This study relied on Swedish data from both the Level of Living Survey (used by Bygren et al and Sundquist et al) and the Panel Study of Living Conditions of the Oldest Old (used by Lennartson and Silverstein). The former includes participants up the age of 74; the latter follows those who have reached the age of 75. A total of 1,246 men and women who had participated in both surveys in 1990–1 and 1992 were selected and followed up for survival until 31 December 2003, during which period 691 individuals died. Account was taken of the frequency of hobby activities (including handicrafts and painting), cultural activities (including attendance at the cinema, theatre, concerns, museums and exhibitions) dancing, playing musical instruments, and choir singing. Hazard ratios were performed using Cox regression analyses. A range of symptoms and diseases, functional status, age, gender, educational level (as a measure of socio-economic position), smoking, alcohol, body mass index were included as potential confounders.
When all the activities were analysed collectively, participation in 0-1 activities tripled mortality risk, while participating in two activities doubled the risk relative to those taking part in six or more activities. Within this, women exhibited a dose-response relationship between overall participation and mortality risk. When the activities were analysed individually, together with age and education, strong associations between hobby and cultural activities in survival were observed, with the former particularly significant for men and the latter for women. When health indicators were taken into account, the association between hobby activities and survival was lost for women, but the relationship between cultural activities and survival uniquely persisted for both men and women. A significant relationship emerged between reading books and survival amongst women even when controlling for cognitive status and education level.
In interpreting these results, the authors accepted the possibility of reverse causation, with health status influencing both participation and mortality. In considering residual confounders, they understood that the socio-cultural activities analysed might serve as proxies for health status, but concluded that this would not explain any gender biases observed. A further confounder might be health behaviour, which was presumed by Bygren et al to exert an influence on the relationship between socio-cultural engagement and health. In a bid to compensate for this, account was taken of smoking, alcohol intake and body mass index, which did not modify the results. And, while socio-economic position was acknowledged as a likely confounder of the relationship between engagement and mortality, it was accepted that using education level as a solitary, dichotomised measure may prove inadequate to the task of capturing this.