U-Statistics for Multiple Censored Outcomes with Varying Frequency, Severity, Attribution

• Activity Number: 244
• Type: Contributed
• Date/Time: Monday, August 5, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Knut Wittkowski*+
• Companies: The Rockefeller University
• Keywords: censoring ; attribution ; frequency ; severity ; composite outcome ; nonparametric

Abstract

When several (repeated, censored) events graded by severity and/or attribution need to be considered (e.g., hospitalization, myocardial infarction (MI), stroke, death), many strategies currently discussed either are not comprehensive enough or require unrealistic assumptions to be made, with often unpredictable consequences for validity.

We present a generalizations of the Gehan/Wei/Knuiman test to graded variables interrelated by complex factor/sub-factor structures to incorporate information about ordinal and hierarchical/factorial relationships between variables. A flexible algorithm allows the univariate partial orderings of cumulative variables (by severity and attribution) to be combined first within and then across disease categories, allowing attribution, severity, time-to-nth-event, and survival to be integrated under weak assumptions without requiring that the relative importance of the outcomes be quantified.

The proposed approach fulfills the ICH E9 criteria for "a predefined algorithm", yet is flexible enough to objectively incorporates information about graded, repeated, and censored events for a wide range of complex diseases.