BackgroundSocial health markers, including marital status, contact frequency, network size, and social support, have shown associations with cognition. However, the underlying mechanisms remain poorly understood. We investigated whether depressive symptoms and inflammation mediated associations between social health and subsequent cognition.MethodsIn the English Longitudinal Study of Ageing (ELSA; n=7,136; aged 50+), we used four-way decomposition to examine to what extent depressive symptoms, C-reactive protein (CRP) and fibrinogen (assessed at an intermediate time point) mediated associations between social health and subsequent standardised cognition (verbal fluency, delayed and immediate recall) including cognitive change, with slopes derived from multilevel models (ELSA: 12-year slope). We examined whether findings replicated in the Swedish National Study of Aging and Care in Kungsholmen (SNAC-K; n=2,846; aged 60+; 6-year slope).FindingsWe found indirect effects via depressive symptoms of network size, positive support and less negative support on subsequent verbal fluency, and positive support on subsequent immediate recall (pure indirect effect (PIE)=0.002 [0.000-0.003]). The positive support-verbal fluency mediation finding replicated in SNAC-K. Depressive symptoms partially mediated associations between less negative support and slower immediate (PIE=0.001 [0.000-0.002]) and delayed recall decline (PIE=0.001 [0.000-0.002]), and between positive support and slower immediate recall decline (PIE=0.001, [0.000-0.001]), which replicated in SNAC-K. We did not observe mediation by inflammatory biomarkers.InterpretationFindings provide new insights into mechanisms linking social health with cognition, suggesting that associations between cognition and interactional aspects of social health in particular, such as social support, are partly underpinned by depressive symptoms.