Abstract
Introduction
Cervical spondylotic myelopathy (CSM) is the commonest cause of spinal cord impairment in the elderly population worldwide. Though recent efforts have uncovered valuable clinical predictors of outcome in patients undergoing surgical decompression, the utility of MRI assessment in this regard remains equivocal. To address this fundamental knowledge gap, it is the objective of this study to quantitatively assess the role of MRI in predicting surgical outcome using multicenter prospective data.
Material and Methods
A total of 278 patients with at least one clinical sign of CSM were enrolled in AOSpine North American Study. Of these, baseline MRI data and modified Japanese Orthopedic Association score (mJOA) assessment at 6 months were available for 101 patients. MRIs were reviewed by three investigators for the location of pathology and for presence or absence ( ± ) of signal change on T1 and T2 imaging. Quantitative analysis of T2 hyperintensity area, sagittal extent, and signal change ratios was also conducted. In addition, spinal canal compromise and spinal cord compression were measured on T2 imaging. The mJOA score was used as the primary outcome measure and was dichotomized to discriminate between patients with mild myelopathy postoperatively (≥ 16) and those with substantial residual neurological impairment (< 16). Univariate analyses assessed the relationship of baseline mJOA and MRI analysis with outcome. Logistic regression modeling followed a conceptual division of variables into three key groups: T1 signal analysis, T2 signal analysis, and anatomical measurements. Inclusion of variables in the final model was based on practical, clinical, and statistical considerations (including Akaike information criterion, AIC; Bayesian information criterion, BIC; area under the receiver operator curve characteristics; AUC). The final model was compared with a model containing only baseline mJOA using a likelihood-ratio test.
Results
In univariate analysis, baseline mJOA (p < 0.0001), spinal canal compromise (p = 0.0322), T2 hyperintensity area (p = 0.0422), and maximum height (p = 0.026) were all significantly associated with outcome. A single variable was used to describe T1 hypointensity ( ± ) and anatomical measurements (spinal canal compromise), and two variables were used to describe T2 hyperintensity signal characteristics (maximum height and Wang signal ratio) in the initial model. These four imaging variables along with baseline mJOA yielded an AUC of 0.849. Reduction of variables to create parsimony resulted in a final model including T1 hypointensity (OR = 0.242; CI: 0.068–0.866), spinal canal compromise (OR = 0.940; CI: 0.90–0.982), and baseline mJOA (OR = 1.743; CI: 1.353–2.245) with an AUC of 0.845, while reducing both the AIC and BIC. The AUC for the baseline mJOA-only model was 0.807. The likelihood-ratio test indicated superior performance of the full model compared with the mJOA-only model (p < 0.0001).
Conclusion
Baseline mJOA is a strong predictor of postsurgical outcome in CSM at 6 months; however, a model inclusive of spinal canal compromise and T1 hypointensity assessment in addition to this provides a superior predictive capacity. This suggests that MRI analysis has a significant role in predicting surgical outcome. It is, therefore, recommended that a thorough MRI analysis be conducted in all patients with CSM considered for surgical treatment.