OBJECTIVES: Single-arm trials (SATs) are becoming more common in an era of precision medicines. SATs typically lack trial randomization due to small sample sizes and limited statistical power, but health technology assessment bodies like The National Institute for Health and Care Excellence (NICE) require comparative effectiveness estimates for reimbursement decision making. Matching-adjusted indirect comparisons (MAICs) are one statistical method used to indirectly compare single-arm evidence to relevant comparators. This research aims to quantify the impact of MAIC parameter inclusion and sample size on propagated economic model uncertainty.

METHODS: We used a NICE technology appraisal (TA781) and supporting publications to create a simulated individual participant data (SIPD) set, conduct MAICs using the published comparator data, calculate parametric survival analysis and populate a partitioned-survival model. Two cuts of the SIPD were selected, the 'full’ sample size (n=174) and an arbitrarily selected smaller subset (‘small’, n=30). For both cuts, MAICs were conducted using all variables (ALL) and high priority variables (HP) only.

RESULTS: For the full [small] samples, the effective sample size (ESS) reduced by 40% [46%] (ALL) and 25% [18%] (HP). The full sample MAICs displayed similar levels of uncertainty in the probabilistic economic model. The HP small sample MAIC displayed more uncertainty than both full sample MAICs. The small sample ALL MAIC displayed the most uncertainty, with probabilistic iterations spread across all four quadrants of the cost-effectiveness plane.

CONCLUSIONS: Where uncertainty is driven by low ESS, we recommend consideration of whether a MAIC is truly informative. If a MAIC is deemed necessary, we recommend preserving ESS by only including HP variables whilst transparently outlining the impact this could have on the MAIC quality and supplementing with analyses, such as naïve comparison and qualitative discussion. We suggest these analyses are presented to decision makers so the original shape of iterations is known.