A Proposed Confidence Ellipse Approach for Benefit-Risk Assessment in Clinical Trials.

Abstract

In clinical development, an independent data safety monitoring committee (IDMC) is often established to ensure the test treatment's integrity, quality, safety, and efficacy under investigation. In clinical trials, IDMC may recommend stopping the trial early due to safety, futility/efficacy, or both after reviewing observed data in the interim based on pre-specified stopping boundaries. In practice, the interim data is often too small to reach clinically meaningful differences with statistical significance (i.e., the observed clinically meaningful difference is reproducible and not purely by chance alone). To provide an overall assessment (or complete clinical picture) of the performance of the test treatment under investigation, the FDA (2023) published guidance on the benefit-risk assessment (BRA) framework to facilitate IDMC decision-making. Several methods have been studied in the literature following the FDA's recommended framework. However, these methods did not consider the uncertainties and heterogeneities. Alternatively, a BRA approach is proposed based on a confidence ellipse of primary safety and efficacy endpoints. The proposed confidence ellipse approach was evaluated both theoretically and via a clinical trial simulation. The results indicate that the proposed confidence ellipse provides consistent and stable metrics, particularly as sample sizes increase. The derived metrics of Benefit-Risk Difference (BRD) and Benefit-Risk Ratio (BRR) showed favorable performance across different scenarios and thresholds. Applied to the TESTING trial data (Lv et al. JAMA. 327(19):1888-98, 2022), our method confirmed and extended the original finding that a reduced methylprednisolone dose offered a more favorable benefit-risk profile. Specifically, the confidence ellipse method highlighted that the reduced dose consistently provided a better balance between efficacy and safety, particularly under stricter criteria for clinical significance. This method validated the original conclusions and provided additional insights into how different dosing regimens perform across various clinical scenarios, potentially offering a more refined tool for optimizing treatment decisions in complex therapeutic contexts.