Abstract: Objective To explore the influencing factors of 5-year recurrence in mixed hemorrhoid patients after Milligan-Morgan hemorrhoidectomy (MMH) and to construct a machine learning based predictive model for recurrence risk,evaluating its performance.Methods Retrospective analysis of 684 patients with Milligan-Morgan surgery for mixed hemorrhoids admitted to Gaochun Hospital Affiliated to Jiangsu University from 2018 to 2020.Patients were followed up for 5 years and divided into a recurrence group (n=139) and a non-recurrence group (n=545).The dataset was split into a training set and a validation set at a 7∶3 ratio,with 29 patient characteristics as independent variables.Clinical features and laboratory indicators were compared between the two groups.LASSO regression and XGBoost algorithms were used for feature selection,followed by multivariate logistic regression to identify independent risk factors.A nomogram prediction model was constructed based on the selected key factors,and its performance was evaluated using calibration curves,decision curve analysis (DCA),and receiver operating characteristic (ROC) curves to assess calibration,clinical utility,and discriminative ability.Results Analysis of preoperative clinical parameters from 684 patients revealed that LASSO regression identified 13 significant factors associated with recurrence:disease duration,age,smoking history,lesion diameter,postoperative infection,foreign body reaction,postoperative constipation,spicy diet,diabetes,postoperative psychological distress,anesthesia recovery time,triglycerides (TG),and white blood cell count (WBC).A Venn diagram was used to identify overlapping variables from both algorithms,and logistic regression analysis further refined the selection to 7 key parameters:spicy diet,postoperative constipation,postoperative infection,foreign body reaction,lesion diameter,postoperative psychological distress,and WBC.Multivariate logistic regression confirmed these as independent risk factors (P<0.05).The nomogram prediction model demonstrated good calibration in the calibration curve,high clinical net benefit in DCA,and excellent discriminative ability in ROC analysis (AUC=0.896,95%CI=0.860-0.931),with an accuracy of 81.0%,sensitivity of 80.4%,and specificity of 83.5%.ConclusionThe machine learning-based predictive model exhibits strong performance in assessing recurrence risk after MMH for mixed hemorrhoids,providing a theoretical reference for recurrence prediction.

Key words: mixed hemorrhoids, Milligan-Morgan hemorrhoidectomy, LASSO, XGBoost