Article type
Abstract
Background. The Cochrane Handbook recommends using intention-to-treat (ITT) analysis to maintain the benefits of randomization in clinical trials. However, in certain situations—such as migraine trials where a high percentage of randomized participants may not experience attacks—authors often choose to analyse treated participants (modified ITT [mITT]) to avoid underestimating the intervention effect.
Aim. To compare the impact of two different analysis methods (ITT and mITT) on effect estimates of a network meta-analysis in migraine trials.
Methods. This secondary analysis is based on a systematic review assessing the comparative effectiveness of nine drug classes for the treatment of acute episodic migraine in adult (CRD42020204703). For this methods study, we focused on three efficacy outcomes —pain-free at 2 hours, pain relief at 2 hours, and sustained pain-free up to 48 hours. For each outcome, we performed network meta-analyses within the R environment (V4.1.0) using ITT (including all randomised participants) and mITT (treated participants) approaches. We compared effect estimates and 95% confidence intervals (CI) of the two approaches.
Results. We included 131 RCTs of low or some risk of bias concerns. Post-randomisation ranged between 0-36%.
Overall, ITT and mITT analyses produced comparable effect estimates for all three outcomes. However, concerning pain relief at 2 hours, three comparisons achieved statistical significance, while one lost significance with ITT compared to mITT. For example, ITT showed a significantly greater effect with acetaminophen versus gepants (risk ratio [RR] 1.34, 95% CI 1.02-1.77) and no statistically significant difference for ditans versus gepants (RR 1.18, 95%CI 0.94-1.48); mITT indicated a similar direction of the effect but no statistically significant difference between acetaminophen and gepants (RR 1.27, 95%CI 0.96-1.67) and a greater effect with ditans (RR 1.27, 95%CI 1.02-1.59). For sustained pain-free up to 48 hours, three comparisons achieved statistical significance with ITT but not with mITT. No significant changes were observed for other comparisons.
Conclusions. ITT and mITT analyses generally yielded comparable effect estimates in network meta-analyses, with ITT showing increased statistical power in a few cases due to a larger participant count. The overall direction of the effect estimate remained consistent, despite variations in statistical significance for certain comparisons.
Aim. To compare the impact of two different analysis methods (ITT and mITT) on effect estimates of a network meta-analysis in migraine trials.
Methods. This secondary analysis is based on a systematic review assessing the comparative effectiveness of nine drug classes for the treatment of acute episodic migraine in adult (CRD42020204703). For this methods study, we focused on three efficacy outcomes —pain-free at 2 hours, pain relief at 2 hours, and sustained pain-free up to 48 hours. For each outcome, we performed network meta-analyses within the R environment (V4.1.0) using ITT (including all randomised participants) and mITT (treated participants) approaches. We compared effect estimates and 95% confidence intervals (CI) of the two approaches.
Results. We included 131 RCTs of low or some risk of bias concerns. Post-randomisation ranged between 0-36%.
Overall, ITT and mITT analyses produced comparable effect estimates for all three outcomes. However, concerning pain relief at 2 hours, three comparisons achieved statistical significance, while one lost significance with ITT compared to mITT. For example, ITT showed a significantly greater effect with acetaminophen versus gepants (risk ratio [RR] 1.34, 95% CI 1.02-1.77) and no statistically significant difference for ditans versus gepants (RR 1.18, 95%CI 0.94-1.48); mITT indicated a similar direction of the effect but no statistically significant difference between acetaminophen and gepants (RR 1.27, 95%CI 0.96-1.67) and a greater effect with ditans (RR 1.27, 95%CI 1.02-1.59). For sustained pain-free up to 48 hours, three comparisons achieved statistical significance with ITT but not with mITT. No significant changes were observed for other comparisons.
Conclusions. ITT and mITT analyses generally yielded comparable effect estimates in network meta-analyses, with ITT showing increased statistical power in a few cases due to a larger participant count. The overall direction of the effect estimate remained consistent, despite variations in statistical significance for certain comparisons.