混合痔（湿热型）外剥内扎术后便秘风险预测模型的构建与验证

Construction and validation of a risk prediction model for postoperative constipation after Milligan-Morgan hemorrhoidectomy in patients with damp-heat type mixed hemorrhoids

摘要:

背景与目的 混合痔是肛肠外科常见疾病，外剥内扎术是其经典治疗方式之一，但术后便秘等并发症仍较常见，影响患者术后恢复及生活质量。目前针对混合痔（湿热型）外剥内扎术后便秘风险的预测研究较少。本研究基于传统Logistic回归及大语言模型（LLM）构建术后便秘风险预测模型，筛选术后便秘相关危险因素，并评价模型预测效能，为术后早期干预提供参考。方法 回顾性收集2022年9月—2024年12月扬州大学附属医院中医科收治的734例混合痔（湿热型）并接受外剥内扎术患者的临床资料。按照7∶3比例随机分为训练集（513例）和测试集（221例）。收集患者人口学资料、合并症、实验室指标及围手术期用药等信息，以术后是否发生便秘为结局指标。采用单因素及多因素Logistic回归分析筛选独立危险因素，构建列线图预测模型，并采用受试者工作特征（ROC）曲线评价模型预测效能。同时基于Llama-3.1-8B-Instruct LLM构建术后便秘风险预测模型，评价其准确率、召回率及F1分数。结果 训练集与测试集基线特征比较差异均无统计学意义（均P>0.05）。训练集中共163例（31.77%）患者发生术后便秘。单因素Logistic回归分析显示，血红蛋白计数降低及贫血与术后便秘发生明显相关（均P<0.05）。多因素Logistic回归分析结果表明，血红蛋白计数降低及合并贫血是混合痔（湿热型）外剥内扎术后便秘发生的独立危险因素（均P<0.05）。基于Logistic回归构建的预测模型在训练集和测试集中的AUC分别为0.78和0.76，具有较好的预测效能。LLM在65个训练轮次、学习率为5.00E-05时性能最佳，其准确率、召回率及F1分数分别为85.99%、89.40%和87.66%。结论 血红蛋白计数降低及贫血是混合痔（湿热型）患者外剥内扎术后发生便秘的重要危险因素。基于Logistic回归和LLM构建的预测模型均具有较好的预测性能，其中LLM展现出更高的预测准确率，可为术后便秘高风险患者的早期识别及个体化干预提供参考依据。

Abstract:

Background and Aims Mixed hemorrhoids are among the most common benign anorectal diseases, and Milligan-Morgan hemorrhoidectomy remains a classical surgical treatment. However, postoperative constipation is still a common complication that adversely affects postoperative recovery and quality of life. Currently, studies focusing on risk prediction of postoperative constipation in patients with damp-heat type mixed hemorrhoids are limited. This study aimed to identify risk factors for postoperative constipation and construct predictive models based on Logistic regression and large language model (LLM) to evaluate their predictive performance and provide references for early postoperative intervention.Methods Clinical data of 734 patients with damp-heat type mixed hemorrhoids who underwent Milligan-Morgan hemorrhoidectomy at the Department of Traditional Chinese Medicine, Affiliated Hospital of Yangzhou University, from September 2022 to December 2024 were retrospectively collected. Patients were randomly divided into a training set (n=513) and a testing set (n=221) at a ratio of 7:3. Demographic characteristics, comorbidities, laboratory indicators, and perioperative medication data were collected. Postoperative constipation was defined as the study endpoint. Univariate and multivariate Logistic regression analyses were performed to identify independent risk factors and establish a nomogram prediction model. Receiver operating characteristic (ROC) curves were used to evaluate model performance. In addition, an LLM-based prediction model using Llama-3.1-8B-Instruct was constructed, and its accuracy, recall, and F1 score were assessed.Results No statistically significant differences were observed in baseline characteristics between the training and testing sets (all P>0.05). Among the training cohort, 163 patients (31.77%) developed postoperative constipation. Univariate Logistic regression analysis showed that hemoglobin level and anemia were significantly associated with postoperative constipation (both P<0.05). Multivariate Logistic regression analysis further demonstrated that decreased hemoglobin level and anemia were independent risk factors for postoperative constipation in patients with damp-heat type mixed hemorrhoids after surgery (both P<0.05). The Logistic regression prediction model achieved AUC values of 0.78 and 0.76 in the training and testing sets, respectively, indicating good predictive performance. The LLM model achieved optimal performance at 65 training epochs with a learning rate of 5.00E-05, yielding an accuracy of 85.99%, recall of 89.40%, and F1 score of 87.66%.Conclusion Decreased hemoglobin level and anemia are important risk factors for postoperative constipation in patients with damp-heat type mixed hemorrhoids undergoing Milligan-Morgan hemorrhoidectomy. Both the Logistic regression and LLM-based models demonstrated favorable predictive performance, while the LLM model showed higher predictive accuracy. These findings may provide a reference for early identification and individualized intervention in high-risk patients with postoperative constipation.