ACCELERATED FAILURE TIME MODELS COMPARISON TO THE PROPORTIONAL HAZARD MODEL FOR TIME-DEPENDENT COVARIATES WITH RECURRING EVENTS

This paper presents an analysis of parametric survival models and compares their applications to time to event data used to validate the approximation for repeated events applying the Proportional Hazard Model (PHM) proposed in Mendes and Fard [Int. J. Reliab., Qual. Saf. Eng. 19(6) (2012) 1240004.1-1240004.18]. The subjects studied do not show degrading failures, allowing the comparison between accelerated failure time models with the PHM. Results showed the applicability of the Weibull model and the versatility of the PHM not only to match the results of the parametric model, but also to allow the implementation of time-dependent covariates, resulting in superior model fit and more insightful interpretation for the covariate hazards. The paper contribution is to present the PHM as a simpler, more robust model to determine the acceleration factor for reliability testing when compared to the formidable task of fitting a parametric model for the distribution of failure. The Kaplan-Meier method may provide misleading guidance for covariate significance when time-dependent covariates are applied; however, relevant graphical screening is supplied. Notwithstanding, the PHM provides additional options to treat the repeated observations applying robust covariance correction for lack of heterogeneity in the fixed effects model or adopting the stratified model that absorbs the error using the stratification concept.