Improving Vancomycin Therapeutic Drug Monitoring With a Deep Learning-Based Two-Compartment Predictive Model:...

Area of Science:

• Pharmacology
• Pharmacokinetics
• Machine Learning

Background:

• Vancomycin requires therapeutic drug monitoring (TDM) for optimal dosing.
• Current 1-compartment models (PKRNN-1CM) use time-series data but may be less accurate than 2-compartment models.
• Previous research suggests 2-compartment models offer superior pharmacokinetic insights.

Purpose of the Study:

• Introduce the pharmacokinetic recurrent neural network-2 compartment model (PKRNN-2CM).
• Enhance vancomycin TDM by integrating a 2-compartment pharmacokinetic framework.
• Improve individualized vancomycin dosage prediction using deep learning.

Main Methods:

• Developed PKRNN-2CM, combining recurrent neural networks with a 2-compartment PK model.
• Trained and evaluated the model using simulated and real-world electronic health record data.
• Compared PKRNN-2CM performance against the established PKRNN-1CM.

Main Results:

• PKRNN-2CM significantly outperformed PKRNN-1CM in predicting vancomycin concentrations on simulated data (RMSE 3.04 vs 4.50).
• Real-world data from 5483 patients showed PKRNN-2CM's superiority (RMSE 5.55 vs 5.65; P=.01).
• PKRNN-2CM provided a more accurate estimate of the area under the concentration-time curve to minimum inhibitory concentration ratio.

Conclusions:

• PKRNN-2CM represents a significant advancement in vancomycin TDM.
• The model demonstrates enhanced accuracy and performance over 1-compartment deep learning approaches.
• PKRNN-2CM holds promise for optimizing individualized vancomycin dosing and has potential for other clinical applications.