Colloquium

Tuesday, April 15, 2008

4:00 PM, BAC 219

Speaker: Harry Khamis, Statistical Consulting Center, Wright State University, Dayton, Ohio

Hosts: John Bailer and Vasant Waikar

Title: Testing the Proportional Hazards Assumption in the Cox Model

Abstract: Sir David Cox’s 1972 JRSS paper, where he introduced the Cox proportional hazards regression model, is the most cited paper in the statistical sciences.  This model is used extensively in health and medical research because of the interest in studying “time to event” data and because of the occurrence of censored observations in such data. 

This presentation will assume no prior knowledge of the Cox model; terminology, notation, and ultimately the Cox model will be developed from first principles.  The key model assumption, with which the data must be consistent in order for the analysis to render valid results, is the so-called proportional hazards (PH) assumption.  This will be defined, and the (rather surprising) results of a comparative study of the most successful tests currently available for diagnosing violations of the PH assumption will be presented.