Comparison analysis of separate and joint models in case of time-to-death event of HIV/AIDS patients under ART follow-up

Abstract

Background:

In clinical and medical studies, longitudinal and time-to-event data are considered important measures of health, and most of the time they arise together in practice. The purpose of this study is to compare the separate and joint models of longitudinal and survival data.

Methods:

A simple random sampling technique was used to select 550 random samples of HIV/AIDS patients who had been under antiretroviral therapy follow-up from January 2007 to October 2017 at Arba Minch General Hospital in Ethiopia. A linear mixed effect model was used to handle the longitudinal outcomes, whereas the parametric accelerated failure time models were applied for the time-to-event data. The Bayesian models were analyzed using Gibbs sampler with noninformative prior distribution. The model selection criteria such as deviance information criteria, Akaike information criteria, and Bayesian information criteria were used to compare the models.

Results:

The result from both separate and joint models suggests that there is dependence between the longitudinal and survival data used in the analysis.

Conclusion:

Based on the findings, we concluded that all proposed Bayesian joint models provide consistent results with high precision compared with the separate models. In application, we recommend incorporating the shape of hazard rate functions that the data reveal with model comparison criteria to select the best model that fit the data.

Keywords::

• separate model
• joint model
• linear mixed effects
• longitudinal
• survival analysis

Acknowledgments

The authors thank the anonymous reviewers for their comments that were useful to improve the write-up of the paper and Arba Minch General Hospital for providing the data sets used in this study.

Disclosure

The author reports no conflicts of interest in this work.