Measures of Disease Occurrence

and association

1
• Basic activity of epidemiology is measuring
disease occurrence. The two main measures
are Incidence and Prevalence
• Much of epidemiology concerned with
discovering etiology (causes)
• Diseases are multi-factorial
• Quantification of the level of increased risk by
measuring the effect measure.
• 3 main measures of association
» Relative risks
» Odds ratios
» Attributable risks 2
 Measures of occurrences
• Quantitative measures
• Expressed as : counts, proportions, ratios, or
rates
Measles Male Female Total

Yes a b a+b
NO c d c+d

3
 Which measure of disease occurrence?

• Assessing occurrence of an exposure, development of a

disability , and frequency of medical intervention, etc

• The type of measure depends on the event studied and

purpose of the measure

• Incidence: risk of disease development. Concern about

primary prevention
• Prevalence : burden of disease

4
 Measures of incidence
• In an outbreak investigation, 590 cases of Salmonella
occurred in a city with population of 800000. The
incidence rate is 73 per 100000 population.
• Among statin users, 9% (24 of 263) developed (Atrial
Fibrillation) AF during a total of 1,352 person-years of
follow-up. In contrast, 15% of nonusers (28 of 186)
developed AF during a total of 747 person-years of follow-
up.
• Corresponding incidence rates of AF per 1,000 person-
years were 18 (95% CI 11 to 26) and 37 (95% CI 26 to 54),
respectively (Yinong et all 2003).
5
 Measures of Incidence
I. Cumulative incidence = Incidence proportion
• It is the proportion of the initial population that develops
a given outcome
# new occurences during a specified period of time
CI   10 n
pop at risk

• Population at risk: population at risk refers to a

population or cohort who are at risk (susceptible) of
developing the outcome at the beginning of specified
period. PAR
• 10n : Population Base
• Occurrences are people who develop the outcome and
not the individual incidents 6
 Cumulative incidence
• Measures individual’s average risk of developing the outcome over a specified
period of time , assuming all persons were followed upon for the same period of
time
• CI best used in closed cohort or fixed population , and where loss to follow up and
deaths are negligible, follow up is short
• For significant loss to follow up or deaths due to competing risks, and for dynamic
population

# new occurences during a specified period of time

II .CI   10 n
• C: censored observation due to loss to follow up, death, or whose
c  follow up time is shortened
because of late entry into studypopulation at risk   
•
2  period
Assumption is that censoring occurred at the mid point ofstudy
• Attack rate is a measure of cumulative incidence, calculated in connection with outbreaks (or
epidemics ) of diseases

7
 Incidence density
• Open cohorts
• Different lengths of follow up (competing risks, loss to
follow up, dynamic cohort)
• It is a rate vs risk for CI
# new occurences during a specified period of time
• ID 
Total person  time units observed
 10n

• The denominator is the sum of known outcome free time

contributed by each of the subjects: person years, person
months, person days.
• Assumptions:
– risk of outcome is constant during the specified period of time;
– censored observations do not differ from non censored regarding risk of
outcome;
– no secular trend are observed.
8
 Incidence density
• A study of prostate cancer was initiated in IOWA. A total
of 1000 man, 55-64 years of age , with no prior evidence
of prostate cancer were enrolled in the study. The men
were followed by investigators for four years. Every
year, men were examined and tested for presence of
prostate cancer. The results are as follows
• 10 cases were confirmed at first exam; 15 additional
cases at second exam, 20 additional cases at third
examination, and 25 cases confirmed at fourth exam.

• Calculate ID of prostate cancer

9
 Estimating total person time units
• In large cohorts, calculating total person time units
would be time consuming
• Alternative formula with the assumption that study
outcome and censored observations are uniformly
distributed and occur on average at the midpoint of
the specified study period
# new occurences during a specified period of time
ID   10n
Z
( Pop at risk - )  specified time period
2
• Solving for the prostate cancer example:

10
 Incidence odds
• Cohort of 450 children followed up to determine
incidence of acute bronchitis
• 22 developed acute bronchitis for the first time
during follow up period
• Incidence odds is
22
 0.051
(450  22)
• The CI = 22 / 450 = 0.049
• Incidence odds equal to CI when outcome is rare
( less than 0.10)
11
 Prevalence
• Number of individual in a population with a disease or
other personal attribute. It is a “status report”of a
population regarding a specific outcome
• Prevalence rates calculated as incidence rates, often
expressed per multiple of the population
• Technically, it is not a rate , it is a proportion as no time
element involved.
• A questionnaire distributed to the workforce of a certain
industry on a particular working day. Of the 1534
workers, 178 reported headaches on that day.
• Prevalence rate or proportion is

Prevalence 178
 0.12or12 per100wor ker s 12
1534
 Measures of disease prevalence

• Point prevalence: at the time of the survey “The

point prevalence of smoking among those aged 60–64
years was 38%”( UHS 2003) .
• A prospective wartime study in four Lebanese
communities found a lifetime prevalence of 27.8% for
major depression.
• Karam et al 2007 reported that 308 (17·0%) of
respondents met criteria for at least one 12-month
DSM-IV/CIDI disorder.
Cumulative/ life time
Period prevalence
prevalence 13
 Prevalence Odds
• 35 people with a given disease in a population
of 500 on a given day:
35
• The prevalence odds is :  0.075
(500  35)
35
P 0.07
• This is equivalent to :  500   0.075
1 P  35  0.93
1 
 500 

14
 Relationship between incidence
and prevalence
• In a stable population, increase in incidence will result
in increase prevalence BUT
• Process mediated by two factors: disease resolution
and disease duration
• With advances in medical treatment , we have an
increase in prevalence.
• If population stable, and incidence and prevalence
are unchanging, (steady state condition),
• Prevalence= Incidence * Duration

15
Factors affecting prevalence

Faster recovery Slow recovery time

High mortality Lower mortality
Decreased incidence Increased incidence
Influx of non cases Exodus of non cases
Exodus of cases Influx of cases
Improve cure rate Better reporting

16
 General characteristics of
incidence and prevalence
Cumulative incidence Point Prevalence

Dynamic concept indicating Static concept indicating

change in disease status over existing status at a particular
time time

Good estimate of risk of An unreliable estimate of risk

disease

Unaffected by recovery or Affected by recovery or death

death

17
 Appropriate measures of disease occurrence

Disease Examples Appropriate measures

characteristics
Clearly defined onset Acute appendicitis Incidence
Single episode/possible Colon cancer
multiple but infrequent Major trauma
episodes, short duration
Terminated by death, Myocardial infarction Incidence
spontaneous resolution or
therapy Influenza
Fracture
+chronic relatively stable or Insulin dependent diabetes Incidence and point
requiring long term therapy Renal failure prevalence
Ill defined onset, relatively Hypertension
stable Deafness Point prevalence
osteoarthritis
Ill defined multiple episodes Asthma
Migraine Point and period prevalence

18
 General health and population
indicators
• Special types of incidence rates: Mortality rate , case
fatality rate, attack rate

• Crude vs specific vs adjusted (standardized rates)

• Proportionate mortality ratio

• Case fatality rate

19
Years of potential Life lost (YPLL)
• Measures impact of premature death on
society
• Used to establish public health priorities
• Weighted towards death at younger ages

20
YPLL example
Standard-life to which all preventable deaths are reflected
 
70 y

A = Sam died of coronary heart attack at age of 55 years

 
Sam’s PYLL = 70 -55 = 15 years

B = Jhon died of alcohol poisoning at age of 28 years

 John’s PYLL = 70 – 28 = 42 years

C = Mary died of stroke at age of 71 years

 
Mary’s PYLL = 70 – 71 = 0 years
 Confidence intervals of measures
of occurrences
• It means we are 95 % sure that the population
parameter lies within the range of value s
indicated by the interval
• A confidence interval is affected by sample
size, variance and confidence level.
• Formula by Alan Agresti and Coull

95%CI  p  1.96 p(1  p ) / n  4

22
 Which measure of association?
• Relative risks
– Risk ratios: Ratio of prevalence or cumulative incidence in two populations (cross-
sectional study)
– Rate ratios: ratios of incidence density rates (cohort study)

• Relative Odds or Odds ratios: Ratio of odds of exposure (case control

study).
– “Significantly elevated odds ratios (OR) of mood, anxiety, and impulse-control
disorders were associated with two (OR 2·0–3·6) or more (2·2–9·1) war-related
traumatic events: Karam et 2007)

• Risk or rate difference: difference between the risk in exposed

population and that in the unexposed population
– Attributable risks
– Population attributable risks
EPHD320-Chaaya
23 2008
• Illustration : Effect of viewing smoking in movies on
adolescent smoking initiation: a cohort study.  The
Lancet, Volume 362, Issue 9380, Pages 281-285

• Adolescents in the highest quartile of exposure to movie

smoking were 2·71(RR) (95% CI 1·73–4·25) times more likely to
initiate smoking compared with those in the lowest quartile.
The effect of exposure to movie smoking was stronger in
adolescents with non-smoking parents than in those whose
parent smoked. In this cohort, 52·2% (30·0–67·3) of smoking
initiation can be attributed to exposure to smoking in movies.
(Dalton) et 2003
24
 Which measure of association?
a. prospective studies
• Framingham study: Annual incidence of myocardial infarction for
individuals with severe systolic hypertension (exposed:180 +
mmHg) and normal systolic blood pressure
(unexposed:<120mmHg)

Myocardial Infarction

Systolic Ht # Present Absent Probability Odds

Yes 1000 18.0 982.0 0.018 0.0183

No 1000 3.0 997.0 0.003 0.0030

25
 Myocardial Infarction

Systolic Ht # Present Absent Probability Odds

Yes 1000 18.0 982.0 0.018 0.0183

No 1000 3.0 997.0 0.003 0.0030

Relative risk (risk Ratio)? Odds ratio ?

18 0. 0183
RR   6.0 OR   6. 1
3 0.003
26
 Relationship between Relative Risk and Odds Ratios

I. Probability (incidence) of event is low or disease is

rare

P
 P, Thus RR ~ OR
1 P

How rare? If proportion of disease is less than 5 %

27
 II. Probability (incidence) is high
Example: Incidence of local reactions in persons aged 40 + years old

Local Reaction

Vaccine YES NO #

Yes 65 (25.3%) 192 257

No 17 (7.1%) 224 241

Relative risk (risk Ratio) ? Odds ratio ?

EPHD320-Chaaya 2008 28
 Measures of association: b. Retrospective studies

• In both prospective and case control studies,

A  D 18  997
OR    6.1
BC 982  3
Reporting results as OR of disease

29
In retrospective (case control) study, a sample of cases and a sample of
controls yields an unbiased estimate of odds ratios.

Example: Case control study to examine the relationships of local reaction (cases) to
influenza vaccination (exposure)
Choose a 50 % sample of cases of local reaction and controls without local reaction
True population OR= 4.5 ; OR in 50 % sample = 4.3
( it is not exactly 4.5 because of rounding)

Vaccine Cases Controls

Yes 65 x 0.5= 33 192 x 0.5=96

No 17 x 0.5= 9 224 x 0.5 = 112

30
It is not necessary that the same sampling fraction apply to cases and
controls. For example, all cases and a sample of 20 % controls could
be chosen. ( as cases are usually less frequent, the sampling fraction
for cases is usually greater than that for controls).

OR = 4.5

Vaccine Cases Controls

Yes 65 x 1.0 = 65 192 x 0.2 = 38

No 17 x 1.0 = 17 224 x 0.2 = 45

31
 In a retrospective (case control ) study, when the control group is a sample of the
total population (rather than only of non cases), the relative odds of exposure is
the relative risk !!

Risk factor Cases Non cases Total

population

Present A B A+B

Absent C D C+D

A A A
Oddsof exp incases Odds exp incases
OR  C OR   C  A  B  RR!!
Oddsof exp innoncases B Odd sin pop A B C
D CD CD

32
 To Summarize, in a case control study

What is the What is calculated To obtain

control
group
Sample of OR=oddsexp cases/ oddsexp noncase OR disease
non- cases (if disease is rare,
OR will be equal
to RR)

Sample of OR=odds exp cases/ odds exp pop RR

total
population

33
 Calculation of Odds Ratios when there are more then two exposure
categories

Cigarettes/day cases Controls Odds Ratios

<10 1071 2414 1.0

11-20 203 420

20-30 55 86

31+ 40 48

Total 1369 2968

EPHD320-Chaaya 2008 34
 Odds Ratios Vs. Relative Risk: Advantages

• OR can be estimated in case-control studies;

• OR can be estimated from a logistic function
• OR of event is the reciprocal of RO of non-event
OR of local reaction in vaccine vs. placebo=4.5
RO of local reaction in placebo vs. vaccine= 1/ 4.5
= 0.222

35
 RR & OR Vs P- Value
• A large RR or OR does not necessarily mean
that the P value is small
• The large RR or OR could have occurred by
chance if the sample size is small
• P value depends both on the magnitude of RR
or OR as well as on the sample size

36
 Attributable Risk (Etiologic Fraction)
 Attributable Risk in Exposed: excess incidence rate due to the
association with the risk fator

AR  I exp  I un exp
 Percent AR exp: Proportion of the incidence among those with the
risk factor due to the risk factor

I exp  I un exp RR  1
AR   100   100
I exp RR

37
Example: Iexp = 0.2, , Iunexp – 0.1 ; RR = 2.0

% ARexp?
2.0  1
AR%   50%
2

50 % of the disease X could be attributed to the exposure

38
 Population Attributable Risk
b   RR  1.0 
ARpop 
b  ( RR  1.0)  1.0

b = Prevalence of exposure in the population ( can sometimes

be estimated by the prevalence of exposure in the controls,
if the control group is representative of the reference
population)
RR= Relative risk , estimated in case control studies by the
Odds ratio
39
 Absolute risk reduction

• The absolute difference between the risk of

the event in the control and experimental
groups.
• ARR = CER - EER.

• ARR can be used to calculate the number

needed to treat (NNT).
• Use this term if the event is bad e.g. death.

40
 Number needed to treat.

• The number of patients who needed to be

treated to prevent the occurrence of one
adverse event (e.g. complication, death) or
promote the occurrence of one beneficial
event (e.g. cessation of smoking).

• NNT = 1/ARR.

41
Consider the following 2 scenarios:
A. Incidence (I) in experimental 3% vs. 4% in control
B. Incidence (I) in experimental 30% vs. 40% in control

RR ARD NNT*
(Absolute risk (number needed
difference) to treat)
Icontrol – Iexp 1/ARD

A 0.75 25% 1% 100

B 0.75 25% 10% 10

*NNT: The number of patients you would need to treat so

that one patient would benefit.

42
 Evaluation of presence of a valid association
1. Chance:
The role of random variation from one sample to another. One
major determining factor is the sample size, n=10, n=100, less
variability in the larger sample studies.
 
To quantify the degree to which variability may account for the
results, conduct appropriate statistical tests (t-tests, Chi-2 tests..)
and measure associated P-values

By convention 0.05 = means that there is no more than a 5% or

1/20 probability of observing a result as that observed or more
extreme due to chance (<0.05 significant results)

One problem, p-value depends on the magnitude of the difference

between the groups and the sample size.
Better to use CI. It provides all the information of the p-value, in
addition the effect of the sample size can be ascertained from the
width of the CI. The narrower the CI, less variability, the stronger is
the confidence in the results. 43
 Evaluation of presence of a valid association (cont’d)

2. Bias
In design or conduct of the study has introduced a systematic bias

Selection bias – noncomparable selection of cases vs. controls or

exposed vs. nonexposed (at start of study: issues of generalizability or
external validity or during the follow-up: issues of internal validity)

Information bias – noncomparable abstraction of information

(instrument/ interviewer/ee bias – recall bias)
 

3. Confounding
Mixing up with a third variable
Control: Design / Analysis

4. Judgment of a cause-effect relationship

 
44