Parametric G‐computation for compatible indirect treatment comparisons with limited individual patient data

Abstract

Population adjustment methods such as matching‐adjusted indirect comparison (MAIC) are increasingly used to compare marginal treatment effects when there are cross‐trial differences in effect modifiers and limited patient‐level data. MAIC is based on propensity score weighting, which is sensitive to poor covariate overlap and cannot extrapolate beyond the observed covariate space. Current outcome regression‐based alternatives can extrapolate but target a conditional treatment effect that is incompatible in the indirect comparison. When adjusting for covariates, one must integrate or average the conditional estimate over the relevant population to recover a compatible marginal treatment effect. We propose a marginalization method based on parametric G‐computation that can be easily applied where the outcome regression is a generalized linear model or a Cox model. The approach views the covariate adjustment regression as a nuisance model and separates its estimation from the evaluation of the marginal treatment effect of interest. The method can accommodate a Bayesian statistical framework, which naturally integrates the analysis into a probabilistic framework. A simulation study provides proof‐of‐principle and benchmarks the method's performance against MAIC and the conventional outcome regression. Parametric G‐computation achieves more precise and more accurate estimates than MAIC, particularly when covariate overlap is poor, and yields unbiased marginal treatment effect estimates under no failures of assumptions. Furthermore, the marginalized regression‐adjusted estimates provide greater precision and accuracy than the conditional estimates produced by the conventional outcome regression, which are systematically biased because the measure of effect is non‐collapsible.

Target estimands for population‐adjusted indirect comparisons

“…ML‐NMR targets marginal estimands more flexibly, and can extend within‐study inferences to the relevant target population for HTA decision‐making. In addition, outcome modeling‐based methods such as ML‐NMR are typically more precise and efficient than weighting methods in estimating marginal treatment effects, particularly where overlap is poor and effective sample sizes after weighting are small 2,34,94 . Another advantage of outcome modeling approaches is that they can estimate conditional and marginal estimands, as the conditional estimates can be standardized into marginals.…”

“…A reduction in precision is natural and necessary, and a function of the “distance” between the covariate distributions; we are trying to learn about a treatment effect in a different study than that in which it was originally investigated. Recent simulation studies show that standardized regression‐adjusted (and in some cases, weighting‐adjusted) estimates of the marginal (log) odds ratio are also more precise than regression‐adjusted estimates of the conditional (log) odds ratio in this context 34,55 . These results are expected to hold for non‐collapsible effect measures in general.…”

“…The covariate‐adjusted marginal estimate accounts for covariate imbalances across studies but MAIC can be adapted to also account for imbalances between treatment arms 32 . Similarly, STC can incorporate G‐computation or model‐based standardization 33 so that a covariate‐adjusted marginal estimate is produced that accounts for covariate imbalances across studies and treatment arms, and has no compatibility issues in the indirect treatment comparison 34 . As per the method proposed by Phillippo et al for ML‐NMR, 2 one can estimate the marginal mean outcomes by treatment group, based on the fitted outcome regression.…”

“…As per the method proposed by Phillippo et al for ML‐NMR, 2 one can estimate the marginal mean outcomes by treatment group, based on the fitted outcome regression. Then, the model‐based predictions can be averaged over the joint covariate distribution of the target trial (that with aggregate‐level data) and contrasted to produce an estimate of the marginal treatment effect that accounts for baseline covariates 34 …”

Some considerations on target estimands for health technology assessment

“…Here lies the value of covariate‐adjusted indirect comparisons 20 , 30 , 31 and network meta‐regression approaches, 32 , 33 such as those examined by Phillippo et al 2 in their original simulation study. These methods explicitly account for differences in patient characteristics between studies, thereby targeting estimands in specific samples or populations.…”