Study designs

Cohort or case-control?

Choose how to sample people and observations for an etiologic question.

Developing and validating a measurement instrument

Establish what a score measures, for whom, and how consistently it does so.

Scoping reviews and evidence maps

Map what evidence exists and how it is organized before making claims about effects.

Quality improvement studies

Study a change in care delivery using explicit measures, context and a comparison suited to the claim.

Survey and questionnaire studies

Plan who can respond, what the questions measure and how responses support the intended inference.

Randomized controlled trial

Assigns participants to groups by chance, so baseline differences are balanced on average.

Retrospective cohort study

Uses existing records to classify a past exposure and follow outcomes that have already occurred.

Prospective cohort study

Enrols people before the outcome happens and follows them forward.

Case-control study

Starts from people with and without the outcome, then compares their prior exposures.

Cross-sectional study

Measures a population at one point in time, often to estimate prevalence.

Interrupted time series

Tests whether the level or trend of an outcome changed after an intervention at a known date.

Difference-in-differences

Compares how outcomes changed in an exposed group against a comparison group over the same period.

Diagnostic accuracy study

Compares an index test against a reference standard to see how well it classifies people.

Cluster randomized trial

Randomizes groups rather than people, so the analysis must account for the grouping.

Stepped wedge cluster randomized trial

Every cluster eventually gets the intervention; randomization decides when, not whether.

Prediction model development

Builds a model to predict an outcome for an individual, which is a different job from explaining a cause.

Matched case-control study

Controls are chosen to resemble cases on chosen factors, which changes both the analysis and the sizing.

Noninferiority and equivalence trials

Asks whether a treatment is no worse than a comparator by more than a prespecified margin.

Prevalence survey

Estimates how common something is at a point in time, with precision rather than a comparison as the goal.

Agreement and reliability studies

Asks whether two raters, methods or occasions produce the same answer, which is not the same as asking whether either is right.

Crossover trial

Each participant receives both treatments in sequence and acts as their own control.

Factorial trial

Tests two or more interventions at once by randomizing each independently.

Pilot and feasibility studies

Asks whether the definitive trial can be run at all, not whether the treatment works.

N-of-1 trial

A randomized crossover run within a single patient, repeated enough times to answer a question about that patient.

Adaptive and group-sequential trials

Allows prespecified changes as data accumulate, which is not the same as changing the trial because of what you saw.

Pragmatic trials

Designed to answer whether an intervention works in ordinary practice rather than under ideal conditions.

Nested case-control study

A case-control study built inside a defined cohort, so the source population is known exactly.

Case-cohort study

Compares all cases with a random subcohort sampled at baseline, which can serve several outcomes at once.

Case-crossover study

Compares a person's exposure just before an acute event with their own exposure at other times.

Self-controlled case series

Uses only people who had the outcome, comparing risk during exposed and unexposed periods within each person.

Target trial emulation

Designs an observational analysis by first writing down the randomized trial it is standing in for.

Instrumental variables and Mendelian randomization

Uses a variable that affects the exposure but nothing else about the outcome as a natural experiment.

Uncontrolled before-and-after studies

Compares outcomes before and after a change with no concurrent comparison group, which is weak but common.

Prediction model validation

Tests an existing model on data it was not built from, which is a different study from developing one.

Prognostic factor study

Asks whether a variable adds information about future outcome, which is neither a causal question nor a full prediction model.

Systematic review and meta-analysis

Answers a question by finding and combining existing studies under a protocol written before the search.

Economic evaluation

Compares interventions on cost and consequences together, to inform whether a benefit is worth its price.

Implementation and hybrid studies

Studies how to get an effective intervention adopted, separately from whether it works.

Qualitative and mixed methods studies

Answers questions about meaning, process and experience that a count cannot reach.

Evaluating a screening programme

Asks whether looking for disease in people without symptoms makes them live longer or better, which survival statistics cannot answer.

Registry and routinely collected data studies

Uses data collected for care or administration to answer a research question it was not designed for.

Case reports and case series

Describes one patient or a group with a shared condition or treatment, without a comparison group.

Retrospective chart review and data abstraction

Abstracts research data from existing medical records, which needs a protocol as careful as any prospective collection.

Controlled before-and-after studies

Compares the change after an intervention in sites that received it with the change in comparable sites that did not.

Regression discontinuity designs

Uses a threshold that decides who gets an intervention, comparing people just either side of it.

Mixed-methods studies and integration

Combines quantitative and qualitative strands in one study, and plans where and how they will inform each other.