Kristof De Vos 2, Martine Neyens 1, Nele Goeyvaerts 1, Vincent Duval 1, Yuki Iwaki 1, Juan-Jose Perez-Ruixo 1, Muriel Boulton 1
1 Johnson & Johnson (, ), 2 Johnson & Johnson, at KU Leuven at the time of the work (, )
Introduction: As outlined by FDA guidance [1], Exposure-Response (E-R) analyses can support dosage selection and adjustments. When a single response is obtained per subject, it is recommended to represent the exposure by a simplified metric such as trough drug concentration (Ctrough), maximum drug concentration (Cmax), or area under the drug concentration-time curve (AUC). Authors [2-7] reported that the outcome of E-R analyses may depend on the selected exposure metrics.
Objectives: To assess the impact of (i) the choice of the exposure metrics, and (ii) a confounder affecting both the PK and the response, on the E-R analysis outcome.
Methods: A modeling platform was set-up to simulate clinical trials, and to analyze the simulated data with E-R analysis. It included the following components:
1. A monoclonal antibody, with its concentrations characterized by a 2-compartment linear population PK model. Patients’ disease condition at baseline was introduced as a confounder with a higher clearance and a lower exposure for more diseased patients.
2. The efficacy endpoint was defined as a binary variable expressing a clinical response over the treatment phase. A parametric time-to-event approach (assuming a Gompertz hazard function) was applied to simulate the probability of experiencing the event at different time-points. Depending on the scenario, a drug (concentration-driven) effect (Emax model) and/or a confounding effect were added. A subject was considered a “responder” if at least one of the simulated probabilities was higher than a random value drawn from a uniform distribution, otherwise the subject was considered “censored”.
3. Per trial, 500 subjects were simulated and randomized to receive placebo or drug treatment in a 1:1 ratio, for a period of 12 cycles. A Q4W dosing regimen was considered and PK steady-state was reached at cycle 4. Full compliance to treatment without drop-out was assumed.
Following scenarios were evaluated:
1. Exposure-response relationship: (i) absence of drug effect (EMAX=0), or (ii) presence of shallow or steep drug effect (EMAX=10, HILL=1 or 10), increasing the baseline hazard of response.
2. Confounder effect: (i) absence of effect on both PK and response, or (ii) presence on both PK and response with effect varying from strong positive confounding (decreasing hazard) to strong negative confounding (increasing hazard).
For each scenario, 500 trials were simulated, and Cmax, Ctrough and AUC at cycle 1 and cycle 4 were calculated. E-R relationships were evaluated using 2 logistic regression models, including treatment arm and exposure-metric as independent variables, (i) with or (ii) without the confounder. A Wald test assessed the statistical significance of the exposure-metric effect.
Results: For scenarios without any drug effect, the type I error, i.e. the risk of false positive E-R relationship, was generally low (< 5%) for the different metrics and confounding effects, using both E-R models. Only with strong positive confounding, the risk of a false positive E-R relationship was higher if the confounder was not included as a prognostic factor in the logistic regression model (10-20%). For the scenario with a shallow drug effect and without confounding, the power, i.e. the probability to detect an exposure effect, was low (<30%). It generally increased when a steep drug effect was simulated. It was the highest for AUC, followed by Ctrough and Cmax. Cycle 4 metrics entailed higher power than cycle 1 metrics (power of 90% for AUC at cycle 4). Finally, for the scenarios with a steep drug effect and confounding effect, the logistic regression models without the confounder showed a clear effect on the power. When the confounder had an opposite impact on PK and response, the probability to detect an exposure effect decreased, while it generally increased when the effects went in the same direction. Including the confounder in the E-R model helped to detect an exposure effect in the presence of negative confounding. Conclusions: The power to detect an exposure effect was dependent on the selected exposure metric. Under the current simulation framework and assumptions, steady-state metrics entailed higher power versus cycle 1 metrics and AUC at cycle 4 was the best predictor. Correcting for potential confounding by including relevant covariates in the E-R models is recommended to better characterize the E-R effect. References: [1] FDA Guidance for Industry. Exposure-response relationships – study design, data analysis, and regulatory applications. April 2003. [2] Wiens MR, French JL, Rogers JA. Confounded exposure metrics. CPT Pharmacometrics Syst Pharmacol. 2024 Feb;13(2):187-191. doi: 10.1002/psp4.13074. Epub 2023 Nov 20. PMID: 37984457; PMCID: PMC10864924. [3] Largajolli A. Considerations when deriving time-averaged exposure for censored subjects for logistic regression exposure-response analyses. PAGE 2023. [4] Dai HI, Vugmeyster Y, Mangal N. Characterizing Exposure-Response Relationship for Therapeutic Monoclonal Antibodies in Immuno-Oncology and Beyond: Challenges, Perspectives, and Prospects. Clin Pharmacol Ther. 2020 Dec;108(6):1156-1170. doi: 10.1002/cpt.1953. Epub 2020 Aug 2. PMID: 32557643; PMCID: PMC7689749. [5] Ruiz-Garcia A, Baverel P, Bottino D, Dolton M, Feng Y, González-García I, Kim J, Robey S, Singh I, Turner D, Wu SP, Yin D, Zhou D, Zhu H, Bonate P. A comprehensive regulatory and industry review of modeling and simulation practices in oncology clinical drug development. J Pharmacokinet Pharmacodyn. 2023 Jun;50(3):147-172. doi: 10.1007/s10928-023-09850-2. Epub 2023 Mar 4. PMID: 36870005; PMCID: PMC10169901. [6] Sancho-Araiz A. Does shrinkage on exposure metrics truly impair E-R analysis? PAGE 2024. [7] Hu C, Zhou H, Sharma A. Landmark and longitudinal exposure-response analyses in drug development. J Pharmacokinet Pharmacodyn. 2017 Oct;44(5):503-507. doi: 10.1007/s10928-017-9534-0. Epub 2017 Jul 20. PMID: 28730565.
Reference: PAGE 34 (2026) Abstr 11966 [www.page-meeting.org/?abstract=11966]
Poster: Methodology - Other topics