Background:
An accurate knowledge of a patient’s risk of cord-level intraoperative neuromonitoring (IONM) data loss is important for an informed decision-making process prior to deformity correction, but no prediction tool currently exists.
Methods:
A total of 1,106 patients with spinal deformity and 205 perioperative variables were included. A stepwise machine-learning (ML) approach using random forest (RF) analysis and multivariable logistic regression was performed. Patients were randomly allocated to training (75% of patients) and testing (25% of patients) groups. Feature score weights were derived by rounding up the regression coefficients from the multivariable logistic regression model. Variables in the final scoring calculator were automatically selected through the ML process to optimize predictive performance.
Results:
Eight features were included in the scoring system: sagittal deformity angular ratio (sDAR) of ≥15 (score = 2), type-3 spinal cord shape (score = 2), conus level below L2 (score = 2), cervical upper instrumented vertebra (score = 2), preoperative upright largest thoracic Cobb angle of ≥75° (score = 2), preoperative lower-extremity motor deficit (score = 2), preoperative upright largest thoracic kyphosis of ≥80° (score = 1), and total deformity angular ratio (tDAR) of ≥25 (score = 1). Higher cumulative scores were associated with increased rates of cord-level IONM data loss: patients with a cumulative score of ≤2 had a cord-level IONM data loss rate of 0.9%, whereas those with a score of ≥7 had a loss rate of 86%. When evaluated in the testing group, the scoring system achieved an accuracy of 93%, a sensitivity of 75%, a specificity of 94%, and an AUC (area under the receiver operating characteristic curve) of 0.898.
Conclusions:
This is the first study to provide an ML-derived preoperative scoring system that predicts cord-level IONM data loss during pediatric and adult spinal deformity surgery with >90% accuracy.
Level of Evidence:
Prognostic Level III. See Instructions for Authors for a complete description of levels of evidence.
