|
Activity Number:
|
325
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 6, 2003 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Risk Analysis
|
| Abstract - #301220 |
|
Title:
|
A Modified Estimator for the Cumulative Covariate Effect in Aalen's Additive Risk Model
|
|
Author(s):
|
Michael Brent McHenry*+ and Chung-Chou H. Chang and Stewart J. Anderson and Howard E. Rockette and Lisa A. Weissfeld and Mark Roberts
|
|
Companies:
|
University of Pittsburgh and University of Pittsburgh and University of Pittsburgh and University of Pittsburgh and University of Pittsburgh and University of Pittsburgh
|
|
Address:
|
Biostatistics, Pittsburgh, PA, 15261-0001,
|
|
Keywords:
|
quadratic programming ; Aalen's Additive Risk Model ; cumulative regression estimator ; linear complimentary problem
|
|
Abstract:
|
Aalen (1989) proposed the ordinary least squares (OLS) method for computing cumulative regression coefficients of Aalen's additive risk model (1980). This method is used to compute the cumulative covariate effect at each event time assuming that the effects of covariates on time to an event are additive rather than multiplicative. However the estimated hazard function via the OLS method is unrestricted, which may result in negative hazard values. One way this problem can be resolved is by setting negative estimated hazard values to zero. Such an ad hoc solution, however, is likely to bias the hazard estimates. We present a modified estimator for the cumulative regression coefficients in Aalen's model using inequality constraints that restrict the estimation so that one can obtain only nonnegative hazard estimates. The method of inequality constrained least squares (ICLS) was developed from the linear complementary problem of Cottle and Dantzig (1968). The problem is efficiently solved with a quadratic programming algorithm offered by Gill et al. (1995). We compare the ICLS estimator to the OLS estimator in a data analysis of bone marrow transplant patients.
|
