BiostatisticsTier 1Medical Sciences concept

Study design types - selection and recognition (cohort, case-control, RCT, cluster, stepped-wedge, ecological)

Core concept

Exam task is usually pattern recognition: given a scenario description, name the design -- know the identifying features, not just the definition
Cohort (prospective or retrospective): groups defined by exposure, followed for outcome -- can calculate incidence, relative risk
Case-control: groups defined by outcome (cases vs controls), look back for exposure -- can only calculate odds ratio, good for rare diseases

Design	Identifying feature	Best for
RCT	Investigator allocates exposure/intervention, ideally randomised + blinded	Causal inference, strongest evidence
Cohort	Exposure first, followed forward for outcome	Rare exposures, multiple outcomes, incidence
Case-control	Outcome first, look back for exposure	Rare outcomes/diseases, quick and cheap
Cross-sectional	Exposure and outcome measured at the same time point	Prevalence, not causation (no time sequence)
Cluster RCT	Groups (e.g. clinics, villages), not individuals, randomised	Interventions that can't be delivered to individuals within a group without contamination
Stepped-wedge	All clusters eventually receive intervention, but timing of crossover is randomised	When withholding an intervention indefinitely is unacceptable
Ecological	Unit of analysis is a population/group, not individuals	Hypothesis generation only -- ecological fallacy risk

Trap: an ecological study finding population-level correlation (e.g. country-level fat intake vs cancer rate) cannot be used to infer individual-level causation -- this is the ecological fallacy**
Cluster randomisation is required, not optional, when the intervention is delivered at a group level (e.g. an education programme rolled out per hospital) -- individual randomisation would cause contamination between arms
Stepped-wedge is chosen specifically as an ethical compromise: everyone eventually gets the (presumed beneficial) intervention, while still allowing randomised comparison of timing

Feeds directly into hierarchy of evidence (RCT > cohort > case-control > cross-sectional > ecological/case series) for a separate note
Case-control's odds ratio (vs cohort's relative risk) is exactly why odds ratio calculation matters - design dictates which statistic is valid
Cross-sectional design cannot establish temporality -- the same limitation underlies why it cannot support Bradford Hill's "temporality" criterion for causation

3 more sections, plus exam facts

Premium unlocks every note across every specialty, and the full exam fact library behind it.