Improving estimates of event incidence over time in populations exposed to other events: application to three large databases.

The Kaplan-Meier (KM) method is commonly used to estimate the incidence of an event over time. It assumes independence between the event of interest and any competing event that precludes the event of interest to occur. However, when the competing event is death without the event of interest, censoring these patients will affect the incidence of the event of interest by modifying the number of exposed patients, so that KM results will be misleading. Three prospective cohorts were studied: (1) 657 renal transplant recipients, (2) 262 children with acute leukemia who received bone marrow transplants, and (3) 8,353 intensive care patients. The main outcome measures were kidney graft loss, leukemia relapse, and ICU-acquired infection, respectively, with death before the main outcome as the competing event. The incidence of each main outcome was overestimated by the KM method. The magnitude of overestimation ranged from 3% to 30%, and varied with baseline patient characteristics and follow-up duration, with most of this variation being related to the rate of the competing event. A competing-risk approach must be used to analyze the risk of events other than death in cohort studies, particularly when mortality rates are high.