CASCADE: Context-Aware Data-Driven AI for Streamlined Multidisciplinary Tumor Board Recommendations in Oncology.

Area of Science:

• Oncology
• Medical Informatics
• Artificial Intelligence

Background:

• Hepatocellular carcinoma (HCC) treatment decisions are complex, often requiring multidisciplinary tumor board (MTB) input.
• Existing therapeutic guidelines may not capture the nuances of individual patient cases for optimal HCC management.
• The integration of advanced computational methods like machine learning (ML) offers potential for improved treatment prediction.

Purpose of the Study:

• To evaluate the efficacy of a machine learning algorithm in predicting treatment recommendations for HCC patients.
• To compare the predictive accuracy of the ML model against established clinical guidelines (ESMO, NCCN).

Main Methods:

• Retrospective analysis of 140 HCC patients discussed at an IRB-approved multidisciplinary tumor board.
• Extraction of clinical and imaging variables for input into a gradient-boosting machine learning algorithm (XGBoost).
• Performance assessment using confusion matrix metrics and Receiver Operating Characteristic (ROC) curve analysis.

Main Results:

• The XGBoost ML model demonstrated predictive capability for all eight treatment recommendations made by the MTB.
• The ML model's predictions showed higher accuracy compared to recommendations derived from ESMO and NCCN guidelines.
• The model successfully integrated diverse clinical and imaging data for predictive modeling.

Conclusions:

• A machine learning model utilizing clinical and imaging data can reliably predict expert-recommended HCC treatments.
• This AI-driven approach shows promise in supporting clinical decision-making, particularly in resource-limited settings.
• ML models offer a valuable adjunct to traditional guidelines for personalized HCC treatment planning.