Developing a machine learning model for predicting venlafaxine active moiety concentration: a retrospective study...

Area of Science:

• Pharmacology
• Data Science
• Clinical Medicine

Background:

• Venlafaxine is a common antidepressant requiring therapeutic drug monitoring.
• Predicting venlafaxine concentration is crucial for optimizing treatment efficacy and minimizing adverse effects.

Purpose of the Study:

• To develop a predictive model for venlafaxine concentration using real-world evidence.
• To leverage machine learning and deep learning techniques for accurate venlafaxine level prediction.

Main Methods:

• Utilized real-world data from 330 patients treated with venlafaxine.
• Identified key predictors including venlafaxine dose, sex, age, hyperlipidemia, and adenosine deaminase.
• Assessed nine machine learning algorithms, selecting eXtreme Gradient Boosting (XGBoost) for the final model.

Main Results:

• The XGBoost model achieved an R-squared of 0.65, predicting venlafaxine concentration.
• Prediction accuracy within ±30% of actual concentration was 73.49% in the testing cohort.
• Subgroup analysis showed 69.39% accuracy within ±30% of the therapeutic range.

Conclusions:

• An XGBoost model effectively predicts blood venlafaxine concentration using real-world data.
• This model can aid clinicians in adjusting venlafaxine regimens for improved patient management.