Linear regression modelling to assess the impact of socio-economic, demographic and health-related variables on wellbeing in the elderly population

Population ageing is a worldwide phenomenon that poses biological, social and economical challenges: prevention of chronic diseases is key to improve health and life expectation, and to reduce the burden of healthcare systems. Self-perceived wellbeing is a predictor of the onset of chronic conditions, however, so far, variables influencing wellbeing have been investigated only in cross-sectional studies, not providing any guidance on which factors can produce positive/negative variations of wellbeing over time. The aim of this work is to investigate which factors, among demographic, socioeconomic and health-related variables, most strongly influence wellbeing changes in the elderly population, using a regression approach. A total of 23622 subjects from the Survey of Health, Ageing and Retirement in Europe study was considered, for which wellbeing was measured by the CASP-12 score in two visits spaced two years apart. The model was developed on a training set by using multiple linear regression, with a subsequent LASSO regularization to select the most relevant variables. The model output was the CASP-12 score at the second visit, while the inputs were the variables collected at the first visit and their differences between the two visits. The model presented good performance, with R2 equal to 0.63 on the test set. The most important factors resulted to be the feeling of being left out, having difficulties in making ends meet and the self-perceived health status. Future work includes the use of linear and non-linear classification methods to predict the direction of CASP-12 changes in time.