Parametric inference on quantile residual life

[摘要] The need for residual life analysis arises in many fields including medicine and life testing. For instance, in medicine, a clinician and a patient would be interested in knowing by how long a new drug can extend the life span of that patient. Problems in remaining life time after surviving up to a certain time are often framed and addressed statistically in terms of mean, hazard rate or quantile. The quantile approach enjoys some practical advantages over the other approaches such as robustness, ease of interpretation, and existence. Most of the methodological work on quantile residual life in the literature has been semi-parametric or non-parametric.However, parametric approaches are expected to be optimal or asymptotically efficient under a correct specification of the model.Furthermore, the parametric approach does not require nonparametric estimation of the probability density function of the underlying distribution under informative or noninformative censoring to evaluate the variance of the quantile estimator. In this dissertation, parametric inference procedures for the quantile residual life under competing and non-competing risks settings are developed for the one-sample, two-sample and regression cases. We adopt the accelerated failure time (AFT) framework to incorporate covariatesfor the regression case.The finite sample properties of the proposed methods are studied through extensive simulations. The simulation results indicate that the proposed methods perform well. The proposed methods are applied to a breast cancer data. PUBLIC HEALTH SIGNIFICANCE: The results established in this dissertation will provide new parametric methods to researchers and investigators in public health who conduct quantile residual life analysis, which will facilitate efficient communication between researchers and stakeholders regarding the efficacy of new interventions.

[发布机构] the University of Pittsburgh