Background: Cancer pain is one of the most common symptoms in cancer patients, and drug decision-making in cancer pain management remains challenges. This study aims to develop machine learning models using real-world clinical data and prior knowledge to support drug decision-making in cancer pain management.
Methods: Clinical records from the Xiangya Hospital information system and a specialized cancer pain platform were used to develop two machine learning models: one for patients newly experiencing pain and one for patients with inadequate pain control. A total of 10,317 clinical records were used for model training, and 1,000 external records were obtained from the Cancer Hospital of the Chinese Academy of Medical Sciences for validation. Model performance was evaluated based on accuracy, AUC, and brier score.
Results: Decision Tree and Gradient Boosting algorithms were selected for the two models, achieving an average accuracy of 98.47% and 94.74%, respectively, with AUCs of 99.62% and 94.74%. External validation accuracy was 97.4% and 93.1%, respectively, with AUCs of 99.83% and 97.01%.
Conclusion: The models proposed in this study can serve as decision support tools for healthcare professionals, assisting physicians in making optimized medication decisions in the absence of pharmacists.
Keywords: Cancer pain treatment; Clinical decision support; Decision tree; Drug therapy; Machine learning.
Copyright © 2024 Elsevier B.V. All rights reserved.