Objectives <p>This study aims to assess the applicability of published polymyxin B population pharmacokinetics (PopPK) models in critically ill patients using an external dataset, with the goal of establishing an evidence-based framework to guide optimal model selection for precision dosing in clinical critical care practice.</p> Methods <p>Nine published PopPK models of polymyxin B were reconstructed. A validation dataset was compiled from data of critically ill patients receiving intravenous polymyxin B. The predictive performance was assessed based on predicted and simulated diagnosis and Bayesian forecasting.</p> Results <p>The data from 132 samples of 72 critically ill patients were collected as the validation dataset. In prediction-based diagnostics, none of the nine PopPK models met the criteria. In simulation-based diagnostics, although no model achieved full adequacy in the tests, four models showed acceptable normality in normalized prediction distribution error plots. Bayesian forecasting significantly improved the predictive performance across all models, particularly when ≥ 2 prior observations were available. Collectively, one model demonstrated good prediction performance.</p> Conclusions <p>The integration of PopPK model with Bayesian forecasting provides a valuable approach for model-informed precision dosing of polymyxin B, facilitating the individualization of dosing regimens in critically ill patients.</p>