Multilevel Mixed-Effects Survival Parametric Models: A Study of Cohorts from Brazil, Argentina and Italy of time to event from chronic diseases in elderly people.

Abstract

Ageing of populations has made the study of the elderly of great importance for social and health planning purposes. The aim of this work was to analyze time to event of populations coming from different cohort studies conducted in Brazil, Argentina and Italy. Chronic diseases such as diabetes mellitus, cardiovascular disease (CVD), obesity and cancer are prevalent diseases associated with aging. First, Kaplan-Meier method was applied to explore survival probabilities for each country and provide estimates of the net probability of death due to each variable. Second, as proportional hazard strong assumption of classical Cox model was not satisfied, flexible parametric survival models (Royston and Parmar, 2002) were fitted for each country in order to estimate covariate effects. This survival model framework use natural cubic splines to model baseline. Finally, the structure of variability from each cohort was taken into account by using a multilevel survival model. The random-effects approach coupled with RP flexible models allowed us to used random effects, including random coefficients of covariates. Conditional hazard ratios were estimated for continuous covariates and conditional predictions (specific to each group) means were compared. In Brazil, survival to death was associated with CVD, hazard ratio (HR) estimate equal to 1.68 (CI: 1.05, 2.69); In Argentina, time to cancer occurrence was associated with High Blood Pressure (HBP), HR estimate equal to 1.71 (CI:1.06,2.77) and in Italy, HR estimates for Gender and Physical Activity were significant. In all countries but in Italy, HBP was associated significantly with a shorter time to death. Whereas RP flexible models are suitable when proportional hazard are not satisfied, multilevel structure should be used when considering the survival time of different studies.