Dynamic Predictions From Longitudinal CD4 Count Measures And Time To Death of HIV/AIDS Patients Using a Bayesian Joint Model

Abstract

Background: Personalised or stratified medicine has played an increasingly important role in improving bio-medical care in recent years. A Bayesian joint modelling approach to dynamic prediction of HIV progression and mortality allows such individualised predictions to be made for HIV patients, based on monitoring of their CD4 counts. This study aims to provide predictions of patient-specific trajectories of HIV disease progression and survival.Methods: Longitudinal data on 254 HIV/AIDS patients who received ART between 2009 and 2014, and who had at least one CD4 count observed, were employed in a Bayesian joint model of disease progression, as measured by CD4 counts, and survival, to obtain individualised dynamic predictions of both processes that were updated at each visit time in the follow-up period. Different forms of association structure that relate the longitudinal CD4 biomarker and time to death were assessed; and predictions were averaged over the different models using Bayesian model averaging.Results: A total of 254 subjects were observed in the dataset with a median age of 30 years (interquartile range, IQR, 26–38). The individual follow-up times ranged from 1 to 120 months, with a median of 22 months and IQR 7 -39 months. The median baseline CD4 count was 129 cells/mm3 (IQR 61–247 cells/mm3). From the joint model with highest posterior weight, subjects whose functional status was working were significantly associated with a higher baseline CD4 count (β = 1.86; 95% CI: 0.65 3.04) whereas subjects who were bedridden were significantly associated with a lower baseline CD4 count (estimated effect β = -3.54; 95% CI: -5.65, -1.39), compared to ambulatory patients. A unit increase in weight of the individual increased the mean square root CD4 measurement by 0.06. The estimates of the association structure parameters from all three models considered indicated that the HIV mortality hazard at any time point is associated with the current underlying value of the CD4 count at the same time point. The model with highest posterior weight also had a time-dependent slope, indicating that HIV mortality is also associated with the rate of change in CD4 count. From both the model-averaged predictions and the highest posterior weight model alone, an increase in CD4 count was predicted at different visit times from the dynamic predictions. It was also found that there was an increase in the width of prediction intervals as time progressed.Conclusions: Functional status, weight and alcohol intake are important contributing factors that affect the mean square root of CD4 measurements. For this particular dataset, model averaging the dynamic predictions resulted in only one of the hypothesised association structures having non-zero weight at the majority of time points for each individual. The predictions were therefore similar whether we averaged them over models or derived them from the highest posterior weight model alone. We also observed that the parameter estimates in the both the CD4 count and survival sub-models showed slight variability between the postulated association structures.