Predicting Postoperative Kyphosis After Cervical Laminoplasty: Decision-Tree Algorithm

Introduction: Approximately 20% of patients develop kyphotic deformity after cervical laminoplasty. Understanding preoperative risk factors for post-laminoplasty kyphotic deformity (PKD) could allow surgeons to develop tools to tailor their treatment according to patient’s risk of complications. We aimed to develop a decision-tree algorithm to evaluate the risk of developing PKD one-year post-laminoplasty based on radiological measurements.

Methods: We retrospectively reviewed cervical spondylotic myelopathy (CSM) patients who underwent laminoplasty with a complete 1-year clinical follow-up at an academic tertiary care center. Radiographic measurements included T1 Slope, C2–7 Cobb angle (CA), C2–7 sagittal vertical axis (cSVA), Occiput to C2 angle (O-C2 angle), Neck Tilt (NT), and C2-C3 disc angle (c2-3 DA). PKD was defined as a loss of cervical lordosis greater than −10° by comparing pre-operative and post-operative C2-7 Cobb angles. Logistic regression analyses assessed the predictive performance of radiographic measurements for PKD. Radiographic measurements, age, gender, and BMI were used to create a decision tree model to predict PKD.

Results: Seventy-six patients (54 males, 71%) met the inclusion criteria, with nine developing PKD. Univariate logistic regressions showed that O-C2 angle (p=0.003, OR 1.23) and C2-3 DA (p=0.009, OR 1.15) were positively associated with PKD. When controlling for age, BMI, and sex, C2-7 Cobb angle (p=0.028, OR 0.837), O-C2 angle (p=0.005, OR 1.23), and C2-3 DA (p=0.01, OR 1.16) remained significant. A decision tree model calculated the importance of the features and identified that a O-C2 angle (< than 27 degrees) and C2-7 Cobb (>than 8.9) offered the best classification performance (AUC: 0.95, CI 0.91-0.99, sensitivity of 88% and specificity of 92%). This algorithm provided a positive likelihood ratio of 11.9 suggesting that patients with a positive test result in both decision tree nodes are approximately 12 times more likely to have PKD.

Conclusion : Radiographical predictors are able identify the risk of PDK with high metric performance. The O-C2 angle, and C2-3 DA in a decision tree algorithm may be pivotal radiographic indicators to evaluate the preoperative risk of PKD.