Presentation (CTI)

OBJECTIVES: This analysis aims to identify decision contexts in which matching adjusted indirect comparisons (MAICs) alter the point estimates but do not reduce decision uncertainty. It also examines why these methods were used in selected NICE technology appraisals (TAs), despite offering limited value for reducing decision uncertainty.
METHODS: We conducted a targeted methods review of MAICs, focusing on methodological constraints such as effective sample size (ESS), bias-variance trade-offs, and the consequences of model misspecification and untestable assumptions in sparse data settings. Key sources included methodological evaluations of MAICs and the associated limitations of this method. We then identified and reviewed NICE TAs where a submission used a a Bucher comparison for indirect evidence that was replaced with a MAIC during the appraisal process.
RESULTS: Methodological literature clearly demonstrates that when there is poor overlap between studies or small study sample sizes, MAICs produce unreliable estimates with unstable standard errors. Importantly, a small absolute ESS is a clear indicator that the assumptions required for MAIC are untenable. Case reviews found that committees in NICE TAs requested MAICs with the goal of addressing baseline imbalances despite clear data limitations for the decision context.
CONCLUSIONS: MAICs can provide more reliable indirect treatment comparisons but only under a specific set of circumstances and assumptions. MAICs do not intrinsically provide the decision maker with a better tool for decision making. In appraisals where data are limited, which is often the case in rare disease and oncology, companies and NICE committees should consider all available comparative evidence as mutually informative rather than each method being an exclusive choice. Our research highlights that in decision contexts with limited data, decision makers should engage with indirect comparison methods critically and avoid defaulting to the use of evidence hierarchy frameworks to inform their preferred choice of method.