The studies included in the meta-analysis reflect a random sample of the relevant

distribution of ORs as effect sizes and the pooled OR estimates the mean effect in this distribution. Study weights were assigned according to the inverse variance. Q values were Navitoclax calculated for estimating heterogeneity as the weighted sum of squared differences between individual study effects. According to the classification of Hartvigsen and colleagues (2004), ORs between 1.50 and 2.00 were considered moderate, and higher ORs were considered strong. ORs were considered statistically significant if the 95% CI straddled 1.00. Publication bias was examined through visual inspection of asymmetry in a scatter plot and Egger’s (1997) constant of regression. A sensitivity analysis was conducted based on trial quality. Only studies with a quality score < 4, ie, those

Ion Channel Ligand Library datasheet with low risk of bias, were included in the sensitivity analysis to explore how methodological quality affects the overall result (Guyatt and Rennie et al 2002). The Statistical Programming Language R, version 2.14.0 was used for all analyses. The electronic searches identified 589 publications, of which 154 were considered potentially relevant and were evaluated as full-text papers. Of these, 146 studies were excluded. Figure 1 presents the flow of the studies through the review and the reasons for exclusions. Searching the reference lists of the eight eligible studies identified another two eligible studies. Therefore 10 studies were included in the review (Schultz et al 2004, Steenstra et al 2005, Dionne et al 2005, Hagen et al 2005, Schultz et al 2005, Shaw et al 2005, Kapoor et al 2006, Lotters and Burdorf, 2006, Turner

et al 2006, Reme et al 2009). Quality: Five studies had a low risk of bias, with AHRQ scores of 2 ( Lotters et al 2006) or 3 ( Schultz et al 2004, Steenstra et al 2005, Kapoor et al 2006, Turner et al 2006). The other five studies all had a moderate risk of bias, with an AHRQ score of 5. The quality criterion related to < 20% loss to follow up was met in only three of the Rebamipide studies ( Hagen et al 2005, Steenstra et al 2005, Kapoor et al 2006). Consensus about quality interpretation was unanimous. Table 1 presents the quality of the studies and Table 2 presents the characteristics of the studies. Participants: The total number of participants in the 10 included studies was 4683. Overall, 59% of the participants were male, although one study listed no gender details ( Schultz et al 2004). The mean age of participants in each study ranged from 35 to 43 years. Outcome: Absence from usual work in a given period was reported using different terms such as ‘not return to work’, ‘sick leave’, ‘work absenteeism’, ‘sickness absenteeism’, and ‘compensated sick leave’. Follow-up time ranged from 3 to 24 months.