代谢相关脂肪性肝病的严重程度预测模型构建与验证

Construction and validation of a severity prediction model for metabolic associated fatty liver disease

目的:分析中重度代谢相关脂肪性肝病(MAFLD)发生的独立危险因素,构建中重度MAFLD危险因素的预测模型并验证其有效性。方法:第一部分,将2022年1~5月在海南医学院第二附属医院体检中心被确诊为MAFLD的278名体检者作为研究对象(训练集),收集人口学资料及化验指标,依据超声结果将其分为轻度MAFLD组(200人)和中重度MAFLD组(78人),单因素、多因素分析筛选独立危险因素。第二部分,运用二分类逻辑回归方程构建中重度MAFLD预测模型,列线图可视化呈现模型。第三部分,收集2022年11~12月我院体检中心被确诊为MAFLD的人群(外部验证集200人)作为中重度MAFLD预测模型的验证人群,分组同前。绘制代表模型区分度的受试者工作特征(receiver operating characteristic curve, ROC)曲线、校准曲线图和模型临床适用度评价模型效能等进行模型内外部验证。结果:中重度MAFLD的危险因素为空腹血糖(FPG)、血尿酸(UA)、甘油三脂(TG)、甘油三酯葡萄糖指数(TyG)、总胆固醇(CHOL)和高密度脂蛋白(HDL-C)。UA[OR=1.021, 95%CI (1.015, 1.027),P<0.001]和FPG[OR=1.575, 95%CI (1.158, 2.143),P=0.004]是中重度MAFLD人群的独立危险因素。可视化列线图模型显示UA是中重度MAFLD发生风险贡献较大的因素。ROC曲线显示训练集、内部验证集和外部验证集的AUC值分别为0.870 1、0.868 6和0.799 1;模型校准度校准后曲线几乎与参考线相重合,Hosmer-Lemeshow检验P>0.05;模型临床适用度绘制的决策曲线(decision curve analysis, DCA)高于两条极端曲线,预测中重度MAFLD患者能从预测模型中获益。结论:UA联合FPG构建的预测模型准确性较高、临床适用度较好,可应用于临床诊断。

Objective:To analyze the independent risk factors for the occurrence of moderate-to-severe metabolic-associated fatty liver disease(MAFLD),to construct a prediction model for moderate-to-severe MAFLD,and to verify the validity of the model.Methods:In the first part,278 medical examiners who were diagnosed with MAFLD in Medical Examination Center at the Second Affiliated Hospital of Hainan University from January to May 2022 were taken as the study subjects(training set),and they were divided into mild MAFLD group(200)and moderate-severe MAFLD group(78)based on ultrasound results.Demographic data and laboratory indexes were collected,and risk factors were screened by univariate and multifactor analysis.In the second part,a dichotomous logistic regression equation was used to construct a prediction model for moderate-to-severe MAFLD,and the model was visualized in a line graph.In the third part,the MAFLD population(200 people in the external validation set)from our physical examination center from November to December 2022 was collected as the moderate-to-severe MAFLD prediction model,and the risk factors in both groups were compared.The receiver operating characteristic(ROC)curves,calibration curves,and clinical applicability of the model were plotted to represent model discrimination for internal and external validation.Results:The risk factors of moderate-to-severe MAFLD were fasting glucose(FPG),blood uric acid(UA),triglycerides(TG),triglyceride glucose index(TyG),total cholesterol(CHOL),and high-density lipoprotein(HDL-C).UA[OR=1.021,95%CI(1.015,1.027),P<0.001]and FPG[OR=1.575,95%CI(1.158,2.143),P=0.004]were independent risk factors for people with moderate to severe MAFLD.The visualized line graph model showed that UA was the factor contributing more to the risk of moderate to severe MAFLD in this model.The ROC curves showed AUC values of 0.8701,0.8686 and 0.7991 for the training set,internal validation set and external validation set,respectively.The curves almost coincided with the reference line after calibration of the model calibration degree with P>0.05 in Hosmer-Lemeshow test.The decision curve analysis(DCA)plotted by the clinical applicability of the model was higher than the two extreme curves,predicting that patients with moderate to severe MAFLD would benefit from the prediction model.Conclusion:The prediction model constructed by combining FPG with UA has higher accuracy and better clinical applicability,and can be used for clinical diagnosis.