Yuchen Wang, PhD: No relevant disclosure to display
Objectives: Characterizing the exposure-response (E-R) relationship is essential for dose selection and benefit-risk assessment in clinical trials.[1] In special populations, accurate estimation of safety remains particularly important but often challenging due to limited data and high variability. In this simulation-based study, we compared the performance of three approaches – dose-exposure-response (D-E-R) analysis, dose-response (D-R) analysis, and cohort-level summary approach – in estimating safety outcomes in special populations.
Methods: We used simulations to demonstrate the difference in performance among the three approaches on the estimation of the estimands of interest. The simulations were based on results from two related studies reported in literature and the associated population PK (PopPK) and safety exposure-response models. [2,3] One of the two studies tested multiple doses, but most of the individuals were not from the population of interest. The other study included only individuals from the population of interest but had a small sample size and tested only one dose level. This data set and models were then used to simulate 1000 trials. We applied the three approaches to each simulated trial. In the D-E-R analysis, a log-linear model was used to describe D-E relationship, and a logistic regression model was used for the E-R relationship. D-R analysis utilized a logistic regression model to establish the D-R relationship. Both the D-E-R and D-R analyses included covariate effect in the model specification. The cohort-level summary approach used the observed incidence rates in each cohort as the estimates, ignoring patient type or any covariates. The true incidence rate at each dose level was calculated using the true PopPK model and the true E-R model, and it was used to assess the bias and mean squared error (MSE) in the estimates for each dose level in the population of interest.
Results: The cohort-level summary approach showed modest bias due to the small number of target population patients in each dose group and the limited demographic diversity in one of the studies. The estimates from the D-E-R and D-R analyses showed low bias in most dose groups, benefiting from the increased sample size through pooling the two studies and the incorporation of information from the other population. The lowest bias from the D-E-R and D-R was observed in the 200 mg group, which had the largest sample size. Slightly higher bias was seen in the 150 mg and 300 mg groups, with only 2 and 1 subjects included, respectively. The estimates from D-R exhibited MSE values up to three times higher than those from D-E-R, and in some cases, even exceeded that, highlighting a substantial difference in estimation precision between the two approaches. D-E-R outperformed D-R due to its ability to incorporate exposure and better account for interindividual variability.
Conclusions: The cohort-level summary approach can be valuable in settings where model assumptions need to be minimized. Both D-R and D-E-R analyses were able to provide estimates for the special population with low bias. Notably, the estimates from D-E-R analysis showed lower MSE, which may translate into increased power to detect covariate effects on the outcome.
Citations: [1] Food and Drug Administration, 2003. Guidance for industry: exposure-response relationships-study design, data analysis, and regulatory applications. https://www.fda.gov/regulatory-information/search-fda-guidance-documents. [2] Fukae M, Baron K, Tachibana M, Mondick J, Shimizu T. Population pharmacokinetics of total and unbound valemetostat and platelet dynamics in healthy volunteers and patients with non-Hodgkin lymphoma. CPT Pharmacometrics Syst Pharmacol. 2024; 13: 1641-1654. doi:10.1002/psp4.13201 [3] Fukae M, Rogers J, Garcia R, Tachibana M, Shimizu T. Bayesian sparse regression for exposure–response analyses of efficacy and safety endpoints to justify the clinical dose of valemetostat for adult T-cell leukemia/lymphoma. CPT Pharmacometrics Syst Pharmacol. 2024; 13: 1655-1669. doi:10.1002/psp4.13203
Keywords: Special Population, Exposure-response, Safety
