Social isolation, depressive symptoms, and incident cardiovascular disease in older Europeans: a multinational cohort study from SHARE

Social isolation, affective well-being, and depressive symptoms are related but distinct psychosocial constructs that may be associated with cardiovascular disease (CVD). Because SHARE Wave 4 did not include a validated loneliness scale, we examined structural social isolation (a five-item index of limited social ties, roles, and participation), elevated depressive symptoms, and a happiness-derived low positive affect indicator (defined from a single happiness item) rather than loneliness itself. We aimed to estimate their associations with incident CVD, describe joint exposure patterns and descriptive attenuation, and explore regional variation across Europe. Using de-identified SHARE data accessed by registered researchers through the SHARE Research Data Center, we analysed 36,091 participants aged > = 50 years from 16 European countries who were free of baseline CVD (5,661 incident events; 195,412 person-years; median follow-up 4.0 years). Social isolation was measured using a five-item structural index. The prespecified primary analysis used a 0–1 versus > = 2 dichotomy (59.8% of the analytic cohort exposed), while continuous, ordinal, and stricter-cutoff specifications were prespecified sensitivity analyses because the threshold was not externally validated. Low positive affect was derived from a single happiness item, and elevated depressive symptoms were defined as EURO-D > = 4. Model A, the primary confounder-adjusted model, adjusted for age, sex, education, and income; Models B and C additionally adjusted for health behaviours and baseline health status and were interpreted descriptively. Complete-case, multiple-imputation, and discrete-time complementary log-log analyses assessed robustness to missing data and interval-censored event timing, and proportional hazards and interaction tests were performed. In the primary confounder-adjusted model (Model A), social isolation (HR 1.18, 95% CI 1.11–1.24), elevated depressive symptoms (HR 1.35, 95% CI 1.27–1.43), and low positive affect (HR 1.11, 95% CI 1.03–1.20) were associated with incident CVD. After additional adjustment for health behaviours and baseline health status (Model C), the associations were attenuated to HR 1.08 (95% CI 1.02–1.15), HR 1.14 (95% CI 1.07–1.21), and HR 1.01 (95% CI 0.94–1.09), respectively, with similar mutually adjusted estimates. The social isolation association was small and definition-dependent: estimates were weaker for the continuous score (HR 1.02 per point, 95% CI 1.00-1.04) and null for the stricter > = 3 cutoff (HR 1.04, 95% CI 0.98–1.11), with no clear graded dose-response pattern. Complete-case, multiple-imputation, and discrete-time analyses produced broadly similar results. In the eight-category joint-exposure analysis, categories combining social isolation with elevated depressive symptoms had higher point estimates than the reference group (HR 1.20, 95% CI 1.10–1.30 for the combined group), but multiplicative ( P = 0.951) and additive (RERI − 0.014, 95% CI -0.151 to 0.124) interaction tests were not statistically significant, and these findings should not be interpreted as evidence of synergy or effect modification. Eight-year cumulative CVD incidence was 25.2% versus 18.2% among socially isolated versus non-isolated participants and 27.1% versus 20.7% among participants with versus without elevated depressive symptoms. In this multinational European cohort, elevated depressive symptoms were the most consistent psychosocial marker of incident CVD across model specifications and sensitivity analyses. A broad binary structural social isolation measure showed a modest, definition-dependent association without clear dose-response evidence. The single-item low positive affect indicator was not independently associated after fuller adjustment and does not address loneliness, which was not measured in SHARE Wave 4. Future studies should incorporate validated loneliness measures and prespecified causal and interaction frameworks.