Artificial Intelligence and Predictive Modeling in Spinal Oncology: A Narrative Review

Area of Science:

• Spinal oncology
• Medical artificial intelligence
• Predictive modeling

Background:

• Artificial intelligence (AI) significantly impacts medicine, particularly spinal oncology, by enhancing patient prognosis for treatment strategies.
• Physician survival predictions are often inaccurate, driving the development of AI-driven predictive models.
• Interpreting and assessing the quality of these complex AI models present significant challenges.

Purpose of the Study:

• To review the stages and challenges in developing AI predictive models in spinal oncology.
• To use the Skeletal Oncology Research Group machine learning algorithms as a case study.

Main Methods:

• A narrative review of relevant scientific literature was conducted.
• The review focused on the methodology of developing and validating predictive models.

Main Results:

• Predictive model development involves six stages: preparation, development, internal validation, presentation, external validation, and implementation.
• Key validation measures include calibration, discrimination, decision curve analysis, and Brier score.
• Many current predictive models lack rigorous validation and transparent reporting, hindering their clinical utility.

Conclusions:

• AI is revolutionizing spinal oncology, but improvements are needed in model validation, reporting, and implementation.
• Understanding how to assess AI model quality is crucial for effective clinical integration and patient care.