Retrospective Competing Risks Analysis
Also known as: retrospective CRA, competing risks survival analysis (retrospective), cause-specific hazard analysis (retrospective), subdistribution hazard analysis (retrospective)
Retrospective competing risks analysis applies competing risks methodology to historical (already-collected) time-to-event data in which subjects can experience one of several mutually exclusive endpoints. It uses the cumulative incidence function and cause-specific or subdistribution hazard models to estimate the probability of each event type while accounting for the fact that occurrence of one event permanently precludes the others. Widely used in oncology, cardiology, and transplant medicine where administrative or registry records are the data source.
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 retrospective competing risks analysis when: (a) time-to-event data already exist in registries, administrative databases, or medical records; (b) subjects can experience multiple mutually exclusive endpoints and the primary endpoint probability would be overestimated by standard survival analysis; (c) research aims are descriptive (CIF estimation), comparative (Gray's test), or prognostic (Fine-Gray model). Do NOT use when the data were collected prospectively with full covariate control — a prospective competing risks design is then preferable. Do not apply if censoring is informative and cannot be modelled, if the competing events are not truly mutually exclusive, or if the retrospective records are too incomplete to reliably reconstruct the event timeline.
Strengths & limitations
- Correctly estimates absolute event probabilities in the presence of competing events, avoiding the upward bias of 1-minus-Kaplan-Meier.
- Exploits large existing registries and administrative datasets, enabling sample sizes rarely achievable in prospective studies.
- The Fine-Gray model directly targets the clinically relevant cumulative incidence, making results actionable for risk prediction and clinical decision support.
- Allows etiologic and predictive questions to be addressed simultaneously using cause-specific and subdistribution hazard models on the same dataset.
- Widely accepted in oncology, cardiology, and transplant research, with established reporting guidelines (e.g., STROBE, TRIPOD).
- Retrospective data cannot guarantee comparability of exposure groups; unmeasured confounding is a persistent threat that propensity-score methods can reduce but not eliminate.
- Event ascertainment depends on historical coding quality — ICD coding changes, incomplete cause-of-death information, and administrative censoring can introduce misclassification bias.
- Covariate data are limited to what was recorded at the time; time-varying confounders and post-baseline covariates are often unavailable.
- Left-truncation (subjects already under follow-up when the database begins) is frequently overlooked, leading to biased hazard estimates if not explicitly handled.
- Fine-Gray subdistribution hazards weight subjects by the inverse probability of censoring; retrospective datasets with complex censoring patterns may violate these weighting assumptions.
Frequently asked
Why can I not just use Kaplan-Meier and censor competing events?
Censoring competing events in Kaplan-Meier analysis treats those subjects as if they remain at risk for the primary event indefinitely, which is biologically impossible once a competing event has occurred. This inflates the estimated cumulative incidence. The cumulative incidence function from competing risks analysis correctly attributes probability mass across all event types, ensuring the estimates sum to the observed overall event probability.
Should I use the cause-specific or Fine-Gray model?
The cause-specific hazard model is preferred for etiologic questions — it estimates the instantaneous hazard among those still at risk and is the natural framework for understanding biological mechanisms. The Fine-Gray subdistribution hazard model is preferred for prediction questions — it models the cumulative incidence directly and is more interpretable for clinical risk scores and decision tools. Ideally, report both to provide a complete picture.
How do I handle left-truncation in registry data?
Left-truncation occurs when subjects entered the registry after the study's nominal time origin (e.g., diagnosis date predates registry enrollment). These subjects should enter the risk set only at the time they became observable, not at time zero. In R this is handled by specifying an entry time in the Surv() object; ignoring it inflates early risk-set sizes and biases hazard estimates downward.
What software implements competing risks analysis?
In R, the cmprsk package provides Gray's test and Fine-Gray regression; the survival package handles cause-specific Cox models; mstate and riskRegression provide additional tools including time-varying covariates. In Stata, stcrreg implements the Fine-Gray model. SAS PROC PHREG with the EVENTCODE option fits subdistribution hazard models from version 9.4 onward.
How large a sample is needed?
As a practical rule, each event type should have at least 10 events per covariate in the model (events-per-variable rule). Because the effective event count for each cause is a fraction of total events, retrospective datasets from large registries are often necessary to fit adequately powered competing risks regression models with multiple predictors.
Sources
- Fine, J. P., & Gray, R. J. (1999). A proportional hazards model for the subdistribution of a competing risk. Journal of the American Statistical Association, 94(446), 496–509. DOI: 10.1080/01621459.1999.10474144 ↗
- Prentice, R. L., Kalbfleisch, J. D., Peterson, A. V., Flournoy, N., Farewell, V. T., & Breslow, N. E. (1978). The analysis of failure time data in the presence of competing risks. Biometrics, 34(4), 541–554. DOI: 10.2307/2530374 ↗
How to cite this page
ScholarGate. (2026, June 3). Retrospective Competing Risks Analysis. ScholarGate. https://scholargate.app/en/epidemiology/retrospective-competing-risks-analysis
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.
- Cox proportional hazardsEpidemiology↔ compare
- Kaplan-Meier AnalysisEpidemiology↔ compare
- Retrospective Cohort StudyEpidemiology↔ compare
- Retrospective survival analysisEpidemiology↔ compare