Mixed-effects varying-coefficient model with skewed distribution coupled with cause-specific varying-coefficient hazard model with random-effects for longitudinal-competing risks data analysis

Tao Lu, Min Wang, Guangying Liu, Guang Hui Dong, Feng Qian

Research output: Contribution to journalArticlepeer-review

3 Scopus citations

Abstract

It is well known that there is strong relationship between HIV viral load and CD4 cell counts in AIDS studies. However, the relationship between them changes during the course of treatment and may vary among individuals. During treatments, some individuals may experience terminal events such as death. Because the terminal event may be related to the individuals viral load measurements, the terminal mechanism is non-ignorable. Furthermore, there exists competing risks from multiple types of events, such as AIDS-related death and other death. Most joint models for the analysis of longitudinal-survival data developed in literatures have focused on constant coefficients and assume symmetric distribution for the endpoints, which does not meet the needs for investigating the nature of varying relationship between HIV viral load and CD4 cell counts in practice. We develop a mixed-effects varying-coefficient model with skewed distribution coupled with cause-specific varying-coefficient hazard model with random-effects to deal with varying relationship between the two endpoints for longitudinal-competing risks survival data. A fully Bayesian inference procedure is established to estimate parameters in the joint model. The proposed method is applied to a multicenter AIDS cohort study. Various scenarios-based potential models that account for partial data features are compared. Some interesting findings are presented.

Original languageEnglish
Pages (from-to)519-533
Number of pages15
JournalJournal of Biopharmaceutical Statistics
Volume26
Issue number3
DOIs
StatePublished - May 3 2016

Keywords

  • AIDS study
  • Bayesian inference
  • competing risks
  • longitudinal data
  • mixedeffects varying-coefficient models
  • skewed distribution
  • survival data
  • varyingcoefficient hazard models

Fingerprint

Dive into the research topics of 'Mixed-effects varying-coefficient model with skewed distribution coupled with cause-specific varying-coefficient hazard model with random-effects for longitudinal-competing risks data analysis'. Together they form a unique fingerprint.

Cite this