代谢相关脂肪性肝病发生脂肪性肝炎的无创诊断模型构建及分析

Abstract

目的 通过临床一般资料、血清学指标及肝脏弹性成像无创检查手段，基于LASSO及Logistic回归建立代谢相关脂肪性肝病(MAFLD)发生脂肪性肝炎的诊断模型，并评估该模型的诊断价值。 方法 纳入2018年1月—2021年12月于广东省中医院诊断为MAFLD且完善肝病理活检的患者为研究对象299例，根据肝病理NAS评分将其分为脂肪性肝炎组(n=170)例和无脂肪性肝炎组(n=129)。先后通过LASSO回归及多因素Logistic回归筛选MAFLD发生脂肪性肝炎的影响因素，并建立无创诊断模型，利用列线图形式可视化，采用加强Bootstrap法进行内部验证，绘制ROC曲线及Calibration曲线，并在MAFLD+NAFLD和MAFLD+cHBVi两个亚组人群中观察模型的诊断效能，并与其他诊断模型进行比较分析。计数资料组间比较采用χ²检验；符合正态分布的计量资料组间比较采用成组t检验，不符合正态分布的计量资料组间比较采用Mann-Whitney U检验。采用多因素Logistic回归分析，筛选最佳诊断因素，构建列线图诊断模型，绘制受试者工作特征曲线(ROC曲线)，计算ROC曲线下面积(AUC)，并进一步采用加强Bootstrap法对模型进行内部验证，绘制Calibration曲线显示校准度。 结果 两组间BMI、ALT、AST、ADA、ALP、GGT、TBA、TCO2、UA、HbA1c比较差异均有统计学意义(P值均 < 0.05)；FibroScan方面，两组LSM及CAP比较提示差异具有统计学意义(P值均 < 0.001)；病理学方面，两组的纤维化等级、脂肪变积分、小叶炎症积分、气球样变积分及NAS总分差异均有统计学意义(P值均 < 0.001)。亚组队列方面，MAFLD+NAFLD有、无脂肪性肝炎组分别为63、48例，MAFLD+cHBVi有、无脂肪性肝炎组分别为90、71例。通过LASSO回归及多因素Logistic回归筛选出LSM、CAP、BMI、AST是判断MAFLD患者是否发生脂肪性肝炎的最佳诊断因素，并以此构建LCBA模型。LCBA模型结果提示，总MAFLD、MAFLD+NAFLD和MAFLD+cHBVi人群的AUC分别为0.816、0.866、0.764(P值均 < 0.001)，ROC曲线对比显示均优于acNASH、HSI、NFS模型。 结论 LCBA模型用于诊断MAFLD患者是否发生脂肪性肝炎的效能稳定，且优于acNASH、HSI、NFS，值得临床推广。

Alternate abstract:

Objective To establish a diagnostic model for steatohepatitis in metabolic associated fatty liver disease (MAFLD) based on LASSO and logistic regression analyses by using general clinical data, serological parameters, and noninvasive liver elastography, and to evaluate the diagnostic value of this model. Methods A total of 299 patients who were diagnosed with MAFLD and underwent liver biopsy in Guangdong Provincial Hospital of Traditional Chinese Medicine from January 2018 to December 2021 were enrolled as subjects, and according to NAS score, they were divided into steatohepatitis group with 170 patients and non-steatohepatitis group with 129 patients. The LASSO regression analysis and the multivariate logistic regression analysis were used to identify the influencing factors for steatohepatitis in MAFLD, and a noninvasive diagnostic model was established, visualized in the form of nomogram, and internally validated by the enhanced Bootstrap method. The receiver operating characteristic (ROC) curve and the calibration curve were plotted for the model, and its diagnostic efficacy was observed in the MAFLD+NAFLD and MAFLD+cHBVi subgroups, which was then compared with other diagnostic models. The chi-square test was used for comparison of categorical data between groups; the independent-samples t test was used for comparison of normally distributed continuous data between groups, and the Mann-Whitney U test was used for comparison of non-normally distributed continuous data between groups. A multivariate logistic regression analysis was used to determine optimal diagnostic factors, and a nomogram diagnostic model was established; the ROC curve was plotted, and the area under the ROC curve (AUC) was calculated; the enhanced Bootstrap method was used for internal validation of the model, and the calibration curve was plotted to show the level of calibration. Results There were significant differences between the two groups in body mass index (BMI), alanine aminotransferase, aspartate aminotransferase (AST), adenosine deaminase, alkaline phosphatase, gamma-glutamyl transpeptidase, total bile acid, total carbon dioxide concentration, uric acid, HbA1c (all P < 0.05). As for FibroScan, there were significant differences between the two groups in liver stiffness measurement (LSM) and controlled attenuation parameter (CAP) (both P < 0.001); as for pathology, there were significant differences between the two groups in fibrosis degree, steatosis score, lobular inflammation score, ballooning degeneration score, and total NAS score (all P < 0.001). In the subgroup analysis, there were 63 patients with steatohepatitis and 48 patients without steatohepatitis in the MAFLD+NAFLD group, and there were 90 patients with steatohepatitis and 71 patients without steatohepatitis in the MAFLD+cHBVi group. The LASSO regression analysis showed that LSM, CAP, BMI, and AST were the best diagnostic factors for the presence or absence of steatohepatitis in MAFLD patients, and the LCBA model was established based on these indices. The LCBA model showed an AUC of 0.816 in the total MAFLD population, 0.866 in the MAFLD+NAFLD population, and 0.764 in the MAFLD+cHBVi population (all P < 0.001), and comparisons based on the ROC curve showed that they were superior to the acNASH, HSI, and NFS models. Conclusion The LCBA model has a stable performance in the diagnosis of steatohepatitis in patients with MAFLD and is superior to acNASH, HSI, and NFS. Therefore, it holds promise for clinical application.