Generalized additive mixed models to discern data-driven theoretically informed strategies for public brain, cognitive and mental health

Background:Social isolation is recognized as a public health emergency. However, little research has tried to discern appropriate public health strategies to prevent its adverse health effects. Rose’s The Strategy of Preventive Medicine provides a theoretical framework for such arbitrations between approaches. Therein, determining the shape of the relationship between risk factor and outcome is crucial for evaluating the potential effectiveness of prevention strategies. However, this shape has not been determined so far. We thus aimed to investigate the nature of the relationship of social isolation with brain, cognitive and mental health. Methods:In this pre-registered analysis, we employed generalized additive mixed models as a data-driven approach to model social isolation’s links to brain, cognitive and mental health outcomes in a well-characterised population-based sample. We compared more complex models with parsimonious linear models to evaluate the evidence for a nonlinear relationship. We derived brain measures from 3T MRIs, assessed cognitive functions with extensive neuropsychological testing and measured social isolation and mental health outcomes using established questionnaires. Overall, we studied over 10,000 (mean age 58a, 53% women) participants at baseline and over 5,500 (mean age 64a, 53% female) at follow-up after ~6years. Results:We found social isolation to significantly predict smaller hippocampal volume, worse cognitive functions and more anxiety and depressive symptoms. Furthermore, the shape of the relationship was firmly linear for all dependent variables except processing speed in which a steeper slope was detected amongst socially isolated individuals. Most of the health effects of social connection are observed in individuals that would not be categorised as socially isolated. Conclusions:Applying advanced statistical methods to a large and well-characterised dataset we provide evidence in support of a shift in focus away from individual-level and towards population-level preventive approaches.