Using Machine Learning to Predict Unplanned Hospital Utilization and Chemotherapy Management From Patient-Reported...

Area of Science:

• Oncology
• Health Informatics
• Machine Learning

Background:

• Adverse chemotherapy effects frequently lead to hospital admissions and treatment adjustments.
• Identifying predictors of unplanned hospital utilization is crucial for enhancing healthcare quality and patient well-being.

Purpose of the Study:

• To evaluate if patient-reported outcome measures (PROMs) improve the predictive performance of machine learning (ML) models.
• To predict hospital admissions, triage events, and chemotherapy modifications using ML models incorporating PROMs.

Main Methods:

• Utilized clinical trial data from 508 participants undergoing chemotherapy.
• Compared six feature sets (clinical data, PROMs, and combinations) across six ML models (logistic regression, decision tree, adaptive boosting, random forest, SVMs, neural network).
• Assessed model performance in predicting hospital admissions, triage events, and chemotherapy changes, considering data imbalance.

Main Results:

• PROMs significantly improved the prediction accuracy for all studied outcomes.
• Random Forest and SVM models demonstrated the highest performance for predicting admissions and chemotherapy changes in balanced datasets.
• Logistic Regression performed best on imbalanced datasets, and data balancing generally led to superior predictive performance.

Conclusions:

• Machine learning models applied to PROM data can effectively predict hospital utilization and chemotherapy management.
• This approach holds potential for improving healthcare planning and enabling personalized cancer treatment.
• The study highlights best practices in ML research by comparing methods for handling imbalanced data.