Assistant Professor Columbia University New York, New York, United States
Introduction: Cervical spondylotic myelopathy (CSM) is the most common type of spinal cord disease in adults. Identifying patients at risk for prolonged postoperative length of stay (LOS) can help optimize perioperative care and resource utilization. This study aims to develop predictive models for prolonged LOS after CSM surgery.
Methods: Data was obtained from the Quality Outcomes Database CSM dataset, a prospective registry of 1141 patients from 14 sites. Prolonged LOS was defined as 3 days or more (75th percentile). Patients with missing data were excluded. Remaining patients were partitioned into training (n=728) and test sets (n=182). Logistic Regression, XGBoost, and Random Forest models were trained to predict prolonged LOS after surgery for CSM.
Results: Overall, 910 patients met inclusion criteria with mean age of 60.5 ± 11.6 years, 47.5% female, mean BMI 30.0 ± 6.4, and mean ASA grade 2.5 ± 0.6. Logistic Regression, XGBoost, and Random Forest models demonstrated excellent performance with mean AUROCs of 0.868 (±0.028), 0.855 (±0.027), and 0.864 (±0.026), respectively. Significant predictors of prolonged LOS were older age (OR 1.03, 95% CI: 1.03–1.04, p=0.041), past shoulder surgery (OR 2.86, 95% CI: 2.54-3.17, p=0.043), and more fused levels (OR 1.92, 95% CI: 1.86-1.97, p< 0.001). In contrast, radicular motor deficits (OR 0.32, 95% CI: 0.28–0.37, p=0.004) and anterior surgical approach (OR 0.07, 95% CI: 0.06–0.08, p< 0.001) were associated with shorter hospital stays.
Conclusion : In this large CSM registry cohort, prior shoulder surgery emerged as a key predictor of prolonged LOS, along with older age and more extensive, posterior fusions. Prior shoulder surgery may lead to functional impairment and increased pain from surgical positioning and taping, which may in turn impair postoperative mobility, delay rehabilitation, and increase the risk of complications. These findings suggest that patients with prior shoulder surgery may be considered a distinct entity with regards to potential resource utilization.