Background Research suggests that social isolation may increase the risk of developing or dying from vascular disease, but the evidence is inconsistent and may be affected by confounding and reverse causation, whereby poor health leads to being more socially isolated. We examined the association between social isolation and vascular disease incidence and mortality in the Million Women Study after accounting for confounding and reverse causation biases.

Methods Among 3 25 770 women (mean age 68 years) without vascular disease or cancer, frequency of contact with family or friends, groups, and number in household was reported and used to calculate a three-item social isolation score. Cox regression was used to estimate the relative risks (RR) and 95% confidence intervals (CI) of coronary heart disease (CHD) and stroke incidence and mortality in relation to social isolation. To reduce reverse causation bias, women who reported fair/poor self-rated health were excluded. Analyses were adjusted for demographic (age, region), socioeconomic (education, area deprivation), behavioural (smoking, alcohol consumption, physical activity and body mass index), disability, and health-related (hypertension, diabetes) risk factors. We assessed the proportion of the association that could be explained by each risk factor by calculating the percentage reduction in the likelihood ratio X² test statistic after each adjustment.

Results During 6 years of follow-up, there were 10 853 incident CHD events, 557 CHD deaths, 6269 incident stroke events and 585 stroke deaths. Compared to the least isolated women, the most isolated women (12% of participants) were not at an increased risk of incident CHD (RR=1.07, 95% CI 1.01 to 1.14, p=0.06) but were at increased risk of CHD mortality (RR =1.80, 1.40–2.31, p<0.0001); they were also at an increased risk of stroke incidence (RR =1.28, 1.18–1.38, p<0.0001) and mortality (RR =1.70, 1.33–2.16, p<0.001). With the exception of stroke mortality, adjustment for confounders, particularly for the behavioural factors, led to large reductions in the X² test statistic (e.g. 92% for CHD incidence; 64% for CHD mortality; 67% for stroke incidence; and 41% for stroke mortality).

Conclusion We found no association between social isolation and CHD incidence, but social isolation was associated with increases in CHD mortality and stroke incidence and mortality. Given the attenuation in these associations after adjustment for health and lifestyle factors, residual confounding cannot however be ruled out.