2023 , Vol. 20 >Issue 09: 923 - 929
DOI: https://doi.org/10.3877/cma.j.issn.1672-6448.2023.09.006
基于超声瞬时弹性成像的多参数决策树模型评估慢性乙型肝炎患者肝纤维化等级
Copy editor: 吴春凤
收稿日期: 2023-05-26
网络出版日期: 2023-12-11
基金资助
杭州市科技发展计划项目(20201203B179)
版权
Ultrasonic elastography based multiparameteric decision tree model for assessment of grade of liver fibrosis in patients with chronic hepatitis B
Received date: 2023-05-26
Online published: 2023-12-11
Copyright
构建及验证基于超声瞬时弹性成像(UTE)技术的多参数决策树模型用于识别慢性乙型肝炎病毒(HBV)感染患者肝纤维化等级。
回顾性收集2019年2月至2022年10月在杭州市西溪医院就诊的530例慢性HBV感染患者的UTE特征及临床资料。将所有患者按7∶3比例随机分为训练集(370例)和测试集(160例),在训练集中采用多因素逐步Logistic回归方法筛选独立预测因子,通过决策树算法建立诊断模型,在测试集中通过受试者操作特征曲线评估UTE及临床特征对轻度、中重度肝纤维化及肝硬化的诊断效能。应用Delong检验评估模型与各独立预测因子诊断效能的差异,应用Hosmer-Lemeshow检验评估模型是否过拟合。
诊断模型在训练集中用于评估轻度肝纤维化、中重度肝纤维化及肝硬化的AUC分别为0.871、0.925和0.952,在测试集中的AUC分别为0.859、0.901及0.936。Delong检验显示无论在训练集还是在测试集中,模型的诊断效能在不同肝纤维化等级中都要高于相关的独立预测因子的诊断效能,且差异均具有统计学意义(P均<0.05)。Hosmer–Lemeshow检验显示模型在识别3种肝纤维化状态中均没有过拟合(P均>0.05)。
基于UTE的多参数决策树诊断模型可以很好地识别HBV感染患者的肝纤维化等级。
高建松 , 陈晓晓 , 冯婷 , 包剑锋 , 魏淑芳 , 潘林 . 基于超声瞬时弹性成像的多参数决策树模型评估慢性乙型肝炎患者肝纤维化等级[J]. 中华医学超声杂志(电子版), 2023 , 20(09) : 923 -929 . DOI: 10.3877/cma.j.issn.1672-6448.2023.09.006
To construct and validate a multiparameter decision tree model based on ultrasound elastography for identifying the grade of liver fibrosis in patients with chronic hepatitis B virus (HBV) infection.
The ultrasound elastography features and clinical data of 530 patients with chronic hepatitis B who visited Hangzhou Xixi Hospital from February 2019 to October 2022 were retrospectively collected. The patients were randomly divided into a training set and a test set in a 7∶3 ratio. Based on the training set, multivariate logistic regression method was used to select features, and a diagnostic model was established by using decision tree algorithm. In the test set, the diagnostic efficacy of ultrasound elastography and clinical features for different levels of liver fibrosis (mild, moderate, and severe liver fibrosis and cirrhosis) was evaluated by receiver operating characteristic (ROC) curve analysis. The Delong test was used to evaluate the differences in the diagnostic efficacy between the model and independent predictors, and the Hosmer-Lemeshow test was used to evaluate whether the model was overfit.
The area under the ROC curve (AUC) of the diagnostic model for diagnosing mild liver fibrosis, moderate to severe liver fibrosis, and cirrhosis in the training set was 0.871, 0.925, and 0.952, respectively; the AUC in the test set was 0.859, 0.901, and 0.936, respectively. The Delong test showed that the diagnostic performance of the model was higher than that of the independent predictive factors associated with different levels of liver fibrosis in both the training and testing sets, and there was a significant difference in the diagnostic performance between them (P<0.05). The Hosmer-Lemeshow test showed that the model was not overfit in identifying the three grades of liver fibrosis (P>0.05).
The diagnostic model developed can effectively identify the grade of liver fibrosis in patients with chronic HBV infection.
Key words: Hepatitis B; Elastic imaging; Serological markers; Decision tree; Hepatic fibrosis
表1 训练集与测试集慢性乙型肝炎病毒感染患者的相关临床特征比较 |
特征 | 训练集(n=370) | 测试集(n=160) | 统计值 | P值 |
---|---|---|---|---|
性别[例(%)] | χ2=0.99 | 0.611 | ||
男 | 232(62.70) | 106(66.25) | ||
女 | 138(37.30) | 54(33.75) | ||
年龄(岁,![]() | 44.86±11.48 | 44.94±12.67 | t=-0.08 | 0.940 |
Ⅲ型前胶原[ng/ml,M(QR)] | 17.00(13.20,23.69) | 14.85(11.15,20.39) | Z=-3.609 | 0.050 |
Ⅳ型胶原[ng/ml,M(QR)] | 22.45(17.45,28.87) | 19.90(16.18,28.10) | Z=-2.062 | 0.530 |
层粘连蛋白[ng/ml,M(QR)] | 21.70(16.67,26.03) | 21.40(15.22,25.58) | Z=-0.927 | 0.354 |
透明质酸酶[ng/ml,M(QR)] | 57.64(42.91,88.59) | 58.03(42.01,58.03) | Z=-0.388 | 0.698 |
APRI[M(QR)] | 0.45(0.30,0.71) | 0.43(0.29,0.67) | Z=-1.111 | 0.266 |
FIB-4[M(QR)] | 1.10(0.79,1.68) | 1.08(0.75,1.57) | Z=-0.757 | 0.449 |
肝硬度[kPa,M(QR)] | 9.00(6.50,13.43) | 8.75(6.65,12.70) | Z=-0.120 | 0.910 |
脂肪衰减参数(db/m,![]() | 232.51±27.93 | 234.69±32.53 | t=-0.74 | 0.462 |
肝纤维化分级[例(%)] | χ2=0.40 | 0.983 | ||
F0 | 38(10.27) | 15(9.38) | ||
F1 | 118(31.89) | 54(33.75) | ||
F2 | 100(27.03) | 43(26.88) | ||
F3 | 32(8.65) | 15(9.38) | ||
F4 | 82(22.16) | 33(20.62) |
注:APRI为门冬氨酸转移酶纤维化指数;FIB-4为肝纤维化4因子指数 |
表2 逐步Logistic回归筛选不同肝纤维化程度的独立预测因子 |
特征 | 轻度纤维化(F0-1 vs F2-4) | 中重度纤维化(F2-3 vs F0-1、F4) | 肝硬化(F4 vs F0-3) | |||
---|---|---|---|---|---|---|
OR(95%CI) | P值 | OR(95%CI) | P值 | OR(95%CI) | P值 | |
性别 | NA | NA | NA | NA | NA | NA |
年龄 | 1.022(1.002~1.043) | 0.034 | 1.047(1.019~1.074) | 0.001 | 1.102(1.066~1.139) | <0.001 |
Ⅲ型前胶原 | NA | NA | NA | NA | ||
Ⅳ型胶原 | 1.080(1.042~1.119) | <0.001 | 1.038(1.011~1.065) | 0.006 | 1.032(1.005~1.059) | 0.019 |
层粘连蛋白 | 1.021(1.000~1.043) | 0.044 | NA | NA | NA | NA |
透明质酸酶 | NA | NA | 1.003(1.000~1.006) | 0.047 | NA | NA |
APRI | 1.493(1.358~1642) | <0.001 | NA | NA | NA | NA |
FIB-4 | NA | NA | NA | NA | 0.656(0.446~0.965) | 0.032 |
肝硬度 | 1.493(1.358~1.642) | <0.001 | 1.520(1.389~1.664) | <0.001 | 1.393(1.273~1.524) | <0.001 |
脂肪衰减参数 | NA | NA | 0.991(0.982~1.000) | 0.045 | 0.987(0.976~0.998) | 0.025 |
注:NA表示不可用,这些变量在多因素Logistic回归模型中被剔除,APRI为门冬氨酸转移酶纤维化指数,FIB-4为肝纤维化4因子指数,F0-4为肝纤维化分级 |
表3 决策树模型在慢性乙型肝炎病毒感染患者训练集和测试集中的诊断效能 |
肝纤维化等级 | AUC(95%CI) | 敏感度(%) | 特异性度(%) |
---|---|---|---|
训练集 | |||
轻度肝纤维化 | 0.871(0.833~0.903) | 75.2 | 82.7 |
中重度肝纤维化 | 0.925(0.893~0.95) | 86.0 | 85.5 |
肝硬化 | 0.952(0.925~0.971) | 87.8 | 91.7 |
测试集 | |||
轻度肝纤维化 | 0.859(0.795~0.909) | 78.0 | 79.7 |
中重度肝纤维化 | 0.901(0.844~0.943) | 81.3 | 87.5 |
肝硬化 | 0.936(0.886~0.968) | 87.9 | 85.0 |
1 |
|
2 |
郝岸华, 杜忠彩, 周林, 等. 慢性乙型肝炎患者肝脏硬度检测和S指数变化及评估肝纤维化程度应用价值探讨 [J]. 实用肝脏病杂志, 2023, 26(5): 630-633.
|
3 |
中国肝炎防治基金会, 中华医学会感染病学分会, 中华医学会肝病学分会和中国研究型医院学会肝病专业委员会. 瞬时弹性成像技术诊断肝纤维化专家共识(2018年更新版) [J]. 中华肝脏病杂志, 2019, 27(3): 182-191.
|
4 |
程丹颖, 李贲, 纪世博, 等. 瞬时弹性成像技术在肝纤维化无创诊断中的应用 [J/OL]. 中国肝脏病杂志(电子版), 2021, 13(4): 9-13.
|
5 |
代海峰, 王甜, 雷迅, 等. 应用瞬时弹性成像技术联合APRI和AAR指数评估慢性乙型肝炎患者肝纤维化临床价值研究 [J]. 实用肝脏病杂志, 2021, 24(2): 196-199.
|
6 |
|
7 |
|
8 |
|
9 |
|
10 |
|
11 |
|
12 |
张明, 祝再然, 赵桂金, 等. PCA-决策树模型评价miR-30a与RPR、AAR、FIB-4、APRI在乙肝患者肝纤维化的应用价值 [J]. 广东医学, 2021, 42(1): 106-110.
|
13 |
|
14 |
|
15 |
|
16 |
|
17 |
|
18 |
|
/
〈 |
|
〉 |