Abstract
Biomarker-guided designs are increasingly used to evaluate personalized treatments based on patients' biomarker status in Phase II and III clinical trials. With adaptive enrichment, these designs can improve the efficiency of evaluating the treatment effect in biomarker-positive patients by increasing their proportion in the randomized trial. While time-to-event outcomes are often used as the primary endpoint to measure treatment effects for a new therapy in severe diseases like cancer and cardiovascular diseases, there is limited research on biomarker-guided adaptive enrichment trials in this context. Such trials almost always adopt hazard ratio methods for statistical measurement of treatment effects. In contrast, restricted mean survival time (RMST) has gained popularity for analyzing time-to-event outcomes because it offers more straightforward interpretations of treatment effects and does not require the proportional hazard assumption. This paper proposes a two-stage biomarker-guided adaptive RMST design with threshold detection and patient enrichment. We develop sophisticated methods for identifying the optimal biomarker threshold, treatment effect estimators in the biomarker-positive subgroup, and approaches for type I error rate, power analysis, and sample size calculation. We present a numerical example of re-designing an oncology trial. An extensive simulation study is conducted to evaluate the performance of the proposed design.
Introduction
In the era of precision medicine, researchers frequently utilize biomarkers to identify patient subgroups most likely to respond to a novel therapy. However, pre-specifying a biomarker cutoff before a trial begins is inherently risky; an incorrect threshold can lead to diluted results or missed treatment effects. Furthermore, traditional survival-based trials rely heavily on the Hazard Ratio (HR), which assumes proportional hazards—a condition often violated in modern immunotherapy where treatment effects may be delayed.
This paper introduces a novel two-stage clinical trial design that adaptively identifies the optimal biomarker threshold to "enrich" a study with patients most likely to benefit. By utilizing Restricted Mean Survival Time (RMST) instead of traditional hazard ratios, the design remains robust even when treatment effects are non-proportional, providing a more intuitive measure of clinical benefit (e.g., average survival time gained).
Methodology
The proposed framework follows a two-stage biomarker-guided adaptive RMST design, as illustrated in Figure 1.
The Two-Stage Workflow
- Stage I (Exploration): Patients are enrolled regardless of biomarker status. At the end of this stage, a prediction method based on piecewise exponential models is used to estimate a biomarker cutpoint ($\hat{c}_0$) where the treatment begins to show a positive effect.
- Stage II (Enrichment): Based on the Stage I findings, the trial either continues with "No Enrichment" (enrolling all patients) or switches to "Enrichment," where enrollment is restricted to the identified biomarker-positive subgroup.
- Final Analysis: The treatment effect is measured using RMST, representing the area under the survival curve up to a specific time point ($t^*$).
Calibration Weighting (CW)
A critical challenge in adaptive enrichment is the "distributional shift" caused by changing enrollment criteria mid-trial. To prevent biased results, the authors introduce Calibration Weighted Estimators. These tools re-weight the data from both stages to ensure the final estimate accurately reflects the treatment effect in the target population. Among the methods tested, the CW G-Formula Estimator demonstrated the highest statistical power and lowest variance.
Evaluation and Key Results
The design was evaluated through a numerical re-design of the JAVELIN Lung 200 oncology trial (avelumab vs. docetaxel) and extensive simulations involving 10,000 datasets.
- Sample Size Efficiency: In the JAVELIN Lung 200 re-design, the proposed enrichment design required only 845 patients to achieve 90% power, compared to 940 patients for a traditional all-comer design—a 10% reduction in required participants.
- Bias Correction and Error Control: Simulations confirmed that "Naive" estimators overestimate treatment effects when enrichment occurs. The proposed CW estimators successfully removed this bias, maintaining a 95% coverage probability. Additionally, the design strictly controlled the family-wise type I error rate below the 2.5% threshold.
- Robustness: RMST proved superior to Hazard Ratios in scenarios with delayed treatment effects, as it does not rely on the proportional hazards assumption and provides a more clinically meaningful interpretation of survival gain.
Discussion and Future Work
While the proposed design offers significant advantages in efficiency and interpretability, it currently focuses on a single continuous biomarker. Future research may extend this framework to handle complex "signatures" involving multiple biomarkers. Additionally, the method's sensitivity to the choice of truncation time ($t^*$) and its reliance on piecewise exponential assumptions for early-stage prediction are areas for further refinement. The authors suggest that integrating this approach into group-sequential designs could allow for even earlier trial termination for success or futility, further optimizing the drug development process.
Summary updated