Bayesian Dosing Simulator ( BDS ): A Pharmacokinetic Modeling Tool for Optimized Antibiotic Therapy

BACKGROUND: Meropenem is widely used to treat hospital-acquired pneumonia in critically ill patients, with efficacy dependent on the time that free concentrations exceed the minimum inhibitory concentration (fT > MIC). In practice, single-sample therapeutic drug monitoring (TDM) may not ensure target attainment, particularly in patients requiring continuous renal replacement (CRRT). Model-informed precision dosing (MIPD) enables individualized, real-time adjustments, but implementation is limited. Herein, we developed a Shiny application for Bayesian meropenem dose optimization. METHODS: A previously published nonparametric model was translated into a parametric Bayesian framework with maximum a posteriori (MAP) updating implemented through a Shiny interface. Plasma pharmacokinetic (PK) data from critically ill patients served for external validation. Posterior predictions were benchmarked against population and individual (MAP) outputs from Pmetrics 3.09 and population median profiles from 1000 Monte Carlo simulations per patient. Predictive performance was assessed using relative median prediction error (rMPE) and relative median absolute prediction error (rMAPE), for the combined cohort overall and stratified by CRRT status. Sparse sampling scenarios (one, two, and three observations per patient) were also evaluated. RESULTS: Eighteen patients were evaluated (non-CRRT n = 13; CRRT n = 5). In Pmetrics, rMPE ranged from -7.7% to 9.4% and rMAPE from 18.5% to 25.5% across renal replacement strata and prediction types. Simulation-based population predictions yielded greater error, with rMPE/rMAPE of -25%/34% in CRRT patients and -21%/30% in non-CRRT patients. Across Bayesian Dosing Simulator (BDS) sampling scenarios, prediction error remained within prespecified limits (rMPE ±20%, rMAPE ≤ 30%), with F20 ranging from 80% to 94% and F30 from 87% to 94%. CONCLUSION: A validated nonparametric meropenem model was successfully implemented within an open-source Bayesian framework yielding predictive accuracy comparable to the reference model with robust performance under sparse sampling, supporting its feasibility for individualized meropenem dosing. Prospective evaluation of its clinical safety and effectiveness is needed.