Choosing nested case-control or case-cohort
Choose risk-set sampling for a specific outcome or a reusable subcohort for several outcomes.
Fit
When it matters, and when it does not
It matters when
- You study a specific outcome and can sample controls at each event time.
- You need time-matched exposure measurements and efficient risk-set comparisons.
It matters less when
- Choose case-cohort when a common random subcohort should support several outcomes. Read: Case-cohort study
- If exposure measurement is cheap for everyone, use the full cohort. Read: Prospective cohort study
Before you start
What to check in your study
- Source cohortDefine entry, follow-up, outcome times and who remains at risk.
- Exposure measurementCheck specimen timing, assay cost and availability.
- Future outcomesDecide whether one sample must support several disease endpoints.
- The sampling probabilities and eligible risk sets are known.How to check: Keep selection records and any matching strata.
- The analysis reflects how people were sampled.How to check: Use conditional risk-set methods or a case-cohort weighted method with appropriate variance.
- Unavailable specimens may create selection bias.How to check: Justify missing at random for imputation; test departures if missingness may depend on unseen exposure.
Plain language
How it works
Nested case-control sampling selects controls from the risk set at each event. A control may later become a case and may be sampled again. Case-cohort sampling selects a random subcohort from the source cohort, plus cases outside it. Sampling can be stratified if probabilities are retained.
Langholz and Thomas compare the designs through efficiency and cost. Barlow and colleagues describe case-cohort analysis. Choose using outcome frequency, specimen logistics and reuse needs. Neither design is a cohort followed backwards. An ordinary unweighted analysis of the selected people can give the wrong answer.
Worked example
A worked example
- Setting
- A biobank has 1,000 people and 40 later cases.
- Question
- Should the team create a reusable assay sample?
- Approach
- A random subcohort contains 100 people. Of the 40 cases, 6 are already in that subcohort.
- Result
- The subcohort is 100 of 1,000 people, or 10 in 100. Adding the 34 outside cases gives 134 distinct people to assay.
- What it means
- With four controls per case, nested sampling has 160 control slots, not necessarily 160 distinct people. Overlap and repeated selection decide the assay count.
In practice
Recent studies PubMed search · not vetted
Watch out
Common mistakes
The one to remember
Nested case-control controls come from those still at risk when each case occurs. A case-cohort subcohort is sampled independently of future case status and can include future cases. Both need sampling-aware analysis.
- Excluding future cases from the subcohort.Instead: Sample independently of future outcome status.
- Treating all control slots as different people.Instead: Count distinct specimens after accounting for overlap.
- Fitting ordinary Cox regression to the sample.Instead: Use the method and variance for the sampling design.
Inside Mallard
How Mallard uses this
Mallard includes this comparison in Learn search and the Atlas. Plans do not link to this comparison automatically.
concept.choosing_ncc_case_cohort, the id a plan cue records.Related Learn pages
Related designs and methods
Evidence
Sources & further reading
- Nested case-control and case-cohort methods of sampling from a cohort: a critical comparison (opens in a new tab) · Am J Epidemiol · doi 10.1093/oxfordjournals.aje.a115471
- Analysis of case-cohort designs (opens in a new tab) · J Clin Epidemiol · doi 10.1016/s0895-4356(99)00102-x