Social isolation in relation to vascular disease incidence and mortality among adults in the UK

Social isolation has recently gained considerable attention as a potential modifiable risk factor for major causes of death and disability such as coronary heart disease and stroke (i.e. vascular disease), and all-cause mortality. However, previous studies have been limited by small sample sizes, reverse causation, and confounding bias. This thesis sought to address these limitations by examining social isolation in relation to vascular disease incidence, vascular disease mortality, and all-cause mortality among two large prospective cohorts of UK adults. To advance current conceptualisations of social isolation as a health risk factor, age, gender, and neighbourhood environments were examined as effect modifiers. Furthermore, independent and joint associations were examined for structural (i.e. social contact frequency and living alone), functional (i.e. social support), and perceived (i.e. loneliness and relationship satisfaction) dimensions of social isolation. Finally this thesis tested the popular claim that social isolation is as bad for health as smoking 15 cigarettes per day. Generally healthy participants of the Million Women Study (N=326,169) and UK Biobank (N=296,913) were prospectively observed for six and seven years, respectively. An index of structural dimensions of isolation was used to measure social isolation. Associations between social isolation and vascular disease incidence and mortality outcomes were estimated using multivariable Cox regression. Analyses were adjusted for personal characteristics, health behaviours, and physiological factors. The percentage attenuation in the likelihood ratio chi-square test statistic indicated the degree to which these factors explained any associations with each outcome. Sub-group analyses and likelihood ratio tests of interaction were used to examine effect modification and interaction. This thesis found evidence of strong associations between social isolation and mortality outcomes but only weak associations with vascular disease incidence. Compared to the least isolated Million Women Study and UK Biobank participants respectively, those who were most isolated had 80% (95% confidence interval: 41% - 131%) and 68% (28% to 121%) greater CHD mortality, 71% (34% to 118%) and 92% (32% to 181%) greater stroke mortality, and 32% (24% to 40%) and 38% (27% to 51%) greater all-cause mortality. Whereas compared to the least isolated participants, the most isolated had 7% (1% to 14%) greater and 6% lower (12% lower to 2% greater) CHD incidence, and 28% (19% to 38%) and 23% (9% to 40%) greater stroke incidence. Associations generally did not change according to age, gender, or neighbourhood socioeconomic deprivation, crime rates, distance to services, and greenspace availability. Health behaviours, namely smoking, explained the largest component of these associations. Among the constituent measures of structural isolation, living alone was more strongly and consistently associated with mortality. After adjusting for explanatory factors and structural isolation, lacking confiding support (i.e. functional isolation) was associated with 15% (6% to 24%) greater all-cause mortality. Perceived dimensions of social isolation were not independently associated with all-cause mortality. Associations between structural isolation and all-cause mortality also tended not to change if participants additionally lacked support or were lonely. Finally, associations between social isolation and all-cause mortality were 5 to 10-fold weaker than those for cigarette smoking. In conclusion, there is stronger evidence to suggest that social isolation is independently associated with mortality versus vascular disease incidence. These associations did not vary by age, gender, neighbourhood factors, or functional and perceived isolation. Cigarette smoking had much stronger associations with mortality than social isolation. Further large-scale studies are needed to examine the role of delays in seeking care as a potential explanatory factor. Future research should also leverage lifecourse epidemiological methods to further examine the role of confounding in the associations observed.