Joint Statistics/Biostatistics & Medical Informatics Seminar
Modelling, estimation and inference with time-varying effects in survival analysis
Thomas Scheike, Ph.D., Department of Biostatistics, University of Copenhagen
Monday, May 19, 2003, 3:30 pm
1221 Computer Science Statistics Center, 1210 W. Dayton St.
Models and methodology that deal with time-varying effects is needed in many practical settings. In medical studies where treatment effects are of interest it is often expected that these treatment effects will be time-varying, resulting, e.g., in an initially strong effect that wears off during the duration of the study. I present some approaches for dealing with time-varying effects in the context of proportional hazards models or additive hazards models. An important hypothesis when time-varying effects are studied is to decide if a covariate effect is significantly time-varying. I show a theoretically justified approach for answering such questions. One important aspect of the suggested methodology is that the testing of time-varying effects can be carried out in terms of successive tests. The theoretical foundation for the work is a study of the semi-parametric proportional and additive hazard models. References LIDA (8:247-262, 2002), Scandinavian Journal of Statistics (28:57-74, 2002) and unpublished work to appear in SJS.
Back to General Departmental Seminar Series