Feldmand. Module 3 Study design

Question 1

Experimental Studies

Answer 1

Non-observational or experimental studies are intervention studies and the basic principle is that the design of the study involves deliberately changing population parameters and assessing the effect.

• Laboratory experiments
– Researcher designs and implements study
• Clinical / Field trials
– Therapeutics
– Vaccine efficacy
– Health promotion programs
• Community intervention trials – Insecticide impregnated bednets

Question 2

Observational Studies: Differences between descriptive and analytic

Answer 2

• Descriptive
– Estimate disease frequency or time trend
– Used to generate hypotheses
– No comparisons or statistical analyses
– Correlational studies, case reports and series

• Analytic
– Identify risk factors for disease
– Estimate and quantify effect
– Suggest possible interventions

Question 3

Types of Analytic Observational Studies

Answer 3

Most common designs you’ll encounter in observational epidemiology (1 -3)
1 • Case-control: Retrospective or prospective – Selection based on case status
2 • Cohort: Retrospective or prospective – Selection based on exposure
3 • Cross-sectional - snapshot

4 • Ecologic – individual level data missing

Question 4

Case-control Studies

Answer 4

• Study participants selected based on disease status: Exposure status in diseased individuals (cases) is compared to exposure status in well individuals (controls)
• Most commonly conducted observational analytic epidemiologic study

Question 5

When should we do a case-control study?

Answer 5

• Can be a useful 1st step to evaluate relationship between an exposure and disease
• When the disease is rare
• When efficiency is needed (i.e., there isn’t time to do a cohort study)

Question 6

Selection of Participants in case study

Answer 6

• Cases: Many sources, including:
– Hospital patients
– Clinical practice
– Diagnostic labs
– Disease registries 

• Controls: The comparability of cases and controls is essential
– Controls should be chosen from population which gives rise to cases
– Can be very tricky!

Question 7

Control Selection - Matching (case study)

Answer 7

• Select controls so they are “matched” to cases on certain characteristics, such as age, race, sex, socioeconomic status and occupation
• Done to adjust for the effect of the “matched” variable
• Frequency matching (aka group matching): select controls so proportion of controls with a certain characteristic is identical to proportion of cases with same characteristic. E.g., If cases are 25% male, make control population 25% male
• Individual matching (aka pairwise matching): select a control for each case, so that case and control are similar with respect to variable(s) of concern. E.g., Match 25 y.o. female case with 25 y.o. female control. Pairwise matching requires a matched analysis
• Potential limitation is that you can no longer assess the effect of the matched variable because cases and controls are deliberately manipulated so there is no difference between the two groups

Question 8

Case-control Studies - Analysis

Answer 8

• Because the participants are selected based on disease status, there is no denominator population with which we can calculate risk
• We estimate risk by comparing the odds of exposure among the diseased to the odds of exposure among the well

Odds ratio = ad / bc

Remember that the reduced formula for the OR ad/bc is the CROSS-PRODUCT ratio in a 2-by-2 table

Question 9

Case-control Studies - Bias

Answer 9

• Case-control studies are particularly prone to bias
– Selection bias: A problem with who gets into your study. Can happen when inclusion of cases or controls somehow depends on exposure of interest
– Recall bias and limitations of recall: When cases and controls recall their experiences differently • It may be difficult to remember things that happened in the past or the information might not exist

Question 10

Case-control Studies – Strengths and Limitations

Answer 10

• Strengths
– Relatively quick and inexpensive
– Particularly well-suited to diseases with long latent periods
– Optimal for evaluating rare diseases
– Can examine multiple etiologic factors for a single disease

• Limitations
– Inefficient for evaluating rare exposures
– Cannot directly compute incidence rates of disease in exposed and nonexposed individuals, unless study is population based
– May be difficult to establish a temporal relationship between exposure and disease
– Is particularly prone to bias, in particular selection and recall bias

Question 11

Cohort Studies

Answer 11

• Study participants selected based on exposure status: Risk of disease in exposed individuals is compared to risk of disease in unexposed individuals
• Often conducted after hypothesis has been explored in case-control study

Question 12

Cohort Studies – Selection of Study Population

Answer 12

Select groups on the basis of whether or not they were exposed: – E.g., To assess effects of an occupational exposure such as a solvent, the exposed group is the group that works with the solvent and the unexposed group might be office staff at the parent company

OR

Select a defined population before any of the members become exposed or before the exposures are identified:
– Selection could be made on basis of some factor not related to exposure, such as community of residence
– Take histories of, or perform blood tests or other assays on, the entire population
– Using the results of the histories or the tests, the population can be separated into exposed and nonexposed groups

Question 13

Retrospective vs. Prospective Cohort Studies

Answer 13

• Cohort studies may be prospective or retrospective, depending on temporal relationship between initiation of study and disease occurrence

• Retrospective cohort
– All relevant events (exposures and outcomes of interest) have occurred when the study is initiated

• Prospective cohort (aka concurrent cohort or longitudinal studies)
– Relevant exposure may or may not have occurred when study initiated but the outcomes have definitely not yet occurred
– The cohort must be followed into the future

Question 14

Cohort Studies - Analysis of RR

Answer 14

• If a positive association exists between exposure and disease, then the proportion of the exposed group that develops disease (risk in the exposed group) is greater that the proportion of the nonexposed group in which disease develops
Riskexposed > Riskunexposed
RR > 1

Question 15

Cohort Studies - Bias

Answer 15

• Bias in assessment of the outcome: The person who determines the disease status of each subject may be biased by knowing the study hypothesis and the exposure status of each subject
• Information bias: If the quality and extent of information obtained is different for exposed and nonexposed subjects, bias can be introduced
• Bias from nonparticipation: Generally, only a proportion of those who are eligible to participate actually do so. How do those who do not participate differ from those who participate?
• Bias from loss to follow-up: The incidence rates (risk) for exposed and nonexposed groups may be difficult to calculate

Question 16

Cohort Studies – Strengths and Limitations

Answer 16

• Strengths
– Temporal sequence between exposure and disease can be clearly established
– Well suited for assessing the effects of rare exposures
– Allow for the examination of multiple effects of a single exposure
– Allow direct measurement of incidence of disease in exposed and nonexposed groups

• Limitations
– Inefficient for the evaluation of rare diseases
– If prospective, can involve following large numbers of individuals into the future and are therefore very resource intensive
– If retrospective, requires availability of adequate records
– The possibility of loss to follow-up in cohort studies

Question 17

Cross-Sectional Studies

Answer 17

• A ‘snapshot’ of exposure and disease is determined because both exposure and disease outcome are determined simultaneously for each subject

• The cases that are identified are prevalent cases, because they existed at the time of the study
– Therefore, this design is aka prevalence study
– When the outcome is serologic evidence of disease, this study design is called a seroprevalence study

Question 18

Cross-Sectional Studies – Strengths and Limitations

Answer 18

• Strengths
– May be first step in assessing an association
– Efficient

• Limitation
– Cannot determine a temporal relationship: Therefore cannot make conclusions about causality
– Because cross-sectional surveys consider prevalent rather than incident cases, the data reflect determinants of survival as well as etiology

Question 19

Ecologic Studies

Answer 19

• May be a first approach in determining whether an association exists
• Rather than studying characteristics of individuals, characteristics of groups are studied
• The results of ecologic studies should be interpreted with caution, but can be useful preliminary indicators to causal hypotheses that should be tested more thoroughly – Ecologic fallacy: the erroneous assumption that associations seen at group levels can be applied to the individual
• Efficient study design because does not require follow- up or direct contact with individual subjects

Question 20

Study validity

Answer 20

• Internal
– Whether the observed result is true
– True only if following have been eliminated
• Bias (systematic error)
• Confounding (3rd var confounds association)
• Random error (probability result due to chance)
• External
– Generalizability

Question 21

Precision vs. accuracy

Answer 21

• Precision
Yes
No
.. .. . ...
. .. .. . ...
No
Precise
– –
AKA reliability, reproducibility Extent to which repeated
measurements of a relatively stable phenomenon fall closely to each other
  • Accuracy
– AKA validity
– Degree to which the results of a measurement correspond to true state of phenomenon being measured

Question 22

How to design a study and what aspects to consider?

Answer 22

• Identify research question and refine to testable hypothesis

• Choose design appropriate to question
– Efficiency
– Population
– Unit of analysis

• Consider
– Sample size (power)
– Sampling scheme
– Choice of statistics
– Resources (time, money, personnel) 
– Ethics
– Data generation and management