Presentation (CTI)

OBJECTIVES: Matching adjusted indirect comparisons (MAIC) and simulated treatment comparisons (STC) are common approaches for pairwise population-adjusted indirect comparisons, where individual patient data (IPD) are available from one trial and aggregate data (AD) from another. However, limited covariate overlap can reduce the efficiency of MAIC, and non-linear outcome models can greatly complicate the implementation of STC for estimating marginal treatment effects needed for decision-making. This research aims to develop a new simple to implement STC approach for estimating marginal treatment effects in the anchored setting in which trials share a common comparator arm.
METHODS: We propose an STC method that uses jackknife pseudo values of the marginal treatment effect estimate, calculated on the linear predictor scale in the IPD trial, as the dependent variable rather than the observed outcome. These pseudo values decompose the marginal treatment effect into patient-level contributions, which can then be combined with baseline characteristics to estimate the treatment effect in the population of the AD trial. Unlike other recent proposals to extend STC, our approach avoids specifying the joint covariate distributions in the AD trial or using computationally intensive techniques such as numerical integration or simulation. Additionally, our approach includes a simple variance estimator to calculate confidence intervals.
RESULTS: In simulated trials of continuous, binary, and time-to-event outcomes, our proposed approach shows less bias than traditional STC for non-linear models and smaller mean squared error than MAIC under both correct and incorrect specification of baseline covariate functional forms.
CONCLUSIONS: By estimating marginal treatment effects with minimal modelling assumptions and without computationally intensive methods, our approach offers a practical and accessible alternative for population-adjusted indirect comparisons, supporting population-level decision making.