Machine learning can predict surgical indication: new clustering model from a large adult spine deformity database

Area of Science:

• Orthopedics
• Spine Surgery
• Medical Informatics

Background:

• Adult Spine Deformity (ASD) management is complex, lacking a definitive algorithm.
• Factors influencing management include quality of life, comorbidities, symptoms, spine geometry, surgical risk, and disability.
• Machine learning offers efficient analysis of complex patient data.

Purpose of the Study:

• To develop a machine-learning algorithm for predicting surgical intervention in ASD patients.
• To utilize baseline data for accurate management predictions.

Main Methods:

• Retrospective analysis of prospectively collected data from 1319 ASD patients.
• Clustering methods to group patients based on similar characteristics.
• Predictive modeling to identify key variables influencing surgical decisions.

Main Results:

• Three distinct patient clusters were identified based on age, spinal alignment, and pelvic incidence.
• Prediction error rates varied across clusters (20-37%).
• Patient-reported outcomes (ODI, SRS-22) and major curve Cobb angle were significant predictors of surgical indication across all clusters.

Conclusions:

• The study successfully identified three patient clusters within ASD.
• Key variables driving management decisions were determined for each cluster.
• The developed algorithm aids in predicting surgical needs for ASD patients.