高尿酸血症Lasso-logistic回归预测模型的建立  

Establishment of Lasso-logistic Regression Predictive Model for Hyperuricemia

作　　者：杨凯迪 朱国军 刘翔[3] 周尚成 王晨阳 刘冰[1] YANG Kaidi;ZHU Guojun;LIU Xiang;ZHOU Shangcheng;WANG Chenyang;LIU Bing(Center of Health Administration and Development Studies,Hubei University of Medicine,Shiyan 442000,China;Director′s Office,Shiyan Railway Construction Hospital,Shiyan 442000,China;Physical Examination Center of Shiyan Renmin Hospital,Shiyan 442000,China;School of Public Health and Management of Guangzhou University of Chinese Medicine,Guangzhou 510000,China)

机构地区：[1]湖北医药学院卫生管理与卫生事业发展研究中心,湖北十堰442000 [2]十堰市铁建医院院长办公室,湖北十堰442000 [3]十堰市人民医院体检中心,湖北十堰442000 [4]广州中医药大学公共卫生与管理学院,广州510000

出　　处：《医学综述》2023年第18期3708-3714,共7页Medical Recapitulate

基　　金：国家自然科学基金(71774049)。

摘　　要：目的建立高尿酸血症Lasso-logistic回归预测模型。方法选取中铁某局2016年健康体检基线数据完整且分别于2018、2020年完成随访的1610名建筑业职工为研究对象。通过Lasso回归筛选高尿酸血症相关预测因子,利用多因素Logistic回归构建模型并用列线图实现模型的可视化。分别采用受试者工作特征曲线(ROC曲线)、曲线下面积(AUC)、校准曲线、临床决策曲线对预测模型进行评估,并采用Bootstrap方法对模型展开内部验证。结果在4年内共有658例(40.87%)高尿酸血症病例。高尿酸血症组男性比例高于非高尿酸血症组(P<0.01),年龄小于非高尿酸血症组(P<0.01),体质量指数、收缩压、舒张压、高血脂比例、基线尿酸、肌酐、丙氨酸转氨酶、天冬氨酸转氨酶、谷氨酰转肽酶水平高于非高尿酸血症组(P<0.05或P<0.01)。基于Lasso-logistic的回归模型纳入5个与高尿酸血症相关的因素,分别为性别、年龄、基线尿酸、收缩压、体质量指数。多因素Logistic回归分析结果显示,性别、年龄、体质量指数、收缩压、基线尿酸是高尿酸血症发生的影响因素(OR=2.795,95%CI 1.349~5.797;OR=0.962,95%CI 0.951~0.974;OR=1.050,95%CI 1.011~1.092;OR=1.013,95%CI 1.004~1.022;OR=1.025,95%CI 1.022~1.028)(P<0.05或P<0.01)。所构建预测模型AUC为0.819(95%CI 0.799~0.840),内部验证AUC为0.817。Hosmer-Lemeshow拟合优度检验显示拟合度较好(P>0.05)。临床决策曲线分析结果提示,阈值概率为0~0.9时,使用预测模型预测高尿酸血症风险有较好的净收益。结论建立的高尿酸血症Lasso-logistic回归预测模型具有较好的预测能力,有助于早期识别高尿酸血症高风险人群。Objective To establish a Lasso-logistic regression predictive model for hyperuricemia.Methods A total of 1610 construction workers from a China Railway Bureau who had completed baseline data of physical examination in 2016 and completed follow-up visits in 2018 and 2020 were included as the research objects.The related predictors of hyperuricemia were screened by Lasso regression,and the model was established by multivariate Logistic regression analysis,and the visualization of the model was realized by nomogram.The prediction model was evaluated by analyzing the receiver operating characteristic curve(ROC),the area under curve(AUC),calibration curve and clinical decision curve,and the Bootstrap method was used for internal verification of the model.Results In the study,658 cases(40.87%)of hyperuricemia were discovered.The proportion of males in the hyperuricemia group was higher than that in the non-hyperuricemia group(P<0.01),while the age was younger than that in the non-hyperuricemia group(P<0.01),and the levels of body mass index(BMI),systolic blood pressure,diastolic blood pressure,hyperlipidemia ratio,baseline uric acid,creatinine,alanine aminotransferase,aspartate aminotransferase and glutamyl transpeptidase were higher than those in the non-hyperuricemia group(P<0.05 or P<0.01).Based on Lasso-logistic regression model,five factors related to hyperuricemia were included,i.e.gender,age,baseline uric acid,systolic blood pressure,and BMI.Multivariate Logistic regression analysis showed that gender,age,BMI,systolic blood pressure and baseline uric acid were the factors affecting the occurrence of hyperuricemia(OR=2.795,95%CI 1.349-5.797;OR=0.962,95%CI 0.951-0.974;OR=1.050,95%CI 1.011-1.092;OR=1.013,95%CI 1.004-1.022;OR=1.025,95%CI 1.022-1.028)(P<0.05 or P<0.01).The AUC of the established prediction model was 0.819(95%CI 0.799-0.840),and the AUC of internal validation was 0.817.Hosmer-Lemeshow goodness-of-fit test showed that the fitting degree was good(P>0.05).The results of clinical decision curve analysis s

关 键 词：高尿酸血症 Lasso-logistic预测模型 列线图 

分 类 号：R589.7[医药卫生—内分泌]