【摘要】 目的:探讨混合痔患者术后早期发生重度疼痛的危险因素,构建预测模型并进行内部验证。方法:回顾性分析2023年1月—2023年12月于宜昌市中心人民医院就诊的760例混合痔患者的临床资料,分为重度疼痛组(n=208)和非重度疼痛组(n=552)。采用单因素分析、Lasso回归、Logistic回归及机器学习相关方法构建临床预测模型。结果:本次共构建了3个Logistic回归模型与7个机器学习模型。在Logistic回归模型中,Lasso回归结合Logistic回归(逐步向前法)构建的临床预测模型预测效果最好,该模型包含了10个预测因子,曲线下面积为0.961、霍斯默-莱姆斯福德检验表现出良好一致性(P>0.05)、决策曲线分析发现阈值概率处于0%～95%之间时模型所获得的净收益大。在机器学习模型中,7个模型原始数据集的预测准确率均高于Logistic回归模型。所有预测因子根据平均特征重要性排名依次为红细胞、白细胞、饮酒、吸烟、居住地、便秘、学历、痔核数、病程、糖尿病。结论:本次研究中共构建了10个临床预测模型,其中1个Logistic回归模型与7个机器学习模型预测准确率均较高,医疗人员可根据研究目的,对比不同模型选择最适合的预测模型进行预测与风险控制。更多还原

【Abstract】 Objective:To explore the risk factors for severe pain in patients with mixed hemorrhoids in the early postoperative period,construct a predictive model,and perform internal validation.Methods:A retrospective analysis was conducted on the clinical data of 760 patients with mixed hemorrhoids who visited Yichang Central People’s Hospital from January to December 2023.The patients were divided into severe pain group(n=208) and non-severe pain group(n=552).Clinical prediction models were constructed using univariate analysis,Lasso regression,Logistic regression,and machine learning methods.Results:A total of3 Logistic regression models and 7 machine learning models were constructed.Among the Logistic regression models,the clinical prediction model constructed using Lasso regression combined with Logistic regression(stepwise forward method) showed the best predictive effect.This model included 10 predictive factors,with an area under the curve of 0.961,Hosmer-Lemeshow test showing good consistency(P>0.05),and decision curve analysis indicating that the net benefit obtained by the model was large when the threshold probability was between 0 % and 95%.Among the machine learning models,the prediction accuracy of the original dataset of all 7 models were higher than that of the Logistic regression model.All predictive factors were ranked according to the average feature importance as follows:red blood cells,white blood cells,alcohol consumption,smoking,residence,constipation,education level,number of hemorrhoids,duration of disease,and diabetes.Conclusion:A total of 10 clinical prediction models were constructed in this study,among which 1 Logistic regression model and 7 machine learning models had high prediction accuracy.Medical personnel can choose the most suitable prediction model for prediction and risk control based on the research purpose and compare different models.更多还原