A Reverse Counting Process for Analyzing Survival Data with Multiple Event Times

• Activity Number: 454
• Type: Contributed
• Date/Time: Wednesday, August 6, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Brian Claggett*+ and Hajime Uno and Lu Tian and Lee Jen Wei
• Companies: Harvard Medical School and Dana-Farber Cancer Institute and Stanford University and Harvard
• Keywords: Survival ; Non-parametric ; Multiple events ; Regression ; Clinical Trials

Abstract

In many long-term clinical studies, it is conventional practice to define multiple distinct clinical endpoints, each of which indicate a worsening of a patient's overall health. For example, major adverse cardiovascular endpoints (MACE) are often defined to include death, myocardial infarction, and stroke, and may further include angina, revascularization, bleeding events, or cardiovascular hospitalization. Despite the variety of clinical outcomes monitored in such trials, the global comparison is often conducted by analyzing only the time to the first occurrence of any of the events. As a result, much of the rich information concerning a patient's ``event history'' is never fully utilized. Problems with time-to-first analysis have been well described, and well-known methods for the analysis of multiple or recurrent events are often model-dependent.

In this paper, we propose a simple "reverse counting process", which extends the Kaplan-Meier estimator to account for multiple events per patient. We show how to utilize these methods for the purposes of non-parametric two-sample comparisons, as well as multivariable regression models to estimate adjusted treatment effects.