Selective Information Borrowing for Region-Specific Treatment Effect Inference under Covariate Mismatch in Multi-Regional Clinical Trials

Multi-regional clinical trials (MRCTs) are central to global drug development, enabling evaluation of treatment effects across diverse populations. A key challenge is valid and efficient inference for a region-specific estimand when the target region is small and differs from auxiliary regions in baseline covariates or unmeasured factors. We adopt an estimand-based framework and focus on the region-specific average treatment effect (RSATE) in a prespecified target region, which is directly relevant to local regulatory decision-making. Cross-region differences can induce covariate shift, covariate mismatch, and outcome drift, potentially biasing information borrowing and invalidating RSATE inference. To address these issues, we develop a unified causal inference framework with selective information borrowing. First, we introduce an inverse-variance weighting estimator that combines a “small-sample, rich-covariate” target-only estimator with a “large-sample, limited-covariate” full-borrowing doubly robust estimator, maximizing efficiency under no outcome drift. Second, to accommodate outcome drift, we apply conformal prediction to assess patient-level comparability and adaptively select auxiliary-region patients for borrowing. Third, to ensure rigorous finite-sample inference, we employ a conditional randomization test with exact, model-free, selection-aware type I error control. Simulation studies show the proposed estimator improves efficiency, yielding 10-50% reductions in mean squared error and higher power relative to no-borrowing and full-borrowing approaches, while maintaining valid inference across diverse scenarios. An application to the POWER trial further demonstrates improved precision for RSATE estimation.

💡 Research Summary

Multi‑regional clinical trials (MRCTs) are indispensable for global drug development, yet estimating a region‑specific average treatment effect (RSATE) in a small target region is challenging because of covariate shift, covariate mismatch, and outcome drift across regions. This paper proposes a unified causal‑inference framework that enables selective borrowing of information from auxiliary regions while preserving valid inference for the RSATE.
The first component introduces an inverse‑variance weighting (IVW) estimator that combines two doubly‑robust estimators: (i) a “small‑sample, rich‑covariate” target‑only estimator that uses all prognostic variables available in the target region (including region‑specific covariates), and (ii) a “large‑sample, limited‑covariate” full‑borrowing estimator that pools all regions but only the common covariates. By weighting each estimator inversely to its estimated variance, the IVW estimator attains the minimum variance under the assumption of no outcome drift.
Recognizing that outcome drift—systematic differences in potential outcomes not captured by observed covariates—may violate the exchangeability assumption, the second component applies conformal prediction to each auxiliary‑region patient. A conformal prediction interval is constructed based on the common covariates; patients whose observed outcomes fall outside the interval are deemed non‑comparable and excluded from borrowing. This patient‑level selection yields a subset of auxiliary data that plausibly satisfies the no‑drift condition, after which the IVW estimator is recomputed on the combined target data and selected auxiliary data.
The third component addresses the inferential challenge introduced by the data‑driven selection. A conditional randomization test (CRT) is employed: treatment assignments are repeatedly re‑randomized conditional on covariates, the selection rule is reapplied, and the test statistic is recomputed to generate its exact finite‑sample null distribution. This model‑free approach guarantees exact type‑I error control while accounting for selection uncertainty, avoiding the anti‑conservative behavior of standard asymptotic tests.
Simulation studies spanning a wide range of covariate‑shift magnitudes, degrees of covariate mismatch, and presence or absence of outcome drift demonstrate that the proposed selective borrowing method (CSB) reduces mean‑squared error by 10–50 % relative to both no‑borrowing and full‑borrowing baselines and improves statistical power by 5–15 percentage points. The conformal‑based selection is especially effective when covariate distributions differ sharply across regions, as it filters out biased auxiliary observations without sacrificing much sample size.
The methodology is illustrated with the POWER trial, a multi‑regional study of enobosarm versus placebo. Treating North America as the target region and South America/Europe as auxiliaries, the CSB estimator yields a substantially tighter 95 % confidence interval for the RSATE compared with either the target‑only or full‑borrowing estimators, providing more precise evidence for local regulatory decision‑making.
In summary, the paper delivers three key innovations: (1) an efficiency‑maximizing IVW combination of target‑only and full‑borrowing doubly‑robust estimators, (2) a conformal‑prediction‑driven patient‑level selection mechanism to mitigate outcome drift, and (3) an exact CRT‑based inference procedure that respects the selection step. Together, these advances enable robust, efficient, and regulatory‑relevant estimation of region‑specific treatment effects in MRCTs, even when the target region is small and heterogeneous. Future work may extend the framework to hierarchical multi‑source settings or incorporate Bayesian priors for further gains.