Department of Statistics Seminar
North Carolina State University

presents

Jason P. Fine

Harvard University

"Analysis of Competing Risks Data with Non-Proportional Hazards Models"

ABSTRACT

With explanatory covariates, the standard semi-parametric analysis for competing risks data models the cause-specific hazard functions under a proportional hazards assumption. The cause-specific hazard functions are quite useful but do not have a natural interpretation in terms of the crude survival probabilities, or cumulative incidence functions, for particular failure types. Applied statisticians may be especially interested in the cumulative incidence in risk-benefit analyses in which the marginal failure probabilities for the different types provide an intuitively appealing summary of risk-benefit tradeoffs for decision makers. We will begin by introducing a well-known prostate cancer clinical trial in which treatment with a potentially life saving chemotherapy had an unexpected and fatal adverse effect (Byar and Green, 1980). Next, basic concepts in competing risks theory, including issues of identifiability and interpretability, will be reviewed. Newer methods which predict the cumulative incidence by combining estimates of the cause-specific hazard functions under the proportional hazards formulation will be presented (Cheng, Fine, and Wei, 1998). Unfortunately, these methods do not allow the analyst to directly assess the effect of a covariate on the marginal probability function. Novel estimation, inference, and prediction procedures for a rich class of non-proportional hazards models in which the cumulative incidence is decomposed via a semi-parametric mixture model with precedents in the cure rate literature will be developed. Simple graphical techniques can be used to select the best fitting candidate model. The methods will be compared and contrasted on the prostate cancer data.

Thursday, January 22, 1998

8:00 - 9:00 am

124 Dabney Hall

Refreshments will be served in 124 Dabney Hall at 7:45.