64th ISI World Statistics Congress - Ottawa, Canada

64th ISI World Statistics Congress - Ottawa, Canada

A joint model of longitudinal data and survival data with detection and downweighting of longitudinal outliers

Conference

64th ISI World Statistics Congress - Ottawa, Canada

Format: IPS Abstract

Keywords: aids, hiv, joint models, mixed, model, outliers

Session: IPS 332 - Innovative Statistical Approaches to address Emerging Challenges in Public Health

Tuesday 18 July 2 p.m. - 3:40 p.m. (Canada/Eastern)

Abstract

Joint modelling of longitudinal and survival data enables us to associate time-dependent covariates with time-to-event outcomes. The linear mixed effects model is usually assumed for the longitudinal outcome. This model focuses on the modelling the mean trajectory over time. However the variability of the longitudinal outcome could be associated with the time-event outcome, for example visit-to-visit variability in CD4 count or viral load of an individual could be associated with the risk of death or HIV-/AIDS associated events. The linear mixed model is replaced by the mixed-effects location scale model in the joint model. This allows simultaneous modelling of between-subject and within-subject variability in addition to modelling the mean trajectory over time. A mixed-effects location scale model allows researchers to study within- and between-person variation in repeated measures. Using a large observational study of HIV patients, we propose to use a mixed effects location scale model for the longitudinal submodel of the joint model. We hypothesize that within patient variability in CD4 count or viral load is associated with the risk of death e.g CD4 count (at the start of ART) variability as a risk factor for death, e.g. increased CD4 count variability is associated with increased risk of death. A secondary objective will be to identify explanatory variables which are associated with the longitudinal variability by extending mixed effects location scale model to incorporate a linear predictor in the mean for the natural log of the variability. We will further explore whether there are some patients with unusual high or low variability (outliers).