[Paper Review] ASIED: A Bayesian Adaptive Subgroup-Identification Enrichment Design

Yanxun Xu, Florica J. Constantine+2 more|arXiv (Cornell University)|Oct 4, 2018

Statistical Methods in Clinical Trials24 references3 citations

TL;DR

This paper proposes ASIED, a Bayesian adaptive enrichment design that identifies predictive biomarkers and dynamically modifies trial eligibility to target subgroups with enhanced treatment effects. Using a multilevel target product profile (LRV and TV), ASIED employs a fully Bayesian random partition model (BayRP) to guide interim decisions on population enrichment, achieving high accuracy in subgroup detection and robust decision-making under heterogeneous treatment effects.

ABSTRACT

Developing targeted therapies based on patients' baseline characteristics and genomic profiles such as biomarkers has gained growing interests in recent years. Depending on patients' clinical characteristics, the expression of specific biomarkers or their combinations, different patient subgroups could respond differently to the same treatment. An ideal design, especially at the proof of concept stage, should search for such subgroups and make dynamic adaptation as the trial goes on. When no prior knowledge is available on whether the treatment works on the all-comer population or only works on the subgroup defined by one biomarker or several biomarkers, it is necessary to incorporate the adaptive estimation of the heterogeneous treatment effect to the decision-making at interim analyses. To address this problem, we propose an Adaptive Subgroup-Identification Enrichment Design, ASIED, to simultaneously search for predictive biomarkers, identify the subgroups with differential treatment effects, and modify study entry criteria at interim analyses when justified. More importantly, we construct robust quantitative decision-making rules for population enrichment when the interim outcomes are heterogeneous in the context of a multilevel target product profile, which defines the minimal and targeted levels of treatment effect. Through extensive simulations, the ASIED is demonstrated to achieve desirable operating characteristics and compare favorably against alternatives.

Motivation & Objective

• To address the challenge of identifying patient subgroups with enhanced treatment response when no prior knowledge exists on whether a therapy works for all-comers or a specific biomarker-defined subgroup.
• To develop a decision-making framework that dynamically adapts eligibility criteria during the trial based on interim data, particularly in the presence of heterogeneous treatment effects.
• To incorporate a multilevel target product profile (LRV and TV) into the decision-making process to balance clinical viability and statistical rigor.
• To create a robust, quantitative, and coherent Bayesian framework for subgroup identification and enrichment decisions that accounts for uncertainty and variability in treatment effects.
• To improve early-phase drug development by enabling data-driven, adaptive decisions that support precision medicine goals without relying on pre-specified subgroups.

Proposed method

• Proposes a Bayesian random partition (BayRP) model to flexibly identify subgroups with enhanced treatment effects by partitioning patients based on biomarker profiles.
• Uses a multilevel target product profile (LRV = lower reference value, TV = target value) to define minimal and desired treatment effect thresholds for decision-making.
• Employs a hierarchical Bayesian model to estimate treatment effects in subgroups and compute posterior probabilities for Go/Stop/Continue decisions at interim analyses.
• Introduces decision rules based on posterior probabilities of treatment effect exceeding LRV and TV, allowing for adaptive enrichment or stopping based on evidence.
• Incorporates sensitivity analysis and simulation-based evaluation to assess operating characteristics under various scenarios.
• Tunes decision thresholds via hyperparameters (ξ₁, ξ₂) to reflect user-specific risk tolerance (e.g., false go or false stop risk).

Experimental results

Research questions

• RQ1 Can a Bayesian adaptive design effectively identify predictive biomarkers and subgroups with enhanced treatment effects when no prior knowledge exists about the treatment’s population-level efficacy?
• RQ2 How can a multilevel target product profile (LRV and TV) be formally integrated into interim decision rules for population enrichment in adaptive clinical trials?
• RQ3 What is the operating performance of the proposed ASIED design in terms of correctly identifying effective subgroups or all-comers while minimizing false positive and false negative decisions?
• RQ4 How does the BayRP model compare to pre-specified subgroup methods in detecting heterogeneous treatment effects under varying biomarker configurations?
• RQ5 To what extent can the decision rules be calibrated to reflect different risk preferences in drug development (e.g., conservative vs. aggressive go decisions)?

Key findings

• In Scenario 2 (effective subgroup with β₀=0.25, β₁=2.55), ASIED recommended the investigational drug for a subgroup with 96% probability and stopped for the drug with 4% probability, demonstrating high accuracy in subgroup detection.
• When the treatment effect in the subgroup increased to β₁=2.83 (Scenario 3), ASIED correctly shifted to recommending the drug for all-comers in 6% of simulations, reflecting adaptive learning.
• For all-comers with moderate effect (Scenario 4, β₀=0.25, β₁=2.55), ASIED recommended the drug for all-comers in 95% of simulations, showing strong performance in detecting broad efficacy.
• In the case of a strong all-comer effect (Scenario 5, β₀=0.25, β₁=2.83), ASIED recommended the drug for all-comers with 100% probability, confirming robustness under high-effect scenarios.
• When no effective subgroup existed (Scenario 1), ASIED correctly stopped the trial in 99% of simulations, indicating low false positive rate in futility detection.
• The design demonstrated high specificity and sensitivity in all scenarios, with final recommendations closely aligning with true underlying treatment effects across diverse settings.