Abstract
INTRODUCTION:Neck pain is a common symptom in patients with cervical spondylotic myelopathy (CSM). Machine learning (ML) can be leveraged to develop prediction tools that can predict post-operative neck pain in patients with CSM.METHODS:We analyzed the prospective Quality Outcomes Database consisting of patients with CSM who underwent cervical spine surgery. 1141 patients were split into an 80% training/20% testing cohort. ML algorithms were evaluated on their performance to predict achievement of the minimum clinically important difference (MCID) in Visual Analogue Scale-Neck Pain (VAS-NP) at 2 years post-operatively given a set of 20 pre-operative features. Hyperparameter tuning was performed with 5-fold cross-validation. Recursive feature selection was used to select 5 key variables for predicting VAS-NP MCID achievement. Model performance was assessed with accuracy, area under the receiver operating curve (AUROC), precision, recall/sensitivity, and specificity. The final model was evaluated on the testing cohort. A GUI was created using the Shiny application software to predict post-operative neck pain.RESULTS:Random forest models demonstrated the best performance for predicting 2 year post-operative neck pain. Random forest model achieved 80% cross-validation accuracy for predicting MCID in neck pain. Recursive feature selection identified age, baseline VAS-NP, baseline modified Japanese Orthopedic Association (mJOA) score, baseline EQ-VAS score, and BMI as the most important variables for predicting post-operative neck pain. Final models using these variables achieved 80% accuracy and AUROC during the testing phase. Finally, we developed the first web interface to allow prediction of post-operative neck pain with user-inputted data.CONCLUSIONS:Baseline scores for VAS-NP, mJOA, EQ-VAS, BMI, and age are important features for predicting post-operative neck pain. ML calculators using these variables can be used in the clinical setting to predict post-operative neck pain with good accuracy.