Kaplan–Meier

Kaplan–Meier
Name	Kaplan–Meier estimator
Type	Statistical estimator
Introduced	1958
Developers	Edward L. Kaplan; Paul Meier
Field	Biostatistics
Applications	Clinical trials; Epidemiology; Reliability engineering

Kaplan–Meier.

The Kaplan–Meier estimator is a nonparametric method for estimating time-to-event survival functions used widely in Clinical trials, Epidemiology studies, Oncology research and Reliability engineering. Developed by Edward L. Kaplan and Paul Meier in 1958 amid expanding interest in longitudinal studies, it provides stepwise estimates of survival probability that accommodate right-censored data arising in follow-up studies and cohort analyses. The estimator underpins many survival analyses reported by organizations such as the World Health Organization, Centers for Disease Control and Prevention, National Institutes of Health, and appears in regulatory submissions to agencies like the Food and Drug Administration.

Introduction

The Kaplan–Meier approach addresses incomplete follow-up typical in studies associated with Randomized controlled trials, Prospective cohort studys, Case-control studys and Registry analyses. It complements hypothesis tests such as the Log-rank test and graphical summaries used in reports by groups including the European Medicines Agency, American Society of Clinical Oncology, International Society for Pharmacoeconomics and Outcomes Research and academic centers like Johns Hopkins University and Mayo Clinic. Historically, the method influenced subsequent work by statisticians at institutions like Harvard University, University of Oxford, Stanford University and University of Cambridge.

Definition and Estimator

Formally, the Kaplan–Meier estimator constructs a step function estimating the survivor function S(t) using observed event times from cohorts such as patients enrolled in Cancer trials, participants in Framingham Heart Study, or components in Space Shuttle test programs. At each distinct event time observed in datasets from groups like British Medical Journal publications or New England Journal of Medicine articles, the estimate multiplies conditional survival probabilities inferred from risk sets assembled similarly to procedures in Cox proportional hazards model analyses. Implementations appear in statistical software developed by vendors including R (programming language), SAS, Stata (software), SPSS, and packages maintained at institutions such as The University of Washington and Carnegie Mellon University.

Assumptions and Properties

The estimator assumes noninformative censoring in settings like multicenter Randomized trials and cohort investigations managed by entities such as Centers for Medicare & Medicaid Services or National Cancer Institute. It presumes independent censoring across groups comparable to stratified analyses undertaken by teams at Memorial Sloan Kettering Cancer Center or Dana-Farber Cancer Institute. Key properties—consistency, right-continuity, and stepwise decrements at observed times—mirror theoretical developments from probability work at Princeton University, Columbia University, and University of Chicago. The estimator aligns with counting-process theory advanced by researchers associated with University of Washington and University of Minnesota.

Variance Estimation and Confidence Intervals

Variance estimates commonly derive from Greenwood's formula, originally appearing in literature connected to authors associated with University of Iowa and cited across journals like Statistics in Medicine and Biometrika. Confidence intervals for the survival function utilize transformations (log-log, arcsine) recommended in guidance from bodies such as International Council for Harmonisation of Technical Requirements for Pharmaceuticals for Human Use and statistical texts used in curricula at Massachusetts Institute of Technology, Yale University, and University of California, Berkeley. Variance estimation supports hypothesis testing frameworks used by trial statisticians at Pfizer, Roche, and Novartis when preparing submissions to regulators like European Medicines Agency.

Comparison and Related Methods

The Kaplan–Meier estimator contrasts with parametric survival models like the Weibull distribution, Exponential distribution models and semiparametric approaches such as the Cox proportional hazards model. Extensions and alternative techniques include life-table methods used historically in analyses by the British Medical Journal and actuarial practice at firms such as Lloyd's of London and Prudential plc. Competing nonparametric procedures and multistate models have been developed by researchers affiliated with Karolinska Institute, Max Planck Society, Johns Hopkins Bloomberg School of Public Health and incorporated into textbooks by authors at Oxford University Press.

Applications in Medicine and Research

Kaplan–Meier curves are ubiquitous in oncology trials reported by groups like European Society for Medical Oncology and American Association for Cancer Research, in cardiology trials published with authors from Cleveland Clinic and Mount Sinai Health System, and in infectious disease studies by teams at Centers for Disease Control and Prevention and World Health Organization. The method supports analyses of transplant outcomes in programs at Mayo Clinic and University of Pennsylvania Health System, vaccine efficacy follow-ups coordinated with Bill & Melinda Gates Foundation projects, and device reliability assessments in aerospace programs by National Aeronautics and Space Administration. It is routinely taught in biostatistics courses at institutions including Harvard T.H. Chan School of Public Health, Columbia University Mailman School of Public Health, and London School of Hygiene & Tropical Medicine.

Limitations and Extensions

Limitations include sensitivity to informative censoring observed in observational studies by groups such as Surveillance, Epidemiology, and End Results Program and inability to adjust for time-dependent covariates without augmentation by models like the Cox proportional hazards model or techniques from Marginal structural model literature. Extensions include competing-risks methods advanced in collaborations involving Stanford University and University of California, San Francisco, multistate models used by researchers at University College London, and machine-learning survival adaptations developed by teams at Google and Microsoft Research. Contemporary work integrates Kaplan–Meier estimators into reproducible pipelines following standards promoted by The REporting of studies Conducted using Observational Routinely-collected health Data initiatives and clinical trial registries such as ClinicalTrials.gov.

Category:Statistical methods