Survival Analysis with Shared Frailty on Dementia-focused New Zealand InterRAI data

Abstract

Survival analysis is a unique branch of statistical approaches tailored for analysing time-toe-vent data, including censored observations, with the primary outcome being the time until an event occurs. The Cox Proportional Hazards model is a popular regression technique in survival analysis due to its semi-parametric nature, which describes the instantaneous potential of observing the event. Whilst the model is suitable under certain conditions, there lies an assumption within this model that researchers can mistakenly avoid. The results can be misleading for time-to-event data that violate this assumption. This thesis aims to enrich the Cox Proportional Hazards model fitting using the Partial Likelihood without dismissing its popular semi-parametric aspect into the Cox model for time-varying covariates. We continue to examine the theoretical foundations of this extended Cox model to allow for random effects that lead towards fitting a mixed-effects Cox model using the Penalised Partial Likelihood. The International Resident Assessment Instrument (interRAI) dataset acts as our source of time-to-event data, where the event corresponds to the death of an individual. This study aims to determine, from interRAI data, the predictors that could increase dementia mortality. The data include individuals with and without dementia, and we further split this data into two age groups to delineate predictors of young-onset and late-onset dementia mortality. We apply the mixed-effects Cox model to each age group based on aspects of their demographic, disease diagnoses, physical health, psychosocial health, and lifestyle elements to create five separate models for each age group. We control each model by gender and ethnicity. Our significant findings mainly concentrate on young-onset dementia subjects. Those who suffer from a history of strokes, a record of falls, and moderate to severely impaired vision can lead to a higher risk of death than young-onset dementia patients who do not suffer from these three previously mentioned illnesses of the same age. Our models also suggest that young-onset dementia patients with a terminating assessment over the age of 65 have a higher potential of death over non-dementia of a similar age range due to unobserved heterogeneity under select health aspects. This higher potential mainly comes from the patients’ disease diagnoses, physical health and psychosocial health.