A machine learning framework for precision dosing of daptomycin across diverse disease states
6 affiliations
- 1Department of Pharmacy, The Second Qilu Hospital of Shandong University, Jinan, China
- 2Department of Clinical Pharmacy, Key Laboratory of Chemical Biology (Ministry of Education), State Key Laboratory of Discovery and Utilization of Functional Components in Traditional Chinese Medicine, School of Pharmaceutical Sciences, Cheeloo College of Medicine, Institute of Clinical Pharmacology, Shandong University, Jinan, China
- 3Cheeloo College of Medicine, Shandong University, Jinan, China
- 4Department of Critical Care Medicine, Qilu Hospital of Shandong University, Jinan, China
- 5Center for Health Outcomes Research and Delivery Science, Children’s National Hospital, Washington, DC, USA
- 6Department of Pediatrics, George Washington University School of Medicine and Health Sciences, Washington, DC, USA
Abstract
The large pharmacokinetic variability of daptomycin across disease states poses a challenge for individualized dosing. Conventional population pharmacokinetic (PopPK) models are often limited to specific conditions, lacking generalizability. This study used a machine learning (ML) framework to predict daptomycin exposure and guide dose optimization across heterogeneous populations. Nine published PopPK models were used to generate 90,000 simulated profiles across patients with infections and/or patients treated with renal transplantation and/or renal replacement therapy. Seven ML algorithms were trained to predict the 24 h area under the concentration–time curve (AUC0–24). Model performance was evaluated using statistical metrics and clinical scenario assessments, and comparison with PopPK models. External validation was conducted with real-world data. CatBoost demonstrated the highest predictive accuracy, identifying dose, creatinine clearance, disease state, and weight as key predictors. The validation cohort comprised 20 patients with infections (501 concentration measurements) and 19 patients receiving renal replacement therapy (513 concentration measurements). In infected patients, the prediction error of AUC0–24 was reduced by 60.1% compared with existing PopPK models (root mean square error [RMSE], 141 vs 353 mg·h/L; mean absolute prediction error, 23.5% vs 58.1%). The ML model correctly classified 75.0% of patients into the appropriate clinical AUC categories (<666, 666–939, and >939 mg·h/L), outperforming PopPK models (average overall accuracy: 30.0%). In patients undergoing renal replacement therapy, ML also improved performance (RMSE, 197 vs 218 mg·h/L; overall accuracy, 52.6% vs 31.6%). The ML model demonstrated accurate prediction of steady-state daptomycin AUC in heterogeneous clinical populations and was successfully validated using real-world data. By enabling accurate pre-treatment exposure prediction, it provides a flexible and clinically useful pathway toward precision dosing of daptomycin.
Related
- A safety-first machine learning framework for precision public health: segmenting vaccine hesitancy among Iranian adults
- Precision targeting using machine learning to optimize medication therapy management delivery in value-based care: A quasi-experimental study using real-world evidence
- Machine Learning-Augmented Precision Dosing: Integrating Pharmacokinetic Modeling With Clinical Decision Support.
- What determined match outcomes at the 2026 FIFA World Cup? An explainable machine learning analysis of penalty area precision and composure under pressure
- An integrated evolutionary optimization and machine learning framework for predicting pharmacokinetics of nanocarrier-encapsulated drugs
BibTeX
@article{zhang2026machine,
title = {{A machine learning framework for precision dosing of daptomycin across diverse disease states}},
author = {Xin-Fang Zhang and Jing Sun and Guo-Xiang Hao and Xiao-Mei Chen and Yue Zhou and Xiao-Jie Wang and John van den Anker and Wei Zhao and De-Qing Sun and Bo-Hao Tang},
journal = {Antimicrobial Agents and Chemotherapy},
year = {2026},
month = oct,
doi = {10.1128/aac.01089-26},
url = {https://doi.org/10.1128/aac.01089-26},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - A machine learning framework for precision dosing of daptomycin across diverse disease states
AU - Xin-Fang Zhang
AU - Jing Sun
AU - Guo-Xiang Hao
AU - Xiao-Mei Chen
AU - Yue Zhou
AU - Xiao-Jie Wang
AU - John van den Anker
AU - Wei Zhao
AU - De-Qing Sun
AU - Bo-Hao Tang
PY - 2026
DA - 2026/10/14
JO - Antimicrobial Agents and Chemotherapy
DO - 10.1128/aac.01089-26
UR - https://doi.org/10.1128/aac.01089-26
AB - The large pharmacokinetic variability of daptomycin across disease states poses a challenge for individualized dosing. Conventional population pharmacokinetic (PopPK) models are often limited to specific conditions, lacking generalizability. This study used a machine learning (ML) framework to predict daptomycin exposure and guide dose optimization across heterogeneous populations. Nine published PopPK models were used to generate 90,000 simulated profiles across patients with infections and/or patients treated with renal transplantation and/or renal replacement therapy. Seven ML algorithms were trained to predict the 24 h area under the concentration–time curve (AUC0–24). Model performance was evaluated using statistical metrics and clinical scenario assessments, and comparison with PopPK models. External validation was conducted with real-world data. CatBoost demonstrated the highest predictive accuracy, identifying dose, creatinine clearance, disease state, and weight as key predictors. The validation cohort comprised 20 patients with infections (501 concentration measurements) and 19 patients receiving renal replacement therapy (513 concentration measurements). In infected patients, the prediction error of AUC0–24 was reduced by 60.1% compared with existing PopPK models (root mean square error [RMSE], 141 vs 353 mg·h/L; mean absolute prediction error, 23.5% vs 58.1%). The ML model correctly classified 75.0% of patients into the appropriate clinical AUC categories (<666, 666–939, and >939 mg·h/L), outperforming PopPK models (average overall accuracy: 30.0%). In patients undergoing renal replacement therapy, ML also improved performance (RMSE, 197 vs 218 mg·h/L; overall accuracy, 52.6% vs 31.6%). The ML model demonstrated accurate prediction of steady-state daptomycin AUC in heterogeneous clinical populations and was successfully validated using real-world data. By enabling accurate pre-treatment exposure prediction, it provides a flexible and clinically useful pathway toward precision dosing of daptomycin.
SN - 1098-6596
SN - 0066-4804
ER -