基于Lasso回归的慢性乙型肝炎发生肝硬化列线图预测模型的构建  

Construction of a Lasso regression-based prediction model for development of cirrhosis in chronic hepatitis B

作　　者：李影 韩可兴 苏倩[1] 徐楠[1] 谢琴秀[1] 郜玉峰[1] Ying Li;Ke-Xing Han;Qian Su;Nan Xu;Qin-Xiu Xie;Yu-Feng Gao(Department of Infectious Diseases,The First Affiliated Hospital of Anhui Medical University,Hefei 230022,Anhui Province,China;Department of Infectious Diseases,The Lu’an Hospital Affiliated to Anhui Medical University,The Lu’an People’s Hospital,Lu’an 237005,Anhui Province,China)

机构地区：[1]安徽医科大学第一附属医院感染病科,安徽省合肥市230022 [2]安徽医科大学附属六安医院感染病科,六安市人民医院感染病科,安徽省六安市237005

出　　处：《世界华人消化杂志》2023年第7期282-289,共8页World Chinese Journal of Digestology

基　　金：安徽省自然科学基金,No.2208085MH204.

摘　　要：背景慢性乙型肝炎(chronic hepatitis B,CHB)呈全球性流行,其进展成肝硬化的过程常被患者忽略.通过临床常规指标来构建肝硬化无创诊断模型,可为肝硬化的早期诊治提供参考价值.目的利用临床常见指标构建肝硬化的列线图预测模型.方法选取2010-2018初次就诊于安徽医科大学第一附属医院和第二附属医院感染科接受肝活检的CHB患者,收集其实验室检查指标并进行组间比较.采用Lasso回归模型筛选对肝硬化具有预测价值的预测因子,并采用多因素Logistic回归分析建立预测模型.采用Bootstrap法重采500次进行模型的内部验证,计算曲线下面积(the area under curve,AUC)以评估模型区分度.绘制决策曲线分析(decision curve analysis,DCA)以评估模型的获益度,校准曲线(calibration curve,CA)以评估模型的校准度.结果共纳入CHB病例1087例,其中并发肝硬化者135例,两组间除乙肝病毒脱氧核糖核酸(the deoxyribo nucleic acid quantification of hepatitis B virus,HBV DNA)定量、谷丙转氨酶(alanine transaminase,ALT)外,其余指标均具有统计学差异(P<0.05).经Lasso回归分析后,最终筛选出的预测变量为年龄、甲胎蛋白(alpha fetoprotein,AFP)、清蛋白(albumin,ALB)、球蛋白(globulin,GLB)、谷氨酰转肽酶(glutamyl transpeptidase,GGT)、血小板计数(platelet,PLT).经多因素Logistic回归分析建立预测模型Logit P=1.26+0.02×年龄+0.001×AFP-0.10×ALB+0.07×GLB+0.004×GGT-0.02×PLT,其中AUC为0.83,95%置信区间(confidence interval,CI)为0.79-0.87.DCA曲线提示使用建立的预测模型能够使患者的净获益增加,CA曲线提示该预测模型的预测效应与实际结果间一致性良好.结论本研究以CHB患者的年龄、AFP、ALB、GLB、GGT、PLT作为预测变量,构建的对CHB并发肝硬化的列线图预测模型具有良好的预测效能,值得临床进一步推广.BACKGROUND Chronic hepatitis B(CHB)is a global epidemic,and its progression to cirrhosis is often overlooked by patients.Noninvasive diagnostic models for cirrhosis,which are developed using common clinical indicators,can provide reference value for the early diagnosis and treatment of cirrhosis in CHB.AIM To construct a prediction model for cirrhosis based on common clinical indicators.METHODS Patients with CHB who underwent liver biopsy at the Department of Infectious Diseases,The First or Second Affiliated Hospital of Anhui Medical University from 2010 to 2018 were selected,and their laboratory test indicators were collected and compared between patients with and without cirrhosis.Lasso regression model was used to screen the variables with predictive value for cirrhosis,and multivariate logistic regression analysis was performed to establish a prediction model for cirrhosis.The area under the curve(AUC)was calculated to assess the discrimination performance of the model.Decision curve analysis(DCA)was performed to assess the benefit of the model,and calibration curve-based analysis(CA)was performed to assess the calibration of the model.RESULTS A total of 1087 CHB cases were included,of which 135 had cirrhosis.All indicators were statistically different between the two groups except for hepatitis B virus(HBV)DNA,alanine transaminase(ALT)(P<0.05).Lasso regression analysi identified the predictive variables as age,alpha-fetoprotein(AFP),albumin(ALB),globulin(GLB),glutamyl transpeptidase(GGT),and platelet count(PLT).A prediction model for cirrhosis was developed by multifactorial logistic regression analysis:Logit P=1.26+0.02×age+0.001×AFP-0.10×ALB+0.07×GLB+0.004×GGT-0.02×PLT.The AUC of the model for predicting cirrhosis was 0.83(95%confidence interval:0.79-0.87).DCA suggested that the use of the developed prediction model resulted in an increased net benefit for the patients,and CA suggested that the predictive effect of the prediction model was in accordance with the actual outcome.CONCLUSION The present

关 键 词：慢性乙型肝炎 肝硬化 列线图 预测模型 Lasso回归 

分 类 号：R512.62[医药卫生—内科学] R575.2[医药卫生—临床医学]