Alternative Weighting Approaches for Anchored Matching-Adjusted Indirect Comparisons via a Common Comparator

Abstract

Background: Adjusted indirect comparisons (anchored via a common comparator) are an integral part of health technology assessment. These methods are challenged when differences between studies exist, including inclusion/exclusion criteria, outcome definitions, patient characteristics, as well as ensuring the choice of a common comparator. Objectives: Matching-adjusted indirect comparison (MAIC) can address these challenges, but the appropriate application of MAICs is uncertain. Examples include whether to match between individual-level data and aggregate-level data studies separately for treatment arms or to combine the arms, which matching algorithm should be used, and whether to include the control treatment outcome and/or covariates present in individual-level data. Results: Results from seven matching approaches applied to a continuous outcome in six simulated scenarios demonstrated that when no effect modifiers were present, the matching methods were equivalent to the unmatched Bucher approach. When effect modifiers were present, matching methods (regardless of approach) outperformed the Bucher method. Matching on arms separately produced more precise estimates compared with matching on total moments, and for certain scenarios, matching including the control treatment outcome did not produce the expected effect size. The entropy balancing approach was used to determine whether there were any notable advantages over the method proposed by Signorovitch et al. When unmeasured effect modifiers were present, no approach was able to estimate the true treatment effect. Conclusions: Compared with the Bucher approach (no matching), the MAICs examined demonstrated more accurate estimates, but further research is required to understand these methods across an array of situations.

Performance of unanchored matching-adjusted indirect comparison (MAIC) for the evidence synthesis of single-arm trials with time-to-event outcomes

“…Second, only 1000 repetitions were conducted for each scenario of data generation due to the labor-demanding process of manually completing part of the K-M curve digitization. Although the number of repetitions matches that of a previous simulation study in the realm of MAIC [5], the possibility of insufficient repetitions to reveal the properties could not be fully ruled out. Third, the same specification of shape and scale parameters of the Weibull distribution was used in the simulation of both A and B arms, which may be reasonable if the populations are adequately homogenous across trials.…”

“…At its core, the MAIC method utilizes the individual-level patient data (IPD) from the trial of an intervention (usually a manufacturer's own product) and the published aggregate data from the trial of a comparator intervention, and re-weights the patients with IPD such that their characteristics are balanced with those of the patients from the aggregate data of the comparator's trial [3]. The weights can be obtained using propensity scores estimated with method of moments or entropy balancing, either of which is calculated using the observed characteristics that need to be balanced [3,5]. The outcome of the patients with IPD calculated with re-weighting is then compared with that of the published aggregate data to obtain the relative effect [3].…”

“…Previous discussion of and studies on the properties of MAIC have focused on anchored analyses of linear-scale outcomes [3][4][5][6]. The properties of unanchored MAIC in the context of time-to-event analysis have not been investigated so far yet the literature is gradually picking up the approach without appreciating the unique profiles of this estimator [11][12][13].…”

Pns311 Predictive-Adjusted Indirect Comparison (Paic): A Novel Method for Population-Adjusted Indirect Comparison

“…Signorovitch et al 10 propose using a method of moments to estimate the model parameters by setting the weights so that the mean covariates are exactly balanced across the two trial populations. Petto et al 19 and Belger et al 22 propose using entropy balancing instead of the method of moments to estimate the weights, claiming that entropy balancing should penalize extreme weighting schemes and provide greater precision. In fact, weight estimation via entropy balancing and the method of moments are mathematically identical.…”

Component network meta‐analysis compared to a matching method in a disconnected network: A case study

“…Goring et al (2016) and Béliveau, Goring, Platt, and Gustafson (2017) discuss Bayesian approaches to link disconnected networks. Another possible approach is to add evidence from nonrandomized trials to a contrastbased network, using propensity score or matching-adjusted indirect comparisons (MAIC) methods (Petto, Kadziola, Brnabic, Saure, & Belger, 2019;Phillippo et al, 2017;Signorovitch et al, 2012;Veroniki, Straus, Soobiah, Elliott, & Tricco, 2016). However, nonrandomized comparisons may be associated with an unclear risk of bias, potentially higher than for RCTs.…”