Pelvic Incidence-Dependent Clustering of Sagittal Spinal Alignment in Asymptomatic Middle-Aged and Elderly Adults: A Machine Learning Approach

Comprehensive Summary

This study aims to develop and validate a partially automated spine classification technique in middle-aged and elderly adult populations. To conduct this, 635 asymptomatic adults were enrolled for morphologic stratification, along with a group of 103 adult spinal deformity (ASD) patients for validation. Machine learning (ML) algorithms then derived sagittal spinal morphology clusters and pelvic incidence-based correction criteria to identify a realignment strategy. The derived correction criteria were retrospectively applied to the ASD cohort, and outcomes were compared between patients whose postoperative alignment met versus did not meet the pelvic incidence–based targets. As a result, the surgical correction criteria computed by the algorithm obtained a significantly lower incidence of postoperative mechanical complications, unplanned reoperation, unplanned readmission, and superior patient-reported outcomes during follow-up. It is important to note that each subject was of Chinese heritage, and the cohorts were overwhelmingly made up of females (76%), which may limit generalizability.

Outcomes and Implications

Post-surgical outcomes from restoring sagittal alignment are subject to high variation, particularly due to an inefficient method for determining realignment strategies. By streamlining and partially automating this process, it should be possible to reduce the cost of treatment, the time spent during preoperative planning, the clinical variation, and increase clinical efficiency. Although the authors do not comment on a timeline for clinical implementation, they do state this process has “the potential to better guide spinal realignment surgery and facilitate individualized interventions.” It can be assumed that further validation with a lower gender disparity and a more even ethnic spread is necessary.