Predicting overall survival in chordoma patients using machine learning models: a web-app application.

Area of Science:

• Oncology
• Medical Informatics
• Biostatistics

Background:

• Chordoma is a rare bone cancer with challenging prognosis.
• Accurate survival prediction is crucial for effective patient management and treatment planning.

Purpose of the Study:

• To evaluate the efficacy of machine learning (ML) survival models compared to the standard Cox proportional hazards (CoxPH) model for chordoma patient survival prediction.
• To identify the best-performing ML model for clinical application.

Main Methods:

• A population-based cohort study using the Surveillance, Epidemiology, and End Results database (2000-2018) of 724 chordoma patients.
• Development and validation of three ML survival models and one CoxPH model.
• Model performance assessed using concordance index (C-index), Brier score, ROC curves, and calibration curves for 5- and 10-year survival probabilities.

Main Results:

• ML models demonstrated superior performance compared to the CoxPH model.
• The DeepSurv ML model achieved the highest C-index (0.795) and superior discrimination (AUC 0.84 for 5-year, 0.88 for 10-year survival).
• DeepSurv showed strong calibration and effective risk stratification, with an implemented web application for clinical use.

Conclusions:

• Machine learning algorithms, especially DeepSurv, are effective for chordoma survival prediction.
• DeepSurv offers excellent discrimination and calibration, representing a valuable tool for clinical decision support in chordoma care.