Furthermore, we conducted linear regression analyses to investigate whether: (1) the percentage of smokers in the workgroup predicts change in smoking status; (2) the average body mass index in the workgroup predicts weight change (change in BMI); and (3) average physical

activity level predicts change in physical activity. To avoid response bias introducing spurious associations, we calculated the number of smokers, levels of body mass index and physical activity as the average of baseline and follow-up values. In other words, we looked at the association between change in score and average score (Bland and Altman, 1986). Potential non-linear effects were evaluated through quadratic terms; these were INCB024360 clinical trial significant with regard to smoking status. In the case of quadratic effects, we centralized the variable for average share of smokers to avoid issues with multicollinearity. All the statistical analyses were performed with SAS Proc Glimmix and Proc GLM, version 9.2 (SAS Institute). Table 1 presents descriptive C646 datasheet statistics of the participant and workgroups at baseline and follow-up. On average, the respondents were 46.5 years old and had worked at their current workplace for approximately 9.5 years

at baseline. 82% of the respondents worked as health care workers, while approximately 7% were managers and 10% held another type of work position (such as janitor and secretary). Respondents had an average baseline BMI of 24.91, which increased to 25.15 at follow-up. Of the respondents who smoked at baseline, 13.75% had quit by the time of follow-up. The analyses on workgroup level illustrate workgroup variation for some variables. For example, in the quartile of workgroups with lowest smoking, only 17% of employees smoke, while 52% smoked in the quartile of workgroups with highest level of smoking. Table 2 presents the results from the multilevel regression models, showing how much of the variation in each outcome

that is explained by workgroup. Three of the eight outcomes were significant at the 0.05 level. Specifically, we found that 6.49% of the variation in baseline smoking status (p < 0.0001; 95% CI: 4.46–10.22), 6.56% of the variation in amount smoked (p = < 0.0001; for 95% CI: 4.59–10.09) and 2.62% in BMI (p = 0.0002; 95% CI: 1.20–3.97) was explained by workgroup. Also, 1.11% of the variation in LTPA was explained by workgroup, albeit only borderline significant (p = 0.0620; 95% CI: 0.43–6.77). In small workgroups, only the variation in smoking and amount smoked was significantly explained by workgroups (results not shown). We found similar results in additional analyses where gender, age and cohabitation status were included as fixed effects (results not shown). Results from the linear regression analyses are presented in Table 3. We found support for two of our three tested outcomes.