Topics in Joint Modeling of Longitudinal Biomarker, Quality of Lifetime, and Survival Data

[摘要] Joint modeling techniques have been developed for analyzing correlated longitudinal and survival data in many studies. It provides consistent and efficient estimates of the parameters even when the longitudinal covariate is measured infrequently and with measurement error. This work focuses on the use of joint modeling to solve two different statistical problems. The first one is about analyzing censored biomarker measurements and survival data under a case-cohort design. The goal is to study how the biomarker level changes over time, and the relationship between longitudinal biomarker measurements and the event time. We suggest a modified likelihood-based approach to adjust the possible bias introduced by the censoring due to detection limits and the measurement error in biomarker measurements.The second topic is about drawing inference for mean quality adjusted lifetime data. We consider continuous health experience and define the quality function with repeatedly measured quality of life score. We propose a consistent and asymptotic normal estimate for the mean quality adjusted lifetime and derive its asymptotic variance. The performances of the proposed methods have been demonstrated in simulation studies and through real dataexamples.Public Health Significance: A case-cohort study is a cost-effective design that is used in many large epidemiological studies. The method proposed in the first part of this dissertation increases the efficiency of the parameter estimation under case-cohort designs, which can lead to a considerable reduction in cost and effort. The evaluation of health benefits is a major public health challenge. The second part of this dissertation presents the estimate of mean quality adjusted lifetime, which can serve as a measurement of health benefits in the comparison of treatments or public health strategies.

[发布机构] the University of Pittsburgh