An Accelerated-Proportional-Additive Regression Model of Mean Residual Life with an Event-Free Fraction for Censored Data

• Activity Number: 596
• Type: Invited
• Date/Time: Thursday, August 4, 2011 : 8:30 AM to 10:20 AM
• Sponsor: WNAR
• Author(s): Chen-Hsin Chen*+ and Wei-Hwa Chang
• Companies: Academia Sinica and Academia Sinica
• Keywords: Cure model ; Life expectancy ; Logistic model ; Martingale process ; Mixture model ; Residual life

Abstract

In demographical and medical studies, it is customary to predict a surviving subject's life expectancy using the mean residual life function. To evaluate the covariate effects on the residual life with censored data, the proportional mean residual life model (Chen & Cheng, 2005), the linear mean residual life model (Chen & Cheng, 2006), and other models have been recently developed. Chen & Cheng (2006) noticed a critical challenge for estimating the mean residual life function when the estimated event time distribution has a level-off tail. In this case, the measure of life expectancy becomes meaningless in the whole study population and makes sense only in the susceptible or non-cured subpopulation. To cope with this phenomenon, we propose a semiparametric mixture model (Kuk & Chen, 1992; Lu & Ying, 2004) combining the logistic model for the event probability with a class of accelerated-proportional-additive mean residual life models. We also provide two useful quantities in predicting the life expectancy for a surviving subject. Inference procedures are developed to take into account of censoring, and also evaluated via simulation studies and an illustrative data analysis.