Bibian van der Voorn

143 GENDER-SPECIFIC HPA AXIS ACTIVITY IN CHILDHOOD assumed that a normal distribution extends no more than 2 SDs from the mean 14 , i.e., when normally distributed, the meanminus 2 SDs should be >0 nmol/L. Data analyses were performed using Review Manager (RevMan) version 5.3.5, 2014. For each study, the standardized mean gender difference (95% CI) in cortisol concentration was calculated by combining the SD with the sample size. Subsequently, fixed-effect meta-analyses were performed first, which assumes that the effect estimate of the group differences was fixed across studies. Second, the results of these analyses were compared with random-effects meta-analysis, which weigh studies of variable sample sizes more equally. We reported any source of bias from each included article conform the PRISMA statement and assessed selection, performance, detection and other biases. Bias was assessed as low, unclear or high ( Figure 1, Appendix 2 ). A sensitivity analysis was done by excluding studies that had ≥ 1 high bias risks. Heterogeneity of the data was assessed by the I 2 statistic, with significance defined as I 2 >50%. Publication bias was assessed through funnel plots. FIGURE 1. Risk of bias graph presenting a summary of the judgements of the accessors concerning risk of bias across all studies included in the meta-analysis. Bias risk is presented as percentage of total studies (n = 58). RESULTS Figure 2 shows the flowchart of the descriptive analysis and meta-analysis. Of the 6,158 titles and abstracts, 414 (7%) were eligible for full-text screening, from which 79 articles (19%) were included. Thirty-one authors of articles with insufficient quantitative data were contacted, of whom 12 responded: six provided the necessary quantitative data, five did not have access to the raw data anymore and one was not willing to participate. Two articles reported the cortisol production rate assessed through 24h-serum sampling, which hampered inclusion in the meta-analysis. The authors of 27 articles that only provided gender-specific data in figures were contacted, but could not be reached. Subsequently, these articles were excluded. Finally, 21 articles were included only in the descriptive analysis, and 58 articles (with data on 16,551 subjects) had sufficient data for inclusion in the meta-analysis.