CENSORING AND ITS IMPACT ON KAPLAN–MEIER SURVIVAL ESTIMATES: INSIGHTS FROM A SIMULATION STUDY

Abstract

Background: Survival analysis is essential in medical research for studying time-to-event outcomes. The Kaplan–Meier (KM) estimator is widely used but its performance depends heavily on censoring. Objective: To examine the impact of varying levels of censoring (10%, 30%, 50%) on KM survival estimates using simulated data. Methods: A simulation study was conducted with 1,000 patients per dataset, assuming survival times followed an exponential distribution with a true median of 12 months. Random right censoring was introduced at three levels: low (10%), moderate (30%), and high (50%). Each scenario was replicated 1,000 times. Kaplan–Meier estimates of median survival were compared against the true value in terms of bias, precision, and confidence interval coverage. Results: At 10% censoring, KM estimates closely matched the true median (bias = –0.2 months; SE = 1.1; CI coverage = 95%). At 30% censoring, bias increased to –1.1 months with reduced CI coverage (92%). At 50% censoring, median survival was underestimated by –3.4 months, SE nearly doubled (2.3), and CI coverage dropped to 85%.Conclusion: The Kaplan–Meier method is reliable under low censoring but underestimates survival when censoring is high. Researchers should report censoring rates, interpret KM estimates cautiously, and consider complementary methods such as Cox regression, parametric survival models, or restricted mean survival time.