DeepTDM: Deep Learning-Based Prediction of Sequential Therapeutic Drug Monitoring Levels of Vancomycin

Area of Science:

• Pharmacology
• Artificial Intelligence
• Critical Care Medicine

Background:

• Therapeutic drug monitoring (TDM) is crucial for optimizing vancomycin dosing in critically ill patients.
• Traditional pharmacokinetic (PK) models and Bayesian methods struggle with the complex, dynamic physiology of these patients.
• Accurate vancomycin level prediction is vital to ensure efficacy and minimize toxicity.

Purpose of the Study:

• To develop a novel deep learning model, GointMLP, for predicting sequential vancomycin TDM levels in intensive care unit (ICU) patients.
• To create DeepTDM, a decision support system for real-time vancomycin dose optimization.
• To enhance clinical decision-making and improve patient outcomes through precise vancomycin management.

Main Methods:

• Utilized a gated recurrent unit (GRU)-integrated joint multilayer perceptron network (GointMLP) model.
• Incorporated regression and classification predictions for dual verification of predicted values.
• Developed the DeepTDM system for real-time vancomycin dose adjustments.

Main Results:

• The GointMLP model demonstrated superior prediction accuracy compared to traditional PK models and other machine learning/deep learning approaches.
• The model's generalizability was validated across local cohorts and the diverse MIMIC-IV dataset.
• The GointMLP system offers robust performance in predicting vancomycin TDM levels.

Conclusions:

• The GointMLP model effectively addresses limitations of current vancomycin TDM methodologies.
• DeepTDM facilitates individualized vancomycin dosing, improving precision and clinical workflows.
• Integration of DeepTDM into practice is expected to significantly enhance patient outcomes in vancomycin therapy.