Kenta Yoshida1, François Mercier2
1Genentech, 2Genentech
Introduction/Objectives: Exposure-response (ER) analysis is an essential component in clinical pharmacology to assess the relationship between drug exposure and therapeutic or adverse outcomes, which is critical for clinical drug development decisions such as dose selection or treatment optimization. Bayesian approaches, such as Markov Chain Monte Carlo (MCMC) algorithm, are particularly suited for ER analysis by enabling natural handling of uncertainty through posterior samples and informative clinical trial simulations, as well as implementing learn-and-confirm cycles. The newly developed R package, BayesERtools [1], is designed to facilitate the use of Bayesian approaches for ER modeling by providing a streamlined user interface to easily execute common tasks including model development, simulations, and plotting, together with a documentation of the workflow in `BayesERbook` [2]. Methods: BayesERtools currently supports four model types: linear and Emax models for continuous endpoints, and linear and Emax logistic regression models for binary endpoints. The `rstanarm` [3] and `rstanemax` [4] packages are used for the model development of linear and Emax models, respectively. The selection of exposure metrics is implemented by comparing models with the approximated leave-one-out (loo) cross-validation model performance from the `loo` [5] package. Selection of covariates in addition to an exposure metric is implemented by using projection predictive feature selection from the `projpred` [6] package (currently not available for Emax models). For simulation at new exposure levels or new population, in addition to standard simulations, the package provides functions to enable marginalized simulations, where the predicted responses are marginalized (or averaged) over the provided population, so that the average response for the population of interest is derived. Results: BayesERtools enabled streamlined workflow for ER modeling with Bayesian framework. Basic model development can be performed with `dev_ermod_*()` family of functions, where * can be `lin`, `emax`, `bin`, `bin_emax`, depending on the model types. Exposure metrics and covariate selections can be performed with `dev_ermod_*_exp_sel()` and `dev_ermod_*_cov_sel()` family of functions. Simulation for new exposure levels can be performed with `sim_er_new_exp()` and `sim_er_new_exp_marg()` functions for standard and marginalized simulations, respectively. Plotting of ER relationships can be performed with `plot_er()` with multiple options, such as including exposure range box plot at the bottom or showing credible intervals of the model coefficient for exposure metrics, with `plot_er_gof()` function using most common options enabled for goodness-of-fit plots. The packages implements other useful functionalities including visualization of the ranking of variable selections from the `dev_ermod_*_cov_sel()` step and generating forest plots for covariate effects. Typical workflow examples are presented in `BayesERbook` [2]. Conclusions: BayesERtools offers a user-friendly platform for conducting Bayesian ER analysis within the R environment. By enabling streamlined model development, evaluation, and simulations, the package can aid in quantitative clinical drug development decisions. References: [1] Yoshida K, Mercier F (2025). BayesERtools: Bayesian Exposure-Response Analysis Tools. R package version 0.2.1, https://genentech.github.io/BayesERtools/ [2] Yoshida K, Mercier F (2025). BayesERbook. Available at: https://genentech.github.io/BayesERbook/ [3] Goodrich B, Gabry J, Ali I & Brilleman S. (2024). rstanarm: Bayesian applied regression modeling via Stan. R package version 2.32.1, https://mc-stan.org/rstanarm. [4] Yoshida K, Navarro D (2024). rstanemax: Emax Model Analysis with ‘Stan’. R package version 0.1.7, https://yoshidk6.github.io/rstanemax/. [5] Vehtari A, Gabry J, Magnusson M, Yao Y, Bürkner P, Paananen T, Gelman A (2024). “loo: Efficient leave-one-out cross-validation and WAIC for Bayesian models.” R package version 2.8.0, https://mc-stan.org/loo/. [6] Piironen J, Paasiniemi M, Catalina A, Weber F, Vehtari A (2023). “projpred: Projection Predictive Feature Selection.” R package version 2.8.0, https://mc-stan.org/projpred/.
Reference: PAGE 33 (2025) Abstr 11761 [www.page-meeting.org/?abstract=11761]
Poster: Methodology - Other topics