Multivariate multilevel modelling of population well-being using three prospective cohorts

Abstract

Health as “a state of complete physical, mental, and social well-being” has been the most widely adopted definition over past decades. However, little empirical research exists identifying common determinants of the multiple facets of population well-being. Moreover, improvements in physical well-being are not necessarily coupled with advances in mental or social well-being. This suggests that the drivers of physical, mental, and social well-being may differ, highlighting the importance of identifying factors that could improve health holistically. To identify factors that prospectively and simultaneously predict multiple facets of population well-being, this research applied multivariate multilevel analyses to model multiple well-being measures simultaneously, using three population-based cohorts across non-Western and Western settings (Hong Kong, the US, and the UK). A key data source for the thesis was the FAMILY Cohort, the first large-scale population-based cohort in Hong Kong aiming to promote population health, happiness, and harmony (3Hs), in accordance with the World Health Organization’s triaxial definition of health. Validating associations found using the US National Epidemiologic Survey on Alcohol and Related Conditions and the UK Understanding Society’s Household Longitudinal Study could provide external validity due to the consistent findings despite differences in social patterning. Building upon the success of recent high-dimensional association studies, the multivariate multilevel model was combined with an environmental-wide association studies (EWAS) strategy, for the first time, to systematically examine and validate over 170 factors covering objective health measures, morbidity, behaviours, mental health, family environment, and social capital. Depressive symptoms, psychiatric disorders, stressful life events, life satisfaction, happiness, stroke, oral health, Parkinson’s disease, and comorbidities were found to be associated with both subsequent physical and mental well-being. External validation using comparable UK data yielded consistent findings. Furthermore, depressive symptoms, psychiatric disorders, and stressful life events were negatively associated with all of the 3Hs. For the first time, common determinants of physical, mental, and social well-being have been detected. In view of the global rise in alcohol consumption, the controversy concerning the health effects of moderate drinking, and the different socio-cultural patterning of alcohol consumption between Western and non-Western settings, the associations between drinking patterns and changes in physical and mental well-being were investigated via multivariate multilevel analyses and validated across two population-representative cohorts. Quitting drinking was associated with a more favourable change in mental well-being in women, identified in Chinese with validation in Americans. Recognising that the health consequences of re-employment for people with disabilities are sparsely documented and that the causal relationship remains unclear, the impact of re-employment was evaluated in a panel of participants of the UK study. Leveraging a natural experimental design and to obtain more causally robust estimates, difference-in-difference in conjunction with propensity score was incorporated in the multivariate multilevel model. Re-employment had a positive impact on physical and mental well-being for working-age people on out-of-work disability benefits. By modelling multiple outcomes simultaneously, the present thesis provides evidence for the common determinants of multiple facets of population well-being and helps prioritise hypotheses for further investigation, with the ultimate goal of holistically improving population well-being.