Personalized venlafaxine dose prediction using artificial intelligence technology: a retrospective analysis based on...

Area of Science:

• Pharmacogenomics and Precision Medicine
• Artificial Intelligence in Healthcare
• Clinical Pharmacy and Pharmacology

Background:

• Venlafaxine dosing varies significantly among patients, necessitating individualized treatment strategies.
• Optimizing venlafaxine dosage is essential for maximizing therapeutic efficacy and minimizing adverse effects.

Purpose of the Study:

• To identify key factors influencing venlafaxine dosage requirements using real-world data.
• To develop and validate an artificial intelligence-based model for predicting personalized venlafaxine doses.

Main Methods:

• Retrospective analysis of depression patients treated with venlafaxine.
• Comparison of seven machine learning models (XGBoost, LightGBM, CatBoost, GBDT, ANN, TabNet, DT) for dose prediction.
• Validation using confusion matrices and Receiver Operating Characteristic (ROC) analysis.

Main Results:

• The TabNet model achieved the highest prediction accuracy (0.80).
• Seven significant variables identified: blood venlafaxine concentration, total protein, lymphocytes, age, globulin, cholinesterase, and blood platelet count.
• High Area Under the Curve (AUC) values for predicting 75 mg (0.90), 150 mg (0.85), and 225 mg (0.90) doses.

Conclusions:

• A robust TabNet model for venlafaxine dose prediction was successfully developed using real-world data.
• The model demonstrates high accuracy, supporting personalized venlafaxine dosing regimens.
• Findings offer valuable clinical guidance for optimizing venlafaxine treatment.