As smoking, obesity, physical inactivity and excessive alcohol consumption remain the largest preventable risk factors for morbidity and mortality in industrialised countries,1^–3 it is important to identify factors that can contribute to these lifestyle risk factors. Health risk behaviours are known to be influenced by multiple genetic, behavioural, psychological and social factors.4^–6 These risk factors have been widely studied, but less is known about the determinants of their clustering7^–9 and whether social characteristics could facilitate the co-occurrence of these factors. Recently, adverse psychosocial working conditions have been shown to contribute to risk factors such as obesity10 and heavy drinking,11 and there is a small amount of cross-sectional evidence suggesting that co-occurrence of lifestyle risk factors might also be affected by psychosocial circumstances.12 13

Lack of social capital is considered to be one of the possible psychosocial determinants of lifestyle risk factors.14^–18 For instance, aspects of social capital, such as low social participation19 20 and lack of institutional trust,21 have been associated with a lower likelihood of smoking cessation. However, earlier studies have rarely focused on social capital in the context of work, although the workplace could be an important source of social capital.22 23 Compared to broader living environments such as neighbourhoods, workplaces may provide more concentrated opportunities for social capital and thus also be useful in the analysis of its essential characteristics.24 However, to our knowledge, no previous studies are available on the potential role of workplace social capital in the co-occurrence of multiple lifestyle risk factors.

Studying determinants of the clustering of lifestyle risk factors is important, because adverse health behaviours tend to aggregate7 25 and may reinforce each other in their effects.12 Hence, in the case of such clustering, the overall effect of the combined risk factors could be underestimated. For instance, without trust, support and justice at work, some employees may exercise less, thereby gaining weight, whereas others may drink and smoke more.13 Underestimation may also take place if the risk factors have synergist effects.26 Indeed, empirical evidence suggests that co-occurrence of multiple lifestyle risk factors is likely to cause a greater health risk than the sum of their independent, additive effects.7

In this large-scale occupational cohort study, we investigated the associations between workplace social capital and the co-occurrence of lifestyle risk factors in a sample of Finnish public sector employees. Our study adds to previous research in four ways. First, to our knowledge this is the first study to examine workplace social capital in relation to co-occurring lifestyle risk factors. Second, the use of both a cross-sectional and prospective design allowed us to observe the changes in this association and look for a temporal link between factors. Third, to eliminate the potential impact of individual response styles, we used organisational-level aggregated scores (ecological level) in modelling the effect of workplace social capital. Fourth, as the clustering of several risk factors is rare, very large datasets are needed to explore co-occurring risk factors; our data allowed for such analysis, as the study covered over 30 000 employees working in nearly 3000 work units.

We formulated two hypotheses:

1. The higher the level of social capital perceived by an employee (individual level), the lower the risk of co-occurrence of lifestyle risk factors.

2. The higher the level of social capital in the work group (ecological level), the lower the risk of co-occurrence of lifestyle risk factors.

METHODS

Study population and study design

The ongoing Finnish Public Sector Study focuses on the entire public sector workforce in 10 towns and 21 hospitals in Finland. These employees hold a wide range of socioeconomic positions from city mayors to semiskilled cleaners, the largest occupational groups being nurses and teachers. The baseline questionnaire was administered in 2000–2002 (n = 48 592; age range 17–64 years). The response rate was 68%. In 2004–2005, a follow-up questionnaire was sent to all identifiable respondents of the baseline survey who were still alive. A total of 35 914 responses were received (response rate 77%). Mean follow-up was 3.59 years (SD 0.81).

In order to study social capital adequately, we excluded employees working in units with less than three employees (n = 808), employees with a missing value for work unit size (n = 53) and individuals with missing data on individual and/or ecological social capital at baseline, on co-occurrence of risk factors at baseline and/or follow-up, and/or on covariates at baseline (n = 2680). The final cohort of this study included 31 373 employees. This sample did not differ substantially from the eligible baseline population (total personnel employed in 2000–2002) in terms of mean age (45.0 (9.2) years in the sample, 45.2 (9.1) in the eligible population), proportion of women (83% vs 82%), occupational status (29% non-manual vs 29% non-manual) and mean individual-level social capital at baseline (3.6 (0.8) vs 3.6 (0.7)).

Assessment of social capital at baseline

Workplace social capital was assessed with a validated and psychometrically tested eight-item measure22 which assesses the three types of social capital: bonding, bridging and linking.27 28 The items were as follows: (1)“We have a ‘we are together’ attitude”; (2)“People feel understood and accepted by each other”; (3)“We can trust our supervisor”; (4)“Our supervisor treats us with kindness and consideration”; (5)“Our supervisor shows concern for our rights as employees”; (6)“People keep each other informed about work-related issues in the work unit”; (7)“Do the members of the work unit build on each other’s ideas in order to achieve the best possible outcome?”; and (8) “People in the work unit cooperate in order to help develop and apply new ideas”. The responses were marked on a 5-point scale (1 = “totally disagree”, 5 = “totally agree”, apart from the seventh item where the categories were 1 = “very little”, 5 = “very much”). A summary score of the responses was produced (Cronbach’s α = 0.88).

Ecological-level social capital scores were also calculated by work unit. Data on administrative units were obtained from the employers’ registers.24 Based on this information, 2967 functional work units, typically at a single location (eg, a school or a hospital ward), were determined. The units at the lowest organisational level were selected if the organisational hierarchy had multiple levels. Median work unit size was 19 employees (interquartile range 12–34; range 3–430). These figures show that even if the work unit size varied between three and 430 people, most work units were relatively small. Overall, 96% of work units had less than 100 employees and 64% of work units had less than 20 employees. Only 8% of the participants were in work units with a less than 50% response rate. The ecological social capital of the work unit (second level) was calculated as the mean of individual (first level) responses of co-workers in the same work unit (self-estimation excluded) and then these mean scores were linked to each member of the work unit. Both individual and ecological-level social capital scores were divided into quartiles for the analysis, the highest quartile indicating the highest level of social capital.

Assessment of co-occurrence of lifestyle risk factors at baseline and follow-up

Risk factors assessed included current smoking, heavy drinking, body mass index (BMI) ⩾25 kg/m² (indicating overweight as defined by WHO29) and physical inactivity. The four risk factors were dichotomised as adherence and non-adherence to public health recommendations (no risk behaviour referred to non-smoking, non- or moderate drinking, BMI <25 m² and physical activity >2 metabolic equivalent task (MET) hours per day).

Smoking was measured both at baseline and at follow-up with the following questions: “Do you smoke or have you previously smoked regularly, that is, daily or almost daily?” and “If you have smoked, do you still smoke regularly?”. From this information, smoking status (non-smoker vs smoker) was derived. Alcohol use was based on the respondent’s report of their habitual frequency and amount of beer, wine and spirit intake. This information was transformed into grams of alcohol per week. As in our earlier studies, a dichotomous variable was created to represent heavy drinking, with a cut-off point corresponding to average weekly consumption ⩾190 g of absolute alcohol for women30 and ⩾275 g for men.31

The respondents reported their average time spent on physical activity per week as well as the intensity of the activity in relation to walking, vigorous walking, jogging and running. The time spent on each activity in hours per week was multiplied by the average energy expenditure of the activity in question, expressed in METs. A summary score of MET hours per week was calculated for each respondent. A volume of activity of less than 2 MET hours per day indicated physical inactivity.32

We calculated the number of risk factors on the basis of these binary variables. The participants with three or four risk factors had a score of 2, those with one or two risk factors scored 1, and those with no risk factors scored 0.

Assessment of background characteristics

Information on age, sex, socioeconomic position (manual, lower non-manual and upper non-manual) and employer (town vs hospital) was derived from employers’ records at baseline. Marital status was drawn from the questionnaire at baseline.

Data analyses

Because the outcome, although numeric, was highly skewed and only had a few values, we did not use linear modelling. As there were no theoretical grounds for the assumption that the odds ratios would remain the same, irrespective of the cut-off point selected for the number of risk factors, and be in line with our earlier research,7 12 we decided to examine the associations between social capital and the co-occurrence of risk factors using multinomial regression analysis. The multinomial models were used to assess the likelihood of having one or two risk factors, and three or four risk factors versus no risk factors. We chose the “no risk factors” category as the comparison category and determined separate odds ratios for all categories of the co-occurrence of risk factors for each quartile of the social capital, with the exception of the comparison category. The analyses were conducted in three phases. First, the models were adjusted for age, sex, marital status and employer. Second, in addition to earlier adjustments, the models were adjusted for baseline co-occurrence of adverse lifestyle risks. Third, the models were additionally adjusted for socioeconomic position.

In order to address the fact that individual employees were nested within work units, all main analyses were performed using multilevel regression analysis. In this analysis, it is assumed that both individuals and work units are randomly sampled, and that there is interdependence between individual and work unit residuals. This type of modelling enables simultaneous examination of the effects of group-level and individual-level variables on individual-level outcomes, while controlling for the non-independence of observations within groups. Multilevel models recognise the existence of data hierarchies by allowing for residual components at each level in the hierarchy. The residual variance is partitioned into a “between work” unit component (the variance of the work unit-level residuals) and a “within work” unit component (the variance of the individual-level residuals).

We expressed the results as odds ratios (OR) and their 95% confidence intervals (95% CI). We used intra-class correlation (ICC) to study the differences in variance in social capital between work units. ICC was 22%, indicating that a substantial proportion of the variance in individual social capital occurred between work units. All statistical analyses were performed with SAS 9.1.3 statistical software (SAS Institute, Cary, NC), applying the Glimmix procedure.

RESULTS

The mean number of lifestyle risk factors by background characteristics is presented in table 1. The mean number of risk factors was 0.9 for women and 1.2 for men. At baseline, 38% of the participants had no risk factors, while 4% had more than two risk factors. The mean of lifestyle risk factors was significantly higher among manual workers (p<0.001), municipal employees (p<0.001) and people living without a partner (p<0.001). During the follow-up, the proportion of overweight employees increased, while the proportion of smokers decreased. At the end of the follow-up, the prevalence of several lifestyle risk factors was higher than at baseline.

The cross-sectional associations between social capital and co-occurring risk factors at baseline are shown in table 2. Exposure to low social capital at the individual level was associated with a higher likelihood of having three or four versus no risk factors, and the excess risk persisted after adjustment for socioeconomic position. A similar trend was observed at the ecological level. The results further showed that the association between low social capital and risk of having one or two versus no lifestyle risk factors was very modest and became weaker after additional adjustments.

Table 3 depicts the results from the multilevel multinomial logistic regression analyses on the prospective associations between social capital (baseline) and the co-occurrence of lifestyle risk factors (follow-up). After adjustment for age, sex, marital status and employer, low social capital at work at time 1 was associated with a higher risk of co-occurrence of health risk behaviours at time 2, both at the individual and at the ecological level. However, the association became non-significant when the level of prevalent co-occurrence of lifestyle risk factors at baseline was controlled for. Thus, no evidence was obtained to demonstrate that social capital predicted future co-occurring risk factors irrespective of baseline risk factor status.

DISCUSSION

Previous research has addressed issues related to the possible influence of psychosocial work environment on lifestyle risk factors. However, to date there has been no evidence of an association between social capital at work and the risk of co-occurrence of lifestyle risk behaviours. The findings of this large-scale cohort study suggest that co-occurrence of multiple lifestyle risk factors may be related to low workplace social capital. However, this association was obvious only in the cross-sectional setting.

Workplace social capital may influence the co-occurrence of multiple lifestyle risk behaviours in several ways. First, social capital may reinforce informal social norms and social control over various deviant health behaviours.33 34 Second, high social integration may contribute to a positive affect through increasing self-care and decreasing the likelihood of having lifestyle risk factors.35 Third, innovative behaviours, such as use of health promoting information and services, can spread more easily in communities with a high level of social capital.35 Fourth, high perceived social capital at work may protect against unhealthy behavioural coping (ie, smoking, comfort eating and heavy drinking) by enhancing the individual’s “healthy” emotional and social coping resources.36

The slightly weaker association between co-worker assessed ecological-level social capital and the likelihood of co-occurrence of lifestyle risk factors may relate to exposure misclassification or measurement imprecision because the work units were relatively large and the data were based on administrative records. Informal work groups might provide a more accurate proxy for ecological-level social capital in some cases, but such data were not available in this study. Thus, the assessment of co-workers might be a less accurate reflection of social capital than an individual’s own assessment. It is also possible that ecological-level social capital increases the likelihood of healthier lifestyle choices only if it also influences the individual’s own perception of social capital.

Due to its prospective nature, our study enables a better understanding of the causal relationships between social capital and the co-occurrence of lifestyle health risk factors than earlier research. This type of study may inform professional practice and enable a clearer understanding of the mechanisms related to social capital, lifestyle risk factors and health. It seems that the adverse health effects of low social capital are, to some extent, behaviourally mediated by clustering of multiple health risk behaviours. It is noteworthy, however, that low social capital predicts subsequent co-occurrence of multiple health risk factors when the impact of baseline health risk factors is not taken into account in the model. The attenuated prospective association between social capital and health risks after adjustment for baseline co-occurrence of lifestyle risks shows how earlier health risk factors tend to track into the future, and that the continuous nature of individual health behavioural hazards nullifies the impact of social capital on health risk factors in the prospective setting.

Our findings point to the individual’s perception of social capital as a correlate of the occurrence of multiple lifestyle risk factors. However, we also found ecological-level evidence that workplace health programs intended to promote social capital may have an impact on co-occurring risk factors. In the future, intervention studies are needed to test whether a change at work unit level might predict change in the likelihood of co-occurring risk factors and whether these predictive associations exist across different subgroups including high- and low-socioeconomic position employees with different social norms and habits.37

Some methodological limitations should be noted. First, the dichotomisation of risk factors enabled assessment of co-occurring risk factors but may have reduced statistical power, thereby underestimating the strength of the associations. Had we chosen lower cut-off points, the proportion of high-risk individuals would have notably increased. Second, the self-report nature of the data makes them subject to recall and response bias. Non-response and misclassification are likely to influence different behaviours to differing degrees. For example, self-reported current smoking is probably more accurate than self-reported alcohol use.38 Third, we treated all health risk lifestyles equally. Some earlier studies have suggested that certain specific health behaviours may have a decisive role in the accumulation of the behaviours.39 For example, in some studies smoking has been found to be at the core of such accumulation.40 Fourth, we do not know whether changes in lifestyle risk factors resulted in a long-term shift in lifestyle risk factors beyond the period covered by the study, and whether there were several changes in the co-occurrence of health risk factors between the two surveys. Fifth, although this a prospective study, a revised causality is still possible (the co-occurrence of multiple lifestyle risk factors may lead to reduced social capital). Sixth, our study population largely consisted of female white-collar public sector employees. These mainly well-educated employees have more healthy lifestyles, a better knowledge of health risks, and a higher level of social capital at work than many other groups of employees. Therefore, the healthy worker effect may affect our data, and the generalisabilily of the findings should be tested using other samples. Finally, although most workplaces were rather small, there were some larger work units with more than 100 employees. Hence, in this study, the concept “social capital” mostly refers to the characteristics of a small group but in some cases may refer to the characteristics of the wider social environment at work. Future studies may explore whether group size and/or density influence the impact of social capital on the possibility of lifestyle risk factors.

Since persons with several behavioural risk factors are likely to have considerable medical expenditure compared to persons without such factors,41 our results may have some practical implications. However, because the individual-level association was not completely supported at the ecological (aggregated) level, and after adjustment for several factors, the association between social capital and the co-occurrence of risk factors was observed only in a cross-sectional setting, this study did not give clear support for interventions to improve workplace social capital as a means of preventing multiple lifestyle risk factors and promoting health.