Predicting adverse drug events in older inpatients: a machine learning study

Area of Science:

• Geriatric Medicine
• Pharmacovigilance
• Computational Health

Background:

• Adverse drug events (ADEs) pose a significant risk to older inpatients.
• Effective prediction models are crucial for preventing medication-related harm in this vulnerable population.

Purpose of the Study:

• To develop a machine learning-based prediction model for ADEs in older Chinese inpatients.
• To identify significant risk factors associated with ADEs in this demographic.

Main Methods:

• Retrospective analysis of 1800 older Chinese inpatients.
• Development and evaluation of seven machine learning models (XGBoost, AdaBoost, CatBoost, GBDT, LightGBM, TPOT, RF).
• Performance assessment using accuracy, precision, recall, F1 scores, and Area Under the Receiver Operating Characteristic Curve (AUC).

Main Results:

• 13.00% of patients experienced ADEs, with antineoplastic agents being the primary cause.
• The AdaBoost model demonstrated the best predictive performance (AUC 0.91, accuracy 88.06%).
• Ten significant risk factors for ADEs were identified, including drug interactions, length of stay, and patient demographics.

Conclusions:

• A novel machine learning model for ADE prediction in older patients was successfully established.
• The developed model can be implemented at the bedside to enhance clinical practice and patient safety.