Who gets better after surgery for degenerative cervical myelopathy? A responder analysis from the multicenter Canadian spine outcomes and research network
Husain Shakil,Nicolas Dea,Armaan K Malhotra,Ahmad Essa,W Bradley Jacobs,David W Cadotte,Jérôme Paquet,Michael H Weber,Philippe Phan,Christopher S Bailey,Sean D Christie,Najmedden Attabib,Neil Manson,Jay Toor,Andrew Nataraj,Hamilton Hall,Greg McIntosh,Charles G Fisher,Y Raja Rampersaud,Nathan Evaniew,Jefferson R Wilson
DOI: https://doi.org/10.1016/j.spinee.2024.09.033
2024-10-17
Abstract:Background context: Degenerative cervical myelopathy (DCM) is the most common cause of acquired nontraumatic spinal cord injury worldwide. Surgery is a common treatment for DCM; however, outcomes often vary across patients. Purpose: To inform preoperative education and counseling, we performed a responder analysis to identify factors associated with treatment response. Study design/setting: An observational cohort study was conducted utilizing prospectively collected data from the Canadian Spine Outcomes Research Network (CSORN) registry collected between 2015-2022. Patient sample: We included all surgically treated DCM patients with complete 12-month follow-up and patient-reported outcomes (PROs) available at 1-year. Outcome measures: Treatment response was measured using the minimal clinically important difference (MCID) in PROs including the Neck Disability Index (NDI) and EuroQol-5D (EQ-5D) at 12 months postsurgery. Methods: A Least Absolute Shrinkage and Selection Operator (LASSO) machine learning model was used to identify significant associations between 14 preoperative patient factors and likelihood of treatment response measured by achievement of the MCID in NDI, and EQ-5D. Variable importance was measured using standardized coefficients. To test robustness of findings we trained a separate XGBOOST model, with variable importance measured using SHAP values. Results: Among the 554 DCM patients included, 229 (41.3%) and 330 (59.6%) patients responded to treatment by meeting or surpassing MCID thresholds for NDI and EQ-5D at 1-year, respectively. LASSO regression for likelihood of treatment response measured through NDI found the variable importance rank order to be baseline NDI (OR 1.06 per 1 point increase; 95% CI 1.04-1.07), then symptom duration (OR 0.65; 95% CI 0.44-0.97). For EQ-5D, the variable importance rank order was baseline EQ-5D (OR 0.16 per 0.1-point increase; 95% CI 0.03-0.78), living independently (OR 2.17; 95% CI 1.22-3.85), symptom duration (OR 0.62; 95% CI 0.40-0.97), then number of levels affected (OR 0.80 per additional level; 95% CI 0.67-0.96). A separate XGBoost model of treatment response measured through NDI, corroborated findings that patients with higher baseline NDI, and shorter symptom duration were more likely to respond to treatment, and additionally found older patients, and those with kyphosis on baseline upright X-ray were less likely to respond. Similarly, an XGBoost model for treatment response measured through EQ-5D corroborated findings that patients with higher baseline EQ-5D, shorter symptom duration, living independently, with fewer affected levels were more likely to respond to treatment, and additionally found older patients were less likely to respond. Conclusions: Our findings suggest patients with shorter symptom duration, higher baseline patient NDI, lower EQ-5D, younger age, living independently, without kyphosis on preoperative X-ray, and fewer affected levels are more likely to respond to treatment. Timing of surgery with respect to patient symptoms is underscored as a crucial and modifiable patient factor associated with improved surgical outcomes in DCM.