Can a Case-Control Study Estimate Risk or Prevalence Directly?

If Half Your Participants Are Cases, Does That Mean Disease Prevalence Is 50%?

You conduct a case-control study involving 200 people with a particular disease and 200 controls. Because half the participants are cases, your dataset shows a disease proportion of 50%.

But the disease may affect fewer than one person in every thousand in the population from which those participants were selected.

Can you report the observed 50% as disease prevalence? Can you calculate the probability of developing the disease among exposed participants? And if your analysis produces an odds ratio of 3.0, does that mean exposed people face three times the risk?

These questions concern a fundamental distinction between the information collected in a case-control study and the population quantities researchers may wish to estimate.

The answer depends on how participants were sampled, which population denominators are available, and whether the study estimates disease odds, incidence rates, or another epidemiological quantity.

Conventional Case-Control Samples Do Not Directly Reveal Population Risk

Why Case-Control Sampling Changes What You Can Estimate

Why Can't You Calculate Disease Prevalence From the Sample?

In a conventional case-control study, researchers deliberately select individuals who have the outcome and individuals who serve as controls.

The number of cases and controls is therefore determined partly by the study protocol rather than by their natural frequency in the underlying population.

Suppose researchers recruit 100 cases and 100 controls. The sample contains 50% cases because the investigators selected equal numbers of participants.

If they had recruited 100 cases and 400 controls, the case proportion would have been 20%. Nothing about the underlying population's disease frequency would necessarily have changed.

Consequently, the sample case proportion cannot ordinarily be interpreted as disease prevalence.

Prevalence describes the proportion of a defined population that has a condition at a specified point or during a specified period. Estimating it requires information about the relevant population denominator and a sampling or ascertainment process that supports population-level inference.

What Is the Difference Between Risk, Incidence Rate, and Prevalence?

These quantities describe different aspects of disease occurrence and should not be used interchangeably.

A case-control study may contain detailed exposure information but lack the population denominators needed to estimate these quantities directly.

That limitation arises from the sampling structure, not from a general inability to study disease occurrence.

What Can a Conventional Case-Control Study Estimate?

One of the principal measures is the exposure odds ratio comparing cases and controls.

In a conventional two-by-two table, the odds ratio can be calculated using the numbers of exposed and unexposed cases and controls.

The odds of exposure among cases are 3.5 times the odds of exposure among controls in this example.

Under suitable sampling assumptions, this exposure odds ratio can identify a corresponding population association measure. The precise interpretation depends on how controls were selected.

It does not follow that the probability of developing disease is 3.5 times higher among exposed individuals.

Why Is an Odds Ratio Not Automatically a Risk Ratio?

A risk ratio compares two probabilities of developing an outcome during a specified period.

An odds ratio compares odds, which are calculated differently from probabilities.

For an event with probability p, its odds equal p divided by one minus p.

In this illustration, the odds ratio and risk ratio are different despite being calculated from the same underlying probabilities.

The difference generally becomes more consequential as the outcome becomes common, although the magnitude also depends on the underlying probabilities.

Researchers should therefore identify the parameter actually estimated rather than use odds, risk, and rate terminology interchangeably.

What Is the Rare Disease Assumption?

When an outcome is uncommon in the relevant exposure groups, disease odds and disease risks are numerically similar.

Under appropriate conventional cumulative case-control sampling, the disease odds ratio may then approximate the disease risk ratio.

For example, if the risk is 2% among exposed individuals and 1% among unexposed individuals, the risk ratio is 2.00, while the odds ratio is approximately 2.02.

This approximation can be useful, but it is not a universal property of case-control studies.

The relevant condition concerns outcome probabilities in the groups being compared, not merely a vague statement that the disease is rare in the overall population.

Researchers should also avoid assuming that an odds ratio can always be interpreted as a risk ratio simply because cases were recruited for an uncommon disease.

Does Incidence-Density Sampling Require the Rare Disease Assumption?

No. This is an important exception to the common explanation of case-control odds ratios.

In incidence-density sampling, controls are sampled from individuals at risk at the times when cases occur.

With appropriate sampling and analysis, the resulting odds ratio can estimate an incidence rate ratio without requiring the outcome to be rare.

This is possible because controls represent the person-time experience that generated the cases.

The estimated association is therefore connected to the incidence rate ratio through the sampling design, rather than through a numerical approximation between disease odds and disease risks.

The distinction matters because incidence rate ratios and cumulative risk ratios are different quantities.

A rate ratio of 2 does not necessarily mean that the cumulative probability of developing disease over a particular period is doubled.

How Does the Control-Sampling Strategy Affect Interpretation?

Different case-control sampling strategies can support estimation of different population association measures.

These interpretations require appropriate sampling assumptions. The table should not be treated as a guarantee that every study using a particular label identifies the stated parameter.

For example, an incorrectly selected control group may fail to represent the intended population experience.

Researchers should establish how controls represent the source population before interpreting the resulting association measure.

Can Case-Control Studies Estimate Absolute Risk With Additional Information?

Sometimes. The inability to estimate absolute risk directly from conventional case-control counts does not mean absolute risk is mathematically impossible to recover under every design.

Suppose a study provides a valid relative association estimate and researchers also have reliable external information about baseline risk in the relevant population.

Under suitable assumptions, the external information may help translate the relative association into estimated absolute risks.

However, this requires careful consideration of the association measure, population comparability, exposure distribution, follow-up period, and confounding.

For instance, an odds ratio can be combined with a known baseline risk to calculate the corresponding risk under an assumed relationship.

The formula illustrates a mathematical relationship, not a general procedure for converting every adjusted case-control odds ratio into a causal absolute risk.

Conditional odds ratios, population heterogeneity, and the noncollapsibility of odds ratios complicate such conversions. Reliable absolute-risk estimation may require a suitable prediction model, covariate distribution, and additional population data.

Can Known Sampling Fractions Help Recover Population Quantities?

Yes, under suitable designs.

If the probabilities of selecting cases and noncases are known, researchers may be able to use sampling weights or likelihood-based methods to recover particular population quantities.

For example, suppose an outcome-defined sample is drawn from a fully enumerated population with known numbers of cases and noncases.

Because the population case fraction is known independently, it may be possible to estimate absolute risks by exposure group using appropriately weighted information, provided the sampling and measurement assumptions are satisfied.

However, knowing the ratio of cases to controls in the research dataset is not equivalent to knowing the sampling fractions.

The latter require information about the underlying eligible population and the probabilities of inclusion.

Researchers should not attempt to reconstruct population prevalence from an arbitrary case-control ratio without those denominators.

Can a Nested Case-Control Study Estimate Incidence Directly?

A nested case-control study samples cases and controls from an underlying cohort.

The parent cohort may contain the population denominators, person-time, and outcome information needed to estimate incidence or cumulative risk.

However, those quantities are not automatically available from the nested case-control sample alone.

If the full cohort's outcome experience and follow-up information are retained, researchers may estimate incidence from the parent cohort while using the nested sample to analyze expensive exposures.

The distinction is between what the sampled case-control dataset identifies independently and what can be estimated using additional cohort information.

Researchers considering a nested rather than conventional case-control design should specify which population quantities will come from the full cohort and which analyses will use sampled controls.

What About Case-Cohort Studies?

Case-cohort designs use a subcohort sampled from an underlying cohort, together with eligible cases identified during follow-up.

With suitable sampling information and analysis, the subcohort can represent aspects of the parent cohort's population experience.

These designs may be particularly useful when several outcomes are investigated using the same subcohort.

They can support relative association estimation and, when sufficient cohort and sampling information exists, contribute to absolute-risk estimation.

However, ordinary unweighted calculations from the sampled subcohort and cases may not recover the desired population quantities.

Researchers must account for the case-cohort sampling structure and use appropriate estimation methods.

The circumstances favoring a case-cohort design therefore depend partly on which outcomes and population measures are required.

Can Case-Control Studies Estimate Prevalence Odds Ratios?

Yes, in certain designs where cases are selected according to prevalent disease status and controls appropriately represent individuals without the condition in the corresponding source population.

Such studies may estimate a prevalence odds ratio under appropriate sampling assumptions.

However, a prevalence odds ratio is not the same as prevalence itself.

It compares the odds of prevalent disease between exposure groups, or equivalently the corresponding exposure odds under the appropriate case-control sampling relationship.

Prevalence associations may reflect both disease occurrence and disease duration or survival.

Consequently, a prevalence odds ratio should not automatically be interpreted as an incidence association or a causal effect on disease onset.

Why Does Disease Duration Matter for Prevalent Cases?

Prevalence reflects both the occurrence of new cases and how long individuals remain affected.

An exposure associated with longer survival or disease duration may be more common among prevalent cases even if its relationship with disease onset is different.

Conversely, an exposure associated with rapid mortality may be underrepresented among living prevalent cases.

This is one reason incident-case designs are often preferable when the primary question concerns factors associated with new disease occurrence.

The issue is distinct from the inability to estimate prevalence directly from an arbitrarily sampled case-control dataset.

Even when a valid prevalence odds ratio can be estimated, its interpretation requires attention to the mechanisms governing disease duration and case ascertainment.

Can Logistic Regression Predict Disease Probability From Case-Control Data?

Not directly from the ordinary fitted probabilities in a conventionally sampled case-control dataset.

Logistic regression estimates associations between predictors and case status, but outcome-dependent sampling changes the case fraction in the analysis sample.

Under suitable assumptions, logistic regression slope parameters can estimate population log odds-ratio associations, while the intercept is affected by case-control sampling.

Consequently, predicted probabilities obtained directly from an uncorrected model may reflect the artificially selected case fraction rather than population disease risk.

Calibration using valid population incidence or prevalence information, known sampling fractions, or other appropriate methods may permit probability estimation.

Such calibration requires explicit assumptions and should be evaluated for the population in which predictions will be used.

A model that discriminates cases from controls well does not necessarily produce accurate absolute-risk predictions.

Does an Odds Ratio Greater Than One Mean the Exposure Causes Disease?

No. An odds ratio quantifies an association under the specified sampling and analytical conditions.

It may reflect a causal relationship, but it can also be affected by confounding, selection bias, measurement error, reverse causation, or random variation.

For example, if an exposure is associated with another determinant of disease, the crude odds ratio may not represent the exposure's causal effect.

Likewise, differential recall may change the measured exposure distribution among cases and controls.

Researchers should therefore distinguish the estimated association from causal interpretation and avoid presenting relative associations as absolute probabilities.

Clear reporting should state the association measure, confidence interval, sampling scheme, and relevant limitations.

Why a 50% Case Fraction Does Not Mean 50% Disease Prevalence

The population prevalence is 1%, whereas the sample case fraction is 50%.

The difference arises because the investigators deliberately sampled cases at a much higher rate than noncases.

Now suppose the 200 sampled cases include 100 exposed individuals, while the 200 controls include 50 exposed individuals.

The investigators may report an estimated odds ratio of 3.00, together with an appropriate confidence interval and assessment of bias.

They cannot infer from the sample alone that exposed people have a 66.7% disease risk or that unexposed people have a 40% disease risk.

Those percentages reflect the deliberately selected sample.

This example demonstrates that a case-control study can produce an informative relative association while lacking the information needed to estimate absolute disease frequency directly.

The limitation is a consequence of outcome-based sampling, not evidence that the association analysis is inherently invalid.

Common Misconceptions About Risk and Prevalence Estimation

How Should You Decide Which Population Measure Your Study Can Report?

Begin by specifying the quantity your research question requires. An association between prior exposure and disease occurrence is not the same as the absolute probability of developing disease.

Then determine whether the available design and sampling information identify that quantity.

In the methods section, describe the control-sampling scheme and intended estimand. In the results, use terminology consistent with the actual parameter estimated.

For example, report an odds ratio as an odds ratio rather than converting it into a statement about percentage increases in risk without justification.

If absolute risks are calculated using external information, explain the source of that information, the estimation method, and the assumptions required.

These distinctions are particularly important when communicating findings to clinicians, policymakers, or the public, who may reasonably interpret statements about risk as statements about probability.

Before Reporting Risk or Prevalence, Verify What Your Design Identifies

Frequently Asked Questions About Case-Control Risk Estimation

Case-Control Studies Estimate Associations, Not Automatically Population Probabilities

How to Cite This Guide

This guide is intended to be read, shared, and used in research, teaching, and academic work. If you draw on its ideas, explanations, or other content, please acknowledge the source by citing the guide. Doing so gives appropriate credit and helps your readers locate the original resource.