Discovering patterns of social connections among older adults: A machine learning cluster analysis

Importance: Issues related to social connection are increasingly recognised as a global public health priority. However, there is a lack of a holistic understanding of social connection and its health impacts as most empirical research focuses on a single or few individual concepts of social connection. Objective: To explore patterns of social connection and their relationship with health and wellbeing outcomes. Design: Observational study using machine learning cluster analysis and regression analysis. Setting: England.Participants: Our analytical sample included 7,706 participants aged 50 or above from the forth wave of the English Longitudinal Study of Aging (2008-2009). Exposure: social connection clusters informed by the cluster analysis.Main Outcomes and Measures: This study considered a number of outcomes related to mental health (depression), hedonic (life satisfaction, pleasure) and eudaimonic (self-realisation) wellbeing, general health (self-reported health) and health behaviour (moderate or vigorous physical activity). We controlled for a number of key confounders identified using directed acyclic graphs, including age, sex, ethnicity, education, social class and wealth.Results: Among 7,706 participants, the mean age was 64.7 (SD=9.6) and 54.9% were females. We identified five clusters including: the highly connected (33.0%), disconnected (12.6%), gapped structure & poor function (14.4%), gapped structure & high function (20.5%) and poor function & mixed quality (19.5%) clusters. All clusters had poorer outcomes compared to the highly connected cluster. The difference was smallest between the highly connected and gapped structure & high function clusters across most outcomes.