A machine learning approach to predict daptomycin exposure from two concentrations based on Monte Carlo simulations.

Area of Science:

• Pharmacology and Computational Biology
• Clinical Pharmacokinetics
• Machine Learning in Medicine

Background:

• Daptomycin, a lipopeptide antibiotic, exhibits concentration-dependent activity.
• Accurate drug exposure estimation is crucial for optimizing daptomycin therapy.
• Machine learning (ML) models show promise in predicting drug exposure compared to traditional methods.

Purpose of the Study:

• To develop and validate an XGBoost ML model for predicting daptomycin's area under the curve (AUC).
• To estimate daptomycin AUC using limited blood samples (pre-dose and 1-hour post-dose) and patient covariates.
• To assess the feasibility of this ML approach for therapeutic drug monitoring (TDM).

Main Methods:

• Simulated 5150 patients using two pharmacokinetic models.
• Trained an XGBoost model on 75% of simulated data, predicting AUC from two daptomycin concentrations and covariates.
• Validated the model on a 25% test set and an independent simulation set, evaluating performance using root mean square error (RMSE).

Main Results:

• The XGBoost model accurately estimated daptomycin AUC using two concentration points and five covariates (sex, weight, dose, creatinine clearance, body temperature).
• Achieved low relative bias (0.43%) and RMSE (7.69%) in the test set.
• Demonstrated strong performance in the validation set with relative bias of 4.61% and RMSE of 6.63%.

Conclusions:

• The developed XGBoost ML model enables accurate daptomycin AUC estimation from sparse data.
• This approach can support clinical decision-making for daptomycin dose adjustments.
• Facilitates future therapeutic drug monitoring studies for daptomycin.