Designing a study II

Research Methods Dent 313

Types of randomized control trials

Simple randomized design Crossover design Factorial studies

Simple randomized design

Randomization to two or more groups Simple and powerful When enough subjects are available and can be recruited

Crossover design

Subjects randomized to one study group After a specific period of time, the same subjects are switched to the other group Advantages

Gives two subjects for the price of one Less variability and more power Each subject serves as his/her own control Increase subject motivation

All subjects will be in treatment group at some stage

Good when you cannot recruit enough subjects

Crossover design

Disadvantages

Bias due to carryover effects leading to failure to ascribe successes or failures to correct group

Carryover effects are due to the 1st treatment but occur

during the 2nd treatment E.g. subject receives antibiotic A for 3 month then Antibiotic B for the next 3 months. Infection appeared in month 4. Is it because antibiotic B failed? or infection appeared in the period of antibiotic A but did not manifest itself until the period of antibiotic B To overcome carryover effect

Washout period: a time with no treatment

Factorial studies

Designed to answer more than one question by randomizing each subject to more than one condition Advantages

Get two studies in the price of one Cost effective

Disadvantages

1st Q: Lack of oral hygiene on caries 2nd Q: Lack of oral hygiene on periodontal disease

Methods of allocating subjects within a randomized design

Randomization with equal allocation Blocked randomization Randomization with unequal allocation Stratified randomization

Randomization with equal allocation

Equal number of person to each treatment group Standard method with highest power Needs enough number of subjects in each group (>20) to minimize the effect of chance in having 2 equal groups in the sample when they are not equal in the population Conditions of subjects have to be similar as well. Thus equal allocation in number is not by itself sufficient

Blocked randomization

Only used

When exact number in each group is needed but the study is too small (4 or 6 subjects per block) In larger studies when temporal changes affecting study enrollment are expected

Enrollment at different times of people with changing

In large multicenter studies

Assignment of subjects is randomized

Blocked randomization

Disadvantages

Staff may figure out the assignment of a subject prior to enrollment (in non-blinded studies) Esp., when all but the last subject of a block have been enrolled

E.g., 4-subject block in two groups A, B. 3 have been

randomized as ABB, thus the last one is to be assigned (not randomized) to A for the two groups to be equal

Overcome by randomly choosing among different size blocks so that staff do not know the size of the block within which the subjects are being randomized

Randomization with unequal allocation

E.g., two-to-one randomization Advantages

Subjects of serious diseases may benefit from unequal allocation

>50% chance of receiving the new treatment when allocated to the larger group

More knowledge about side effect when allocating >50% of subjects to the treatment group Losing power Harder to reject the false Null hypothesis Inconsistency with equipoise principles

Disadvantages

Investigator beliefs that the 2 groups are equal Investigators may believe that one group is superior to the other

Stratified randomization

Preferred when an equal distribution of baseline prognostic factors is needed

Baseline prognostic factors

Sex and age

Unequal distribution of baseline factors may lead to confounding Avoided by randomizing persons within groups of important baseline factors Advantage: variability is decreased, thus power increased Disadvantage: only possible with one or two associated baseline factors

Types of observational studies

Cross-sectional studies Prospective cohort studies Case-control studies Nested case-control studies Ecologic studies

Cross-sectional studies

Easy and fast Information is collected from subject at a single point of time Used to answer descriptive questions

What is the prevalence of a disease? Prevalence is the proportion of individuals in a population who have a specific disease or condition at a particular moment of time

Used to determine frequency of risk behavior Useful in estimating sample size

Cross-sectional studies

Not good in answering analytic questions

An association found may go in either

Risk factor may cause the outcome or vice versa Effect-cause or reverse causality E.g., either direction: alcohol and depression E.g., one direction: smoking & facial wrinkles

Prospective cohort studies

Sample is assembled prior to development of the outcome and followed over time Subjects are evaluated to make sure that they do not already have the outcome being studied Provide much stronger evidence in support of a causal relationship Reduce the possibility of reverse causality Minimizing recall bias

Information about risk factor is collected ahead of

disease development Recall bias is a problem with case control studies developing disease make subject remember an exposure

Prospective cohort studies

Can be used to calculate incidence rate Incidence rate is the number of new cases of a particular condition in an atrisk population per unit time Longitudinal study Length of follow up time is based on how long it takes to develop the disease

Prospective cohort studies

Disadvantages

Take a long time to perform esp., when the disease develops slowly Costly and inefficient for studying uncommon diseases (fewer persons will develop the disease) Bias due to loss of subjects to follow up Period of temporal changes may influence results

Introduction of new instruments Change in clinical practice

Answer of research question may become less relevant when the study is complete

Case-control studies

Subjects are assembled based on whether they have experienced the outcome (cases) or not (controls) Frequencies of risk factors are compared between cases and controls

Case-control studies

Advantages

Efficient especially for studying uncommon diseases

Cases and controls must originate from the same population Disadvantages
Cannot be used to determine prevalence and incidence Selection bias: loss of cases/controls prior to their selection (a case died prior to assembly of cases, then the sample wouldnt be representative) Recall bias

Cases are more likely to remember exposures than controls E.g., cases with cancer may report previous exposures because they have been more aware about their health and subjected to many previous tests

Case-control studies

Can be matched and unmatched Matching

Individual matching

Each case is individually matched with one or more controls E.g., 45 yrs old man as case matched with 45 yrs old
man as a control

Frequency matching

Controls are matched to cases as a group Similar distribution of cases and controls on each matched variable E.g., males with range of 20-40 yrs account for 30% in
both groups

Case control studies

Advantage of matching

Eliminate confounding Increasing difficulty and cost of identifying controls esp., with limited number of potential controls

Disadvantage of matching

E.g., 45 yrs old male with prostate cancer

Matching for a variable will not enable to study its impact on the outcome
E.g., matching for smoking to study the effect of diesel fumes on lung cancer

Best to avoid matching except in small studies where it is difficult to adjust statistically for all possible confounders unless if matching is used

Case-control studies

How many controls per case to enroll

Greatest efficiency with equal number Adding additional controls when

enough cases cannot be obtained such as in rare

Nested case-control studies

A case-control study where cases and controls are drawn from the subjects enrolled in a prospective study

Cases and controls from same population Information on risk factors has been collected prior to development of disease no recall bias When the outcome is death (you cannot examine the dead) Not viable unless information about risk factors is collected at the beginning of the prospective study

Consider banking serum and cells at the beginning of

Ecologic studies

Collect data at the aggregate rather than individual level Aggregate levels

Neighborhood, city, state, country E.g., Water fluoridation on frequency of dental visits no data were collected from individuals When data do not exist on individual level When the primary focus is the well-being of an entire community Best to generate hypothesis which can be tested by other study design

Used

Specifying a hypothesis

What are you hoping to prove before data collection Hypothesis

Null form

There is no difference
There is a difference

Alternative form

Study hypothesis is stated in both the null and alternative forms

Statistical analysis is based on inferential reasoning

Inferential reasoning

Assessing the probability that an association found in a sample could have occurred by chance if there were no true association in the population If the probability that the association could have occurred by chance falls below (p<0.05), null hypothesis is rejected and alternative hypothesis is accepted

A study with more than one question

Collecting data on more than one outcome

Collecting data on additional risk factors for the same outcome does not answer multiple questions

Multiple outcomes

Different stages of the same disease process

Smoking on angina, MI and death

Different disease processes influenced by the same risk factor Outcomes unrelated to one another

Smoking on gingival health, lung cancer and heart disease

Sample size

Depends on the statistical test used Discussed later

Ethical approval

Human research committees

Protect the rights of subjects Insures that the subjects fully informed Insures that subjects have consented to
participate Insures that the risks are reasonable an much less than the new knowledge/ benefit that the study will provide Insures that confidentiality is maintained