Kaplan–Meier survival analysis overestimates cumulative incidence of health-related events in competing risk settings: a meta-analysis

Abstract

ObjectivesKaplan–Meier survival analysis overestimates cumulative incidence in competing risks (CRs) settings. The extent of overestimation (or its clinical significance) has been questioned, and CRs methods are infrequently used. This meta-analysis compares the Kaplan–Meier method to the cumulative incidence function (CIF), a CRs method. Study Design and SettingWe searched MEDLINE, EMBASE, BIOSIS Previews, Web of Science (1992–2016), and article bibliographies for studies estimating cumulative incidence using the Kaplan–Meier method and CIF. For studies with sufficient data, we calculated pooled risk ratios (RRs) comparing Kaplan–Meier and CIF estimates using DerSimonian and Laird random effects models. We performed stratified meta-analyses by clinical area, rate of CRs (CRs/events of interest), and follow-up time. ResultsOf 2,192 identified abstracts, we included 77 studies in the systematic review and meta-analyzed 55. The pooled RR demonstrated the Kaplan–Meier estimate was 1.41 [95% confidence interval (CI): 1.36, 1.47] times higher than the CIF. Overestimation was highest among studies with high rates of CRs [RR = 2.36 (95% CI: 1.79, 3.12)], studies related to hepatology [RR = 2.60 (95% CI: 2.12, 3.19)], and obstetrics and gynecology [RR = 1.84 (95% CI: 1.52, 2.23)]. ConclusionThe Kaplan–Meier method overestimated the cumulative incidence across 10 clinical areas. Using CRs methods will ensure accurate results inform clinical and policy decisions.