Choosing a cohort or case-control design
Choose cohort follow-up or case-control sampling from your question, the available population, and the cost of measuring exposure.
Fit
When it matters, and when it does not
It matters when
- You have an exposure-outcome question and must choose how people enter the study.
- Measuring exposure in everyone would be costly, so sampling cases and controls may save work.
- You need to decide whether the available data can give absolute risks or only a relative comparison.
It matters less when
- Your aim is to estimate how common a condition is at one time. Use a prevalence survey. Read: Prevalence survey
- You can assign the intervention by chance and need its effect. Start with a randomized trial. Read: Randomized controlled trial
- You want to predict an individual's future outcome. Start with prediction-model development. Read: Prediction model development
Before you start
What to check in your study
- How common the outcome and the exposure areA rare outcome favors sampling cases and controls. A rare exposure favors a cohort.
- The effect measure you needsummaryMeasureA cohort gives risks, rates and their ratios. A case-control study gives an odds ratio, and absolute risk only with the sampling fractions.
- The source population and its time originBoth designs need one. Controls are sampled from it, never from whoever is easy to reach.
- What measuring exposure costs, and whether records already hold itCostly exposure data favor sampling. Records made before the outcome avoid recall bias in either design.
- How controls would be sampledsamplingStructureThe linked case-control and nested case-control pages explain control sampling. Check the scheme before choosing the effect measure.
- A missing-data plan that starts from an assumptionIn either design, decide why exposure or outcome values may be missing before choosing a method. The assumption picks the method, not the share missing.
- In a case-control study, controls come from the population that produced the cases.How to check: Ask of each control: had they developed the outcome, would they have been a case here?
- Exposure is recorded before the outcome, or recalled equally well by cases and controls.How to check: Prefer records made before the outcome. Cases often search their memory harder, called recall bias.
- In a cohort, follow-up and outcome checks are the same in every exposure group.How to check: Compare loss to follow-up and testing rates between groups.
- In either design, missing values follow the assumption you named.How to check: Records are often fuller for cases than for controls, and people lost from a cohort differ from those who stay. So missing completely at random (MCAR) is rarely plausible. Missing at random (MAR) means the gaps depend only on recorded factors, such as age, site and case status. Under MAR, multiple imputation with the outcome in the model can help, but it is not automatically better than complete records. If exposed people are less likely to have their exposure recorded than unexposed people like them, missing not at random (MNAR) is plausible. Then add a tipping-point analysis, which shifts the imputed values until the conclusion changes.
Plain language
How it works
Start from the population, exposure contrast and outcome horizon. A cohort follows an eligible population and counts later outcomes. A case-control study samples cases and controls from a source population. Either can use existing records: retrospective does not mean case-control.
A cohort suits direct risk estimates and several outcomes. Case-control sampling helps when outcomes are rare or exposure tests are costly. Nested sampling keeps the comparison within a defined cohort. Use the linked case-control pages for control selection and analysis rules.
Pooled case and control counts cannot determine the matched estimate. Matched sets need conditional logistic regression, not an odds ratio from pooled counts.
Worked example
A worked example
- Setting
- A biobank follows 20000 adults. Over ten years, 200 develop pancreatic cancer. A new blood marker is costly to measure.
- Question
- Should the team test every stored sample or sample cases and controls?
- Approach
- A full cohort can estimate absolute risk directly. A nested case-control design can reduce assay work while keeping controls within the source cohort.
- Result
- Testing 200 cases and 800 sampled controls uses 1000 samples: 19000 fewer than the full cohort's 20000.
- What it means
- The smaller assay workload favors nested sampling. The sampling scheme still determines which effect the analysis can estimate.
In practice
Recent studies PubMed search · not vetted
Watch out
Common mistakes
The one to remember
Control sampling determines what a case-control odds ratio estimates. A convenient control group cannot be repaired by a more complicated regression.
- Calling any study that uses existing records a case-control study.Instead: Retrospective describes when data were collected. A case-control study is defined by sampling on the outcome.
- Choosing between the designs before counting how common the outcome is.Instead: Count the expected cases first. If few would occur, sample cases and controls. If many would, and exposure is recorded for everyone, analyze the whole cohort.
- Dropping people whose exposure is unknown, in either design.Instead: Those left may not reflect exposure in the source population, which is selection bias. Keep them, and use multiple imputation under a stated assumption.
- Sizing a matched case-control study with a formula for unmatched groups.Instead: Use a method built for matched sets, such as Dupont's, and keep the matching in the analysis.
Inside Mallard
How Mallard uses this
For a typed observational question, Mallard may offer this comparison beside a cohort or case-control design. The comparison does not change the chosen design.
concept.choosing_cohort_case_control, the id a plan cue records.How to report it
STROBE is the guideline for both designs this page compares, with a separate checklist for cohort and for case-control studies. Report which design was chosen and why, the source population either way, and use the checklist for the design actually run rather than a combined one. A reporting guideline is not a design manual: use it to make the study transparent, not as a substitute for careful design. Check the target journal's own requirements as well, which this page cannot know.
Related Learn pages
Related designs and methods
Evidence
Sources & further reading
- Case-control studies: Increasing scientific rigor in control selection (opens in a new tab) · Res Pract Thromb Haemost · doi 10.1016/j.rpth.2023.100090
- Observational studies: cohort and case-control studies (opens in a new tab) · Plast Reconstr Surg · doi 10.1097/PRS.0b013e3181f44abc
- Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies (opens in a new tab) · BMJ · doi 10.1136/bmj.39335.541782.AD
- Multiple imputation for missing data in epidemiological and clinical research: potential and pitfalls (opens in a new tab) · BMJ · doi 10.1136/bmj.b2393