Written and maintained by CASRAI Editorial Board
Last updated
A case-control study is an observational research design that starts from the outcome and works backward. Investigators identify a group of people who already have a condition or outcome of interest (the cases) and a comparison group of people who do not have it (the controls), then look back at each group’s history to see whether prior exposure to a suspected risk factor differed between the two groups. Because the outcome has already occurred by the time the study begins, exposure information is gathered retrospectively — from medical records, existing databases, biological samples, or participant recall — rather than measured going forward in time.
This backward-looking structure is what distinguishes a case-control study from a retrospective cohort study, where participants are still grouped by exposure status even though the data happens to already exist. In a case-control study, group membership is defined by outcome (case or control), and exposure is the variable being reconstructed after the fact.
Why researchers choose a case-control design
The case-control design exists to solve a specific practical problem: studying rare outcomes efficiently. If a disease or event occurs in only a small fraction of a population, a forward-looking cohort study would need to enroll and follow an enormous number of people to accumulate enough cases to analyze — often impractical in terms of time and cost. A case-control study sidesteps this by deliberately recruiting people who already have the outcome, so the sample is enriched with the very individuals a cohort study would have to wait years to identify.
This makes case-control studies especially well suited to:
- Rare diseases or outcomes, where a cohort large enough to observe a meaningful number of cases would be infeasible to assemble.
- Diseases with a long latency period between exposure and outcome, where a prospective cohort would require years or decades of follow-up.
- Hypothesis-generating research on multiple possible exposures at once, since a single case-control study can examine many candidate risk factors in the same set of cases and controls, whereas a cohort study is generally structured around one exposure of primary interest.
- Outbreak investigations, where speed matters and a retrospective look at recent exposures among affected versus unaffected individuals can identify a likely source quickly.
The tradeoff for this efficiency is that case-control studies are generally considered to sit lower on the hierarchy of evidence for establishing causation than well-conducted cohort studies or randomized controlled trials, because exposure is measured after the outcome and is therefore more vulnerable to bias in how it is recorded or recalled.
How a case-control study is structured
1. Defining and selecting cases
A case is a person who has the outcome under investigation, defined by an explicit, objective, and reproducible case definition — typically a specific diagnosis, confirmed by a defined diagnostic standard (a pathology result, a clinical criteria set, a registry entry), applied consistently to every potential case. Cases are commonly identified through hospital records, disease registries, clinics, or population surveillance systems. The precision of the case definition matters enormously: a definition that is too loose will mix genuinely different conditions together and dilute any real association; one that is too narrow may limit how far the findings generalize.
2. Selecting controls
Controls are people without the outcome, but selecting them well is widely regarded as the most difficult and consequential step in case-control design. The guiding principle is that controls should represent the exposure distribution of the population that produced the cases — in other words, if a control had developed the outcome, they would have been eligible to be identified as a case in this same study. Common control sources include:
- Population-based controls, sampled from the general population that gave rise to the cases (e.g., via random-digit dialing, voter or driver’s license registries, or neighborhood sampling) — generally considered the strongest option because they best represent the source population, but often harder and more expensive to recruit.
- Hospital- or clinic-based controls, drawn from patients being treated for other, unrelated conditions at the same facility — more convenient to recruit and often more willing to participate, but at risk of not representing the general population if the reason they are in the hospital is itself related to the exposure under study.
A poorly chosen control group is one of the most common sources of a spurious or distorted association in a case-control study, and reviewers scrutinize control-selection methodology closely for exactly this reason.
3. Matching
Many case-control studies match each case to one or more controls on characteristics likely to confound the exposure-outcome relationship, such as age, sex, or geographic region, so that cases and controls are comparable on everything except exposure and outcome status. Matching can be done individually (one or more specific controls per case, sharing matched characteristics) or by frequency (the control group as a whole mirrors the case group’s distribution on the matched variables). Matching improves statistical efficiency and controls confounding from the matched variables directly, but it comes with a real analytic obligation: once matching is used, the matched variable can no longer be examined as an independent risk factor, and the statistical analysis (commonly conditional logistic regression) must account for the matched design or the resulting estimates will be biased.
4. Measuring exposure retrospectively
With cases and controls identified, investigators then collect exposure history for both groups using the same methods and the same effort — interviews, questionnaires, medical records, pharmacy or claims data, biomarkers from stored specimens, or environmental records. Applying identical exposure-ascertainment methods to cases and controls is essential; any systematic difference in how thoroughly or accurately exposure is measured between the two groups introduces bias directly into the result.
Calculating and interpreting the odds ratio
Because participants in a case-control study are selected on outcome rather than sampled from a defined population followed over time, the true incidence or absolute risk of the outcome generally cannot be calculated directly from the study. Instead, the standard measure of association is the odds ratio (OR): the odds of exposure among cases divided by the odds of exposure among controls. Data are typically summarized in a 2×2 table:
| Exposed | Unexposed | |
|---|---|---|
| Cases | a | b |
| Controls | c | d |
The odds ratio is calculated as OR = (a × d) / (b × c). An OR of 1 indicates no association between exposure and outcome; an OR greater than 1 suggests the exposure is associated with higher odds of the outcome; an OR less than 1 suggests the exposure is associated with lower odds. When the outcome under study is rare in the source population (a common justification for using a case-control design in the first place), the odds ratio closely approximates the relative risk that a cohort study would have produced — a relationship known as the rare disease assumption. When the outcome is not rare, the odds ratio will tend to overstate the magnitude of the relative risk, and this distinction matters when interpreting or communicating case-control findings.
Recall bias, selection bias, and other common pitfalls
The retrospective structure that makes case-control studies efficient is also what makes them vulnerable to several well-documented sources of bias:
- Recall bias — cases, precisely because they have experienced the outcome, often think harder about their past exposures and recall them more completely (or, in some cases, less accurately) than controls, who have no comparable reason to scrutinize their own history. A parent of a child with a birth defect, for example, may recall and report medication use during pregnancy more thoroughly than a parent of an unaffected child. This differential recall between groups can create or exaggerate an apparent association that does not reflect a true difference in exposure.
- Selection bias — if the process used to identify cases or, especially, controls is related to the exposure itself, the resulting sample will not represent the true exposure-outcome relationship in the source population. Hospital-based controls are a frequent source of this problem when the conditions bringing them to the hospital are themselves linked to the exposure under study.
- Confounding — as with any observational design, a confounding variable associated with both the exposure and the outcome can distort the apparent association unless it is controlled for through matching, stratification, or multivariable statistical adjustment.
- Reverse causality and temporal ambiguity — because exposure and outcome are both assessed after the outcome has occurred, it can be harder to establish with confidence that the exposure genuinely preceded the outcome, particularly for exposures that could plausibly change as a result of early, undiagnosed disease.
None of these pitfalls are unique to case-control studies — similar issues affect other observational designs — but the retrospective, outcome-first structure of a case-control study makes recall bias and control-selection bias particularly characteristic weaknesses that reviewers and readers should specifically look for. The Newcastle-Ottawa Scale is the most widely used tool for appraising case-control (and cohort) study quality, including how rigorously controls were selected and matched.
Nested case-control studies
A nested case-control study is a case-control study conducted within an existing cohort. Rather than assembling cases and controls independently, investigators start with a cohort that has already been enrolled and followed (often for an unrelated primary purpose, or with biological specimens banked at baseline), identify the cases who developed the outcome during follow-up, and sample a set of controls from among the cohort members who remained outcome-free at the time each case occurred. Because both cases and controls were part of the same defined cohort from the start, control selection is less prone to the population-representativeness problems that affect a standalone case-control study, and if biospecimens were collected at enrollment — before the outcome occurred — exposure biomarkers can sometimes be measured directly rather than relying entirely on recall, reducing recall bias as well. Nested designs are common in cancer epidemiology and other fields where large prospective cohorts with banked baseline specimens already exist.
Case-cohort studies: a related but distinct alternative
A case-cohort study is a second design built from an existing cohort, and it is often confused with a nested case-control study, but the two select their comparison group differently. In a case-cohort study, investigators draw a single random subcohort from the entire baseline cohort — sampled without regard to who later becomes a case — and compare it against all cases that occur anywhere in the full cohort during follow-up, including any subcohort members who themselves later become cases. Because the subcohort is a simple random sample of the baseline population rather than a set of controls matched individually to each case, it can serve as the comparison group for more than one outcome studied within the same cohort, and it can be assembled before follow-up is complete. The tradeoff is analytic: because subcohort members can overlap with cases and the sampling is not matched to specific case events, standard logistic regression does not produce valid estimates from a case-cohort dataset — the analysis requires specialized methods designed for this overlapping sampling structure rather than the conditional logistic regression typically used for a matched nested case-control study. Case-cohort designs are common in occupational and cancer epidemiology, where a single banked subcohort is reused as the comparison group across a series of separate analyses of different outcomes.
Case-control study vs. cohort study
Case-control and cohort studies are the two principal analytical designs used in observational epidemiological research, alongside cross-sectional studies, and the choice between them is one of the most common design decisions in clinical and epidemiological research. The essential difference is the direction of inquiry: a cohort study selects participants by exposure status and follows them forward to observe whether the outcome develops, while a case-control study selects participants by outcome status and looks backward to reconstruct exposure history.
| Feature | Case-control study | Cohort study |
|---|---|---|
| Participants selected by | Outcome status (case vs. control) | Exposure status |
| Direction | Backward (retrospective exposure assessment) | Forward (prospective, or retrospective using existing records) |
| Best suited to | Rare outcomes, multiple candidate exposures, outbreak investigations | Rare exposures, multiple possible outcomes, establishing incidence |
| Can calculate incidence/absolute risk directly? | Generally no | Yes |
| Primary measure of association | Odds ratio | Relative risk (or hazard ratio) |
| Characteristic bias risks | Recall bias, control-selection bias | Loss to follow-up, confounding |
| Typical cost and duration | Lower cost, faster | Higher cost, often longer (especially prospective) |
Neither design is inherently superior; the right choice depends on how rare the outcome is, how many candidate exposures are being investigated, and how much time and budget are available. CASRAI’s guide to clinical study design types places case-control studies within the broader landscape of observational and interventional designs, and the prospective vs. retrospective study comparison covers the related but distinct question of when data collection happens relative to when a study begins. For a fuller side-by-side breakdown of both designs, including cost, duration, and characteristic bias profiles, see CASRAI’s dedicated case-control study vs. cohort study comparison.
Reporting case-control studies
The STROBE (STrengthening the Reporting of OBservational studies in Epidemiology) statement identifies cohort, case-control, and cross-sectional studies as the three main analytical designs used in observational research, and provides a dedicated reporting checklist with case-control-specific items covering how cases and controls were defined and selected, how matching was handled in the analysis, and how potential sources of bias were addressed. Following STROBE when writing up a case-control study, and consulting it during design, helps ensure the study reports the information a reader needs to judge its validity — particularly the control-selection and matching details that are so often where case-control studies go wrong. See CASRAI’s STROBE Checklist for Observational Studies for the full 22-item guideline broken out item-by-item.
Research-administration considerations: IRB review, consent, and secondary data use
Because most case-control studies reconstruct exposure from records that already exist — hospital charts, disease registries, claims data, or biobank specimens collected for another purpose — they raise a specific, recurring set of research-administration questions that a prospective cohort or trial does not face in quite the same way.
IRB review and waiver of consent
A retrospective case-control study built entirely from existing records or specimens is still human subjects research under the Common Rule (45 CFR 46) if it involves identifiable private information, and it still requires IRB review. What frequently differs from a prospective study is whether informed consent must be obtained from each case and control individually. Because re-contacting everyone identified through a historical chart review or registry is often impracticable — and because analyzing already-existing records typically carries minimal incremental risk — investigators frequently request a waiver of the consent process under 45 CFR 46.116(f). An IRB may grant that waiver only if it makes four specific findings: the research involves no more than minimal risk; the waiver will not adversely affect the rights and welfare of the subjects; the research could not practicably be carried out without it; and, where appropriate, subjects will be provided with additional pertinent information after their participation. This four-part test is distinct from — and more demanding than — the separate provision that waives only the signed-consent-form requirement (45 CFR 46.117(c)); the two are commonly conflated in practice but are not the same waiver, and a protocol should specify which one it is requesting.
Chart review, secondary data use, and HIPAA
Where a case-control study uses protected health information (PHI) held by a covered entity — as most hospital-based chart-review case-control studies do — HIPAA’s Privacy Rule applies independently of, and alongside, the Common Rule consent analysis. An IRB or a separately constituted Privacy Board can waive the requirement for individual HIPAA Authorization under 45 CFR 164.512(i) if it documents that use of the PHI involves no more than minimal privacy risk (an adequate plan to protect identifiers, an adequate plan to destroy them at the earliest opportunity absent a documented justification to retain them, and written assurance against further use or disclosure) and that the research could not practicably be conducted without the waiver and without access to the PHI. Where full identifiability is not needed, using a Limited Data Set under a Data Use Agreement, or fully de-identifying the extract under the Safe Harbor or Expert Determination method (45 CFR 164.514), can avoid the Authorization/waiver analysis for that dataset altogether. Because a case-control study is unusually dependent on how completely and accurately exposure was originally documented, protocols should also address — separately from the regulatory analysis — whether the existing record is reliable enough to support the exposure ascertainment the design requires.
Frequently asked questions
What is a case-control study in simple terms?
It is a study that starts with people who already have a condition (cases) and compares them to similar people who do not have it (controls), then looks back at each group’s history to see whether a suspected exposure was more common among the cases.
What is the main disadvantage of a case-control study?
Its retrospective structure makes it particularly vulnerable to recall bias (cases and controls may not recall or report past exposures with equal accuracy) and selection bias in how controls are chosen, and it generally cannot calculate the true incidence or absolute risk of the outcome.
Why use an odds ratio instead of relative risk in a case-control study?
Because participants are sampled based on outcome status rather than drawn from a defined population followed over time, the true incidence of the outcome usually cannot be calculated, so relative risk cannot be computed directly. The odds ratio can be calculated from the same 2×2 exposure-by-outcome data and, when the outcome is rare, closely approximates what the relative risk would have been.
What is a nested case-control study?
A case-control study conducted within an existing cohort, where cases are the cohort members who developed the outcome during follow-up and controls are sampled from cohort members who did not, at the time each case occurred. It reduces some of the control-selection problems of a standalone case-control study and can allow exposure biomarkers to be measured from specimens banked before the outcome occurred.
How are controls chosen in a case-control study?
Controls should represent the exposure distribution of the population that produced the cases — meaning that if a control had developed the outcome, they would have been eligible to be identified as a case in the same study. Common sources are population-based sampling (generally the strongest but hardest to execute) and hospital- or clinic-based recruitment (more convenient but at greater risk of not representing the general population).
Does a case-control study need IRB approval?
Yes. If it involves identifiable private information about living individuals, a case-control study is human subjects research under the Common Rule and requires IRB review, even when it is built entirely from existing records. What often differs from a prospective study is whether individual informed consent is required — many retrospective, records-based case-control studies qualify for a waiver of consent under 45 CFR 46.116(f) because re-contacting each case and control is impracticable and the incremental risk of using existing records is minimal, but the IRB must still make that finding; a waiver is not automatic.
What is a case-cohort study, and how is it different from a nested case-control study?
Both are built inside an existing cohort, but they select their comparison group differently. A nested case-control study samples a small number of controls specifically matched to each case at the time the case occurred. A case-cohort study instead draws one random subcohort from the entire baseline cohort, independent of who later becomes a case, and reuses that same subcohort as the comparison group for all cases — which lets a single subcohort support separate analyses across multiple outcomes, at the cost of requiring specialized statistical methods rather than standard conditional logistic regression.
