B1171
Title: A longitudinal model with individual repeated measures as predictors of a Coxian phase-type survival distribution
Authors: Conor Donnelly - Queens University Belfast (United Kingdom) [presenting]
Lisa McCrink - Queens University Belfast (United Kingdom)
Adele Marshall - Queens University Belfast (United Kingdom)
Hannah Mitchell - Queens University Belfast (United Kingdom)
Abstract: Previous research has found that there exists a relationship between an individual's survival time and the trajectory of some related, repeatedly observed, time-varying covariate. For instance, the dynamic nature of the haemoglobin (Hb) levels within individuals suffering from chronic kidney disease has a strong association with the individuals' time until death. The repeated observations on individuals longitudinal response are analysed by employing a linear mixed effects model, utilised to generate not only a population-average trajectory for the longitudinal response over time but, more importantly, individual-specific trajectories. The individual deviations from the population-average trajectory, represented by the latent random effects, are subsequently incorporated within a Coxian phase-type distribution so as to investigate their effect on disease progression and thus on the survival of renal patients. The Coxian phase-type distribution is a special type of Markov model, shown previously to accurately represent the time until an event occurs, with the capability to estimate the rate of deterioration of individuals through sequential, unobserved states of the disease before the event is realised. By linking these longitudinal and survival processes, greater insight is offered into how end-stage renal disease progresses and the effect of changing Hb levels on this progression.
