Simple Techniques for Estimation of Additive Rate Regression Models

• Activity Number: 545
• Type: Contributed
• Date/Time: Thursday, August 2, 2007 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Computing
• Author(s): M. Brent McHenry*+ and Stuart R. Lipsitz and Debajyoti Sinha
• Companies: Bristol-Myers Squibb Company and Brigham and Women's Hospital and Medical University of South
• Keywords: exponential / piecewise exponential distributions ; method-of moments ; Poisson regression ; additive hazard rate regression model

Abstract

If the distribution of the failure time is exponential, Aitkin et al. (2005) showed that the MLE of the additive hazard regression model can be obtained using a Poisson linear model; we extend this result to the piecewise exponential distribution. However, when the hazard rates are close to 0 and/or censoring is high, there often exists convergence problems in the Newton-Raphson algorithm since the MLE is on or close to the boundary of the parameter space. Thus, we also propose a weighted least squares method-of-moments technique to obtain consistent and asymptotically normal regression parameter estimates. Simulations show that the weighted least squares estimate does not suffer from the convergence problems of the MLE. The piecewise exponential version provides estimates similar to semiparametric models, but are much simpler to compute than methods of Aalen (1989, 1993).