B0412
Title: Predicting depression in old age: combining life course data with machine learning
Authors: Carlotta Montorsi - Luxembourg Institute of Socio-Economic research (Luxembourg) [presenting]
Abstract: With ageing populations, understanding life course factors that raise the risk of depression in old age may help anticipate needs and reduce healthcare costs in the long run. We estimate the risk of depression in old age by combining adult life course trajectories and childhood conditions in supervised machine learning algorithms. Using data from the survey of health, ageing and retirement in Europe (SHARE), the performance of six alternative machine learning algorithms is implemented and compared. The performance of the algorithms is analysed using different life-course data configurations. While similar predictive abilities are obtained between algorithms, the highest predictive performance is achieved when employing semi-structured representations of life courses using sequence data. The Shapley additive explanations method is used to extract the most decisive predictive patterns. Age, health, childhood conditions, and low education predict most depression risk later in life. Still, new predictive patterns are identified in indicators of life course instability and low utilization of dental care services.
