Assessing the predictive utility of preoperative canal diameter, foraminal stenosis, and cord-lamina angle for postoperative C5 palsy: a matched cohort validation study.
OBJECTIVE Postoperative C5 palsy is a debilitating complication of cervical spine surgery with unclear etiology. A previous study of 12 patients proposed a three-variable model to predict postoperative C5 palsy based on MRI measurements of the anteroposterior diameter of the spinal canal (APD), foraminal diameter (FD), and cord-lamina angle (CLA). The initial paper reported an area under the receiver operating characteristic (ROC) curve (AUC) of 0.97, with a sensitivity of 91% and specificity of 100%. This study aimed to validate the three-variable model and its individual components in a larger patient cohort. METHODS Adult patients who underwent anterior or posterior cervical discectomy and fusion at C4-5 with subsequent C5 palsy were retrospectively identified (2010-2023). Postoperative C5 palsy was defined as a new or increased deltoid/biceps weakness on manual motor testing. Patients with C5 palsy were propensity score-matched 1 to 3 to a control group based on age, sex, Charlson Comorbidity Index score, procedure type, and levels fused. A biostatistician used the APD, FD, and CLA MRI measurements to calculate ROC curves, sensitivities, and specificities for single- and three-variable predictive models of postoperative C5 palsy. RESULTS A total of 42 patients with C5 palsy were matched to 126 controls. There were no significant differences in preoperative APD (mean 9.20 vs 9.45 mm, p = 0.426), minimum FD (mean 1.82 vs 1.90 mm, p = 0.609), or maximum CLA (mean 38.8° vs 38.5°, p = 0.867) between the C5 palsy and control groups. The APD had an AUC of 0.455, sensitivity of 0.912, and specificity of 0.148. The minimum FD had an AUC of 0.501, sensitivity of 0.765, and specificity of 0.330. The maximum CLA had an AUC of 0.513, sensitivity of 0.765, and specificity of 0.348. When applied collectively, the three-variable predictive model performed similarly to chance, yielding an AUC of 0.519 (95% CI 0.410-0.628), sensitivity of 0.676, and specificity of 0.452. CONCLUSIONS In a large, well-matched validation cohort, the preoperative MRI parameters of the APD, FD, and CLA were not predictive of postoperative C5 palsy, both individually and as a three-variable model. The etiology of C5 palsy remains poorly understood and reliable preoperative predictors have yet to be identified. The previously proposed three-variable model demonstrated inconsistent predictive utility for postoperative C5 palsy, and the model should be interpreted with caution before more rigorous external validation.