Bibian van der Voorn

142 CHAPTER 10 Appendix 1 presents the full search strategy, which was based on the following index terms or free-text words: ‘cortisol’ or ‘glucocorticoid’, and ‘sex difference’ or ‘sexual characteristics’, and ‘child’ or ‘adolescent’. Studies in children with (psycho)pathology, on synthetic glucocorticoids, or with risk for abnormal HPA axis activity (e.g., a history of maltreatment) were excluded. An English language restriction was applied for abstracts of published articles. No restrictions for year of publication or study design, apart from reviews and case reports, were applied. The review protocol was based on the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) statement. DATA COLLECTION Two independent assessors (BvdVand JJH) screened6,158 titles and abstractswithout consideration of outcomes. Studies were not assessed blindly. Disagreement between assessors was discussed until consensus was reached. When gender differences were analyzed without reporting on cortisol levels for boys and girls separately or when data were only presented in graphs, authors were requested for additional quantitative data. Data were stratified into two age groups: <8 yr (prepubertal) and between 8–18 yr (peri-/postpubertal). Ideally, stratification would have been based on pubertal staging according to Tanner. Unfortunately, only aminority of the included studies reported on the subjects’ Tanner stages. Because pubertal onset before age 8 years is considered to be pathologic 12 , we chose 8 yr as cut off for stratification. When articles reported on serial cortisol measurements, we included only data on the youngest assessment age. When cortisol levels were reported prepubertally as well as peri-/postpubertally within the same individual, we included one sampling moment for each stratified group. When articles reported on the same study population, we included the article with the lowest bias risk. When articles reported on dynamic tests of HPA axis activity, we only included baseline cortisol. We only included the control subjects of case-control studies. If known, we excluded female subjects on oral contraceptives. When gender differences were described but not quantified, the articles were included in the descriptive analysis rather than the meta-analysis. META-ANALYSIS When necessary, we converted serum and salivary cortisol levels into nmol/L, and 24h-urine cortisol levels into µg/24h. When means ± SDs were not reported, the SD was calculated based on the following assumptions: the 95% CI is 3.92 SDs wide (2 × 1.96); the inter-quartile range is 1.35 SDs wide; the range is 4 SDs wide; the SD is the SE multiplied by the square root of the sample size 13 . To assess parametricity, we