Document resource
Background and Importance β-lactam antibiotics are cornerstone agents in intensive care units, yet their pharmacokinetics show marked variability in critically ill patients, often resulting in suboptimal exposure. Population pharmacokinetic (popPK) models combined with Bayesian estimation can improve dosing precision; however, clinical implementation remains limited by the lack of user-friendly tools. Web-based applications may bridge this gap and support antimicrobial stewardship programs.Aim and Objectives To develop two interactive Shiny applications based on published popPK models for meropenem and piperacillin-tazobactam, and to validate their predictive performance before and after Bayesian adjustment using real-world ICU data.Material and Methods Two R Shiny applications, MeroDose-UCI and PiperDose-UCI, were developed to perform population simulations and Bayesian dose optimisation. Each integrates a published and externally validated popPK model (Boonpeng et al., 2022 for meropenem; Sukarnjanaset et al., 2019 for piperacillin-tazobactam). Retrospective validation was conducted using plasma concentration data from ICU patients. Predictive performance was assessed by comparing observed versus predicted concentrations for both population and individual (Bayesian) estimates. Accuracy and precision were evaluated using the coefficient of determination (R 2), mean absolute error (MAE) and mean relative error (MRE). Visual Predictive Checks (VPCs) were additionally performed to evaluate model predictive distribution against observed data.Results Thirty-six plasma samples (14 meropenem and 22 piperacillin) from 26 patients (13 per antibiotic) were included.Population model for meropenem: R2 = 0.80, MAE = 3.61 ± 5.38 mg/L, MRE = 38.6 ± 22.8%; after Bayesian adjustment, R2 = 0.95, MAE = 1.15 ± 1.80 mg/L, MRE = 11.5 ± 8.6%.Population model for piperacillin-tazobactam: R2 = 0.86, MAE = 9.31 ± 19.33 mg/L, MRE = 38.4 ± 19.5%; after Bayesian adjustment, R2 = 0.99, MAE = 2.21 ± 4.43 mg/L, MRE = 9.1 ± 5.0%.VPCs confirmed that both models adequately captured the central tendency and variability of observed concentrations.Conclusion and Relevance Two Shiny applications were successfully developed and validated for Bayesian-guided TDM of meropenem and piperacillin-tazobactam. Bayesian adjustment and VPC analysis confirmed the robustness and predictive reliability of the implemented models. These validated tools provide a practical and innovative solution to support pharmacists in optimising β-lactam therapy in critically ill patients.Conflict of Interest No conflict of interest