Assessing the association between all-cause mortality and multiple aspects of individual social capital among the older Japanese

Few prospective cohort studies have assessed the association between social capital and mortality [1‐5]. The studies did not use the same social capital indicators [1‐5]. Some of these studies used proxy measures of social capital [6, 7], such as crime rate [1], electoral participation [1, 5] or volunteer activity [3‐5]. There are several components of social capital, such as social network, participation, trust, reciprocity and volunteering [8]. Previous studies on social capital and mortality did not simultaneously use various components of social capital and their results were not fully consistent. In Finland, the association between mortality and individual social capital variables obtained by factor analysis (leisure participation, interpersonal trust and residential stability) was examined [2]. In men, leisure participation was associated with reduced all-cause mortality. In women, leisure participation and interpersonal trust were associated with reduced all-cause mortality. In a Swedish study, survival analyses showed that both neighbourhood social capital variables (election participation rate and crime rate) were significantly associated with mortality for males older than 65 years old but not for females [1]. Another study showed that living in a neighbourhood with the lowest level of social capital (volunteering, participation, political activities) was associated with significantly higher mortality than living in a neighbourhood with the highest level of social capital in England [5]. In contrast, among adults diagnosed and hospitalized with serious illnesses in the U.S, neighbourhood social capital (network density) was detrimental [4]. In addition, other neighbour social capital variables (social support, participation, volunteering, violence) did not significantly affect mortality [4]. In New Zealand, non-significant associations between neighbourhood social capital (volunteering) and mortality for both male and female were observed [3].

Studies on mortality and social capital have been conducted only in Western countries. However, social capital measurements developed in Western countries may not necessarily be equally applicable to Asian countries because of their different culture [9]. Although general trust has been broadly used as a measurement of social capital [10], it is known that intense ties within a family or group, often observed in collectivist cultures, prevent trust from developing beyond family or group boundaries [11‐13]. In Japan, a relatively collectivist society with intense group ties, human relationships are based on mutual assurance within group members rather than mutual trust between members from different groups [11, 13]. These cultural differences could potentially affect findings on the associations between social capital and health outcomes. In this respect, a social epidemiological study using various social capital indices in a non-Western cultural setting is important.

There is still debate about the precise definition and measurement of social capital [8, 14, 15]. Bourdieu defined social capital as "the aggregate of the actual or potential resources which are linked to possession of a durable network of more or less institutionalised relationships of mutual acquaintance and recognition"[16] and which focuses on the resources of individuals [8]. It is important to determine the association between individual social capital and health, because individual social capital indexes are components of aggregated measurements of community social capital [17]. Additionally, individual measures of social capital are not subject to the common problems arising from using area measurement in epidemiological studies, such as definition of a relevant areas [18, 19]. No study has used various measures of individual social capital as a predictor of mortality in a non-Western country. The aim of the present prospective cohort study was to assess the influence of individual social capital on all-cause mortality among older Japanese.

The average follow-up period was 4.29 years (SD = 0.75). During 27,571 person-years of follow-up for men and 29,561 person-years of follow-up for women, 790 all-cause deaths in men and 424 in women were observed. The incidence rate per 1000 person-years (IR) of death was 28.7 in men and 14.3 in women. Table 1 and 2 show the distribution of the number of deaths and IR according to covariates and social capital variables. Participants with low social capital in terms of generalised reciprocity and social network tended to have higher IR.

Table 3 shows the univariate and covariates adjusted mortality HRs for the different social capital variables among men. The results of multiple imputation models and non-imputation models were similar, particularly in the univariate models, but the 95%CIs were wider in most of the estimates obtained from the imputation models. In the univariate models using multiple imputation, lower social capital was significantly related to higher mortality in one general trust variable (people try to be helpful: HR = 1.42 (95%CI = 1.01-2.00)), all generalised reciprocity variables (listen to someone's concerns: HR = 1.59 (95%CI = 1.24-2.04); look after someone: HR = 1.49 (95%CI = 1.12-2.00)) and four social network variables (volunteer: HR = 1.78 (95%CI = 1.34-2.37); sports: HR = 1.89 (95%CI = 1.28-2.80); leisure: HR = 1.64 (95%CI = 1.21-2.20); meet friends rarely: HR = 1.99 (95%CI = 1.72-2.31)). When adjusting these models for covariates, only one low social network variable was found to be related to higher mortality (meet friends rarely: HR = 1.30 (95%CI = 1.10-1.53)), while the respective findings for two other social network variables (volunteering and leisure) were marginally not significant

Table 4 shows the univariate and covariates adjusted mortality HRs for the different social capital variables among women. The results of multiple imputation models and non-imputation models were also similar, particularly in the univariate models, but the 95%CIs were wider in most of the estimates obtained from the imputation models. In the univariate multiple imputation models, lower social capital was significantly related to higher mortality in all generalised reciprocity variables (listen to someone's concerns: HR = 2.31 (95%CI = 1.49-3.58) and look after someone: HR = 1.71 (95%CI = 1.18-2.47)) and four social network variables (sports: HR = 2.32 (95%CI = 1.41-3.82); leisure: HR = 2.24 (95%CI = 1.36-3.68); meet friends rarely: HR = 2.41 (95%CI = 1.31-4.45); having no friends: HR = 3.40 (95%CI = 2.10-5.52)). In the covariate adjusted multiple imputation analysis, only one lower social network response related to higher mortality (having no friends: HR = 1.81 (95%CI = 1.02-3.23)), while findings for one generalised reciprocity variable (listen to someone's concerns) and one social network variable (leisure) were marginally not significant. Interestingly, the response indicating lower social capital in one general trust variable was significantly related to lower mortality (most people cannot be trusted; HR = 0.65 (95%CI = 0.45-0.96)).

To the best of our knowledge, this study is the first prospective cohort study to assess the relationships between various social capital measures and mortality. In addition, this is the first cohort study on the relationship between social capital and mortality in a non-Western country. The present study showed that the structural social capital variable (friendship network) was a good predictor for all-cause mortality among older Japanese. Among men, it was the frequency of meetings with friends that was important, with those meeting their friends rarely having higher mortality, while was it was the lack of friends that was indicative of higher mortality among women. In addition, low general trust was related to lower mortality among women, suggesting that general trust has a different meaning among older Japanese women than among men.

Our results suggested the existence of culture differences in the association between trust and health. In addition, it is possible that the specific questions used to measure trust may also play a role. In our study, the question about general trust ("Generally speaking, would you say that most people can be trusted?") measures the trust for strangers, not group members [28]. In Japan, a relatively collectivist society with intense group ties, human relations are based on mutual assurance between group members rather than mutual trust between out-group members [11, 13]. The systems of mutual assurance, monitoring and sanctioning, within groups make the Japanese society safe and stable though closed [11, 13]. In such a society with strong ties, people can relatively easily obtain social support [29]. Though Japanese society is gradually changing recently because of globalisation, older people have lived in this traditional type of society for a long time throughout their life-course. Our results could suggest that Japanese older women who did not trust others would adapt well to the collectivist society with intense group ties and benefit from the society. In this Japanese older generation, men tended to work outside while women were predominantly housewives, therefore, men had to communicate with out-group members during their work and this may have contributed to developing their general trust towards strangers. In contrast, lower general trust measured by the question "Would you say that most of the time people try to be helpful?" tended to be associated with higher mortality among both men and women. This may be a more applicable question for measuring general trust among older Japanese, as this cohort may refer to their group members as "people" when answering this question. Previous cohort studies have not used general questions on trusting people to measure social capital. In Finland, a prospective study determined the beneficial effect of trust on all-cause mortality among women aged 30-99 years, not men [2]. Their measure of trust was based on the number of and trust in close friends. In this study, we did not use the factor/principal component analysis to check the association between detailed, not combined, social capital variables and mortality. As the results, various social capital variables were included into the models and different association of trust questions on mortality were shown though this method had the possibility of a type 1 error. Further prospective research assessing trust and mortality with considering various culture backgrounds is needed.

Our results are partially consistent with those of previous studies. Social network and social support are positively associated with health. A meta-analysis of social relationships and mortality determined that strong structural social relationships (social network) and functional social relationships (social support) increased the likelihood of survival [30]. In line with this, our study showed significant associations between friendship network and mortality (meet friends rarely for men; HR = 1.30 (95%CI = 1.10-1.53), having no friends for women; HR = 1.81 (95%CI = 1.02-3.23)); however, social support was not significantly associated with mortality. Although we considered diagnosed diseases and excluded from our study people with limitations in basic activities of daily living, it is possible that people included in the study may have had latent fatal diseases and consequently needed some help; this may have affected our results about social support and mortality. The concept of generalised reciprocity is based on the assumption that when people provide resources, good turns will be repaid at some unspecified time in the future, perhaps even by an stranger [31]. It does not entail tit-for-tat calculations in which individuals can be sure that a good turn will be repaid quickly and automatically [31]. Therefore, we used variables about the provision of social support as generalised reciprocity variables. In our study, generalised reciprocity showed marginal though non-significant association with mortality (HR = 1.27 (95%CI = 0.95-1.70) for men and HR = 1.57 (95%CI = 0.96-2.55) for women). A prospective cohort study in Finland showed that leisure participation was significantly though marginally associated with reduced all-cause mortality (HR = 0.94 (95%CI = 0.89-1.00) for men and HR = 0.96 (95%CI = 0.91-1.00) for women). In our study, covariate adjusted HRs of leisure participation were again marginal, though non-significant (HR = 1.29 (95%CI = 0.94-1.77) for men and HR = 1.54 (95%CI = 0.92-2.57) for women). The meaning of volunteering varies in the societies of different cultures [8] and relationships between volunteering and mortality are not consistent across studies [3‐5]. In our study, covariate adjusted HR of volunteer participation was marginal though non-significant for men (HR = 1.30 (95%CI = 0.95-1.77)) and non-significant for women (HR = 1.06 (95%CI = 0.38-2.95)).

There are several plausible pathways linking social capital to health [32]. At first, social capital may affect individual health by influencing health-related behaviours through promotion of more rapid diffusion of health information and by exerting social control over deviant health-related behaviours [32]. Second, higher social capital may promote health by increasing access to local services and amenities [32]. Good access to service such as transportation, clinics and community health centres could improve health. Third, there are associations between social capital and psychological distress [33, 34]. Social networks and social support can buffer the negative effects of life events on mental health [34]. Fourth, the communities with higher social capital produce more egalitarian patterns of political participation that result in the implementation of policies which ensure the security of all its members [32].

The results of this study have important public health implications. Among older Japanese, structural social capital variable related to friendship network were found to be significantly associated with mortality regardless of various covariates. This result suggests the possibility that public investment to promote social network may reduce the mortality among older people.

Our study has a number of limitations and strengths. The follow-up period (4.29 years) was relatively short. There was a potential bias caused by latent fatal disease though we considered diagnosed diseases and limitations in basic activities of daily living at baseline. In addition, the response rate was 50.4%; therefore, the results may have been affected by selection bias. Hanibuchi et al. previously conducted ecological analysis that assessed associations between community-level social capital and response rate using another data set from the AGES project and found that higher response rates were significantly associated with higher social capital [35]. Therefore, respondents of this study might have higher social capital than non-respondents. Although our results showed significant effects of some dimensions of social capital on all-cause mortality, this low response rate might have attenuated that association. In addition, compared with government data, our study respondents tended to be younger. It could be argued that healthier and younger people tend to respond to our questionnaire while people with higher risks of mortality tend to not participate. If so, this might have contributed to an underestimation of the association between poor social capital and mortality. As a strength, the present study used various social capital variables. Although the validity and reliability of the social capital variables were not been directly examined, a previous study using AGES project data checked the association between social capital variables based on our survey and voting rate, rate of volunteer registration and rate of social participation based on public social survey data [35]. Mean response of trust variable measured by our survey significantly associated with rate of volunteer registration in each community. Similarly, social network variables were significantly associated with voting rate.

In conclusion, friendship network, a measure of individual social capital, was a good predictor for all-cause mortality among older Japanese. In addition, low general trust was related to lower mortality among women. Further studies examining the different effect of social capital between Western and non-Western countries are needed.