Anne Fleur Kortekaas-Rijlaarsdam

CHAPTER 2 32 Moderators Because the number of studies was limited and at least ten studies are recommended for reliable meta-regression (Borenstein et al., 2009), we performed meta-regression analyses only for math outcomes. For math accuracy we tested the following moderators: (1) Demographic moderators age (year); gender (percent male); (2) disorder related moderators percent children diagnosed with ADHD-Inattentive subtype; percent diagnosed with comorbid ODD or CD; parent-rated ADHD severity (standardized [mean divided by SD] baseline ADHD symptom ratings on standardized questionnaires, for an overview of questionnaires used to assess severity of ADHD symptoms, see supplementary material Table E2.2), and (3) study characteristics including release system (immediate versus extended release, the latter including transdermal); duration of the study conditions (days); time of measurement (post-dose, in hours); medication dosage (mg); titration method (clinical titration versus fixed dosages). For math productivity we tested demographic and disorder related moderators: age, gender, percent diagnosed with ADHD-Inattentive subtype, and study characteristics: release system, trial duration, and titration method. Insufficient number of studies reported on comorbid learning disorders. In case of doubt, authors were contacted. Moderators were explored using meta-regression between the study samples’ effect sizes for academic performance and the selected moderators. Mediator and moderator effects were studied separately for math accuracy and math productivity using meta- regression with a random model (method of moments) (Borenstein et al., 2009). Statistical Analysis Statistical analyses were performed using SPSS version 21.0 (IBM, 2012) and Comprehensive Meta-Analysis software V3.0 (Borenstein, 2014). Because accuracy and productivity measures are proportional measures that require effect sizes for binary data (Borenstein et al., 2009), risk differences (MPH minus placebo) were calculated. The standard errors of the risk difference were calculated (Twisk, 2010) because included articles commonly reported on the number correct and the number completed and therefore the reported p-values and standard deviations were not applicable to the calculated risk differences. Effect sizes were calculated for math accuracy, math productivity, reading accuracy and reading number attempted. In supplementary material Table E2.1, we provide a narrative description of the studies reporting on spelling. The derived effect sizes were weighted by their inverse variance to account for differences in sample size and error of measurement (Borenstein et al., 2009). As heterogeneity may have been introduced by using data from studies with different designs (i.e. differences in treatment duration, dosages) and different participants (e.g.