Data - Knowledge driven machine learning model for cancer pain medication decisions.

Area of Science:

• Oncology
• Medical Informatics
• Machine Learning

Background:

• Cancer pain is a prevalent and challenging symptom in cancer patients.
• Optimizing drug selection for cancer pain management presents significant difficulties for healthcare providers.

Purpose of the Study:

• To develop and validate machine learning (ML) models for supporting drug decision-making in cancer pain management.
• Leverage real-world clinical data and prior knowledge to enhance medication selection for cancer pain.

Main Methods:

• Two ML models were developed using clinical records: one for new pain and one for inadequate pain control.
• Decision Tree and Gradient Boosting algorithms were employed, with 10,317 records for training and 1,000 for external validation.
• Model performance was assessed using accuracy, Area Under the Curve (AUC), and Brier score.

Main Results:

• The models achieved high performance, with average accuracies of 98.47% and 94.74% and AUCs of 99.62% and 94.74% respectively.
• External validation demonstrated strong results with accuracies of 97.4% and 93.1%, and AUCs of 99.83% and 97.01%.

Conclusions:

• The developed ML models can function as valuable decision support tools for healthcare professionals.
• These tools can assist physicians in making optimized medication decisions, particularly when pharmacists are unavailable.