Can a study be both cohort and case-control?

No — these are mutually exclusive designs based on whether sampling is by exposure or outcome status. Nested case-control studies are case-control analyses within a defined cohort: the cohort provides the source population, but the analytic design is case-control.

Why do case-control studies report odds ratios?

Case-control studies sample by outcome, so incidence cannot be calculated directly. The odds of exposure among cases compared to controls estimates an odds ratio. When the outcome is rare in the source population, the OR approximates the risk ratio.

Is a retrospective cohort the same as a case-control study?

No. Retrospective cohort studies define exposed and unexposed groups at a baseline time point using records, then compare outcome rates forward from that point. Case-control studies sample by outcome and look backward for exposure history. The timeline diagram clarifies the distinction.

Strata Academy

Case-Control vs Cohort Studies: Framework Selection Guide

Q: Which STROBE checklist should I use?

Match the checklist to the study design described in methods. STROBE has separate versions for cohort, case-control, and cross-sectional studies. If authors mislabel their design, apply the checklist that fits what they actually did.

Temporal direction, rare outcomes, confounding, and matching ROBINS-I, NOS, and STROBE to the correct observational design

Quick answer

Cohort studies identify groups by exposure and follow forward for outcomes; case-control studies identify groups by outcome status and look backward for exposures. Cohort designs suit incidence and relative risk; case-control designs suit rare outcomes and estimate odds ratios. Neither uses ROB 2 — use STROBE for reporting, ROBINS-I or NOS for risk of bias.

1. Core design difference

Observational studies measure associations without random allocation. The fundamental split is temporal direction. Cohort studies start with exposure (or risk factor) status and follow participants forward to compare outcome incidence between exposed and unexposed groups.

Case-control studies start with outcome status — cases with the disease or event, controls without — and retrospectively compare prior exposure frequency. The researcher looks backward in time from outcome to exposure.

This distinction determines which effect measure is natural: cohort studies estimate incidence, risk ratios, or hazard ratios; case-control studies estimate exposure odds ratios. With rare outcomes (typically <10% incidence), the odds ratio approximates the risk ratio.

Students frequently mislabel retrospective cohort studies as case-control. If the study identifies a cohort by exposure at baseline and follows records forward for outcomes, it is a retrospective cohort — even though data collection looks backward in calendar time.

Cohort — exposure defined first, outcome observed later
Case-control — outcome defined first, exposure history queried
Cross-sectional — exposure and outcome measured simultaneously
Nested case-control — case-control study within a defined cohort (efficient for expensive exposures)

2. When cohort design is appropriate

Cohort studies suit questions about incidence, time-to-event, and multiple outcomes from one exposure. Prospective cohorts allow planned measurement of confounders at baseline before outcomes occur.

Examples: following smokers and non-smokers for lung cancer incidence; registry-linked cohorts comparing drug exposure groups for cardiovascular events.

Strengths: direct estimation of risk, clear temporal sequence, ability to calculate absolute risks and attributable fractions. Weaknesses: inefficient for rare outcomes (requires large samples and long follow-up), loss to follow-up threatens validity.

Appraise with STROBE cohort checklist for reporting, ROBINS-I for intervention/exposure risk of bias, or Newcastle–Ottawa Scale (NOS) cohort version for quality scoring in systematic reviews.

3. When case-control design is appropriate

Case-control studies are efficient for rare outcomes — you enrich the sample with cases rather than waiting for events in a large exposed cohort. They are common in aetiology research (e.g. environmental exposures and rare cancers).

Controls must represent the source population that gave rise to the cases. Hospital controls may differ systematically from community controls — selection bias distorts odds ratios.

Recall bias arises when cases remember exposures differently from controls, especially for subjective exposures (diet, medication use). Blinding interviewers and using objective records (prescription databases) mitigates but does not eliminate this.

Matching cases to controls on confounders (age, sex) improves efficiency but requires appropriate analysis (conditional logistic regression) — matched crude comparisons can be misleading.

4. Framework selection by design

STROBE provides separate checklists for cohort, case-control, and cross-sectional studies. Use the checklist matching the actual design — not whichever the authors claim in the title.

ROBINS-I applies to non-randomised studies of interventions and exposures with comparative groups. It can be used for both cohort and case-control designs when the question is about causal effects of an exposure or intervention.

Newcastle–Ottawa Scale has distinct cohort and case-control forms used in systematic reviews. NOS star ratings are widely used but treat them as structured prompts, not definitive quality scores.

ROB 2 never applies to observational designs. If you see non-random allocation, stop reaching for ROB 2.

Cohort reporting → STROBE cohort checklist
Case-control reporting → STROBE case-control checklist
Intervention/exposure causality → ROBINS-I
Systematic review quality scoring → NOS (design-specific form)
Diagnostic accuracy case-control → QUADAS-2 (high applicability concern)

5. Bias patterns to recognise

Selection bias: non-representative controls (case-control), healthy worker effect (cohort), or inclusion restricted to survivors. Ask whether the study population matches the target population for your clinical question.

Confounding: unequal distribution of prognostic factors across exposure groups. Check what was measured at baseline and what was adjusted in multivariable models. Unmeasured confounding remains a limitation in all observational designs.

Information bias: differential misclassification of exposure (case-control recall) or outcome (detection bias in cohorts with differential surveillance). Objective record linkage strengthens both designs.

Immortal time bias: specific to cohort studies where exposed time is misclassified — common in pharmacoepidemiology when exposure starts at prescription but person-time before prescription is misallocated.

6. Decision guide for students

Designing research: choose cohort when you can follow groups forward and outcomes are not extremely rare; choose case-control when outcomes are rare and exposure data can be collected retrospectively with acceptable bias risk.

Appraising published papers: identify design from methods, not title. Draw a timeline diagram — when was exposure defined relative to outcome?

For systematic reviews: pre-specify which observational designs are eligible. Pooling cohort risk ratios with case-control odds ratios without conversion or sensitivity analysis is methodologically unsound.

Users' Guides to the Medical Literature (Sackett et al.) and CEBM study design resources provide additional frameworks for teaching critical appraisal of observational evidence.

Interactive version (quizzes, walkthroughs) loads when JavaScript is enabled.