A machine learning-driven model predicts patient-reported dysphagia after instrumented cervical fusion

Aim: We aim to develop a machine learning (ML)-driven predictive algorithm for patient-reported dysphagia after instrumented cervical fusion. Additionally, we aim to identify features for the prediction of dysphagia and to develop a web-based risk calculator for outcome prediction. Methods: We identified consecutive adults who underwent instrumented cervical fusion at a single institution between 2013-2020. We developed regression-based and ML-based prognostic models and assessed model performance using discrimination and calibration. Additionally, we identified patient features driving the performance of the most effective model. Results: Nine hundred and forty-seven patients were included in this study. There were 62 cases of dysphagia. The gradient boosting model was well-calibrated and demonstrated the highest discrimination of all tested models. The most important features for model performance included: anterior approach, deformity, revision procedure, bipolar disorder, diabetes mellitus, depression/anxiety, male sex, and myelopathy. Conclusion: We report a ML-driven model that accurately predicts patient-reported dysphagia after instrumented cervical fusion. Prediction of dysphagia risk may inform preoperative counseling and appropriate risk stratification. Furthermore, this model may identify modifiable risk factors that may be addressed preoperatively to reduce the risk of dysphagia after cervical fusion.