Classical survival analysis is based on the assumption that censoring (causing incomplete observations) does not affect the risk of the studied event. When the assumption is not fulfilled, if censoring either precludes the occurrence of a event or alters the probability of occurrence of that event, a competing risk situation exists. The statistical methodology for analyzing competing risks has been rapidly developed during the last decades. Still, competing risks seem not often to be recognized in clinical research, which may bias the outcome in these studies.
Koller et al. (1) examined how competing risk issues were handled in high-impact medical journals by reviewing 50 articles, where time-to-event endpoints were analyzed, published in Ann Intern Med, BMJ, JAMA, Lancet, NEJM and PLOS Medicine during 2007-2010.
In 37 of the 50 articles (74%), the definition of at least one endpoint implied competing risks. In 35 of these (70%) at least one competing risk issue was observed. Only 9 of the 50 articles (18%) discussed competing risks as a possible issue, and only 2 (4%) presented correct risk estimates instead of the biased ones from classical methods.
The authors conclude that “A better recognition of competing risks in the clinical community is needed”.
1. Koller MT, Raatz H, Steyerberg EW, Wolbers M. Competing risks and the clinical community: irrelevence or ignorance? Stat Med 2012;31:1089-1097.