Epidemiologic STUDY DESIGNS
Types of Epidemiologic Research
Experimental (implies randomization)
Community intervention trials
Randomization - groups under study will be similar on measured and unmeasured variables
· Strength of the association
· Dose-response effect
· Lack of Temporal Ambiguity
· Consistency of Findings
· Biological Plausibility
· Coherence of Evidence
· Specificity of the Association
Different ways to describe study designs (other than experimental vs. observational):
Deals with timing of investigation of exposure and outcome
· Exposure known prior to disease Forward directionality
· Exposure determined AFTER disease Backward directionality
· Exposure and Disease determined simultaneously Nondirectional
Relates to whether the study is initiated before or after the outcome.
Retrospective the disease/outcome has already occurred when the study begins
Prospective the disease/outcome occurs sometime after the initiation of the study
be a mixture or retrospective and prospective (Mixed timing)
We now go into the specific of the major study design:
Assess therapeutic effects/health benefits
Intention to treat analysis (analyze-as-randomized, effectiveness) vs. Treatment-received analysis (efficacy)
Observational; Can be prospective or retrospective
Which do you think usually provides more accurate information, prospective or retrospective?
Ways in which a cohort study might be designed:
1. Sample population, divide into exposed and nonexposed
2. Identify an exposed population find an external, nonexposed group for comparison
Advantages of cohort studies:
Of observational studies, tends to be least prone to bias
Can address several diseases in same study
With retrospective, can be relatively low cost, quick
Can be used with rare exposures
Disadvantages of cohort studies
Loss to follow-up a potential problem
Usually larger sample size than case-control
Prospective can be expensive & time-consuming
Prospective inefficient with rare diseases, diseases of
Identify cases, then sample those without disease (controls)
Ideally should be from a clearly defined source population
Controls should also be from the same source population
Can be retrospective, prospective, or mixed
Case-Control Studies - continued
Relative to cohort, quicker
Usually smaller sample size
Better with rare diseases than cohort
Better with diseases of long latency periods
Can evaluate multiple exposures
Only one outcome can be studied (in contrast to cohort)
Does not allow direct assessment of risk
More susceptible to certain biases
Not efficient for studying rare exposures
Use of incident vs. prevalent cases in case-control studies
Incident cases preferred
With prevalent cases, exposure-disease relationship may be affected by prognostic indicators of duration of disease.
Choosing controls in a Case-Control Study
Controls should be from same source population as cases.
Population-based vs. hospital/clinic-based
Hospital vs. Population-based Case-control studies
Hospital Advantages (relative to population-based)
· Controls easily accessible
· Less expensive, less time consuming
Hospital Disadvantages (relative to population-based)
· Controls less representative of source population
· Controls are ill; possibly ill due to exposure of interest
· Reye syndrome relation with aspirin
· Creutzfeldt-Jakob Disease relation with animal products
Most frequently used study design.
Provides a snapshot of a population
Variables measured at a single point in time
· NHANES National Health and Nutrition Examination Surveys
· BRFSS Behavioral Risk Factor Surveillance Survey
· EPI Expanded Program on Immunization Surveys
· MICS UNICEF Multiple Indicator Cluster Survey
The target population could be the general population, individuals belonging to an organization (Medical Society), patients at a hospital (patient satisfaction survey), etc., etc., etc.
· Tends to be an inexpensive study design
· Can assess several exposures and diseases simultaneously
· Can generate hypotheses
· Cannot establish whether exposure preceded disease
· In general, deals with prevalent disease (potential for survivor bias)
· Need large sample sizes for rare conditions
Example of a Cross-Sectional Study Peripheral Vascular Disease in Scotland
2 types described both are variants of population-based case-control study: Case-Cohort and Nested Case-Control
Start with a cohort study; identify those with disease (cases) at a latter point in time, select subsample from nondiseased as controls using exposure information collected at the start of the study.
Example: gastric cancer and H. pylori infection
· Cohort of 9,775 men in Taiwan with blood samples at baseline.
· During study, 29 developed gastric cancer
· Selected 220 controls from those with no gastric cancer
· Compared frequency of H. pylori infections at baseline.
Nested Case-Control Study
Variation of Case-Cohort. Identify cases as they occur, select control(s) for each case. Also called density-type case control study.
Example of nested case-control: Cohort study of 5000 telephone employees
Also called ecologic correlations/studies; comparison based on summary or aggregate data from groups (not individuals). Subject to ecologic fallacy.
Deals only with the proportion of deaths due to a cause.
3,520 deaths among nuclear power plant workers
Do not know size of population at risk