OBJECTIVES  Matching-adjusted indirect comparison (MAIC) is a novel comparative effectiveness approach to address biases that can appear in traditional indirect comparison (IC) methods when patient characteristics differ across trials. We examined three unanswered MAIC methodological questions and applied the proposed solutions to a comparison of ADHD treatments. METHODS Using individual patient data from two randomized controlled trials (RCTs) comparing guanfacine (GXR) vs placebo and published summary statistics from four RCTs comparing atomoxetine (ATX) vs placebo, MAIC was used to reweight the GXR data so that observable GXR patient characteristics matched those of ATX patients. Change in ADHD-RS-IV scores was the primary endpoint. Comparative efficacy results were evaluated for their sensitivity to changes in the following three MAIC specifications: variable selection using regression-based methods, statistical moments matched (i.e., mean vs mean and variance), and matching on placebo-arm outcomes. RESULTS Both treatments decreased ADHD-RS-IV scores relative to placebo (-17.9 GXR vs -10.7 placebo; -14.6 ATX vs -5.8 placebo). In the baseline MAIC specification adjusting for patient baseline characteristics and placebo arm outcomes, GXR produced larger decreases in ADHD-RS-IV scores than ATX (Δ: -3.9, p<0.004). The results were insensitive to adding variables to the matching algorithm (Δ: -3.8, p<0.023), or matching only covariate means rather than both means and variances (Δ: -3.6, p<0.006). Applying MAIC without matching placebo arm outcomes indicated a slightly greater decrease in ADHD-RS-IV scores for ATX, but there was no statistically significant difference between GXR and ATX (Δ: 0.6, p<0.649). CONCLUSIONS In this study, MAIC results were insensitive to variable selection via regression and the statistical moments matched, but matching the placebo arms altered the results. Matching placebo arm outcomes is valid when unobserved trial-specific factors have a differential impact on a trial’s treatment and control arm outcomes; this was likely the case in this GXR-ATX study.