(I-566) Distal Junctional Angle and Pelvic Incidence are Significant Predictors of Validated Pain Outcomes Following Fusion for Cervical Myelopathy: A Machine Learning Analysis

Introduction: Cervical laminectomy and fusion is a common operation that has been shown to alter the natural history of cervical spondylotic myelopathy. However, a significant proportion of these patients have neck pain both before and after surgical intervention, which negatively impacts patient reported outcome metrics (PROMs). At present, the variables driving reductions in the visual analog scale (VAS) after surgery have not been well delineated.

Methods: All patients undergoing decompression and C2-T2 instrumented fusion were identified at a major, quaternary care center between 2013 and 2020. An array of demographic and imaging-based variables were extracted. These included age, sex, pelvic incidence (PI), fracture risk assessment tool (FRAX), distal junctional (T2-T4) angle (DJA), thoracic kyphosis, sagittal vertical axis (SVA), and an array of PROMs. Gradient boosting machine learning models were used to construct predictive, patient-level models for achieving >50% VAS reduction at >3 months after surgery. Shapley values were then computed in order to delineate the relative contributions by each variable to the overall model.

Results: A total of 175 patients were included in the series, with a mean follow-up of 275 (SEM +/- 17 days). Gradient boosting and Random Forest models had outstanding predictive capacity in identifying patients with >50% improvement in VAS at long-term follow-up (mean C-statistic=0.98, and 0.96, respectively). The most important variables contributing to these models were PI, DJA (T2-T4 angle), inflammatory arthritis, and FRAX. Increased PI, the presence of an inflammatory arthritis, and lower DJA were strong, negative predictors that patients would experience a reduction in VAS. Higher FRAX scores were associated with increased likelihood of experiencing improvement in pain; however, these patients had greater VAS scores at baseline.

Conclusion : Loss of thoracic kyphosis adjacent to the construct and increased PI were significantly predictive of persistent pain at long-term follow-up. Further studies are needed to better identify the relationship between construct design, cervicothoracic alignment, and pain after these operations.

How to Improve Patient Care: -Loss of thoracic kyphosis from T2-T4 is associated with worse PROMs at long-term follow-up and can be considered along with T1 slope in determining where to end the construct (T1 vs. T2 vs. T3, etc.)