Bayesian Nested Case-Control Study
Also known as: Bayesian NCC, Bayesian nested case-referent study, Bayesian sampled case-control within cohort
A Bayesian nested case-control study embeds a case-control sampling scheme within a defined prospective cohort and then estimates exposure-outcome associations using Bayesian inference. Cases are individuals in the cohort who develop the outcome of interest; controls are sampled from the risk set at the time each case is identified. The Bayesian framework allows incorporation of prior knowledge — from earlier studies, expert opinion, or biological plausibility — and produces full posterior distributions for effect estimates rather than single-point estimates with confidence intervals.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use a Bayesian nested case-control when (1) you have access to a well-characterised prospective cohort with stored specimens or recorded exposures, (2) full cohort analysis is cost-prohibitive or impossible due to expensive laboratory assays, (3) you wish to formally incorporate prior evidence into estimation rather than rely solely on the current data, and (4) the outcome is rare enough that standard cohort analysis would require enormous samples. This design is especially valuable when sample sizes are modest and you want to borrow strength from prior studies rather than ignoring them. Do NOT use it when no pre-existing cohort exists (use a traditional case-control or prospective cohort instead), when the outcome is too common for case-control efficiency gains, when meaningful prior information is unavailable and the Bayesian machinery adds complexity without benefit, or when the regulatory context requires frequentist inference only.
Strengths & limitations
- Preserves the temporal and population advantages of the parent cohort while dramatically reducing the cost of exposure assessment.
- Risk-set sampling ensures controls are drawn from the same underlying population as cases, minimising selection bias.
- Bayesian inference provides full posterior distributions, enabling direct probability statements and straightforward incorporation of prior knowledge.
- Particularly powerful when prior studies exist: borrowing strength across studies improves precision without requiring a new large trial.
- Handles sparse data better than purely frequentist conditional logistic regression through regularisation imposed by priors.
- Requires an existing, well-maintained cohort with complete follow-up data — a significant infrastructural prerequisite.
- Prior specification is subjective; poorly chosen priors can bias results, and transparent sensitivity analyses are essential but add reporting burden.
- MCMC estimation is computationally intensive and requires statistical expertise to implement, diagnose, and interpret correctly.
- The design captures only the subset of cohort information used in sampled sets; full cohort analysis is more efficient if exposure measurement costs allow it.
Frequently asked
How does a nested case-control differ from a standard case-control study?
In a standard case-control study, cases and controls are recruited from potentially different populations, creating selection bias risk. In a nested case-control, both cases and controls are drawn from the same well-defined prospective cohort, ensuring they share the same underlying population and follow-up structure. Exposure information was recorded before the outcome occurred, eliminating recall bias for historical exposures.
How many controls per case should I sample?
One to ten controls per case is the conventional range. Statistical efficiency increases up to about four controls per case and gains little beyond ten. The optimal number depends on the ratio of case to control costs: if exposure measurement is expensive, fewer controls are preferable; if controls are cheap to recruit and measure, five or more can be cost-effective. For rare outcomes with small case counts, more controls improve power.
What prior distribution should I use for the odds ratio?
A common default is a weakly informative normal prior on the log odds ratio, such as N(0, 1) or N(0, 2.5), which rules out implausibly large effect sizes while allowing the data to dominate. If a meta-analysis of previous studies exists, its pooled estimate and standard error can directly parameterise an informative prior. Always run a sensitivity analysis comparing results under different prior choices — if conclusions change substantially, report both.
Do I need to account for matching in the Bayesian analysis?
Yes. Matching in risk-set sampling is accounted for by using conditional logistic regression as the likelihood within the Bayesian model. Ignoring the matched structure by using standard logistic regression can induce bias and underestimate variance. In practice, MCMC samplers such as Stan or JAGS accept a conditional likelihood specification that correctly handles the matched case-control sets.
Is Bayesian nested case-control accepted in regulatory submissions?
Regulatory agencies (FDA, EMA) have published guidance increasingly open to Bayesian approaches, particularly for pharmacoepidemiology and medical device studies where prior data are strong. However, pre-specification of prior distributions and comprehensive sensitivity analyses are required. In contexts where only frequentist results are accepted, the design can still be run as a classical nested case-control and the Bayesian layer omitted.
Sources
- Thomas, D. C. (1977). Addendum to: Methods of cohort analysis: Appraisal by application to asbestos mining. Journal of the Royal Statistical Society, Series A, 140(4), 469–491. link ↗
- Wakefield, J. (2007). Disease mapping and spatial regression with count data. Biostatistics, 8(2), 158–183. DOI: 10.1093/biostatistics/kxl008 ↗
How to cite this page
ScholarGate. (2026, June 3). Bayesian Nested Case-Control Study. ScholarGate. https://scholargate.app/en/epidemiology/bayesian-nested-case-control
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Bayesian Case-Control StudyEpidemiology↔ compare
- Bayesian Cohort StudyEpidemiology↔ compare
- Case-control studyEpidemiology↔ compare
- Cohort StudyEpidemiology↔ compare
- Nested case-controlEpidemiology↔ compare