Employees who were sick-listed in January 1st 2002 were excluded from the analysis. Sickness absence days were counted until December 31st 2004, even if the employee remained on sick-leave thereafter. The number of sickness Fedratinib absence episodes between January 1st 2002 and December 31st 2004 was

also counted for each employee, distinguishing between short episodes (1–21 days) of uncertified absence, and long episodes (>21 days) of mostly certified sickness absence. Earlier, the sick-leave was assessed on the individual level by the total number of sickness absence days in the period January 2000 through December 2001. Statistical analyses The number of absence days was skewed MAPK Inhibitor Library price to the right [mean 48.9 days, standard deviation (SD) 82.8 days; median 18.0 days]. selective HDAC inhibitors Normal distribution was approximated after logarithmic transformation: mean 2.9 (SD = 1.5) and median 2.9. The prospective associations between psychosocial work conditions and the log-transformed number of sickness absence days were analyzed with multiple linear regression (SPSS for Windows, version 15) controlling for earlier sick-leave and psychological distress. The linear regression models fitted the number of sickness absence days in men and women well

but explained little (12–14%) of the variance in the number of sickness absence days. To examine the prospective associations between psychosocial work conditions and sickness absence episodes, a Poisson regression model was computed using GENLOG for general log-linear analysis in SPSS

for Windows version 15. The Poisson distribution implies that the variance is equal to the mean (μ). The Poisson model showed a good fit for the number of long episodes. The variance in the number of short episodes of absence, however, was greater than the mean resulting in overdispersion. Therefore, a zero-inflated negative binomial distribution was estimated for short absences using Transition Data Analysis version 6.4f (Blossfeld Progesterone and Rohwer 2002). The negative binomial distribution proved to be a better fit for the number of short sickness absence episodes. In the negative binomial model and the Poisson regression model earlier sick-leave and psychological distress were adjusted for. Results Of the distributed 395 questionnaires, 265 (67%) were returned to the occupational health service. Twenty-one questionnaires were excluded because they were not complete. Thus, a total of 151 employees (64 men and 87 women) were not eligible for analysis. These non-participants were 39.2 [standard deviation (SD) = 7.1] years of age, had 6,271 sickness absence days and a total of 732 sickness absence episodes, of which 686 short episodes and 46 long episodes, during follow-up. 244 participants (103 men and 141 women) were 39.0 (SD = 8.