腹腔镜胆囊切除效费比聚类分析及其人工智能机器学习预测模型构建  

Clustering analysis of cost-effectiveness ratio and construction of AI machine learning predictive model for laparoscopic cholecystectomy

作　　者：马玉爱[1] 赵金燕 龚雪屹 MA Yuai;ZHAO Jinyan;GONG Xueyi(Department of General Surgery One,Zhongshan People's Hospital,Zhongshan 528403,Guangdong,China)

机构地区：[1]广东省中山市人民医院普外一科,广东中山528403

出　　处：《右江医学》2024年第8期726-732,共7页Chinese Youjiang Medical Journal

基　　金：中山市医学科研项目(2020A020575);中山市社会公益与基础研究项目(210321133948041)。

摘　　要：目的探讨腹腔镜胆囊切除手术的效费比,并通过聚类分析及人工智能机器学习方法构建预测模型。方法回顾性分析中山市人民医院腹腔镜胆囊切除术206例患者的临床资料。将患者住院总费用、总住院时间、住院至手术时间、术后住院时间、术前止痛药物应用次数、术后止痛药物应用次数等维度作为效费比指标进行聚类分析分类指标,分为三类,效费比优、中、差;以CT炎症情况、CT病灶情况、CT病灶部位、病理诊断、CA19-9、AFP、术前血糖、ALB、AST、ALT、PLT、HBG、RBC、WBC、CT距手术时间、伤口疼痛程度、手术时间、术前血压、CT病灶大小、住院至手术时间、术前禁食时间等指标作为变量,进行多分类logistic回归分析及机器学习进行预测模型构建,包括逻辑回归、线性支持向量机、支持向量机、决策树、随机森林、K近邻等分类器进行模型拟合。结果多分类logistic回归分析显示模型拟合卡方值为156.986,P<0.001,似然比ALB、WBC、CT距手术时间、住院至手术时间、高血压、肾脏疾病、手术人员差异有统计学意义,P值分别为0.001、0.019、0.029、<0.001、0.005、0.027、<0.001,模型拟合分类总符合率为74.8%(即准确率为0.748)。多层感知机模型K折验证评分0.461,预测评分0.802;逻辑回归模型K折验证评分0.437,预测评分0.726;支持向量机模型K折验证评分0.529,预测评分0.755;决策树模型K折验证评分0.462,预测评分0.585;随机森林模型K折验证评分0.529,预测评分0.726;K近邻模型K折验证评分0.388,预测评分0.623。结论分析腹腔镜胆囊切除手术效费比中,机器学习建模可有效预测效费比指标,可用于术前评估,按效费比优化医疗资源分配。Objective To investigate the cost-effectiveness ratio(CER)of laparoscopic cholecystectomy and construct predictive model using clustering analysis and AI machine learning method.Methods A retrospective analysis was conducted on clinical data from 206 patients who underwent laparoscopic cholecystectomy in Zhongshan People's Hospital.Total hospitalization cost,total hospitalization time,time from admission to surgery,postoperative hospitalization time,preoperative analgesic application frequency,and postoperative analgesic application frequency were served as cost-effectiveness ratio indicators,which were divided into three categories:excellent cost-effectiveness ratio,moderate cost-effectiveness ratio,and poor cost-effectiveness ratio.Variables such as CT inflammation,CT lesion condition,CT lesion location,pathological diagnosis,CA19-9,AFP,preoperative blood glucose,ALB,AST,ALT,PLT,HBG,RBC,WBC,time from CT to surgery,wound pain severity,surgery duration,preoperative blood pressure,CT lesion size,time from admission to surgery,and preoperative fasting time were used for multi-class logistic regression analysis and machine learning model construction.Models were fitted by classifiers included logistic regression,linear SVM,SVM,decision tree,random forest,and K-nearest neighbors.Results Multi-class logistic regression analysis showed that the fitted chi-square value of the model was 156.986,P<0.001.The likelihood ratios of ALB,WBC,CT to surgery time,hospitalization to surgery time,hypertension,kidney disease,and surgical personnel were statistically significant,and P values were 0.001,0.019,0.029,<0.001,0.005,0.027,<0.001,respectively.The overall conformity rate of the model fitting classification was 74.8%(i.e.,the accuracy rate was 0.748).The K-fold validation score of multi-layer perceptron model was 0.461,and prediction score was 0.802;the K-fold validation score of logistic regression model was 0.437,and prediction score was 0.726;the K-fold validation score of the SVM model was 0.529,and prediction score was 0.

关 键 词：腹腔镜胆囊切除手术 效费比 聚类分析 人工智能 机器学习 

分 类 号：R657.4[医药卫生—外科学]