Retrospective Kaplan-Meier Analysis — Historical Survival Curve Estimation
Retrospective Kaplan-Meier Survival Analysis · Also known as: retrospective KM analysis, retrospective survival curve estimation, historical Kaplan-Meier, retrospective KM estimator
Retrospective Kaplan-Meier analysis applies the Kaplan-Meier product-limit estimator to time-to-event data drawn from existing records — medical charts, registries, or administrative databases — rather than from a prospectively followed cohort. The method estimates the probability of surviving (or remaining event-free) beyond any given time point while accounting for participants whose follow-up ended before the event occurred (censored observations). It is among the most commonly reported analyses in clinical oncology, cardiology, and surgery.
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 KM analysis when: (1) a clinical question concerns time-to-event outcomes and data already exist in records or registries; (2) prospective follow-up is not feasible due to time, cost, or the rarity of the condition; (3) the sample size from historical data is adequate to detect clinically meaningful differences in survival. Do NOT use when: event and censoring times cannot be reconstructed reliably from records; censoring is likely to be informative (e.g., patients with worsening disease are disproportionately lost); the cohort suffers severe survivor bias or immortal-time bias introduced by the retrospective design; or when a parametric survival model or competing-risks framework better matches the clinical question. Prefer a prospective design when high internal validity is required and prospective follow-up is feasible.
Strengths & limitations
- Rapid and cost-effective — leverages existing clinical data without requiring years of prospective follow-up.
- Non-parametric: makes no assumption about the shape of the underlying survival distribution, making it robust across diverse clinical contexts.
- Intuitive graphical output — the step-function survival curve is easily interpreted by clinicians and reviewers.
- Handles censored observations correctly, including administrative censoring at the data extraction date.
- Large sample sizes are often achievable from registries, increasing power to detect moderate differences in survival.
- Susceptible to selection bias: patients whose records exist and meet extraction criteria may differ systematically from those who do not.
- Confounding by indication — treatment groups in retrospective data are not randomly assigned, so baseline differences can distort survival comparisons.
- Record completeness: missing event dates, undocumented deaths, or gaps in follow-up introduce error that cannot be recovered post-hoc.
- Immortal-time bias can inflate apparent survival if the time between cohort entry and treatment initiation is inadvertently excluded from risk exposure.
- Does not provide causal estimates; any survival difference between groups may reflect unmeasured confounders rather than the effect of the exposure or treatment.
Frequently asked
How is retrospective KM analysis different from prospective KM analysis?
The mathematical method is identical — both use the Kaplan-Meier product-limit estimator. The difference lies in data collection: in prospective studies, patients are enrolled and followed forward in time under a defined protocol; in retrospective studies, event and follow-up times are reconstructed from records that already exist. Retrospective designs are faster and cheaper but carry higher risks of selection bias, confounding, and data quality problems that prospective designs can control through protocol and randomization.
Can I use KM analysis when patients have competing events (e.g., death from other causes)?
Standard KM treats competing events as censored observations, which overestimates the cumulative incidence of the primary endpoint when competing risks are substantial. For outcomes such as cancer-specific mortality in elderly cohorts, a competing-risks analysis using the cumulative incidence function (CIF) and the Fine-Gray subdistribution hazard model is more appropriate. KM is adequate when competing events are rare or the research question concerns overall rather than cause-specific survival.
How should I handle immortal-time bias in a retrospective study?
Immortal-time bias arises when the time between a patient's cohort entry and the defining event (e.g., starting a treatment) is counted in the exposed group's risk time even though no event can occur during that window. The remedy is to define cohort entry at the same calendar date for all subjects, or to use a time-varying covariate Cox model that assigns exposure status dynamically rather than at baseline.
What sample size do I need?
Power in survival studies depends on the number of observed events, not on the number of enrolled subjects. As a rule of thumb, at least 10–15 events per group are needed for a reliable KM curve, and 20–30 events per group for stable log-rank testing and Cox regression. Online tools (e.g., nQuery, G*Power) or formulas based on the log-rank statistic can calculate the required number of events given the expected hazard ratio and censoring rate.
Do I need to report this according to a specific guideline?
Yes. Retrospective observational survival studies should follow the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) checklist, which covers cohort definition, follow-up description, handling of missing data, and statistical methods. Journals in oncology and cardiology also commonly require the REMARK (Reporting Recommendations for Tumor Marker Prognostic Studies) checklist when survival is linked to a biomarker.
Sources
- Kaplan, E. L., & Meier, P. (1958). Nonparametric estimation from incomplete observations. Journal of the American Statistical Association, 53(282), 457–481. DOI: 10.1080/01621459.1958.10501452 ↗
- Clark, T. G., Bradburn, M. J., Love, S. B., & Altman, D. G. (2003). Survival analysis part I: Basic concepts and first analyses. British Journal of Cancer, 89(2), 232–238. DOI: 10.1038/sj.bjc.6601118 ↗
How to cite this page
ScholarGate. (2026, June 3). Retrospective Kaplan-Meier Survival Analysis. ScholarGate. https://scholargate.app/en/epidemiology/retrospective-kaplan-meier-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
- Log-Rank TestSurvival↔ compare
- Retrospective Cohort StudyEpidemiology↔ compare
- Retrospective survival analysisEpidemiology↔ compare