Comparison of the Efficacy of Random Forest and Logistic Regression Models in Predicting Postoperative Dysphagia in Patients Undergoing Anterior Cervical Surgery

doi:10.12104/j.issn.1674-4748.2025.15.021

Abstract：Objective: To establish a predictive model for postoperative dysphagia in patients undergoing anterior cervical surgery based on random forest and logistic regression, and to compare their predictive efficacy. Methods: A retrospective study was conducted on 200 patients who underwent anterior cervical surgery at the hospital between June 2021 and June 2024. The patients were divided into two groups based on whether they developed dysphagia: the dysphagia group and the non-dysphagia group. Univariate analysis was used to compare the clinical data between the two groups. Logistic regression and random forest algorithms were used to construct predictive models for postoperative dysphagia in patients undergoing anterior cervical surgery. The predictive value of the two models was compared using the area under the receiver operating characteristic curve (AUC). Results: Univariate analysis showed that age, gender, smoking status, comorbid chronic obstructive pulmonary disease, comorbid diabetes, duration of preoperative pain, Hamilton Anxiety Scale (HAMA) score, postoperative prevertebral soft tissue swelling, and intraoperative use of cervical plates were significantly associated with postoperative dysphagia (all P < 0.05). Multivariate logistic regression analysis showed that age, gender, smoking, comorbid chronic obstructive pulmonary disease, duration of preoperative pain, HAMA score, and postoperative prevertebral soft tissue swelling were independent risk factors for postoperative dysphagia (all P < 0.05). The random forest algorithm model showed that the top six factors ranked by importance were HAMA score (GiNi=14.45), age (GiNi=12.59), smoking (GiNi=5.25), postoperative prevertebral soft tissue swelling (GiNi=3.75), gender (GiNi=2.84), and duration of preoperative pain (GiNi=2.12). ROC comparison results showed that the predictive performance of the multivariate logistic regression model was higher than that of the random forest model, with an AUC of 0.877, sensitivity of 0.924, and specificity of 0.719. Conclusion: Factors such as HAMA score, age, smoking, postoperative prevertebral soft tissue swelling, gender, and duration of preoperative pain are of great significance in predicting postoperative dysphagia in patients undergoing anterior cervical surgery. The risk prediction model for postoperative dysphagia in patients undergoing anterior cervical surgery built based on multivariate logistic regression has good predictive performance and can help clinicians prevent the occurrence of postoperative dysphagia.