Matched Kaplan-Meier Analysis — Survival Estimation in Matched Cohorts
Matched Cohort Kaplan-Meier Survival Analysis · Also known as: KM analysis in matched cohorts, propensity-matched survival curves, matched survival analysis, paired Kaplan-Meier
Matched Kaplan-Meier analysis estimates and compares survival functions in groups that have been pre-balanced through individual or propensity-score matching. By applying the Kaplan-Meier product-limit estimator to matched cohorts or matched pairs, investigators can visualize time-to-event outcomes while controlling for confounders that would otherwise distort treatment or exposure comparisons in observational data.
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 matched Kaplan-Meier analysis when you have an observational time-to-event dataset, a clear exposed/unexposed or treated/untreated contrast, and a set of measured confounders that can be balanced through matching. It is especially appropriate when sample sizes are sufficient to achieve balance after matching (losing unmatched units) and when the clinical question concerns absolute survival differences rather than hazard ratios. Do not use it when matching discards so many subjects that the resulting sample is too small for stable estimates, when important confounders are unmeasured (unmeasured confounding cannot be addressed by matching), when the proportional hazards assumption is of primary interest and a Cox model would be more efficient, or when the data arise from a randomized trial where simple Kaplan-Meier curves without matching are already unbiased.
Strengths & limitations
- Controls for measured confounders without imposing a parametric model on the survival distribution.
- Produces intuitive visual survival curves that are easy to communicate to clinical audiences.
- The matching step makes the comparison groups transparent and the balance verifiable before analysis.
- Flexible matching strategies (exact, caliper, propensity-score) accommodate diverse confounding structures.
- Restricted mean survival time (RMST) summaries from the KM curves avoid the proportional hazards assumption entirely.
- Only measured confounders can be balanced; unmeasured confounders remain a threat to causal inference.
- Matching discards unmatched controls, reducing effective sample size and statistical power.
- Standard Kaplan-Meier confidence intervals and log-rank tests assume independence; ignoring the matched structure inflates Type I error.
- When many subjects are unmatched, the matched sample may not represent the original population, limiting generalizability.
- Propensity-score models must be correctly specified; misspecification leads to residual imbalance that the KM curves cannot reveal.
Frequently asked
Can I use the standard log-rank test after matching?
No — or at least not without caution. Matched pairs are correlated, violating the independence assumption of the standard log-rank test. The stratified log-rank test (each pair is a stratum) or inference based on restricted mean survival time differences with robust variance are the appropriate alternatives. Using the standard log-rank test on matched data typically produces p-values that are too small.
How do I choose between matched KM analysis and a matched Cox model?
Matched KM analysis is preferred when you want to visualize absolute survival probabilities or compute RMST differences without imposing the proportional hazards assumption. A matched Cox model (with a robust sandwich variance or stratified by matched set) is preferred when you want a hazard ratio, need to adjust for residual confounders after matching, or when the proportional hazards assumption holds. Both approaches can be complementary in the same paper.
How many controls per case should I match?
One-to-one matching is simplest and avoids heterogeneity within matched sets. Matching up to four or five controls per case can improve power when controls are plentiful and cases are scarce, but gains in efficiency diminish beyond a ratio of 1:4. The matching ratio should be specified before data analysis.
What if my matched groups still show imbalance after matching?
If SMDs remain above 0.10 for important covariates, additional adjustment is needed. Options include re-specifying the propensity model, tightening the caliper, or complementing the matched KM analysis with a covariate-adjusted Cox model. Reporting residual imbalance honestly and discussing its potential impact is essential.
Is matched KM analysis the same as propensity-score matched survival analysis?
Propensity-score matching is the most common method used to create the matched cohort, but matched KM analysis is also compatible with exact matching, Mahalanobis distance matching, or fine stratification. The KM estimation step is the same regardless of how the matched cohort was constructed.
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 ↗
- Austin, P. C. (2014). Pointwise confidence intervals for restricted mean survival time in a propensity-matched analysis. Statistics in Medicine, 33(14), 2659-2671. link ↗
How to cite this page
ScholarGate. (2026, June 3). Matched Cohort Kaplan-Meier Survival Analysis. ScholarGate. https://scholargate.app/en/epidemiology/matched-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
- Matched case-control studyEpidemiology↔ compare
- Matched Cohort StudyEpidemiology↔ compare
- Propensity Score MatchingResearch Statistics↔ compare
- Survival AnalysisResearch Statistics↔ compare