In subgroup and sensitivity analyses, data were sufficient to assess the effect of study length, comorbidities, IQ, crossover design, unfair dose comparisons, and data imputation. Findings of these analyses were generally robust (appendix pp 479–91). Because of a paucity of data, we could not assess the effect of gender, age (children vs adolescents), low risk of bias, medication status, and industry sponsorship. Sensitivity analyses investigating the effect of different maximum doses confirmed the results of the primary dose analysis (appendix pp 492–575).
