PNS311 PREDICTIVE-ADJUSTED INDIRECT COMPARISON (PAIC): A NOVEL METHOD FOR POPULATION-ADJUSTED INDIRECT COMPARISON

Abstract

Population-adjusted indirect comparison methods are increasingly used to compare treatment outcomes across separate clinical trials and to inform health technology assessment. A novel method for population-adjusted comparisons, predictive-adjusted indirect comparison (PAIC), is presented and a comprehensive simulation study is carried out. The rationale behind PAIC is the following: Individual-level data are used to model the link between an outcome and the available prognostic variables and effect modifiers via a generalised linear model. Then, a large number of pseudo-populations are generated by forward sampling from the published aggregate-level values. The outcomes for a trial conducted in a population exchangeable with that of the trial with aggregate-level data are predicted. The simulated datasets are used to estimate an unbiased average treatment effect. A comprehensive simulation study has been carried out where the statistical properties of PAIC are evaluated and benchmarked against those of other recently proposed approaches for population-adjustment (MAIC and STC). The comparative vulnerability of each population-adjustment method to failures in assumptions/model specification is also assessed and compared to that of standard indirect comparison approaches such as the Bucher method. PAIC consistently produces less biased and more accurate estimates across and within scenarios. All population-adjustment methods provide unbiased estimates when all effect modifiers are accounted for and when these interact with treatment in the same way in both trials. PAIC and STC are less biased than MAIC under effect modifier misspecification. The Bucher method is the gold standard when covariates are balanced but is clearly biased and inappropriate when they are not. MAIC displays large variance over replicates as weighting considerably reduces the effective sample size. Population-adjustment methods should be used with caution as their assumptions seem hard to be met in practice. Notwithstanding, PAIC is a very useful tool for comparative effectiveness research without head-to-head trials.