An evaluation of the effect of a healthcare intervention (or an exposure) must consider multiple possible outcomes, including the primary outcome of interest and other outcomes such as adverse events or mortality. The determination of the likelihood of benefit from an intervention, in the presence of other competing outcomes, is a competing risks problem. Although statistical methods exist for quantifying the probability of benefit from an intervention while accounting for competing events, these methods have not been widely adopted by clinical researchers.
(1) To demonstrate the importance of considering competing risks in the evaluation of treatment effectiveness, and (2) to review appropriate statistical methods, and recommend how they might be applied.
We reviewed 3 statistical approaches for analyzing the competing risks problem: (a) cause-specific hazard (CSH), (b) cumulative incidence function (CIF), and (c) event-free survival (EFS). We compare these methods using a simulation study and a reanalysis of a randomized clinical trial.
Simulation studies evaluating the statistical power to detect the effect of intervention under different scenarios showed that: (1) CSH approach is best for detecting the effect of an intervention if the intervention only affects either the primary outcome or the competing event; (2) EFS approach is best only when the intervention affects both primary and competing events in the same manner; and (3) CIF approach is best when the intervention affects both primary and competing events, but in opposite directions. Using data from a randomized controlled trial, we demonstrated that a comprehensive approach using all 3 approaches provided useful insights on the effect of an intervention on the relative and absolute risks of multiple competing outcomes.
CSH is the fundamental measure of outcome in competing risks problems. It is appropriate for evaluating treatment effects in the presence of competing events. Results of CSH analysis for primary and competing outcomes should always be reported even when EFS or CIF approaches are called for. EFS is appropriate for evaluating the composite effect of an intervention, only when combining different endpoints is clinically and biologically meaningful, and the treatment has similar effects on all event types. CIF is useful for evaluating the likelihood of benefit from an intervention over a meaningful period. CIF should be used for absolute risk calculations instead of the widely used complement of the Kaplan-Meier (1 − KM) estimator.
From the *Department of Medicine, Johns Hopkins University School of Medicine, Baltimore, MD; and Departments of †Health Policy and Management, and ‡Biostatistics, Johns Hopkins University Bloomberg School of Public Health, Baltimore, MD.
Supported by the Agency for Healthcare Research and Quality under task order HHSA290–2005–0034-I-TO2-WA4 (Project ID 20-EHC).
The authors of this manuscript are responsible for its content. No statement may be construed as the official position of the Agency for Healthcare Research and Quality of the US Department of Health and Human Services.
Reprints: Ravi Varadhan, PhD, The Center on Aging and Health, Suite 2–700, 2024 E. Monument St, Baltimore, MD 21205. E-mail: email@example.com.