Matched case-control study
Why this level?
Specify the research question: intervention, diagnostic accuracy, prognosis, aetiology, or screening intervention. Other question types are outside this hierarchy. NHMRC 2009 sourceMatching is a sampling decision rather than a different design, so the level is the one any case-control study carries. The matching has to be carried into the analysis; unmatched, it introduces the bias it was chosen to remove.
Where the NHMRC 2009 hierarchy names this design, it places it here:
| Research question | Level | When |
|---|---|---|
| Intervention | III-2 | An observational comparison uses concurrent controls. Statistical adjustment does not change this design level. |
| Aetiology | III-3 | A case-control study examines the exposure in people selected by outcome status. |
What question does this help answer?
The same question an unmatched case-control answers, an odds ratio for an exposure, but with the matched factors held constant by design rather than adjusted for afterwards.
When to use it
Matching earns its place when a confounder is strong and hard to adjust for, when controls are plentiful but expensive to characterise, or when calendar time must be held fixed because both exposure and outcome drift over the study period.
When not to use it
Do not match on the exposure itself or on anything affected by it: that introduces bias rather than removing it. Do not match on a factor you want to study, because matching removes the ability to estimate its effect. And do not match reflexively, since an unmatched design with adjustment is often simpler and no less valid.
How it works in plain language
Cases are identified, then controls are drawn to match them on the chosen factors, commonly one, two or more controls per case. The matching must then be carried into the analysis with a conditional method, typically conditional logistic regression, which compares exposure WITHIN each matched set. An ordinary logistic regression that ignores the matching answers a different question and gets the standard errors wrong.
What matched sets look like
Why Mallard may recommend it
Mallard may recommend a matched case-control design when the outcome is rare, when a strong confounder such as age or admission period would otherwise dominate, and when cases and eligible controls can both be identified from the same source population.
Key assumptions and design requirements
- Controls come from the same source population that produced the cases, and would have been counted as cases had they developed the outcome.
- Matching factors are chosen before the analysis and are not affected by the exposure.
- The analysis is CONDITIONAL on the matched sets; an unmatched analysis of matched data is a mistake, not a simplification.
- How controls were sampled is stated, because density (risk-set) sampling and sampling from those who never became cases imply different estimands.
- Sizing uses a method for MATCHED data, not an unmatched two-proportion formula.
What to watch out for
Matching does not remove confounding by the matched factor; it removes the ability to STUDY it, and the matching must be carried through into the analysis with a conditional method. Matching on something affected by the exposure introduces bias rather than removing it, and over-matching on a factor close to the exposure discards information.
Clinical example
Question: Is a pre-discharge medication review associated with 30-day readmission? Design: Identify readmitted patients as cases, match each to two patients discharged in the same month who were not readmitted, and compare exposure within matched sets using conditional logistic regression.
How to report it
No reporting guideline dedicated to matched case-control designs was found in a PubMed search on 2026-09-16. STROBE is the guideline to use, with its case-control checklist, which asks for the matching criteria, the number of controls per case, and how the matching was addressed in the analysis. Be specific on the last: say that the analysis kept the matched sets rather than pooling them. 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.
- Power calculations for matched case-control studies (opens in a new tab) · Biometrics
- Power and sample size evaluation for the McNemar test with application to matched case-control studies (opens in a new tab) · Stat Med · doi 10.1002/sim.4780110909
- 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