A Machine Learning Algorithm for Predicting 6-Week Survival in Spinal Metastasis: An External Validation Study Using...

Area of Science:

• Oncology
• Machine Learning
• Biostatistics

Background:

• Existing survival prediction algorithms for spinal metastasis lack external validation for nonsurgical treatments.
• Advances in cancer care necessitate updated survival prediction models.
• The Skeletal Oncology Research Group machine learning algorithm (SORG-MLA) was developed for 6-week survival prediction.

Purpose of the Study:

• To externally validate the 6-week SORG-MLA for patients with spinal metastasis.
• To assess the model's consistency in survival predictions.
• To evaluate the algorithm's utility in a Taiwanese cohort.

Main Methods:

• Model performance was assessed using discrimination (AUC), calibration, Brier score, and decision curve analysis.
• Model consistency was evaluated by comparing 6-week, 3-month, and 1-year survival predictions.
• A Taiwanese cohort was used for external validation.

Main Results:

• The SORG-MLA demonstrated good discrimination (AUC 0.78) and prediction accuracy (Brier score 0.11).
• The model showed suboptimal consistency between 6-week and 90-day survival predictions (78%).
• Decision curve analysis indicated suitability for clinical use across all threshold probabilities.

Conclusions:

• The 6-week SORG-MLA is a validated tool for predicting survival in spinal metastasis patients.
• The algorithm supports informed decision-making for clinicians and patients.
• An online tool is available for practical application of the SORG-MLA.