Methodological issues In this study, we excluded the relatively unhealthy workers at baseline from the study subjects of this study. The results of the sensitivity tests in the two alternative study groups supported the validity of the decision, despite a loss of statistical power. Including them into study subjects of this study (alternative study group 1) would have significantly LY3039478 in vivo underestimated the synergistic effects between job control and social support at work in both men and women. At the same time, the results in the

group (Table 6) suggests that a statistical adjustment of the baseline health conditions was not enough to remove their impact on the psychological job characteristics and general psychological distress at follow-up. We reported Thiazovivin mouse the two (80 and 95%) CIs of the Rothman’s synergy index in consideration of a potential Type II error. In this study, all of the synergy indexes between job control and social support at work on psychological distress were non-significant at the alpha level of 0.05. However, they were significant

at the alpha level of 0.20 in women (Tables 4, 5). Also, in men, when the sample size was almost doubled (i.e., in the alternative study group 1), the 80% CIs of the synergistic indexes became clearly above or below unity (Table 6). All of these indicate that an injudicious application of the typical alpha Reverse transcriptase level (0.05) to interaction significance tests could obscure a possible synergism. As mentioned before, low statistical power in interaction tests (Greenland 1993; Marshall 2007; Selvin

1996) should be considered. In addition, Rothman (1978) warned that a quantitative interval estimation of synergy index should not be confused with a significance test (in which typically the alpha level of 0.05 is employed). Hogan et al. (1978) also reported that the CIs of synergy index based on a simple asymptotic approach (Hosmer and Lemeshow 1992) could be unduly conservative in comparison with alternative approaches. More importantly, we think that a synergism between two exposures should be judged based on an array of information such as a strong theoretical hypothesis, a significant difference between the results under no-interaction assumption and under an interaction assumption as presented in Tables 3, 4 and 5, and confidence intervals considering a type II error, not solely based on the significance test (at the alpha of 0.05) of synergy index. Implications for risk assessment, job stress models, and interventions The most important lesson from this study is that the risk assessment of the combination of low job control, high job demands, and low social support at work on common mental disorders needs to be conducted with full consideration of their interactions and study context (Johnson and Hall 1996; Kasl 1996; Schaubroeck and Fink 1998).