Research Investigator Bristol Myers Squibb, China (People's Republic)
Disclosure(s):
Yizhen Guo: No financial relationships to disclose
Objectives: The exposure-response (E-R) based Clinical Utility Index (CUI) integrates E-R relationships across efficacy and safety endpoints to provide a balanced benefit-risk (B-R) assessment [1]. By utilizing longitudinal or time-to-event (TTE) E-R models, the CUI framework enables B-R assessment over time. We developed an R Shiny app for the CUI framework based on TTE E-R models to integrate key stakeholders' perspectives (e.g., clinicians, regulators) and inform dose optimalization in early clinical development.
Methods: In the CUI platform, an oncology asset profile was simulated. Progression-free survival (PFS) and time to the first Grade 3+ adverse event (Gr3+ AE) were considered the efficacy and safety endpoints. The TTE E-R relationship for both endpoints were described using Cox proportional hazards (CPH) model, where increased average concentration at steady-state (CAVG) was associated with a lower risk of disease progression and a higher risk of Gr3+ AE. The probability of each endpoint at various follow-up time cut-offs (e.g. 6 and 12 months) and CAVG level can then be predicted based on the CPH models. The utility function for each endpoint was defined to connect the probability of the endpoint with a utility score. The function consisted of two parameters: location parameter (μ) and shape parameter (β), allowing for both linear and S-shaped utility definitions. The CUI at each CAVG level was calculated by multiplying the corresponding utility scores together using weighted approach. The R Shiny app can interactively update the CUI assessment with changes in parameters, including weight, utility function (β and μ), and follow-up time cut-offs. The impact of these parameters on optimal dose selection among three dose levels (low, middle, high) was evaluated.
Results: Three scenarios were simulated. In the base scenario, equal weights were assigned to safety and efficacy, a linear utility relationship (μ = 0.5; β = 1) was assumed for both endpoints, and the follow-up time cut-off was at 6 months. In this scenario, the high dose was associated with the highest probability of subjects achieving the optimal exposure range (±30% of optimal exposure) and the max CUI. In the second scenario, greater weight was given to safety endpoint, and a significant drop in the utility score was assumed when probability of safety event exceeded 20% (μ = 0.2; β = 3). Under these conditions, low dose emerged as the optimal dose at 6 months. However, when the follow-up period was extended to 12 months, the middle dose was identified as the optimal dose. This shift underscores the importance of considering TTE E-R relationship in CUI framework. It is worth noting that the presented scenarios are conditional on the setting of CPH E-R models (e.g. baseline hazard and hazard ratio), which varies between different assets and may lead to different assessment result, but it is beyond the focus of this work.
Conclusions: The TTE E-R model-based CUI framework could be sensitive not only to the weighting and utility functions but also to the time window within which the optimal dose decision is made. The R Shiny application can facilitate cross-functional discussions on CUI assessment, thereby prospectively informing dose selection and study design considerations.
Citations: [1] Cheng Y, Chu S, Pu J, et al. Exposure-Response-Based Multiattribute Clinical Utility Score Framework to Facilitate Optimal Dose Selection for Oncology Drugs. J Clin Oncol. 2024;42(35):4145-4152.
