text.

The legend tells the reader how the data are categorized. How the data are divided into
categories is extremely important since the meaning can change depending on how the data
are displayed.

In epidemiology, a direction arrow should be included if it is not obvious. North is assumed

to be the top of the page and therefore if a map is oriented in the same direction, a direction
arrow is optional. In contrast, cartographers feel that a direction arrow is integral to a map.
Therefore, placement of the direction arrow is at the discretion of the map maker but prob-
ably should not be very prominent.

The scale bar is identical to the direction arrow. This tells the reader the linear distance, the
scale. This is a typical component of a map when the image is not common, though in epi-
demiology this piece of information is often omitted since it is rare that distance is the point
of the map. In contrast, cartographers considered this a necessary component of maps.

Once a map has been make with user friendly software, it is exported in an appropriate ile
format. Two general types of ile formats are raster and vector.

Raster Versus Vector Images

Raster images are composed of a grid of pixels. Each pixel has a speciic color value as-
signed to it. A pixel is usually very small so that a seemingly smooth image is created. If
this raster image is magniied, then the individual boxes in the grid can be seen. Conse-
quently, a raster image has an optimal viewing size.

A computer screen generally shows 72 pixels per inch and most printers work with 600-
2400 pixels per inch. The term ppi, pixels/inch, dots/inch, samples/inch, or spi all mean the
same thing.

In contrast, vector images use mathematical formulas to describe shapes. Because of this,
there is no optimal size of a vector image. The image can be printed or displayed at the
highest resolution of the device such as a computer screen or printer.

130
 Clip art image of a man

Blowup of clip art image of man: the arrow is vector and the man’s shirt is raster

On a map, these two types of images are displayed differently depending on the map com-
ponent of a point, line or polygon.

Comparison of the vector and raster data

131
Advantages of Raster and Vector Images
Advantages of vector images include the resolution of the image is independent of the size.
The outline of an image is also smooth. Compared to raster images, the ile size is smaller.

Advantages of raster images is that this in the only format that will show subtle gradients
of color and details. Raster images also allow color correction much easier than a vector
image.

Summary
In a presentation of study results, a map may only be one part of the presentation. In con-
trast, the map may tell the whole story. For example, the series of maps about obesity that
the Centers for Disease Control and Prevention (CDC) provide compelling evidence about
the United States obesity epidemic. One of the most important lessons a beginning map-
maker learns is to know the purpose of the map. In other words, what the map is supposed
to tell the viewer. Once that has been identiied, the smart epidemiologist uses the different
elements in a thoughtful way, confers with a cartographer, and uses resource books written
to help non-map making profession create a map. Even a design professional can assist in
the choice of colors to communicate the purpose of the map.

Concept Reinforcement
1. What choices were made in the obesity maps with regard to basic components of a
map?

2. Do you think the obesity maps would be as compelling if counties were used as the
areal unit instead of states? (There are on average 62 counties per state with Dela-
ware only having 3 and Texas having 254.)

3. Do you think a direction arrow or a scale should have been included on the obesity
map? Explain

132
 Chapter 28 - Use of Tables in Reporting Results
Chapter Objectives
• State ive standard features of a table.

• Describe the purpose of a table.

• Describe the elements necessary for clear table heading and title

Epidemiologists collect data, analyze it, and present it. From all that work, knowledge may
come. Tables are used to present data. A well-constructed table communicates to the reader
the essence of the study. In medium to large studies, the irst table in an article is a charac-
teristics table. Information about the participants are summarized so that readers can ind a
description of the participants at a glance. Additional tables typically present study results
or other details that are best shown in a tabular form.

Purpose of a Table
The general purpose of a table is to include important data beyond what can be simply re-
ported in the text. The maximum size of a table should be no more than a single page, held Unambiguous
as when reading the text. Because the maximum number of lines is signiicantly greater Clear. In terms of
than the maximum number of columns, the data are displayed such that the greatest number data, this means
of items is naturally assigned to the columns. The order of data in the table has considerable there is a clear result.
lexibility. Ordering is typically alphabetical, chronological, geographical or according to
magnitude of the items. The most conspicuous position in a table is top and left.

Rules of humb for Displaying Data

A table should tell us what the numbers represent and how the numbers were obtained or
calculated. In other words, the table should be an eficient display of meaningful and un-
ambiguous data.

The following table is adapted from a paper written by Jane McElroy and colleagues,
called “Potential Exposure to PCBs, DDT, and PBDEs from Sport-Caught Fish Consump-
tion in Relation to Breast Cancer Risk in Wisconsin. This paper was published in the jour-
nal, Environmental Health Perspectives in 2004.

133
 Table 1. Characteristics of Women with Breast Cancer and Controls
Aged 20-69 years
Cases Controls
Characteristic (N=1456) (N=1276)
N % N %
Age at reference date (years)
< 40 104 7.1% 102 8.0%
40-49 389 26.7% 343 26.9%
50-54 263 18.1% 196 15.4%
55-59 255 17.5% 210 16.5%
60-64 240 16.5% 204 16.0%
65-69 205 14.1% 221 17.3%
Family history
None 1142 78.4% 1100 86.2%
Positive‡ 304 20.9% 168 13.2%
Unknown 10 0.7% 8 0.6%
Recent alcohol consumption
None 243 16.7% 219 17.2%
1-6 drinks/week 1022 70.2% 931 73.0%
7 or more drinks/week‡ 189 13.0% 126 9.9%
Parity
0-1 342 23.5% 276 21.7%
2 467 32.1% 373 29.2%
3 or more‡ 646 44.4% 626 49.1%
Age at First Birth (years old)
< 20 236 16.2% 243 19.0%
20-24 613 42.1% 581 45.5%
25-29‡ 297 20.4% 207 16.2%
≥30‡ 131 9.0% 95 7.5%
Nulliparous 179 12.3% 150 11.8%
Age at menopause (years old)†
<45 192 23.2% 205 27.6%
45-49 187 22.6% 170 22.9%
50-54 233 28.2% 219 29.5%
55-69‡ 95 11.5% 65 8.8%
Unknown 119 14.3% 84 11.3%
Education
< High School diploma 75 5.2% 83 6.5%
High School diploma 621 42.7% 534 41.9%
Some College 388 26.7% 356 27.9%
College degree 366 25.1% 300 23.5%
Unknown 6 0.4% 3 0.2%
* Adjusted for age.
† Among postmenopausal women only.
‡ Statistically significant

Researchers generally display data in a similar fashion.

There are several rules of thumb for displaying data in a table.

In constructing a table, the names of the columns and rows should be clear and short. In
addition, subordinate information should be indented. In the example table above, all the
characteristics are listed at the far left edge of the table. The categories of the characteristics
are all indented to spaces.

134
One of the irst overarching rules of thumb is that the table should be understandable by
itself. In other words, it can stand alone. In the example above, the title tells the reader that
information in this table is comparing two groups of women—women with and without
breast cancer. Several demographics about these women are compared by stating the num-
ber and percent of women in each category.

Another rule of thumb is the data in a table are closely related to other data in the table.
The reader can look at the table and summarize the facts. In the example above, all the data
are about differences and similarities in lifestyle choices and risk factors associated with
breast cancer.

Next, column and row headings should be brief, unambiguous, and self-explanatory. Table
footnotes should be used only to enhance the clarity of the headings. The symbol † means
among postmenopausal women only. This makes sense for the characteristic describing
age at menopause. Although the row heading should be clear, in this case, “age of meno-
pause” meaning that it could only apply to women who have gone through menopause, the
symbol linked to the footnote adds clarity.

Any variation in the row or column width should be related to ease of reading and logical.
In some cases, it is easier to read two lines of text and in come cases it is easier to have one
line of text in a wide column. In the example, two lines of text are needed for the title of the
table. That makes sense since none of the characteristics’ names are very long.

The inal rule of thumb is that units need to be included in the table. Notice in the example
table that in all cases the units are provided. In this case, that means years or years old are
the units but units are whatever was reported.

Summary
Tables are a common method for epidemiologists to present data. Using tabular format of-
ten simpliies the data such that readers can understand the material. Several rules of thumb
have been developed in displaying data. These rules of thumb allow readers of epidemio-
logical research to understand complex data in a logical and clear manner.

Concept Reinforcement
1. What does a subordinate position of data tell the reader?

2. Characteristics tables are typically Table 1 in most epidemiological research ar-

ticles. Why is that?

3. How many lines would it take to write out the data presented in the example?
Based on that guess, what is another advantage of using tables that was not de-
scribed in the text.

135
136
 Chapter 29 - Graphs and Charts
Chapter Objectives
• List the types of graphs and charts

• Discuss the pros and cons of the different types of graphs and charts

• State the optimal choice for each type of chart

It is often easier to understand study results when these data are presented graphically
rather than in a narrative format. Graphs can also be easier to understand than tables. When
looking at most types of graphs make sure to pay attention to the horizontal (x-axis) and
vertical (y-axis) axis’ scale. Occasionally there will be a scale on both the left and right side
of the vertical axes.

137
Case Study Description
The following topic is used as an example for all but one of the charts shown in this lesson.
Asthma is a chronic respiratory disease that obstructs the airway due to inlammation of the
bronchial tubes. This inlammation causes the bronchial tubes to swell. The swelling nar-
rows the airway and makes breathing dificult. An asthma attack can lead to death. Asthma
is the leading cause of school absenteeism from chronic conditions. It is also the leading
cause of hospitalizations for children under the age of 15 in the US. For those with Medic-
aid, approximately 25% of all children with asthma will make at least one ER visit a year.
The New York City Department of Health and Mental Hygiene’s Childhood Asthma Initia-
tive is a public health effort to reduce asthma morbidity among children 0-18 years of age.

Charts (also known as diagrams)

The graphical chart provides a visual display of data that would normally be represented
in a table format. Ideally, a chart displays the result in a more reader-friendly format than a
table or text such that the idea is more easily conveyed. Presenting data in the format of a
chart should allow the reader to quickly grasp what the data mean.

A chart has 3 basic components: labeling that deines the data (title, axis labels, legend),
scale (x and y axis range), graphical elements (bars, lines, points).

Bar Chart (Histogram)

Bar charts are one of the most popular types of charts. This chart often displays relation
between one or more categorical variables with one or more quantitative variables repre-
sented by the length of the bars. The categories are usually on the x-axis (horizontal axis).
More than one set of data mandates the use of a legend.

Comparison of asthma hospitalization rates in children aged 0-14, NYC, 1997-2000

138
In this example, not only are bars used to show the relative magnitude of the hospitalization
rates by year, but the color scheme is graduated from dark (highest rate) to light (lowest
rate).

It is recommended that no more than seven numbers are included on the x-axis scale. For
fewer than ive bars, consider using data labels of each bar instead of y-axis scale. This is
shown above. If more than 8-10 categories, use a “rotated bar chart” which means the cat-
egories are displayed on the y-axis and the values are on the x-axis. This is shown below.
Ideally, the data should be sorted on signiicant variable. (This was not done in the example
below.)

Asthma hospitalization rates by neighborhood in NYC

The legend should be placed either inside the plot area or horizontally under the chart title,
not outside of the plot area. This minimizes the size of the area given to displaying these
data.

Stacked bar charts have limited use and are rarely an effective display of data. These work
best when showing primary comparisons across series among categories.

139
 Total charges for asthma hospitalization by payer and age, NYC, 2000

Although it seems to be visually interesting, rarely is a 3-d bar chart the best choice in un-
ambiguously displaying data and therefore should be avoided.

Line Chart
Line charts should only be used when a period of time is involved. This type of chart is ide-
al for showing trends. The time dimension is almost always shown on the x-axis from left
to right. Be careful of scaling effect when more than one variable is presented on a chart.
Usually the variable with the larger scale will appear to have a greater degree of variation
and the smaller scale when appear more “lat,” even though the percent change may be the
same. When several time series are presented in the same chart using black and white print,
mixing solid, dotted and dashed lines for each variable can help distinguish between them.

Trends in asthma hospitalization rates, US Northeast US, NYC

140
Pie Chart
Of all types of charts, pie charts should be avoided since they lack clarity. In addition, 3-d
pie charts are worse than 2-d pie charts because adding the “depth” dimension adds no
information. It seems that the 3-d pie chart is used for aesthetic purposes not to accurately
and eficiently display information. If one insists on using a pie chart, then these should
only be used for displaying proportions and only if the data add up to some meaningful
total.

Percent distribution of asthma hospitalizations by age group, NYC, 2000

Boxplots
Boxplots display distribution and interval-level variables. The median and four quartiles
are displayed. Boxplots are best used for comparing the distribution of the same variable
for two or more groups or two or more time points. Boxplots are also effective in display-
ing a single case compared to a large number of other cases.

Figure. Concurrent hourly arithmetric mean BC (black carbon) concentration (µgm-3)

measured at three South Bronx sites and Hunter College (HC). Line in the box is a me-
dian, box size 25th -75th percentile. Circles are outliers (measurements outside the 10th
and 90th percentile).

141
 Parts of a box plot

Summary
Epidemiology is rich in data. Displaying these data in such a manner that allows for easy
understanding is a skill that is honed in epidemiology. The use of tables and charts is one of
the most common means by which epidemiologist report results since these graphics allow
for a considerable amount of data to be shared in an organized manner. By describing the
data through graphics, a clear presentation of data is likely to result. A clear presentation
improves communication of important research indings.

Concept Reinforcement
1. Explain the difference between a histogram and a line chart.

2. Discuss why pie charts are usually avoided when presenting epidemiologic data.

3. List the components of a box plot and describe when box plots are useful in pre-
senting information

142
 Chapter 30 - Person, Time, and Place
Chapter Objectives
• Describe the component of person

• Describe the component of time

• Describe the component of place

There are three primary components of a descriptive epidemiologic study: person, time and
place. These are essential when documenting an epidemiologic outbreak and monitoring
the spread of a disease. Person refers to the individuals of interest. These may be people
who have an exposure, develop a disease, or are otherwise of interest to the epidemiolo-
gist. Each person who develops a disease or has an exposure is a case. Time refers to the
time frame over which the disease or illness occur. Place is the location of the participants.
Depending on the purpose of a study, all three of these pieces of information may need to
be collected.

Components of Person
Of the three fundamental components of descriptive epidemiology, describing participants
has received the most attention. Below are a few common characteristics.

Age
Several characteristics are important to collect about participants. Age is important. The
human biological clock phenomenon means that as a person ages s/he becomes more vul-
nerable to illness. Lifestyle also inluences some chronic diseases, such as Type II diabetes.

The United States is graying. This means the number of people over 65 years old is in-
creasing. About 35 million Americans are 65 years old or more in mid 2000s. By 2030, the
number is projected to double. With this phenomena, more epidemiological studies will be
needed on the topic of “retirement syndrome,” the bereavement process, and other health
related issues among older adults.

Sex and Race

Other characteristics are sex and race. Predictors of diseases may vary by both of these
characteristics. Further, lifestyle choices may also be different according to sex or race.
Therefore, information about these two characteristics is important to collect.

143
Marital Status
Marital status is generally categorized as single or never married, married or living with a
partner, separated, divorced and widowed. The married category has changed over time. In
the 1950s, the term “living with a partner” would not have appeared on a survey since there
was social sigma associated with this situation. However, these days that social stigma is
signiicantly reduced and about 11 million people reported that they were living with an
unmarried partner in the U.S. Census.

Epidemiological studies have shown that married couples have lower morbidity and mor-
tality, especially men. People who have never married are less likely to be overweight
whereas being married is linked to obesity, especially for men.

What is it about being married that confers protection against some diseases? One protec-
tive hypothesis is that healthier people are more likely to get married and stay married.
Another protective hypothesis is that the state of marriage makes a positive contribution to
health, including lifestyle choices or inancial wellbeing.

Socioeconomic Status
The inancial well being of people has been noted since the beginning of time. It is well
known that wealth is associated with health. Impoverished people have higher death rates
than wealthy people. The advantage of social class is graduated so that each change in
social class confer improved health. These statistics are at the group level. This means that
with any class of people there will be some who die young or become severely ill. How-
ever, as a group the patterns holds regarding increased social position linked with better
health.

Components of Place
Another component of descriptive epidemiology is gathering data about place. It is well
known that geographical variation is seen on the global scale.

Among Counties Variation

Both chronic diseases such as cancer and heart disease and infectious diseases such as
malaria and cholera vary by county. Life expectancy relects geographic variations in the
illnesses and diseases faced by each county.

144
 Comparison of life expectancy in 2000 and 2025 by country

Within Country Geographic Variation

Within the United States, there is considerable geographic variation in climate, environ-
mental pollution, and clustering of different types of people.

Urban and Rural Diferences

Living in the city or country is associated with different exposures. Obviously, city people
live in more crowded conditions where the spread of infectious diseases occurs more read-
ily. People living in agricultural rural areas are more likely to be exposed to pesticides.

Regional Variations
Patterns of illness or diseases can also be observed in larger regions of the country. For
example AIDS rates are higher in the Northeast than other regions of the United States and
this pattern has remained constant over time. The south leads the county in obesity. More
people in the south are obese than in other regions of the county.

145
 Components of Time
The inal component of descriptive epidemiology is gathering data about time. The time
component associated with a disease is known as temporal aspect of disease. In some cases
examining the time dimension can help understand the disease process.

Cyclic Fluctuations
Many cases of seasonal trends in health conditions are observed. For example birth rates
Fluctuation increase in early summer. Feeling depressed also shows an increase in early summer. Cy-
An increase or de- clic luctuation can occur within a year as shown with birth data and depressive symptoms.
crease in the number It can also occur within decades. For example, pneumonia-inluenza shows periodic epi-
of cases over time. demics every few years.

Many disease that show cyclic lucutations related to changes in lifestyle behaviors. For
example, an increase in Malaria can not occur in winter when the mosquito cannot live.

Secular Time Trends

Secular time trends means that changes in the frequency of a disease is gradual and takes a
long period of time. For example, lung cancer has increased in women with the change in
smoking behavior of women. In the mid 1990s few women smoked. With the increase in
smoking behavior by women in the late 1990s, the rate of lung cancer has increased.

Summary
Person, place and time are important characteristics of the participants. Depending on the
epidemiological study, the importance of each of these may vary. Sometimes, knowing
the location of a person does not add any useful information to the study. The same is also
true for the time component and some of the characteristics of the person. Epidemiologists
always give thoughtful consideration of each of these three characteristics when designing
their studies.

Concept Reinforcements
1. What could a high rate of a disease in a geographic area suggest to the researcher?

2. What do you think is the most important characteristic of a country related to life
expectancy ?

3. Do you think there is any disease or health condition that shows no international
variation? If no, explain; if yes, which one and explain.

146
 Chapter  - Case-control Study Design
Chapter Objectives
• Explain the concept of counterfactual

• Compare the three types of causality

• Describe the Bradford Hill Criteria

Epidemiology is the study population health. Epidemiological researchers use various

measurements determine exposures. A key concept in determining exposures is causality.

Crossover
Determining Causality Design
Epidemiologic researchers prefer to be able to determine the cause of the illness, disease, Participants assume
injury, or health outcome. However, it is not possible to determine the cause with 100% both the treatment
conidence for a couple of basic reasons. One reason is that it is unethical to expose people and control experi-
to a known health risk to study the effects of that exposure. In the current political climate, ence.
extraordinary efforts have been instituted to protect study participants. The second reason
it is dificult to determine cause that humans are biologically complicated creatures unique
individual experiences. Because of the variety of experience and the level of biological Participants are
complexity, teasing out the cause-effect path is daunting, to say the least. randomly assigned
to one or the other
No two people are alike. Even if a researcher takes into account as many experiences and group, irst.
exposures as humanly possible, there will still be an uncountable number of experiences
and exposures that differ between any two people. Dr. Kenneth Rothman uses the counter-
The subjects act as
factual ideal to explain the dificulty of determining causal relationships. In an ideal com-
their own controls.
parison, a person will be compared to him or herself. In one state, the person is not exposed
and in the other state the person is exposed to whatever is begin tested. If this could hap-
pen, researchers would be able to determine the effect of exposure because the exposure
would really be the only difference between the two situations. However, we do not have Counterfactual
two identities or parallel existences. Consequently, this situation is called counterfactual.
Contrary to the facts.
Another facet of the counterfactual situation described above is the necessity of having
the exposed and unexposed experiences happen simultaneously so that a time dimension
A condition which
does not inluence the study results. The closest that an epidemiologic study comes to this
cannot be fulilled.
possibility is the crossover design. In this design, the participant is randomly placed in ei-
ther the treatment or control group. After a certain amount of time, the participants switch
groups. In other words, the treatment groups become the control group and visa versa. This
design comes close to achieving a counterfactual comparison. However, the design falls
short because a person can only be in one study group at a time. The time component may
affect the study results since the passage of time may introduce different factors other than
the treatment.

147
To reiterate, in a counterfactual study, each exposed participant would be compared to his/
her unexposed self while facing simultaneous experiences. The number of exposed partici-
pants who develop the disease would be compared to the number of unexposed participants
who develop the disease. The difference would indicate the effect of the exposure. For ex-
ample, in a group of 100, let’s say we found 1 in 4 (25% exposed participants developed
the disease and of this same counterfactual group of 100 people, 1 in 10 (10% unexposed
participants developed the disease. The causal effect of the exposure would be 15% (25%–
10% = 15%. The causal effects of the exposure is the percent of people who were exposed
to the variable and who developed the disease minus the group of people who were NOT
exposed to the variable and developed the disease. The difference is disease incidence be-
tween these two groups is the causal effect.

Since the ideal of a counterfactual experience cannot occur, epidemiologists carefully de-
sign their studies to minimize the differences between the comparison groups. The goal is
to ind an unexposed population that relects a counterfactual situation as closely as pos-
sible. Suppose we use the same examples as above and, instead of having a counterfactual
group, we compare our exposed group to another group of people who were unexposed.
The gold standard for researchers is for the unexposed group to simulate all the important
experiences of the exposed group. Using the same disease rates as above, a lack of a true
counterfactual experience precludes the researcher from being 100% sure that the 15%
difference between the two groups is absolutely due to the exposure. The solution that
epidemiologist use as guidelines to determining a causal link are the Bradford-Hill
criteria, which we will discuss later in this chapter.

Causal Relation
Let’s look a little closer at the task of identifying causal relation in scientiic research. Cau-
sality can be categorized into three fundamental types in order to decreasing strength or
suficiency as a cause. They are 1 a suficient cause, 2 a necessary cause, and 3 a risk
factor.

Suficient cause is rarely observed in epidemiology. The factor must be present before the
disease and it always predicts the occurrence of the disease. Genetic abnormalities are the
best example of this type of causality.

A necessary cause often involves an external exposure. This factor must be present for the
disease to develop but not everyone who is exposed to this factor will develop the disease
or illness. Infectious diseases are good examples of necessary cause. For example, expo-
sure to the tubercle bacillus may result in the person developing tuberculosis. However,
some people such as Mary Mallon, also known as Typhoid Mary, are carriers of the disease
but do not suffer any symptoms.

148
 Mary Mallon, also known as Typhoid Mary, carried the disease typhoid and
infected people even though she did not get sick from the disease.

The weakest causal type is a risk factor. When the factor is present, the probability of
the disease increases. A risk factor is neither a necessary or suficient cause of the disease.
Smoking is a classic risk factor for lung cancer. It is not a suficient cause since some peo-
ple who never smoke are diagnosed with lung cancer. In other words, smoking does NOT
always predict the occurrence of lung cancer therefore it cannot be classiied as a suficient
or necessary causal agent. However, smoking is considered a very strong risk factor for
lung cancer. For heavy male smokers, the probability of developing lung cancer is 20 times
higher compared to non-smoking men.

20-Year Lag Time Between Smoking and Lung Cancer

Cigarettes Lung
Smoked Cancer
Per Person Deaths
Per Year (Per
100,000
4000 People)
Cigarette
Consumption 150
(men) Lung
3000 Cancer
(men)
100

2000

50

1000

1900 1920 1940 1960 1980

Year

Correlation between smoking and lung cancer in US males, showing a 20-year time lag
between increased smoking rates and increased incidence of lung cancer

Image courtesy of the National Institutes of Health

149
The debate about whether or not smoking causes lung cancer can be considered in light
of these deinitions. Often the tobacco company’s spokesperson will use the deinition
of suficient or necessary cause when speaking about the lack of evidence that smoking
cigarettes causes lung cancer. In contrast, the majority of researchers feel conident saying
smoking causes lung cancer, even though it is the weakest type of causality.

Risk
The word risk is commonly used in epidemiology and is often applied to individuals. Risk
is interchangeable with the term probability. The risk of an individual being diagnosed with
an illness is the same as the probability of an individual being diagnosed with an illness.
However, except for the on-lines tools that allow individuals to determine his/her particular
risk based on certain medical histories and lifestyle behaviors, it is common to study the
risk of disease at the population level. In this case, the proportion of people who developed
the disease during a speciied period of time represents the risk of the disease in the popula-
tion during the speciied time frame.

The risk of The number of individuals who were diagnosed with the disease
a speciied during a speciied time period divided by
population
is equal to The total number of individuals in the speciic population

Before continuing, let’s review basic math terminology for fractions. The top number is
known as the numerator and the bottom number is known as the denominator.

Typically, iguring out the numerator in a study is quite straightforward. However, the
denominator is much trickier and the use of the wrong population in the denominator will
result in useless study results. Keep this in mind as you are designing epidemiological ex-
periments and analyzing the data you collect.

he Bradford Hill Criteria

The public wants to know “what caused my disease.” To help answer that question from
epidemiologic study results, in 1965 Sir Austin Bradford Hill described the nine conditions
needed to establish a causal relationship.

Nine Conditions Required to Establish a Causal Relationship:

• Strength of association—the relationship must be clear as deined by the size of the
association measured by appropriate statistical tests. For example, the more highly
correlated hypertension is with a high sodium diet, the stronger is the relation be-
tween sodium and hypertension.

150
 • Consistency—observation of the association must be repeatable in
different populations at different times and in different settings. Protection of human
subjects’ deinition
• Temporality—the exposure must always precede the outcome (disease) Human subject means a living individual
in time. about whom an investigator (whether profes-
sional or student) conducting research obtains
• Plausibility—the explanation must make sense biologically. However, (1) data through intervention or interaction
with the individual, or (2) identiiable private
studies that disagree with established understanding of biological pro- information.
cesses may force a re-evaluation of accepted beliefs.

• Dose-response relationship—increasing amount of exposure increases Intervention includes both physical pro-
the risk. cedures by which data are gathered (for
example, a blood draw) and manipulations of
the subject or the subject’s environment that
• Speciicity –a single cause produces a speciic effect. This is considered are performed for research purposes.
to be the weakest criteria. Absence of speciicity in no way negates a
causal relationship.
Interaction includes communication or
• Coherence – explanation should coincide with natural history and biol- interpersonal contact between investigator
and subject.
ogy of the disease.

• Experiment– any related research that is based on experiments. Private information includes information
about behavior that occurs in a context in
• Analogy – sometimes a commonly accepted phenomenon in one area which an individual can reasonably expect
can be applied to another area. that no observation or recording is taking
place, and information which has been pro-
vided for speciic purposes by an individual
and which the individual can reasonably
Concept Extension: Institutional Review Boards expect will not be made public (for example,
a medical record). Private information must
The Research Act of 1974 deined Institutional Review Boards (IRBs). IRBs be individually identiiable (i.e. the identity
of the subject is obvious or may readily be
are independent ethics committees that have been formally designated to ap- ascertained by the investigator or associated
prove, monitor, and review biomedical and behavioral research involving hu- with the information) in order for obtaining
mans. The purpose of IRBs is to protect the rights and welfare of the study the information to constitute research involv-
ing human subjects.
participants. In the United States, the U.S. Food and Drug Administration and
the Department of Health and Human Services’ regulations have empowered
IRBs to approve, require modiications, and/or disapprove research. The rea- Research means a systematic investigation,
including research development, testing and
son that IRBs were developed was because of research abuses earlier in the evaluation, designed to develop or contrib-
20th century. ute to generalizable knowledge. Activities
that meet this deinition constitute research
One well-known case of research abuse occurred during WWII in which Nazi for purposes of this policy, whether or not
they are conducted or supported under a
physicians performed experiments on prisoners of war. Some of these abuses program that is considered research for other
were exposed during the Nuremberg Trials. purposes (for example, some demonstration
and service programs may include research
In the United States, the Tuskegee Syphilis Study was an unethical and sci- activities).

entiically unjustiiable project. In 1932, the Public Health Service and the
Tuskegee Institute began a project to study the natural course of syphilis. The Minimal risk means that the probability and
study enrolled poor, illiterate black men from rural Alabama (399 with syphi- magnitude of harm or discomfort anticipated
in the research are not greater in and of
lis and 201 without syphilis). There was no participant’s informed consent. In themselves than those ordinarily encountered
fact, the participants were deceived about the study. Those with syphilis did in daily life or during the performance of
not receive proper treatment when penicillin became available and the ac- routine physical or psychological examina-
tions or tests.
cepted treatment.

151
 Examination, Tuskegee study

Image courtesy of the National Archives and Records Administration, US Government

The IRB (institutional review board) at almost all institutions, such as universities and
hospitals that engage in human research, review research projects that involve humans to
ensure that ethical standards outlined by the Federal Policy for the Protection of Human
Subjects (45 CFR (Code of Federal Regulations) 46) are adhered to in conducting research
using human subjects.

Code of Federal Regulations, seen at the Mid-Manhattan Library. Editions of Title 3,

on the President, are kept on archive. Notice that for the irst year of each new
presidency, the volume is thicker.

152
Code of Federal Regulations, Title 45: Public Welfare, Department of Health and
Human Services, Part 46: Protection of Human Subjects (45 CFR 46)
Who must follow this policy?

Any research conducted or supported by any federal agency or department. There are also
some other types of research that must comply but we won’t go into the details of those.

46.116: General requirements for informed consent

No investigator may involve a human being as a subject in research covered by this policy
unless the investigator has obtained the legally effective informed consent of the subject or
the subject’s legally authorized representative.

46.116a: Basic elements mandated to be included in the consent form

(1) A statement that the study involves research, an explanation of the purposes of the re-
search and the expected duration of the subject’s participation, a description of the pro-
cedures to be followed, and identiication of any procedures which are experimental;

(2) A description of any reasonably foreseeable risks or discomforts to the subject;

(3) A description of any beneits to the subject or to others which may reasonably be ex-
pected from the research;

(4) A disclosure of appropriate alternative procedures or courses of treatment, if any, that

might be advantageous to the subject;

(5) A statement describing the extent, if any, to which conidentiality of records identifying
the subject will be maintained;

(6) For research involving more than minimal risk, an explanation as to whether any com-
pensation and an explanation as to whether any medical treatments are available if inju-
ry occurs and, if so, what they consist of, or where further information may be obtained;

(7) An explanation of whom to contact for answers to pertinent questions about the re-
search and research subjects’ rights, and whom to contact in the event of a research-
related injury to the subject; and

(8) A statement that participation is voluntary, refusal to participate will involve no penalty
or loss of beneits to which the subject is otherwise entitled, and the subject may discon-
tinue participation at any time without penalty or loss of beneits to which the subject
is otherwise entitled.

153
Who is exempt from having to obtain IRB approval? 46.101b

1) Research conducted in established or commonly accepted educational settings, involv-

ing normal educational practices, such as (i) research on regular and special education
instructional strategies, or (ii) research on the effectiveness of or the comparison among
instructional techniques, curricula, or classroom management methods.

(2) Research involving the use of educational tests (cognitive, diagnostic, aptitude, achieve-
ment), survey procedures, interview procedures or observation of public behavior, un-
less: (i) information obtained is recorded in such a manner that human subjects can be
identiied, directly or through identiiers linked to the subjects; and (ii) any disclosure
of the human subjects’ responses outside the research could reasonably place the sub-
jects at risk of criminal or civil liability or be damaging to the subjects’ inancial stand-
ing, employability, or reputation.

(3) Research involving the use of educational tests (cognitive, diagnostic, aptitude, achieve-
ment), survey procedures, interview procedures, or observation of public behavior that
is not exempt under paragraph (b)(2) of this section, if: (i) the human subjects are elect-
ed or appointed public oficials or candidates for public ofice; or (ii) federal statute(s)
require(s) without exception that the conidentiality of the personally identiiable infor-
mation will be maintained throughout the research and thereafter.

(4) Research involving the collection or study of existing data, documents, records, patho-
logical specimens, or diagnostic specimens, if these sources are publicly available or if
the information is recorded by the investigator in such a manner that subjects cannot be
identiied, directly or through identiiers linked to the subjects.

(5) Research and demonstration projects which are conducted by or subject to the approval
of department or agency heads, and which are designed to study, evaluate, or otherwise
examine: (i) Public beneit or service programs; (ii) procedures for obtaining beneits
or services under those programs; (iii) possible changes in or alternatives to those pro-
grams or procedures; or (iv) possible changes in methods or levels of payment for ben-
eits or services under those programs.

(6) Taste and food quality evaluation and consumer acceptance studies, (i) if wholesome
foods without additives are consumed or (ii) if a food is consumed that contains a food
ingredient at or below the level and for a use found to be safe, or agricultural chemical
or environmental contaminant at or below the level found to be safe, by the Food and
Drug Administration or approved by the Environmental Protection Agency or the Food
Safety and Inspection Service of the U.S. Department of Agriculture.

Complying with all the elements in this policy is time consuming for researchers and it typ-
ically takes 2-6 months to obtain IRB approval from their institution. The IRB has dozens
of research projects to review at each meeting. If they have any questions about a particular
project, they generate a letter listing their questions, directives or concerns. The researcher
has to respond to every item in the letter and resubmit the application to the IRB. The appli-
cation is put into queue again. This back and forth exchange can happen numerous times.

154
Many institutional IRBs meet every two weeks. So the cycle on any formal communica-
tion is at best one month. Probably the biggest area of back and forth communication is the
elements of the consent letter and/or introductory letter. There is no template for this and
therefore every submission has the potential to generate new or different concerns from the
IRB, even for essentially identical studies.

A large literature exists that describes people’s perception of risk and how they commu-
nicate using probability terms. For example, people who have had a health problem are
more likely to rate it higher on the probability scale than someone without the experi-
ence. Events that are more common are more likely to be ranked higher than less common
events. Memorable experiences are more likely to rank higher on the probability scale than
less memorable experiences. Unfortunately, even though we can describe these and many
more differences, there is not a foolproof system we can use to predict how someone will
rank the probability so that we could institute a systemic correction to an individual’s im-
precise risk communication.

Summary
In a perfect world, the researcher would prefer to be able to igure out the causes of injury,
illness, and disease. However, they lack a counterfactual model and are left with less than
perfect tools to determine the causal factors in diseases. Generally, causal factors are those
that can be shown to increase the likelihood of a disease occurrence. Epidemiologists work
with three basic forms of causal relationships: necessary cause, suficient cause and risk
factor. In 1965, Sir Austin Bradford Hill described the nine conditions needed to establish
a causal relationship. These include strength of association, temporality, consistency, plau-
sibility, dose-response relationship, speciicity, coherence, experiment, and analogy.

Concept Reinforcement
1. Explain the concept of counterfactual.

2. List and describe the three types of causality.

3. List and describe the Bradford Hill criteria.

155
156
 Chapter 2 - Sampling Error and Chance
Chapter Objectives
• Compare random and nonrandom sampling

• Describe simple random sampling

• Deine systematic error and random error

Regardless of the source of the data, an important aspect of epidemiology is developing an

understanding of these data. In this chapter, we will explore the components of sampling.

Most health research collects data from individuals. Researchers are interested in speciic
details about individuals, but must also be careful about protecting the privacy of the peo-
ple in their studies. Oddly, the researcher is very interested in knowing lots of information Population
about an individual but is not interested in the identity of the individual. Except for medi-
ALL individuals
cal reports, called case study in which a unique medical situations is described and a case
in a population
series in which several examples of a unique medical situation is described, the individual
data is combined and group results are reported.

It is important to know the difference between two terms used in epidemiology: population Sample
and sample. A population means ALL individuals. The image below shows a school of ish.
The population includes all members of the school. A sample would include a subset of the A subset of
school of ish. For example, a sample of this school of ish might be ive or ten members a population
of the school.

School of yellow-in goatish (Mulloidichthysvanicolensis)

Image courtesy of the US Government

157
Simple Random Sample
This is the basic type of probability sampling plan. Every individual in the sampling frame
has an equal chance of being selected for the study. A common technique used to select
individuals from a list is to use a random digit generator. Each person is assigned a ran-
dom number. The list is sorted in numerical order using the random digit column. If the
researcher wants to approach 416 potential respondents from a list of 1,200 individuals, the
sample will consist of the irst 416 names on the list after it has been randomized.

Random and Systematic Error

In sampling, the error occurs by the very nature of participant selection, and depends on
who was and, more importantly, was not included in the study. Errors can be classiied as
either systematic or random. Systematic error is nonrandom error in which the values tend
to be inaccurate in a particular direction. For example, measuring participants’ heights with
their shoes on would be a systematic error. Everyone would be slightly taller than if their
height was determined without shoes. If participants were measured in bare feet, there
would still be random error. Random error means that some of these data would be too
high and some would be too low. Random error can occur by chance or lack of knowledge
of all relevant factors.

Chance
Many people believe in chance. Chance can play a fundamental role in both biological and
physical phenomena. On the other hand, some believe that any situation could be a predict-
ed if all the relevant factors were known. Therefore, we can deine random variation as ig-
norance of all the relevant factors that predict a disease. In the commonly used example of
tossing a coin, predicting heads or tails could be determined by applying the physical laws
associated with the toss. However, the complexity of the calculation and the lack of neces-
sary information available on the street (such as wind turbulence in the microenvironment)
in essence transform this predictable situation into chance or guesswork. Consequently, in
epidemiology variation is treated as random until it can be explained, whether the variation
is based upon chance or based on insuficient information about relevant factors. All data
sampling is subject to random error.

Dice is a game of chance. For example each time you throw the pair of dice,
you have only a certain chance of getting a pair of sixes.

158
Another important component of random variation in epidemiology studies is the process
of study subject selection, called sampling. The random variation is known as sampling er-
ror. In sampling, the error occurs by the very nature of participant selection, especially who
was and, more importantly, was not included in the study. Similarly, measurement error
can occur due to the random variation in potential biological experiences of the popula-
tion. Reducing random error can be accomplished in two ways: increasing sample size or
improving the information collected.

Lack of Systematic Error: External Validity

An important component of the study design revolves around the researcher’s ability to
generalize the indings from the sample of people studied to the appropriate population.
Being able to generalize a study to a larger population is called external validity.

Migraine study example

It was once thought by medical professionals that migraines were associated with intel-
ligence. The rationale was that people who used their brains more were more likely to get
headache. The parallel was made to those who exercise being more likely to have muscle
or joint aches. This theory was debunked in the 70s. It seems this fallacy arose because
migraine sufferers that the physicians saw did not represent the population of migraine
sufferers. Less intelligent migraine sufferers were much less likely to visit a doctor. This
systematic error, the sample studied was systematically more intelligent, led to a lack of
external validity in the early migraine studies.

Most studies describe the world in single group studies, for example a survey of students in
a school about their dietary habits and their current weight. A study can describe a unique
setting where the results only apply to that setting (e.g., students who eat lunch from the
salad bar at a school are more likely to be overweight). A study may also describe a typical
setting where the results could represent a more general phenomenon (e.g., students who
skip breakfast are more likely to be overweight). Choosing the correct sampling process is
critical for enhancing external validity.

Lack of Systematic Error: Internal Validity

Validity is the best available approximation of truth of a given population. Internal valid-
ity refers to the validity of inferences (cause-effect relation) in studies.Other way internal
validity is stated is whether the researcher is measuring what she intends to measure. There
are multiple threats to internal validity. One of the threats to internal validity is inconsis-
tency in the comparison groups. Participants bring to the research a myriad of character-
istics, including the willingness to participate. If the characteristics of the participants are
not similar between the groups, a dificulty arises for the researcher in determining if the
differences in the groups are the reason for the study result or has no inluence on the study
result.

159
Summary
In data collection when the researcher relies on the participant to provide the information,
a difference between fact and reported fact can occur. This is called error. If the error has
a pattern to it, it is called systematic error. If a survey has systematic error, then it does not
have external validity. External validity means that the results are not believable. Another
type of systematic error is internal validity which means the researcher is measurable what
is intended to be measured. Maintaining both internal and external validity are necessary
parts of a useful study.

Concept Reinforcement
1. Explain the difference between systematic error and random error.

2. In what way would having only volunteers in a study be a problem?

3. Is it possible to ever remove all errors from a study? Explain

160
 Chapter 3 - Sampling Frame
Chapter Objectives
• Deine a sampling frame

• Compare the use of lists

• Explain the denominator in the sampling frame

For probability sampling to happen, all individuals of the population of interest have a
known, non-zero chance of being selected for the sample. For this to happen, a list of these
individuals needs to be compiled. This is known as the sampling frame. There are four Sampling Frame
basic types of basic probability sampling plans: simple random sampling, systematic sam- A list of the indi-
pling, stratiied random sampling, and cluster sampling. viduals from which a
sample will be drawn
Probability
Probability theory tells us that the data gathered from a random sample can be generalized
to the population. The measure of this variation is called standard error. Imagine a study of
a big city with 10,000,000 inhabitants. Of the ten million inhabitants, 500,000 responded to
a survey about whether the city has enough green space. The study results showed that 75%
(± 3% which relects the standard error) of the inhabitants prefer the option of adding more
green space. Probability theory tells us that if all ten million inhabitants had been asked,
approximately 75% would also have responded by preferring the adding more green space.
In general, the larger the sample, the closer the estimate will be to the true population value.

There are four basic types of probabilistic sampling plans: simple random sampling, sys-
tematic sampling, stratiied random sampling, and cluster sampling. These are described
below.

161
Simple Random Sample
This is the most basic type of probability sampling plan. Every individual in the sampling
frame has an equal chance of being selected for the study. A common technique to select
individuals from a list is to use a random digit generator. Each person is assigned a random
number. The list is sorted in numerical order using the random digit column. The researcher
then selects the number of subjects for the study from the list in numerical order.

Due to the unique marine environment surrounding the Channel Islands, the Sanctuary is
home to a diverse array of marine life, making the region highly valuable to scientiic
research. The Channel Islands National Marine Sanctuary routinely conducts research to
monitor, preserve and protect the sanctuary’s rich resources. In 1998, the CINMS partici-
pated in a regional monitoring survey of the Southern California Bight coordinated by the
Southern California Coastal Water Research Project (SCCWRP). Trawl and sediment
samples from randomly selected sights around the islands were collected to measure the
distribution and health of the island’s marine life.

Image courtesy of the US National Oceanographic and Atmospheric Administration

Systematic Sampling
This is similar to the simple random sample. In this plan, every nth name on the list is
selected to be part of the study. For example, using an example of 1,200 students, the list
would be divided into 300 groups. To make sure that every student has an equal chance of
being selected, for the irst group, the selected student would be randomly chosen. After
the irst student is chosen, the 40th student down the list would be chosen. That means if the
number 10 was randomly selected, then student 10 would be chosen, then student number
50 (10+ 40), and so on until the sample is complete.

One advantage of this system is the relative ease of executing the selection process. The
randomization procedure only happens once. The major disadvantage to this plan is the
possibility that the list may have some type of recurring pattern or cycle. As an illustration,
if the researcher uses a list of houses that is a sequential list of houses from block to block
and each block has the same number of houses, it could happen that only the corner house
on each block is selected. People in these houses might differ from those in the midblock.
For example, the corner lot may be the most attractive house on the block and therefore

162
more expensive to rent or own. Another illustration that shows where this type of sampling
can go wrong is newspapers. If every seventh paper is chosen, it could happen that this
plan selects only Sunday papers. Checking the list for patterns before the sampling plan is
implemented is the best method to avoid this problem.

Stratiied Random Sampling

The third type of probability sampling is known as stratiied random sampling. The popula-
tion is divided into two or more strata (a particular subgroup within the population, such
as male and female) and then either a simple or systematic sampling from each stratum is
taken. It is necessary to have an adequate number of potential participants within each stra-
tum to use this strategy. The primary reason researchers prefer a stratiiedrandom sampling
plan is that the results may have less variation (less error).

Cluster Sampling
In the other three types of sampling, a sample frame is required. However obtaining a list of
potential participations may not be possible. In the above example of high school students,
it would be virtually impossible to obtain a list of all high school students in Montana.
However, a list of all high schools can be readily obtained. This is known as the cluster.
First, a sampling of the cluster is done by either simple or systematic sampling. In this case,
a sample is obtained from the list of high schools. After the sample of clusters is chosen,
a simple, systematic or stratiied random sample of individuals are selected within each
school. For this to work, the researchers would need to be able to obtain a list of all high
school students in the selected schools. The advantage of this style of sampling is that a
sampling frame (complete list of all individuals) is not necessary. The disadvantage is more
variation in the study results. Typically, a larger number of individuals are recruited to the
study using this sampling plan to offset the larger error in study results.

How Do the Sampling Frame and Samples Relate to One Another?

The sample is a subset of the sampling frame. Remember that the sampling frame is the list
of individuals in the population of interest. This may be a small population or a very large
population. The sample should be representative of the larger population based on key
characteristics, such as socio economic status, gender, race, and other parameters deined
by the research team. The statistical methods required to do these analyses often require
comparison of the sample to the sampling frame. In math terms, the sample is the numera-
tor of the fraction used to make the comparison and the sampling frame is the denominator.

163
Summary
The sampling frame is an important component of any epidemiologic study. The sam-
pling frame is the overall population that the researcher wishes to study. This may be a
small group of people, such as a classroom full of students, or a very large group, such as
the insect population of the Earth. Sampling frames have certain characteristics that may
lead the researcher to use different types of sampling techniques. The most basic sampling
technique is the simple random sample. Systematic sampling is a relatively simple type of
sample where the selection process is based on a pattern, such as every seventh name on a
list. Stratiied random sampling is slightly more complex because it involves breaking the
sampling frame into subgroups, then using random or systematic sampling to draw samples
from each subgroup. There are times when a sampling frame cannot be deined. In these
situations, it is possible to do cluster sampling. Cluster sampling involves developing a list
of the populations, called the cluster. The cluster is sampled, and the samples are further
subjected to random, systematic, or stratiied sampling methods. Once the data are all col-
lected, the researchers use statistical methods to analyze the results. The sample relates to
the sampling frame in the same way the numerator relates to the denominator in a fraction.

Concept Reinforcement
1. List the three types of sampling that use lists and describe how lists are used in
each type of sampling.

2. Describe the concept of a sampling frame and why it is important in epidemiology.

3. Explain the denominator in the sampling frame.

164
 Chapter 4 - Confounding and Effect Modifier
Chapter Objectives
• Deine confounder

• Deine effect modiier

• List examples of confounders

Two terms need to be used correctly when reporting study results: association and cause.
An epidemiologic study is rarely able to demonstrate the cause of a disease. It is typically
only able to show associations.

Cause Versus Association

A cause is the reason something happens. There should be two chromosomes in 23 pairs,
or a total of 46 chromosomes, in a healthy baby. Trisomy disorders occur when an extra
chromosome is made during replication. The speciic characteristics of the baby with tri-
somy disorder are linked to the chromosome. For example, Down syndrome is caused by
a chromosome 21 trisomy, the presence of all or part of an extra 21st chromosome. In other
words, chromosome trisomy 21 causes Down syndrome.

An extra chromosome on chromosome 21

An association is a relation between two variables. In epidemiology, one variable is the risk
factor and the other variable is the disease. An association is an indication of the probability
of the risk factor being present along with the disease.

If 25% of obese men are rich, 40% of tall men are rich, and being obese does not alter
the probability of being rich, then no association exists between weight and inancial well
being. However, if the prevalence of inancial well-being differs in obese men versus tall

165
men, there may be an association between weight and inancial well-being. If the weight
and height increase together, it is known as a positive association. If the values go in op-
posite directions, e.g., weight decreases as height increases or visa versa the association is
considered negative.

Association between height and weight associated with wealth

Confounders
Is the association real? One common cause of misinterpretation of study results by not
distinguishing between confounding and effect modiier. Confounding is when the results
suggest a factor as explaining the health outcome but in reality some other factor explains
the health outcome. This other factor “goes along” with the irst factor. This problem is also
known as the inluence of the lurking variable (also known as confounder).

Characteristics of Confounders

Diagram that shows the association among exposure, confounder and health outcome

166
When thinking about a confounder the following two points need to be true.

1. The confounder is not on the causal pathway between exposure and disease.

2. The confounder must be associated with both the exposure and the health outcome.

An example might help clarify this concept.

Case Study: Lung Cancer and Bartenders

Several studies using occupational and underlying cause of death data ascertained from
death certiicates found that bartenders have a high rate of lung cancer. The question that
researcher asked was, “What is it about bartenders that may lead to a high lung cancer
death rate?”

One possible explanation is that bartenders are more likely to smoke cigarettes. However,
it may be that bartenders drink alcohol more than people in other occupations and therefore
the reason for the increased lung cancer death rates is directly related to the alcohol con-
sumption of this profession and has nothing to do with cigarette smoking. The hypothesis:
there is an association between exposure (cigarette smoking) and a disease (lung cancer).
Is alcohol consumption a confounder?

First, we ind that cigarette smoking is statistically associated with lung cancer. Cigarette
smoking is also associated with alcohol drinking. Therefore, alcohol drinking fulills one
criterion for potential confounding. Alcohol drinking is also associated with lung cancer.
Alcohol drinking fulills the second criterion for potential confounding. However, in the
analysis when an adjustment is made for alcohol consumption, the association between
drinking alcohol and lung cancer is no longer there. The original association is explained
by the fact the more smokers also drink alcohol. The conclusion would be that drinking
alcohol is not associated with lung cancer.

Case Study: Down Syndrome and Birth Order

In another example, a study evaluated birth order as a risk factor for increased incidence
of Down syndrome.

Below is the table of study results:

167
Now let us add another facet to this study.

A third factor, which is related to both the exposure (risk factor) and health outcome, and
which accounts for some of the observed relationship between the two is a confounder. A
confounder is not the result of the exposure. In this case given that we are looking at the
association between the child’s birth rank (exposure) and Down syndrome (health out-
come), is the mother’s age a confounder? Given that we are looking at the association be-
tween the mother’s age (exposure) and Down syndrome (health outcome), is the birth rank
a confounder?

168
 Confounding example, part one

Birth order is associated with maternal age but is not a risk factor in younger mothers.

Confounding example, part two

Birth order is associated with maternal age but not a risk factor for Down syndrome in
younger mothers. Therefore, it is deinitely not a confounder.

Case study: Employment Salaries by Sex

An example in which the lurking variable (confounder) plays a role in the strength of the
inding. In a real life example, National Science Foundation (NSF) conducted a study of
bachelor’s degree graduates with a science or engineering degree in 1977 and 1978. NSF
reported that women earned 77% of the average salary of men. However, when the data
were recalculated by taking into consideration the lurking variable of job type, the average
job salary was at least 92% that of men. By combining all science and engineering jobs, the
lower life science salaries in which more women were employed confounded the results.
This example stresses the importance of gathering the right information.

Residual Confounding
Several situations occur in which confounding may be important. The three examples be-
low are known as residual confounding.

Confounding by Unmeasured Variables

This one is the easiest concept to understand, but can also be the most evasive. The study
results may vary or change directions, if an unknown variable is added to the equation. If
the researcher had thought to collect data on this variable, it would not be “unknown.”

169
Case Study: Antioxidant Vitamin Intake and Cardiovascular Disease Risk
Observational studies have shown a protective effect of antioxidant vitamin intake and the
risk of developing cardiovascular disease (CVD). In other words, those who take Vitamin
E are less likely to have heart attacks. However, randomized clinical trials (RCT) have
shown no effect. One theory as to why an observational study’s result differs from the
RCT’s results is due to confounding by behavioral and social factors acting across the life
course. For example, childhood social class may be an important confounder. In order for
an observational study to deal with the confounder, it needs to be measured perfectly. Fail-
ure to measure these life course factors could be the reason for the different results between
observational and RCT studies.

Confounding by Unexamined Variables

Similar to confounding by unmeasured variables, confounding by unexamined variables
means that the researcher had obtained the data but did not include it in the analysis. For
example, in the case of mortality risk and plasma ascorbic acid, physical activity is consid-
ered a confounder. In one published study, data was gathered on physical activity but these
data were not included in the model. If physical activity is truly a confounder, then omit-
ting the confounder may have a sizable effect on the risk estimates for the relation between
ascorbic acid and mortality.

Confounding by Measurement Error

In this case, the variable has been measured but not well. Food intake is an area in which
there is almost always measurement error. Several techniques have been used to try to ob-
tain good data on food intake but all methods have been found to have measurement error.
Examples of food intake methods include daily logs, food frequency questionnaires, and
duplicate diet (you ix 2 plates of food and one plate goes into a collection bin that is given
to the researcher for nutrient analysis).

In the antioxidant vitamin intake and cardiovascular disease risk example, measurement
error, may be present. The intake quantity of antioxidant vitamins in the observational stud-
ies was derived from food frequency questionnaires. Accurate reporting of food intake is
known to contain measurement error and therefore the reported results may be driven by
confounding by measurement error.

Residual confounding is an important concept to keep in mind; however, it should not

paralyze the researcher. Doing a thorough literature review is a sensible step in obtaining
information on potential confounders so that this data may be collected and included in the
statistical models.

170
Efect Modiier
Another type of variable that can inluence the study indings is an effect modiier. When
the degree of association between an exposure and the health outcome changes according
to the value or level of a third variable, this is called an effect modiier. Sex is a common
effect modiier. For example, a hypothetical risk of a disease is 40 per 10,000 for women
and 10 per 10,000 for men. The degree of association between risk and exposure changes
from 40 to 10 per 10,000 based on sex. Therefore, sex is considered an effect modiier in
this case.

One of the most familiar effect modiiers that has had signiicant public policy is alcohol
consumption and driving. Driving is a risk factor for injury. Alcohol consumption is a risk
factor for injury. However, drunk driving signiicantly increases the risk of injury.

It is an effect modiier.

Unlike confounding, in which the determination of a confounder is based on scientiic rea-

soning, an effect modiier can be assessed statistically. One useful method to evaluate the
possibility of an effect modiier is to stratify the data. For example, in a hypothetical study
it was shown that roofers had a two fold increased risk of knee disorders than other types
of construction workers, such as electricians or painters. However, when these two groups
(plumbers and other construction workers) were stratiied by normal weight and over-
weight, the overweight category showed a three fold increased risk and the normal weight
category showed a slight increased risk for plumbers. Weight was an effect modiier.

Summary
In observational studies, especially where the researcher does not have control over all the
characteristics of the population under study, lurking factors can signiicantly inluence the
indings of a study. Therefore, it is imperative that the context be considered in a study. A
thorough reading of the literature can help the researcher be aware of potential confound-
ers. Sometimes, it is just common sense thinking and applying that knowledge.

Concept Reinforcement
1. What is one piece of information that the researcher should collect in any observa-
tional study of lung cancer?

2. Epidemiologists are very likely to say something is associated with a health out-
come but rarely that something causes a health outcome. Why do epidemiologists
shy away from statements of causality?

3. What is the commonality between an effect modiier and a confounder?

171
172
 Chapter 5 - Types of Bias
Chapter Objectives
• Deine selection bias

• Deine recall bias

• Deine misclassiication bias

• Explain Berkson’s bias

• Deine ecological fallacy

In designing a study and collecting information about participants, there is always some
degree of error. Nothing is ever perfect. If it is random error, in other words errors that oc-
cur by chance, then the researcher is not as concerned. However if there are error in the
design of the sampling plan or in the data collection that is not merely chance, then the
results of the study maybe invalid. Sometimes all the researcher can do is note the type of
error in the discussion of the published paper and let the readers decide about the validity
of the study results. In this chapter, we are going to describe several types of errors which
are also known as bias.

Bias
Bias is a systemic error. There is no set formula for interpreting the bias of a study. Howev-
er, a set of questions should always be asked when reading or hearing about study results.

• How were the participants selected? In other words, how representative is the target
population with regard to the hypothesis.

• Are those that participated different from those that did not participate in important
ways?

• How accurate was the exposure and outcome variables measures?

173
Selection Bias
Who is given the survey in the irst place? For example, a researcher is interested in the de-
termining the reasons (predictors) of knee injuries in adolescents (aged 14-18 years) in San
Bernardino Valley, California. The researcher sends a survey to all high school male ath-
letes since the researcher knows this is the most injury prone group. This sampling design
has selection bias since the researcher did not ask females. It is not unreasonable to con-
sider that females may have different experiences that predict knee injuries. Furthermore,
the researcher did not ask students who were not athletes. For both males and females who
had sustained a knee injury in this population (athletes and non-athletes), the researcher did
not gather complete information about predictors. The researcher did gather information on
one sub-group, male athletes.

Referral Bias
If the population of sick people comes from one type of hospital, this may introduce refer-
ral bias. For example, in a famous study called the National Perinatal Collaborative Project
(1959-1974) 12 prestigious hospitals participated in the project. The purpose of the project
was to follow pregnant women and their babies for several years to determine risk factors
associated with pregnancy and children’s health. Many indings were reported but after
several years, it was discovered that some of the health outcomes of the babies were much
higher rates than the general population. This was because the hospitals that participated in
the research were more likely to have women with high risk pregnancies as patients. This
is an example of referral bias—high risk pregnant women were referred to these participat-
ing hospitals, not because of the research study but because they were considered the best
hospitals in the region.

Interpretation Bias
If the question is not clearly written, respondents may answer the question differently than
was intended. For example, “Is your work made more dificult because you have a child in
daycare?”

A “no” response may mean, “No, I do not have a child in daycare” or “No, my work is not
more dificult”

This interpretation bias can also occur in a medical setting. It is also called diagnostic bias.
For example, if a radiologist is asked to look at a chest x-ray for signs of lung cancer and
the radiologist knows the patient is a long time older heavy smoker, s/he may be more
likely to examine the x-ray much more closely for any sign of cancer than for a young
patient without a history of smoking. This behavior is good for clinical practice but is not
good for research since it introduces bias. Participants are treated differently based on their
smoking status.

174
Recall Bias
Asking participants about an event that happened long ago can create recall bias. For ex-
ample, many people believe that someone with an illness often ask “why me?” It is com-
mon for patients to think back on their lives very carefully and remember details that they
think might be associated with the disease. For example, it was found that mothers whose
children have had leukemia were more likely than mothers of healthy children to remember
details of diagnostic X-ray examinations during pregnancy.

Self-selection Bias
Another type of bias can occur if the participants volunteer for a study. This is known
as self-selection bias. The question is whether the volunteers are the same or different in
some important ways from a random sample of participants. A similar type of bias is when
participants recommend their friends or relatives for the study. This is problematic from
an analytical perspective since it is well known that friends and relatives are more similar
than strangers. Again this violates the goal of a random sample of the population in which
the results can be generalized to the population.

Interviewer Bias
No matter how professional an interviewer acts, there is always to possibility that the re-
spondent is providing answers to the questions based on this relationship. For children this
is readily apparent. Children will often say whatever the adult wants to hear. For adults
this can take the form of pleasing the interviewer, avoiding answering an embarrassing
question honestly, giving answers quickly and without thought to inish the interview, etc.
All these permutations can inluence the study indings since the “true” description of risk
factors may not be accurately relected in the participants’ responses.

Berkson’s Bias
Berkson’s bias, also known as admission bias, was irst described in 1946. The underly-
ing concept is that people with more than one disease or condition are more likely to be
hospitalized than people with just one disease or condition. Berkson’s bias occurs when a
researcher is exploring the relation between two diseases. An overestimation of exposed
cases is likely to occur.

In addition to the study design and the data collection biases, there are also a couple of
biases associated with screening.

Lead Time Bias

Even if someone is screened and found to have the disease, early detection does not pro-
long life. For example, clinical trials have shown that screening for lung cancer with chest
x-ray will detect tumors in the lungs earlier; however the average survival time does not
change with this earlier diagnosis. In other words, a person with lung cancer will live the
same period of time with or without the screening test. The lead time bias can give the im-
pression of the screening best being better since it prolongs life but in reality it just lets the
patients know sooner about disease state.

175
Length Bias
For a slowly progressing disease, screening may ind people with the disease at an earlier
time. However, these slowly progressing diseases have better prognosis than quickly pro-
gressing diseases. In other words, patients with slowly progressing diseases will take lon-
ger to die from their disease. The mistaken idea is that the screened patients do better but
this is a false perception. The screening identiies the disease earlier so there is a perceived
increase in length of survival time. This increased length is just an artiice of the disease
progression. In other words, the detection and treatment may not change the number of
people diagnosed and the number of people who die from the disease every year.

Ecological Fallacy
It is important, when analyzing and interpreting data, to be precise in describing what the
data tell you. Ecological fallacy is the result of an error of reasoning that occurs when the
characteristics of a group of people are used to draw conclusions about an individual. Eco-
logical fallacy occurs when people apply statistical information incorrectly. For example,
a study of people who wear glasses shows that they have above-average intelligence. The
ecological fallacy occurs when someone uses the information about the group to draw a
conclusion about an individual who wears glasses. The person wearing the glasses may or
may not have above-average intelligence.

Summary
A bias in the study design and data collection cannot be ixed. There is no analytical tech-
nique to correct for bias. Researchers spend considerable time thinking about the study
design and data collection to minimize the different types of bias. Even with best efforts, if
bias creeps into the study, the researcher will mention it in the discussion part of the pub-
lished manuscript so the reader can decide about the validity of the study results.

Concept Reinforcement
1. What are some questions that you think your peers would be reluctant to give the
“true” answer in a telephone interview? What could the interviewer say to encour-
age participants to answer “truthfully?”

2. Explain Berkson’s Bias and its importance to health research.

3. How does length bias and lead time bias differ?

176
 Chapter 6 - Generalizability
Chapter Objectives
• Discuss the importance of generalizability

• Deine Type I error

• Deine Type II error

• Describe the characteristics of participants in a study to increase generalizability

Generalizability
Generalizability is making predictions based on past, recurring experiences. We expect
that the things that happen in the past will continue to happen in the future if the conditions
remain the same. In epidemiology, researchers strive to collect generalizable data by using
robust sampling techniques. These techniques are important to selecting a sample that is
representative of the overall population. The results of the study are generalizable to the
population only if the sample is representative of the overall population. Several situations
can occur so that the study cannot be generalized.

Researchers use three types of generalizability. The irst of these involves researchers de-
termining if a speciic treatment or intervention will produce the same results in different
conditions. In this type of generalizability, the researchers need to determine whether some
factor beyond the treatment generated the result they are studying. This helps establish the
lexibility of the treatment, or its ability to adapt to new situations. The more lexible the
treatment, the more adaptable it is to different situation.

A second form of generalizability focuses on measures rather than treatments. In this case,
a result is generalizable if it produces the same results with different forms of measurement.

The third form of generalizability is related to the test subjects. The results of the test may
be internally valid, meaning they are valid within the sample group, but not generalizable
to the larger population because of differences between the test group and the overall popu-
lation. In order for the results of a study to be generalizable, the sample group must relect
the characteristics of the larger population.

177
Characteristics of Populations
The populations of interest in epidemiologic studies have particular characteristics that
are of interest to the researcher. Participants also have other characteristics that must be
taken into account when designing the study and selecting the test group. There are general
population characteristics, such as the racial and ethnic composition of the population, the
percentage of men and women, the distribution of ages, socio-economic status, and marital
status. There are also characteristics speciic to the question the researchers asks, such as
disease status, family history of disease, exposure history, etc.

Shows the ive-year (2001-2005) legal immigration rate per country’s total 2000
population, deined as all those who received legal permanent residence in all categories,
including regular immigrants, refugees and asylees, diversity lottery winners,
NACARA/HRIFA beneiciaries, and others. This should be a good representation of
the actual immigration rate, as the adjustment of status backlog was similar in size
in 2000 and in 2005.

Immigrants from “Czechoslovakia (former)” included in data for Czech Republic, those
from “Soviet Union (former)” included in data for Russia, and those from “Unknown
country” conlated with Palestinian Territory. Corrections have been made for anomalous
data for Congo (Brazzaville) vs. Congo (Kinshasa), Niger vs. Nigeria, and Guinea vs.
Guinea-Bissau. Immigration from “Korea” equated with South Korea, with that
from North Korea assumed to be negligible.

Think about whether the following study is generalizable. In this study, the larger popula-
tion consists of high school graduates who decide to attend technical school. This popula-
tion has the following characteristics: 18-21 years old, 75% male, 25% female. The re-
searchers want to study why these people chose to attend technical school. The sample
selected for the study consists of high school graduates ages 18-30 and is divided evenly
between men and women.

Compared to the study population, all potential high school graduates range in age from
18-21 years old and are composed of 75% men and 25% women. The sample population
has an age range of 18-30 years old and is divided evenly between men and women. The
only thing that is consistent is that these two groups both consist of high school graduates.

178
The results obtained from the sample group will not be generalizable to the overall group
because of the fundamental differences in the characteristics of the test group and the over-
all population. One means of iguring out if the makeup of the participants is similar to the
overall population is to run a statistical test. If the differences are big enough, it is said to
be statistically signiicant.

Error
All statistic tests contain some error. An important concept to understand when working
in epidemiology is statistical error as it relates to signiicance testing. There are two basic
types of error: Type I and Type II.

Type I error is essentially a false positive. The probability of a Type I error occurring is
shown using the Greek letter alpha (α). It is called the Type I error rate.

Type II error is the opposite – a false negative. The Type II error rate, which is the prob-
ability of a Type II error, is shown using the Greek letter beta (β).

This concept can be shown using a 2x2 table. The columns indicate someone with a disease
or without. The rows indicate someone with or without a risk factor.

Disease, Illness, Injury

Positive Negative
Risk Factor
Positive True Type I Error
Negative Type II Error True

A 2x2 table includes four options.

• A person can have the disease and the risk factor. This is known as true positive. This
means there are not any errors in the testing.

• A person does not have the disease and does not have the risk factor. This is known
as a true negative. This means there are not any errors in the testing.

• A person does not have the disease but the test for the risk factor shows that the per-
son has the risk factor. This is known as False positive. It indicates type I error.

• A person has the disease but the test for the risk factor shows that the person does not
have the risk factor. This is known as False negative. It indicates type II error.

In any test, the magnitude of type I and type II errors vary. In the world of statistics and
statistical testing there is always a balance between type I and type II errors. A large Type I
error could result in a research reporting an association between a risk factor and a disease
when one does not exist. In contrast a large Type II error could result in a researcher report-
ing no association between a risk factor and a disease when one really exists.

179
Summary
In order for epidemiologic studies to be generalizable, the sample group must relect the
larger population, both in composition and in terms of the speciic question being asked.
For example, if the question of interest is risk of cancer due to exposure to a certain chemi-
cal, the sample population must relect the same characteristics as the overall population
for cancer incidence and potential exposure. Epidemiologists face two speciic types of
error when studying populations. The irst is Type I error, which is essentially a false posi-
tive. The second is Type II error, which is a false negative. In both cases the true situation
is not represented.

Concept Reinforcement
1. Explain the importance of generalizability in epidemiologic studies

2. Describe the different between Type I and Type II error.

3. Explain how to improve generalizability by selecting the proper sample population.

180
 Chapter 7 - Publication Bias
Chapter Objectives
• Describe publication bias and its implications

• Describe afiliation of author and its implication

• Describe implication of industry funded research and ownership of data

Having papers published is an important part of many researchers’ jobs. The dissemination
of research indings advances the ield. However, publishing study results is not straight-
forward.

Publication Bias
Publication bias is the tendency, on average, for studies to be published in which the results
are signiicant and for studies that did not ind a signiicant inding to not be published.
This tendency may result in scientists choosing to report only signiicant indings and not
submitting manuscripts for publication in which the indings were not signiicant.

Steps Towards Publication

The researcher takes many steps to achieve a publication. The irst step after a study is
completed and the data are analyzed is to decide the journal that will receive their sub-
mission. Journals are informally ranked from top tier to fourth tier. The most prestigious
journals are in the top tier. Having a paper published in a top tier journal carries much for
prestige to the researcher and often is more widely read than a fourth tier journal.

After the researcher decides upon the journal, the manuscript is formatted to match the
requirements of the journal. There are often word limits for the manuscript and details re-
garding the layout of the manuscript.

The manuscript is submitted to the journal and the editors look it over. This is the irst place
where publication bias can be felt. Editors may feel that non-signiicant indings are less
likely to be interesting to their readership and therefore reject the manuscript. However, if
the editors think it is interesting and it matches the journal’s purpose, they then send it out
for peer review.

181
A peer review process consists of two or three researchers reading the manuscript carefully
and making comments and suggestion to improve it. The reviewers return the manuscript
back to the journal editors and indicate whether the manuscript should be 1) accepted, no
revisions, 2) accepted with minor revisions, 3) in need of major revisions or 4) rejected.
Most manuscripts are given category 3 or 4 status. It is extremely rare to have a manuscript
accepted without some changes suggested. This is the second point in the publication pro-
cess where a manuscript with a non-signiicant inding may be rejected.

After the peer review process, the authors of the original manuscript have the opportunity
to revise and resubmit their draft or try another journal with their manuscript.

A manuscript is not published in respected scientiic journals without the author disclosing
a conlict of interests and stating their afiliation. The afiliation means the place where the
authors work. A conlict of interest means that the author must disclose any money received
from an institution or business that directly relates to the study. For example, if the re-
searcher is studying the eficacy of a drug manufactured by a pharmaceutical company and
the same pharmaceutical company gives the author a consulting fee, this would be dis-
closed on a conlict of interest statement.

182
Industry Funded-Research
In many cases, private companies (industry) sponsor research at independent research in-
stitutions or universities. The purpose of this research is usually to answer a speciic ques-
tion about a product or service provided by the company. The company who sponsors the
research provides the funding for the research and helps deine the research question being
asked by the study. The researcher should independently perform the research and analyze
the data to answer the question that was posed. Sometimes there are concerns about the
inluence of the company on the results of the study. There have been cases where the com-
pany has coerced the researcher into either designing the study to ensure a favorable result
for the company or trying to quash unfavorable results. These cases have resulted in skepti-
cism regarding the results of industry-sponsored research because of concerns about bias.

Sir Richard Doll

Sir Richard Doll is a famous epidemiologist. He, along with Ernst Wynder, Bradford Hill,
and Evarts Graham, showed that smoking was linked to lung cancer and increased the risk
of heart disease in the 1950s. He also studied asbestos and lung cancer, radiation and leu-
kemia and alcohol consumption and breast cancer.

After his death in 2005, it was discovered that he had received consultancy payments for
chemical companies whose products he defended in court. He received many payments.
For example, one payment was for $1500 per day from Monsanto, a chemical company,
which began in 1976 and continued to 2002. Most of these payments were never disclosed
to the public. Sir Richard Doll has his defenders and those who feel his work is tainted
since he did not disclose his conlict of interest.

Ownership of Data
Ownership of data is a critical issue in scientiic studies. Universities tend to be very open
in terms of the dissemination of research results. University faculty members are encour-
aged to publish their research results and report them at scientiic meetings. In fact, the
publication record of a faculty member is a key measure of success and performance. Many
times, however, industry-sponsored research results in the company owning the research
results. This can lead to suppression of the results for reasons of conidentiality or if they
are not favorable to the company. In other situations, it can lead to inventions made during
the course of the study not being further studied or commercialized.

183
Summary
Dissemination of study results to the wider audience of the research audience is critical for
advancement of science. However, researchers depend on companies who are businesses to
publish their indings, often for a nominal fee. The company relies on subscriptions to their
journals to offset the costs of production. Editors may feel it is necessary to accept ind-
ings that are signiicant rather than non-signiicant indings to keep the readership numbers
high, thereby leading to publication bias. One recent response to this well-known dilemma
is that a few journals have added a section speciically for non-signiicant indings. Be-
cause research is expensive, great care has to be taken by the researchers to make sure their
indings are not the result of a conlict of interest.

Concept Reinforcement
1. Do you think there is any problem with a researcher being paid as a consultant by a
company in which the researcher is evaluating a device the company makes?

2. Can you think of any way in which pharmaceutical companies can get research
done on their drugs without having any conlict of interest situations?

3. What do you think about Sir Richard Dolls’ situation?

184
 Chapter 8 - Crude Rate, Numerator and Denominator
Chapter Objectives
• Calculate crude rate

• Identify numerator and denominator

• Identify the proper denominator for a particular study

Once the data have been collected, the researcher publishes results in peer-reviewed scien-
tiic journals. There are several standard means to report results. The crude rate of a disease
is one of those.

Crude Rate
A crude rate applies to the participants in the study without consideration of any character-
istics of the participants. As the name implies the crude rate is generally the irst impression
of the results. Using the crude rate to compare studies or compare data should be done with
extreme caution since the characteristics of the populations may differ. One of the most
common differences between places is the age structure. In other words the percent of
young, middle aged, and older people. Sometimes, the percent of males and females is very
different by place. The crude rate does not take any of these differences into consideration.

Crude rates are used 1) when number of deaths/diseases is not known for the subgroups
of the population. In other words, the overall death count is known but not the number of
deaths by age groups; 2) the population size of the subgroups is unknown. In other words,
the overall population under study is known but not the number of people in each age
group; 3) when the number of people dead/diseased is too small. If the numbers are small,
the statistics are not very reliable since a few additional deaths can change the rates by
quite a bit.

Crude mortality rate by country. The crude death rate for the world is about
8.23 per 1000 people.

185
Calculation of Crude Mortality Rate

Crude mortality rate (in a speciied year) formula is number of deaths in the speciied year
÷ mid year population (in the speciied year) x 1000

Three pieces of information are needed to calculate the mortality rate:

A Deined Period of Time

It is necessary to have a deined population, with an estimate of the size of the population
during the time period. Since people are always being born and dying, the population is
never constant. Therefore, the population in any given year is the population at mid-year.

Count
The count is the number of deaths occurring over the designated period of time.

A Multiplier
The proportion can be multiplied by 1,000, 10,000 or 100,000, though is usually reported
“per 1000.” The reason to multiply by some number is that it is earlier to understand num-
ber of deaths as a whole number instead of a fraction. For example, a 0.0081 crude mortal-
ity rate is less understandable than 8.1 deaths per 1000 people.

186
Crude Mortality Rate Examples
Kuwait, United States, and Japan

Oldest twins in modern history, Narita sisters, died at age 107

In the table below, Kuwait has the lowest crude mortality rate, substantially lower than
the other countries listed, yet it also has the shortest life expectancy. This is an example in
which the crude rate only provides a partial picture. By looking at the age distribution of
the countries, you can see that Kuwait has substantially fewer people living to old age and
a considerable number of children dying, compared to the other listed counties.

Crude mortality rate, life expectancy and age structure of selected counties,
2007 estimates
Percent population by age group
Crude death Life expec-
Countries 0-14 yrs 15-64 yrs 65+ yrs
rate tancy
Kuwait 2.4 77.4 27% 70% 3%
United
8.26 78.0 20% 67% 13%
States
Japan 8.98 83.5 14% 65% 21%
United
10.09 78.7 17% 67% 16%
Kingdom

187
Example: he Diference Between Crude Rates in 2 Hypothetical Towns
In Northville, population of 100,000, 400 people died in 2008.

In Southville, population of 100,000, 700 people died in 2008.

The crude mortality rate is 4 per 1,000 in Northville and 7 per 1,000 in Southville, so these
data tell us that Southville has a higher mortality rate. However, is there more to this story?

Age structure and mortality counts in Northville and Southville,

population of each 100,000
Age groups (in years) Northville number of deaths Southville number of deaths
<20 154 30
20-39 120 28
40-49 55 33
50-59 20 68
60-69 14 126
70-79 12 212
80-89 10 108
≥90 15 95
Total 400 700

Let’s look at the age structure of the towns with the death counts, in the table. What does
this table tell you? For some reason young people in Northville were more likely to die
than those in Southville. In this situation, the story about death would focus on the reasons
the young are dying in Northville because we all expect death to be a more common oc-
currence as we age.

Numerator and Denominator

When doing population studies, it is important to understand the relationship between the
sample population and the entire population of interest. The relationships are often shown
using fractions, which consist of numerators and denominators. The numerator is the sam-
ple population and the denominator is the overall population.

Sample Population
Total Population

188
Identifying the Proper Denominator
It is very important to ind the proper denominator when designing, performing and analyz-
ing the results of an epidemiologic study. The irst step is to deine the question the study
will ask. Once that has been decided, the proper study population can be deined. If your
question relates to the effects of pesticide exposure to farmer health, your study population
(denominator) will be farmers. You will need to include farmers that use pesticides and
farmers that do not use pesticides if you want to do a comparison. If you simply want to
study the incidence of disease in farmers who use pesticides, you can include only those
farmers who use pesticides.

Summary
Epidemiologists use many statistics to describe their study results. One type of result is a
crude rate. This rate does not take into consideration any characteristics of the populations.
It just shows the number or count of people with the disease, illness or heath outcome of
interest in the numerator and the population at risk at midyear in the denominator. Crude
rates have limited usefulness if other characteristics of the study population are important.
When presenting data, epidemiologists take care in iguring out the denominator. They ask
themselves, “Who makes up the population at risk?”

Concept Reinforcement
1. Explain crude rate and how to calculate it.

2. Describe how epidemiologic data are represented in a fraction.

3. Explain how to identify the proper denominator for an epidemiologic study.

189
190
 Chapter 9 - Direct Standardization
Chapter Objectives
• Describe direct standardization

• Write the formula for calculating direct standardization

• Describe the situation in which this calculation is used

Standardized Rates
Although crude rates sometimes need to be used, the accepted practice is to standardize
(also known as adjusted) the rates. Typically, the adjustment is based on age or some other
population characteristic such as race or gender. Once this has been calculated, compar-
ing rates makes more sense. With that said, it is also very important to look at the speciic
rates for any interesting differences or changes. In other words, our hypothetical towns of
Northville and Southville would show some interesting speciic age group rate differences
that would not be apparent when looking at the overall mortality rate. A standardized rate
differs from the crude rate in that the inluence of some extraneous variable, typically age,
has been removed.

Age structure and mortality counts in Northville and Southville

Age groups (in years) Northville number of deaths Southville number of deaths
<20 154 30
20-39 120 28
40-49 55 33
50-59 20 68
60-69 14 126
70-79 12 212
80-89 10 108
≥90 15 95
Total 400 700

Two methods are commonly used to standardize the mortality rates, the direct method
and the indirect method. In these two methods, standardization is just done by obtaining
a weighted average. An example that students experience using weighted average is their
class grade. This grade is almost always based on weights. For example, at the beginning
of the year, the teacher tells the students that quizzes are worth 30%, homework are worth
20%, exams are worth 30% and inal exam is worth 20%. The inal class grade is computed
using weighted averages: (0.3 times“averaged quiz grade”) + (0.2 times“averaged home-
work grade”) + (0.3 times“averaged exam score”) + (0.2 times“inal exam grade”).

191
 The irst of two steps in standardization involves categorizing the populations under study
into groups (also known as stratums (aka: strata)) based on some characteristic, such as
age groups. The second step is calculating the stratum-speciic rates. The third step is using
a reference population.For example, if we are interested in comparing all cause mortality
rates in two states, we will use the mortality rate by age groups from these two states. These
rates would be standardized to a second population—often data from the US census. It
can be state, national or world population rates. If the researcher is doing an age standard-
ization (probably the most common standardization), then population proportions by age
groups is used as the reference population.

Direct Standardization
In the direct method, a single population is used to standardize (weight) the rates for all
Crude rate comparisons. This is done by taking the sum of the products of stratum-speciic mortality
Remember the crude rates in the speciic population being standardized and the stratum speciic proportions of
rate does not take those strata in a standard population. Using the data below, multiply each stratum speciic
into consideration mortality rates in Florida by the US population proportion for each stratum, and then add
any characteristic these numbers for the mortality rate of each state.
of the population,
such as age struc- (6.96 * 0.01) + (0.33 * 0.05) + (0.15 * 0.07)….+ (129.05 * 0.02) = 7.5 per 1000 for Florida
ture. Therefore for
Florida the crude (5.25 * 0.01) + (0.25 * 0.05) + (0.13 * 0.07)….+ (126.17* 0.02) = 7.0 per 1000 for Cali-
mortality rate would fornia
be 168961/17384430
= 9.7 per 1000 If the direct standardization rate and the crude rate do not differ, this tells the researcher that
people and for the stratum speciic variable (in this case age) was not different between the two popula-
California the crude tions. It is important to calculate both the crude and the standardization rates for this reason.
mortality rate is
232468/35842038 =
6.5 per 1000 people

When the crude rate is higher than the directly standardized (age-adjusted) rates, this means
that the age structure of the state is more unfavorable compared to the standardized popula-
tion (e.g., United States age structure). In developed countries without the experience of
infectious disease epidemics or war related deaths, age is one of the best predictors of
death. Therefore, in this case, we would expect a higher proportion of people to die who are
older and can conclude that the place with a higher crude rate relects an older population.

192
To generalize these indings, we can say that if the direct age adjustment (direct standard-
ization using age) produces a lower mortality rate, than the population of interest has a
more unfavorable age distribution than the population that was used as the standard. In
contrast, if the direct age adjustment produces a higher mortality rate, than the population
of interest has a more favorable age distribution than the population that was used as the
standard.

Summary
Direct standardization is one means of presenting study results so that other study results,
that are also standardized in the same way can be compared to each other. To be able to
directly standardize, data by stratum has to be available. If the crude rate and the direct
standardization rates do not differ, this tells the researcher that the stratum speciic variable
is similar between the comparison groups. Although one number is the reported rate, it is
also important to look at the age speciic rates to identify any interesting patterns.

Concept Reinforcement
1. In which situation is it important to standardize rates?

2. Is there any situation in which crude rates are just as good as standardized rates?

3. What information is required to standardize rates?

193
194
 Chapter 40 - Indirect Standardization
Chapter Objectives
• Describe indirect standardization

• Write the formula for calculating indirect standardization

• Describe the situation in which this calculation is used

In comparing the incidence or mortality rates between different places, it is rare that the
characteristics of the two places are similar. If the important characteristics of two places
such as the age structure, are similar, then a crude rate works just ine. However, if the char-
acteristics are different or suspected to be different, then the rates need to be standardized
so they can be compared to each other.

Standardized Rates
The accepted practice in epidemiology is to standardize (also known as adjust) the rates.
The direct and indirect methods of standardization are the two most commonly used tech-
niques for standardizing data. Using either of these methods, standardization is just ob-
taining a weighted average. An example that students experience using weighted average
is their class grade. This grade is almost always based on weights. For example, at the
beginning of the year, the teacher tells the students that quizzes are worth 30%, homework
are worth 20%, exams are worth 30% and inal exam is worth 20%. The inal class grade
is computed using weighted averages: (0.3 times “averaged quiz grade”) + (0.2 times “av-
eraged homework grade”) + (0.3 times “averaged exam score”) + (0.2 times “inal exam
grade”).

Indirect Standardization
The indirect method is a little more complicated than direct standardization. It is used when
stratum-speciic rates are unavailable in the population of interest. It is also used when the
number of deaths (or counts of disease) per stratum is small, say less than 5 or 10. Finally,
the indirect method is used in occupational studies: studies that compare the death rate in
an industry to the general population death rate. The question that the researcher asks is
whether working in a particular occupation is more hazardous or not. Using these statistics,
studies are produced which rank the most hazardous occupations.

195
The U.S. Bureau of Labor Statistics calculated the most hazardous jobs in 1994. Below are
the top 20 most hazardous jobs.

Occupation Relative Risk Leading Fatal Event

Average All Jobs 1.0 Homicide and Accidents
Fishers 21.3 Drowning
Timber Cutters 20.6 Struck by object
Airplane Pilots 19.9 Airplane crashes
Structural Metal Workers 13.1 Falls
Taxi Cab Drivers 9.5 Homicide
Construction Workers 8.1 Vehicular, Falls
Roofers 5.9 Falls
Electric Power Installers/Repairers 5.7 Electrocution
Truck Driver 5.3 Highway Crashes
Farm Occupations 5.1 Vehicular
Police, Detectives, Supervisors 3.4 Homicide, Highway Crashes
Non-construction Laborers 3.2 Vehicular
Electricians 3.2 Electrocution
Welders and Cutters 2.4 Falls, Fires
Guards 2.3 Homicide
Ground Keepers and Gardeners 1.9 Vehicular
Carpenters 1.6 Falls
Auto Mechanics 1.1 Highway Crashes, Homicide
Supervisors, Proprietors, Sales 1.0 Homicide
Cashiers 0.9 Homicide

196
Calculation of Indirect Standardization
In the indirect method, two steps are used to calculate the rate. For simplicity, we will cal-
culate a state’s mortality rate using indirect standardization. Incidence rate can be used in
the same way.

First, the sum of the product of stratum-speciic rates in the standard population and the
proportion representation of those strata in the population being standardized is used to
produce expected deaths. Then, the actual deaths in the population being standardized are
divided by the expected deaths to produce standard mortality rates (SMR). SMR = actual
deaths divided by expected deaths.

To determine the expected number of deaths, multiply the population of the state by the
stratum speciic mortality rate and divide by 1000, since the above stratum speciic rates
are per 1,000 and we want the expected total number of deaths.

((220,837 x 7.37) + (869,278 x 0.32) + (1,061,594 x 0.16)….+ (380,203 x 152.89))/1,000

= 194,575

Divide the state’s actual number by the expected number (168,961 / 194,575) and then
multiply by 100 = 87%. This is known as the standard mortality ratio (SMR). This means
that deaths in Florida are 87% of the United States rates and for California they are 119%
of the United States rates.

To review, the direct standardization gives an expected rate (standardized rate). This can
be compared to the crude rate. An indirect standardization gives an expected number of
deaths. This can be compared to the actual number of deaths in another population.

197
Summary
Public health researchers often used surveillance data routinely collected by governmental
agencies, such as death certiicates with underlying cause of death. As well as surveillance
data, most governments undertake a census periodically, thus allowing researchers to
apply methods discussed in this chapter to understand some trends in the health of a
population.

Concept Reinforcement
1. Why is indirect standardization used instead of direct standardization method?

2. If the characteristics of two places are similar, does an epidemiologist need to stan-
dardize the rates? Explain

3. What is compared between populations in which indirect standardization is used?

198
 Chapter 41 - Relative Risk
Chapter Objectives
• Construct a two by two table

• Write the formula for determining relative risk

• Explain the meaning of the point estimate

Epidemiologist use statistics to report study results. For certain study designs, one statistic
that is commonly used is called relative risk. Cohort and randomized controlled trial stud-
ies report relative risk. Relative risk is deined as the ratio of the risk of disease or death
among the exposed, to the risk among the unexposed.

Relative Risk
The relative risk can be measured by dividing the incidence rate in the exposed group by
the incidence rate in the unexposed group. The value calculated is called a point estimate.
A point estimate gives a single value along a scale. If the relative risk is 1.0, the risk of
the outcome of interest is exactly equal for both the exposed and the unexposed groups.
Therefore, the exposure did not have any effect on the health outcome. The further the rela-
tive risk is from 1.0, the greater the magnitude of effect. For example, a relative risk of 2.0
means that there is a two-fold increased risk for the health outcome in the exposed group
compared to the unexposed group. Less than 1.0 implies a decrease risk. For example, a
relative risk of 0.5 means the exposed group is 50% less likely to have the disease as the
unexposed group. Another way to say increased or decreased risk is that an effect observed
by a point estimate of less than 1.0 would be phrased as “less likely” and an effect observed
by a point estimate of greater than 1.0 would be phrased as “more likely.”

In epidemiology, it is not very useful to calculate a point estimate without also calculating
a conidence interval. The formula for this calculation is beyond the scope to this lesson;
however, interpretation of a conidence interval is within the scope of this lesson. If the
conidence interval includes 1.0, there is no statistical difference between the intervention
and control group. For example a point estimate of 1.3 with a 95% conidence interval
of 0.8 – 1.5 means that there is no statistically signiicant increased risk. Similarly, if the
point estimate is 0.6 and the conidence interval is 0.4 – 1.2, this also means there is not a
decreased risk for the exposed group because 1.0 in within the conidence interval. In con-
trast, if 1.0 is not included in the conidence interval, the difference between the exposed
and unexposed group would be considered statistically signiicant. If the point estimate is
greater than 1.0, then the exposed group would be at a statistically signiicant increased risk
and if the point estimate was less than 1.0, then the exposed group would be at a statisti-
cally signiicant decreased risk compared to the unexposed group.

199
 2 point estimates and 95% conidence intervals

Analysis of Study Data: Use of Two by Two Table

One important tool used to evaluate the association between exposure and disease is a two
by two table. This is used in many calculations.

Disease Status
Present Absent Total
A (exposure, disease B (exposure, but no
Present A + B
Exposure present) disease)
Status C (no exposure, disease D (no exposure, no
Absent C + D
present) disease)

A + C (total number B + D (total number N (sample

Total
with disease) without disease) total)

The columns represent disease status (with or without the disease) and the rows represent
exposure status (present or absence of exposure). The convention for two by two tables
should be constructed so that the irst row should refer to those with the exposure of interest
and the irst column should refer to those with the disease.

This type of table cross-classiies exposure status and disease status. The total number of
participants with the disease is A + C and the total number without the disease is B + D. In
addition, the total number with exposure is A + B and those without exposure is C + D. The
entry in cell A refers to those with the exposure and the disease present. The cell B refers to
those with exposure but no disease. Cell C refers to those who were not exposed but have
the disease. And inally cell D refers to those without the exposure and without the disease.

200
Order to Filling in the 2x2 Table by Study Type
In a cohort study design, the total number of study participants who have the exposure (A +
B) and who do not have the exposure (C + D) is illed in irst. In a cohort study design, the
participants are followed over time. Cells (A, B, C, D) are illed in at the conclusion of the
study or when the disease status is determined at some point during the study.

In a case control study design, the total number of study participants who have the disease
(A + C) and without the disease (B + D) is illed in irst. At the conclusion of the study, the
exposure status (A, B, C, D) is illed in.

In the cross-sectional study design, enrolling a sample of participants (N) would be deter-
mined irst. Next each participant’s exposure and disease status would be identiied (A, B,
C, D). The totals (A + B) and (C + D) would be calculated last.

Calculation of Relative Risk

Relative risk is calculated with the following formula:

This formula shows a ratio of the exposed to the unexposed. This is the essence of a relative
risk calculation.

201
Summary
One statistic used by epidemiologist to describe study results is relative risk. This is a rela-
tive measure since it compares the exposed group to the unexposed group. Filling out a
2x2 table is helpful in understanding the calculation. If there is no difference between the
two groups, the point estimate is 1.0. Only statistically signiicant differences between the
exposed and unexposed group will have a conidence interval that does not contain one.

Examples of point estimates and conidence intervals, statistical signiicance

Conidence Statistically sig-
Point estimate Includes 1.0
Interval niicant
1.75 1.2 – 2.3 No Yes
2.80 2.3 – 4.9 No Yes
1.22 0.8 – 1.4 Yes No
0.76 0.2 – 0.9 No Yes
0.37 0.3 – 0.5 Yes Yes
0.94 0.7 – 1.6 No No

Concept Reinforcement
1. What is the purpose of a two by two table?

2. Why do ields need to be illed in in different order depending on study type?

3. What is important about statistically signiicant indings?

202
 Chapter 42 - Odds Ratio
Chapter Objectives
• Construct a two by two table

• Write the formula for determining odds ratio

• Explain the meaning of the point estimate

For case control studies, odds ratio is the statistic used to explain the differences between
exposed and unexposed groups. In other words, odds ratio is the ratio of the probability of
occurrence of an event to the probability of the event not occurring.

Odds Ratio
The odds ratio measures the odds of exposure of a speciied disease. Similar to relative risk,
an odds ratio of 1.0 means that the odds of exposure is equal between cases and controls
and that the risk factor is not associated with the disease. An odds ratio greater than 1.0
suggests an increase risk of the disease based on a particular exposure. In contrast an odds
ratio less than 1.0 suggests a lower risk for the disease based on a particular exposure.

Odds ratios should be interpreted with caution. One reason is that a case-control study is
retrospective by design with only one period of observation. Only prospective studies can
determine rates and consequently risks. The reason that a case-control study cannot de-
termine rates and true risks is based on the way in which the study groups are assembled.
In a case-control study, the total number of participants who are exposed (A + B) and
the total number of participants who are not exposed (C + D), do not represent the total
population of exposed and non-exposed individuals. Consequently there is no appropriate
denominator for the population at risk and therefore no way to directly determine disease
rates. However, with that said, if three conditions are met, the odds ratio provides a good
approximation of the relative risk.

• T
he controls are representative of the target population in the frequency of the
exposure of interest

• The cases are representative of all cases with regard to severity and diagnostic criteria

• The frequency of the disease in the population is small.

203
In the exposed group, the odds of a disease (exposed with the disease divided by exposed
without the disease) provides similar results as risk (exposed with the disease divided by
total number exposed to the disease) of a disease when the proportion of study participants
with the disease is small and those with and without the disease are from the same relevant
population.

Disease Status
Present Absent Total
A (exposure, disease B (exposure, but no
Present A + B
Exposure present) disease)
Status C (no exposure, disease D (no exposure, no
Absent C + D
present) disease)
A + C (total number B + D (total number
Total N (sample total)
with disease) without disease)

Odds Ratio Calculation

The formula for calculating the odds ratio is:

Formula for odds ratio

The important point to remember is that in a well-designed case-control study, the odds
ratio is a good estimate of the risk ratio that could have been obtained from a more costly
and time consuming prospective cohort study.

The same interpretation applies for the odds ratio as it does for the relative risk. A point
estimate gives a single value along a scale. If the odds ratio is 1.0, the odds of the outcome
of interest is exactly equal for both the exposed and the unexposed groups. Therefore, the
exposure did not have any effect on the health outcome. The further the odds ratio is from
1.0, the greater the magnitude of effect. For example, an odds ratio of 2.0 means that there
is a two-fold increased risk for the health outcome in the exposed group compared to the
unexposed group. Less than 1.0 implies a decrease risk. For example a odds ratio of 0.5
means the exposed group is 50% less likely to have the disease as the unexposed group.

204
Absolute Versus Relative Measure
Risk can be measured in two ways, as an absolute measure or as a relative measure.

Absolute Measure
Absolute risk provides the number at risk. It is an actual number. For example, as a simple
example, if 10 women in a town of 10,000 adult women are diagnosed with cervical cancer,
the absolute risk of cervical cancer in the town is 10 per 10,000. However, the people in the
town have different lifestyles, genes, living conditions which absolute risk does not take
into consideration.

With more calculations, the attributable risk can be calculated. Attributable risk means the
number of people who get the diseased due to the exposure. On the lip side, it can also
be used to indicate the number of people who would not get the disease if they were not
exposed. Population attributive risk tells us how much of the disease or health outcome can
be attributed to a speciic exposure in the total population. The population attributable risk
is useful for policy makers.

The population attributable risk depends on the magnitude of the relative risk and the
prevalence of exposure in the population. In other words, if not very many people have
the disease, then the population attributable risk will not be very high. In contrast, com-
mon exposure may account for a large attributable risk. If only 2 people chew betel nut in
a rural city, population of 50,000, and 1000 people are diagnosed with lip cancer from this
population, then the relative risk from chewing betel nuts will be small. Similarly, if only
2 people contract a rare lip disease from a population of 50,000, the prevalence would be
very small. In both of these hypothetical cases the population attributable risk would be
small. In contrast smoking cigarettes is common in men in China (~60%) and the risk of
lung cancer is also high in smokers (95% of lung cancer occurs among smokers). There-
fore, the population attributable risk of lung cancer from exposure to smoking for men in
China would be quite high.

Relative Measure
Relative measures (such as odds ratio or relative risk) indicate how strongly an exposure
is associated with a particular disease. The relative measure does not give an indication of
the impact of the exposure associated with the incidence of the disease in the population.
A large statistically signiicant point estimate does not translate into more people with the
disease; it does mean that the likelihood of disease associated with the exposure is high.

205
Summary
In case-control studies, the common statistic to report study results is odds ratio. This is an
estimate of relative risk. It is a measure of the odds of the exposure of a given disease. Both
relative risk and odds ratio are examples of relative measures since they provide informa-
tion on the magnitude of the risk of disease from the exposure. Absolute measures provide
useful information for policy makers because these statistics indicate the number of people
affected by the exposure.

Concept Reinforcement
1. What is the difference between relative risk and absolute risk?

2. Why is prevalence important to know when calculating absolute risk?

3. Why is it possible to use odds ratio as an estimate of relative risk?

206
 Chapter 43 - Confidence Interval
Chapter Objectives
• Deine conidence interval

• Explain the relation between sample size and conidence interval

• Explain the use of 95% versus other percentages for CI

Sample Size Considerations

How big does a sample need to be? The bigger the sample size, the smaller the random er-
ror and the higher the likelihood of inding an effect. However, the bigger the sample size,
the more expensive (and possibly more time consuming) the study. Researchers weigh
these two competing interests: reduction of random error and resource allocation.

If a sample is too small in a probabilistic sample, the researcher will run into one of two
problems. He/she may fail to ind any meaningful result or not be able generalize the ind-
ings to the group or population.

If the sample size is too big, then:

1. t here has been a waste of resources since the researcher need not have gathered all
that data.

2. t here has been a waste of respondents’ time since some of their data were not
needed.

3. a very large study has the power to ind signiicant results for very small, even
clinically inconsequential, differences.

When iguring out how many people to survey, the researcher realizes that as the popula-
tion size increases, the percentage of people needed to achieve a high level of accuracy
decrease rapidly. In other words, to obtain the same level of conidence in the study results,
for larger population the researcher would need a smaller percentage of people surveyed
and for a smaller population the researcher would need a larger percentage of people sur-
veyed.

207
As noted above, all data have some error (variations). The technique to show the amount of
variation in the data is to report the conidence interval. This means that in the distribution
of data, the lower conidence interval point and upper conidence interval point in which
95 out of 100 samples of the speciied size will be within the speciied percent error of the
study response. A rule of thumb is the larger the study size, the narrower the conidence
interval. A second rule of thumb is the rule of diminishing return. This means that the re-
searcher needs signiicantly more study participants to reduce the margin of error from 2
percent to 1 percent than from 5 percent to 4 percent. Just as a refresher, error is the amount
of data variability around the reported result. Error can be shown as + and – value. For ex-
ample if the researcher says men’s ear’s length at age 90 is 75.7 mm with 95% conidence
interval of 71.2-80.2, that would mean that the error is ± 4.5 mm. The ear lengths could be
as long as 80.2 or as short as 71.2, and in 100 studies 95 out of 100 studies, the men would
have an ear length between 71.2 and 80.2.

95% Versus 99% Conidence Interval

There is not any reason that researchers use 95% CI instead of say 94% or 96%. It is just
the convention. When a study reports 99% CI, this indicates that the researchers want to
be very sure of their results. As a refresher, 99% CI means that in 100 studies, all but one
study would ind the same range of results. That is pretty impressive.

Example of 95% CI

Age-adjusted breast cancer rates with 95% conidence intervals in 15 counties

in Missouri

208
Using breast cancer data for 15 counties in mid-Missouri, the breast cancer rate varied
from 184 per 100,000 people to 88 per 100,000 people. However, the conidence intervals
for these counties mostly overlap each other. For example, Audrain and Boone counties
have slightly different breast cancer rates. Boone County’s conidence interval is com-
pletely within Audrain County’s conidence interval. When conidence intervals overlap,
this means that the results from overlapping conidence intervals are similar.

Summary
Conidence interval is an important concept when analyzing statistical data, including
those generated by epidemiologic studies. Conidence interval is used to show the amount
of variation in the data. The larger the study sample, the narrower the conidence interval of
the study. Additionally, the rule of diminishing returns applies to reducing the conidence
interval. It is easier to reduce the conidence interval from 5% to 4% than from 2% to 1%.
Conidence intervals are typically stated in percentages with 95% conidence interval the
most common. When conidence intervals overlap, this means that the results are similar.

Concept Reinforcement
1. Explain the concept of conidence interval.

2. Describe how sample size and conidence interval are related.

3. Discuss why 95% is used for conidence intervals.

209
210
 Chapter 44 - Evidence-Based Research
Chapter Objectives
• Describe the characteristics of evidence-based research

• Describe the beneits of using evidence-based research

• Describe the 2010 Healthy People Program

Epidemiologists often have many ideas about what to study and what programs may be
helpful. Historically, someone could make a suggestion and launch a program. However,
these days to launch a program more and more funding agencies want the researcher to
show that the program has a good chance of being successful. Funding agencies are re-
quiring that the researcher show evidence of success. This can be evidence from a similar
program, possibly with a different population or it can shown by using evidence from other
research that explore the underlying characteristics of the population.

Evidence-based Research
Evidence-based research is research that builds on the results of prior research and takes
advantage of existing research and knowledge to develop an effective plan for answering
the question of interest.

Evidence-based research is beneicial because it guides researchers as they are developing

new research programs. They are able to look at the results of previous research while de-
veloping new research studies.

Evidence-based research is becoming more and more popular as criteria for funding pro-
grams. It is also being used more and more in the medical ield to determine the best prac-
tices for taking care of patients.

Healthy People 2010

Every 10 years the U.S. Department of Health and Human Services provides science-based
national objectives for promoting health and preventing disease. Each state either adapts
the Healthy People objectives or modiies them for their states.

211
Healthy People 2010 is a list of speciic health goals for the United States for which prog-
ress can be measured over time. In 1980, the irst set of national health objectives was iden-
tiied. At the beginning of each decade, the previous decade’s health objectives are evalu-
ated and new goals are set. Healthy People 2010 is the current set of health objectives for
the nation. In 2009, meetings are being held to create the Healthy People 2020 objectives.

To create the Healthy People objectives, scientiic studies over the previous 10 years are
reviewed, new knowledge gained during the previous decade, trends in health, and techno-
logical innovations are incorporated into the new objectives.

Healthy People 2010 Goals

There are two broad goals of Healthy People 2010:

1. Improve quality of life and increase life expectancy

2. Eliminate health disparities

Disparity deinition

The National Institute of Health deines disparities as the “differences in the incidence,
prevalence, mortality, and burden of disease and other adverse health conditions that exist
among speciic population groups in the United States.”

No “rule” exists to say at what point a difference between the overall average and the race/
ethnic/speciic population average indicates disparity. However, the Urban League of Pitts-
burgh suggested a 30% difference. This amount is more than would likely occur by chance.

The top health outcomes that have signiicant disparities—meaning differences in inci-
dence, prevalence, and mortality-in order are cardiovascular disease, cancer, diabetes,
HIV/AIDS, infant mortality, asthma, and mental health.

Healthy People 2010 Health Outcomes

For Healthy People 2010, 28 health outcomes have been identiied to be measured and
followed to determine improvement in the two broad goals. These health outcomes range
from chronic diseases such as cancer to family planning to food safety.

• Access to Quality Health Services

• Arthritis, Osteoporosis, and Chronic Back Conditions

• Cancer

• Chronic Kidney Disease

• Diabetes

• Disability and Secondary Conditions

• Educational and Community-Based Programs

212
 • Environmental Health

• Family Planning

• Food Safety

• Health Communication

• Heart Disease and Stroke

• HIV

• Immunization and Infectious Diseases

• Injury and Violence Prevention

• Maternal, Infant, and Child Health

• Medical Product Safety

• Mental Health and Mental Disorders

• Nutrition and Overweight

• Occupational Safety and Health

• Oral Health

• Physical Activity and Fitness

• Public Health Infrastructure

• Respiratory Diseases

• Sexually Transmitted Diseases

• Substance Abuse

• Tobacco Use

• Vision and Hearing

Within each of these health outcomes, speciic objectives have been set. Overall, more than
450 objectives have been written.

213
Summary
With the improved technological capacity for rapid dissemination of scientiic indings,
evidence-based research is becoming a requirement by funding agencies. It is also becom-
ing more important in medical care. Along with evidence-based research, is the use of a
unifying framework towards improved public health is the Healthy People 2010 goals. It
is updated every 10 years using evidence of the health of the population as well as innova-
tions from the previous decade.

Concept Reinforcement
1. Explain evidence-based research.

2. State the beneits of evidence-based research.

3. Discuss the goals of the Healthy People 2010 program.

214
 Chapter 45 - Public Policy
Chapter Objectives
• Deine policy

• List ive mechanisms used by government to implement policy change

• Describe Heberlein’s three broad ixes to problems

Policy is a set of underlying principles that are used to guide programs at the municipal,
state, national or international level. In this case, we are interested in policies that inluence
health promotion, protection, and services.

Many public policies are linked to health. Income support (assistance), environmental pro-
tection, housing support, and transportation policies are all examples of policies that can be
linked to improved health of the population.

Mechanism for Implementing Policy

The list of how to implement a policy is quite long. Policy implementation may use several
mechanisms simultaneously depending on the whether the policy is being considered on
a larger scale, whether it is a newly enacted policy, who is responsible for the policy, etc.

One tool to implement a policy is establishing a policy as law. A classic public health
policy is the required use of seat belts by drivers in most states. However, if a law is enacted
and there is not any enforcement or regulation, the compliance may be less than perfect.
Consequently, not only are some public health policies enacted as laws but they also have
an enforcement component. For example, if a driver is seen without a seat belt, then in
many states, the driver can be ined.

Sometimes public policy is supported through taxation. The Medicaid program is a federal
program that supports many health programs. An example of one program supported by
federal dollars, which are really our tax dollars, is the breast cancer and cervical screening
program to women who cannot afford these tests. Another example is the WIC (Women,
Infants, and Children) nutritional program. One part of this program is to provide prenatal
care to pregnant women.

215
hree Strategies for Fixing a Societal Problem
Heberlein, a rural sociologist, wrote a paper in 1974 describing three strategies to ix a so-
cietal problem. He argued that the three ways to resolve any societal problem are through
a technological ix, a cognitive ix or a structural ix. He used the example of intervening
to reduce the pollution from 2-cycle boat engines on an inland lake. The technological ix
is to design an engine that emits virtually no contaminants into the water. The cognitive ix
is to educate the public on the pollution caused by 2-cycle motors and encourage their re-
duced use. The structural ix is removing the boat ramp so that boaters are unable to launch
their motorized boats. This provocative idea can be applied to public health problems too.
However, underlying these ixes is the need for money to implement them.

Delay Between Epidemiology Study Results and Policy Illustration

In a perfect world, policy should be based completely on evidence. However, this is rarely
the case. Epidemiology studies are the source of the evidence considered when making
policy. Consider the following classic epidemiology study.

In 1950, Richard Doll and Bradford Hill published a study in the British Medical Journal
that showed that smokers are at a 50 times greater risk for lung cancer.

By 1964, the Surgeon General announced to headline news that smoking was deleterious
to one’s health. This decision was based on reviewing more than 7,000 scientiic articles,
many of which were epidemiology studies.

By the beginning of the 21st century, places across the United States are banning smoking
in public places such as restaurants, bars, and taverns. Lots of reasons can be cited as to
why smoking is still common when unequivocal detrimental health effects are directly as-
sociated with smoking—both to the smoker, to the fetus, to the family exposed to second
hand smoke. Both social pressure (i.e. to be “cool”) and addiction contribute to people
continuing to smoke. Further, the media and advertising efforts have also inluenced con-
tinued use of tobacco. For example, movies are now featuring cigarette brands using close
up shots and the number of actors smoking in movies has skyrocketed.

216
Precautionary Principle
The precautionary principle was formalized in Racine, Wisconsin at the 1998 Wingspread
Conference. The following statements comprise the reasoning behind the precautionary
principles as determined by the conference attendees. The key elements of the precau-
tionary principle include taking precaution in the face of scientiic uncertainty; exploring
alternatives to possibly harmful actions; placing the burden of proof on proponents of an
activity rather than on victims or potential victims of the activity; and using democratic
processes to carry out and enforce the principle-including the public right to informed con-
sent. Several aphorisms in the English language also support the precautionary principle.
For example, “an ounce of prevention is worth a pound of cure” or a medical aphorism
“irst, do not harm.”

Although this principle seems simple and reasonable, application of the principle has prov-
en to be dificult. Critics of the precautionary principle say that it is impractical because the
launching any new technology has inherent risks and potential for negative consequences.
Additionally, the implementation of the precautionary principle would put additional eco-
nomic burden on the proponents of the proposed activity.

Summary
The concept of public good is pervasive in health policies. Some polices target those in
need. To participant in a federal or state sponsored program almost always is based on
inancial need. Most laws related to public health have the scales tipped towards the good
of all whereas the rights of the individuals are considered less important. A once powerful
concept, called the precautionary principle, argued that when in doubt about the harm that
something might inlict on the population, a more conservative approach should be taken.

Concept Reinforcement
1. Of Heberlein’s three ixes, which one do you think is most effective for public
health issues and why?

2. What basic tenet of Americans is in conlict with the Precautionary Principle?

3. To what degree should your rights as a citizen be infringed upon for the good of
all? For example, should you be required to get a vaccine to go to school? Is there a
clear line between the rights of the individual versus the public good? Who should
decide what is suficient?

217
218