Nephrocast-V: A Deep Learning Model for the Prediction of Vancomycin Trough Concentration Using Electronic Health...

Area of Science:

• Pharmacology and Computational Medicine
• Artificial Intelligence in Healthcare
• Critical Care Medicine

Background:

• Vancomycin is essential for treating serious gram-positive bacterial infections, including methicillin-resistant Staphylococcus aureus.
• Achieving and maintaining therapeutic vancomycin trough concentrations is clinically challenging, impacting treatment efficacy.
• Individualized dosing is crucial for optimizing vancomycin therapy in critically ill patients.

Purpose of the Study:

• To develop and validate a deep learning model for predicting vancomycin trough concentrations 2 days in advance.
• To assess the model's ability to recommend optimal vancomycin dosing adjustments.
• To support personalized therapeutic drug monitoring for vancomycin in intensive care unit (ICU) patients.

Main Methods:

• Utilized electronic health record (EHR) data from ICU patients (2016-2024).
• Engineered a deep learning model combining Long Short-Term Memory (LSTM) and Multi-Head Attention layers.
• Incorporated patient demographics, vitals, labs, medications, and dosing history as model features.

Main Results:

• The deep learning model achieved a Mean Absolute Error (MAE) of 3.15 mg/L and Root Mean Square Error (RMSE) of 4.17 mg/L.
• Model performance was comparable to that of a critical care pharmacist using Bayesian dosing software.
• Non-adherence to model-based dose recommendations correlated with non-therapeutic vancomycin levels.

Conclusions:

• Deep learning models show significant potential for individualizing vancomycin therapeutic drug monitoring.
• AI-driven predictions can support clinicians in optimizing vancomycin dosing strategies.
• This approach may enhance the effectiveness and safety of vancomycin treatment in critical care settings.