Machine Learning-Based Survival Prediction Tool for Adrenocortical Carcinoma

Area of Science:

• Oncology
• Medical Informatics
• Biostatistics

Background:

• Adrenocortical carcinoma (ACC) is a rare and aggressive cancer with unpredictable outcomes.
• The S-GRAS score, combining clinical and histopathological factors, offers good prognostic value for ACC patients.

Purpose of the Study:

• To enhance prognostic classification for ACC by developing advanced machine learning (ML) models.
• To create an individualized risk prediction tool for ACC patients.

Main Methods:

• Developed and validated ML models using a large training cohort (n=942) and an independent validation cohort (n=152).
• Constructed sixteen ML models based on individual clinical variables from the S-GRAS dataset.
• Developed a web-based tool for accessible, individualized risk prediction.

Main Results:

• Top-performing ML models (Quadratic Discriminant Analysis, Light Gradient Boosting Machine, AdaBoost Classifier) accurately predicted 5-year overall mortality and 1- and 3-year disease progression.
• Achieved high F1 scores in both training and validation cohorts for key prognostic endpoints.
• The developed web tool provides instant risk estimation for mortality and disease progression.

Conclusions:

• S-GRAS parameters are effective for predicting ACC outcomes when utilized with robust ML models.
• The accessible web application facilitates personalized management decisions for ACC patients by providing instant risk assessments.