Presentation (CTI)

OBJECTIVES: Population-adjusted indirect comparisons (PAICs) are often used to support health technology assessments for rare diseases. However, prior research comparing the performance of anchored matching-adjusted indirect comparisons (MAIC) and simulated treatment comparisons (STC) is conflicting. Additionally, there is limited research comparing these methods in unanchored settings.
METHODS: A simulation study was conducted to compare the performance of four methods: MAIC (adjusting for mean only), MAIC (adjusting for mean and variance), STC (plug-in), and STC (standardization). Data were simulated for an unanchored setting in which there was individual-level patient data (IPD) on treatment A from one single-arm trial and aggregate-level data (AgD) on treatment B from another, each with a sample size of 100. Four scenarios were examined in which we varied: 1) the degree of covariate overlap; 2) the event rate in the IPD; 3) the strength of prognostic factors and effect modifiers; and 4) the standard deviation of covariates in the AgD. The target estimand was the marginal odds ratio comparing treatment A versus B in the AgD study population. Performance metrics included bias, empirical standard error, model standard error, and coverage, with each scenario replicated 2000 times.
RESULTS: STC (standardization) performed well across all scenarios. When covariate overlap was low, MAIC performed poorly in terms of bias and precision. Adjusting for the variance in the MAIC reduced precision without yielding any meaningful gains in bias reduction. In scenarios where the marginal and conditional outcomes differed, STC (plug-in) was biased. When the event rate in the IPD was low, all methods had increased bias.
CONCLUSIONS: Results from this simulation study highlight situations in which MAIC and STC (plug-in) can perform poorly. To compliment the findings from these more conventional methods, STC (standardization) should be routinely conducted when performing unanchored PAICs.