切换至 "中华医学电子期刊资源库"

中华临床医师杂志(电子版) ›› 2025, Vol. 19 ›› Issue (02) : 129 -139. doi: 10.3877/cma.j.issn.1674-0785.2025.02.005

临床研究

基于自动化机器学习建立肝硬化并发急性肾损伤的风险评估模型
陈健1, 夏开建2, 徐璐3, 高福利1, 徐晓丹1, 王甘红3,()   
  1. 1. 215500 江苏常熟,江苏省常熟市第一人民医院消化内科
    2. 215500 江苏常熟,江苏省常熟市医学人工智能与大数据重点实验室
    3. 215500 江苏常熟,江苏省常熟市中医院(常熟市新区医院)消化内科
  • 收稿日期:2025-01-20 出版日期:2025-02-15
  • 通信作者: 王甘红
  • 基金资助:
    常熟市科技发展计划项目(CS202019、CSWS202316)苏州市应用基础研究(医疗卫生)科技创新项目(SYWD2024059)苏州卫生信息与健康医疗大数据学会项目(SZMIA2402)常熟市科技计划(社会发展)项目(CS202452)

Development of a risk assessment model for acute kidney injury in cirrhotic patients based on automated machine learning

Jian Chen1, Kaijian Xia2, Lu Xu3, Fuli Gao1, Xiaodan Xu1, Ganhong Wang3,()   

  1. 1. Department of Gastroenterology, Changshu First People's Hospital, Changshu 215500, China
    2. Department of Information Engineering, Changshu Key Laboratory of Medical Artificial Intelligence and Big Data,Changshu City, Changshu 215500, China
    3. Department of Gastroenterology, Changshu Hospital Affiliated to Nanjing University of Chinese Medicine, Changshu 215500, China
  • Received:2025-01-20 Published:2025-02-15
  • Corresponding author: Ganhong Wang
引用本文:

陈健, 夏开建, 徐璐, 高福利, 徐晓丹, 王甘红. 基于自动化机器学习建立肝硬化并发急性肾损伤的风险评估模型[J/OL]. 中华临床医师杂志(电子版), 2025, 19(02): 129-139.

Jian Chen, Kaijian Xia, Lu Xu, Fuli Gao, Xiaodan Xu, Ganhong Wang. Development of a risk assessment model for acute kidney injury in cirrhotic patients based on automated machine learning[J/OL]. Chinese Journal of Clinicians(Electronic Edition), 2025, 19(02): 129-139.

目的

急性肾损伤(AKI)是肝硬化患者严重且常见的并发症,本研究旨在基于自动化机器学习(AutoML)建立肝硬化并发AKI的预测模型。

方法

从一家医院回顾性收集因肝硬化住院的患者数据,分为AKI组和非AKI组,共纳入22个潜在预测变量,构建并内部验证LASSO回归模型及AutoML模型。通过受试者工作特征曲线下面积(AUC)、敏感度、特异度及准确率等指标评估模型性能,并选取表现最优的模型。使用特征重要性分析、SHAP值散点图和力图对模型进行可视化解释。最后,在Python-Streamlit框架中将最佳模型开发为一款网络应用程序,并在另一家医院的独立测试集上进行外部验证。

结果

共纳入559名肝硬化患者中,其中159例(28.44%)患者出现AKI。在验证集中,GBM模型表现最佳,优于LASSO回归模型和其他AutoML模型,其AUC值为0.93(95%CI:0.86~0.99)。在测试集中,GBM模型的准确率、敏感度、特异度和AUC分别为89.60%、80.00%、93.33%和0.91(95%CI:0.83~0.97)。变量重要性分析显示,β2微球蛋白、碱性磷酸酶、尿素氮、总胆红素、门静脉直径、凝血酶原活动度和MELD评分这7个因素对预测肝硬化合并AKI具有重要影响。SHAP值散点图和力图可视化展示了关键变量对AKI预测的影响。

结论

基于AutoML构建的预测模型,在肝硬化并发AKI风险的早期预测中展现了出色的预测性能及良好的使用便捷性。

Objective

Acute kidney injury (AKI) is a severe and common complication in patients with cirrhosis. This study aimed to develop a predictive model for cirrhosis-related AKI using automated machine learning (AutoML).

Methods

The data of patients hospitalized for cirrhosis were retrospectively collected from a hospital. The patients were divided into AKI and non-AKI groups. A total of 22 potential predictive variables were included to construct and internally validate LASSO regression and AutoML models. Model performance was evaluated using metrics such as the area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and accuracy to select the best-performing model. Feature importance analysis, SHAP scatter plots, and force plots were used for visual interpretation of the model. Finally, the best model was developed into an online application using the Python-Streamlit framework and externally validated on an independent test set from another hospital.

Results

A total of 559 cirrhotic patients were included, among whom 159 (28.44%) developed AKI. In the validation set, the GBM model outperformed the LASSO regression model and other AutoML models, achieving an AUC of 0.93 (95%confidence interval: 0.86~0.99). In the test set, the GBM model achieved an accuracy of 89.60%, sensitivity of 80.00%, specificity of 93.33%, and AUC of 0.91 (95%CI: 0.83~0.97). Variable importance analysis revealed that β2-microglobulin, alkaline phosphatase, blood urea nitrogen, total bilirubin, portal vein diameter,prothrombin activity, and MELD score were the seven key factors influencing the prediction of AKI in cirrhotic patients. SHAP scatter plots and force plots visually demonstrated the impact of these key variables on AKI prediction.

Conclusion

The predictive model developed using AutoML demonstrated excellent predictive performance and user-friendly convenience in the early prediction of AKI risk in cirrhotic patients.

图1 研究流程图
表1 训练集与验证集基线资料比较[n(%)]
变量 训练集(n=351) 验证集(n=83) 统计值 P
Child_Pugh 评分 χ 2=3.650 0.161
A 级 141(40.2) 32(38.6)
B 级 133(37.9) 25(30.1)
C 级 77(21.9) 26(31.3)
性别 χ 2=0.014 0.905
女性 164(46.7) 40(48.2)
男性 187(53.3) 43(51.8)
腹水(%) χ 2=2.401 0.301
256(72.9) 55(66.3)
轻- 中等量 78(22.2) 25(30.1)
大量 17(4.8) 3(3.6)
蜘蛛痣(%) χ 2=0.208 0.649
273(77.8) 62(74.7)
78(22.2) 21(25.3)
水肿(%) χ 2=0.743 0.69
N 224(63.8) 50(60.2)
S 102(29.1) 28(33.7)
Y 25(7.1) 5(6.0)
TBIL [μmol/L,MQ1Q3)] 23.94[13.68,59.85] 22.23[15.39,56.43] Z=0.789 0.836
胆固醇 [mmol/L,MQ1Q3)] 7.86[6.52,9.96] 7.71[6.36,9.98] Z=0.371 0.531
白蛋白 [g/L,MQ1Q3)] 35.40[32.60,37.60] 35.00[31.75,38.45] Z=0.468 0.607
碱性磷酸酶 [U/L,MQ1Q3)] 111.90[82.15,162.00] 105.20[78.75,153.45] Z=0.815 0.595
谷草转氨酶 [U/L,MQ1Q3)] 106.00[75.95,143.38] 96.10[71.30,129.43] Z=1.692 0.094
甘油三酯 [mmol/L,MQ1Q3)] 1.22[0.95,1.65] 1.20[0.90,1.55] Z=0.541 0.469
血小板计数(109/L,x±s) 250.30(93.52) 250.58(101.71) t=-0.022 0.981
肝硬化病因(%) χ 2=6.203 0.184
血吸虫病 68(19.4) 23(27.7)
乙肝 183(52.1) 34(41.0)
丙肝 32(9.1) 7(8.4)
NAFLD 28(8.0) 11(13.3)
其他 40(11.4) 8(9.6)
年龄 [ 岁,MQ1Q3)] 66.00[56.00,77.00] 69.00[57.00,75.50] Z=0.092 0.953
静息收缩压 [mmHg,MQ1Q3)] 130.00[120.00,140.00] 130.00[122.00,140.00] Z=-0.962 0.155
空腹血糖(%) χ 2=0.287 0.592
正常 289(82.3) 71(85.5)
升高 62(17.7) 12(14.5)
BMI [kg/m2MQ1Q3)] 22.80[20.40,25.35] 22.80[20.45,24.85] Z=0.325 0.77
PVD [cm,MQ1Q3)] 1.37[1.17,1.59] 1.37[1.21,1.58] Z=-0.003 0.894
PTA [%,MQ1Q3)] 43.47[32.83,55.47] 40.52[32.09,53.19] Z=0.863 0.484
BUN [mmol/L,MQ1Q3)] 6.38[4.54,8.63] 6.13[4.77,9.71] Z=-0.703 0.817
β2 微球蛋白 [mg/L,MQ1Q3)] 3.42[2.76,4.48] 3.67[3.00,5.28] Z=-1.876 0.058
MELD 评分 [ 分,MQ1Q3)] 18.64[12.73,24.69] 18.25[12.72,25.96] Z=-0.482 0.694
图2 LASSO回归分析中肝硬化患者AKI风险预测因子的系数路径图。图a为回归系数变化趋势。随着λ值增大,各预测变量的回归系数逐步缩小。图b为二项式偏差值变化趋势。二项式偏差值随着λ值的增加先减少,达到最小值后再逐步上升
图3 LASSO模型在训练集和验证集中的校准曲线及DCA曲线。图a为模型在训练集上的校准曲线;图b为模型在验证集上的校准曲线;图c为模型在训练集上的DCA曲线;图d为模型在验证集上的DCA曲线 注:图a和b为模型在训练集和验证集上的校准曲线;图c和d为模型在训练集和验证集上的DCA曲线
图4 LASSO回归模型在训练集与验证集中的ROC曲线图。图a为模型在训练集上的ROC曲线图;图b为模型在验证集上的ROC曲线图
表2 验证集中不同ML模型性能比较
图5 不同AutoML模型ROC曲线对比
图6 GBM模型在验证集中的变量重要性分析和学习曲线。图a为变量重要性贡献图;图b为学习曲线图
图7 基于GBM模型开发的App在测试集上的性能评估。图a为使用App在测试集上的ROC曲线图;图b为App在测试集上预测的混淆矩阵图
图8 GBM模型在测试集上的SHAP特征分析
图9 测试集中GBM模型的力图分析。图a为预测为发生AKI的概率为71%;图b为预测为不发生AKI的概率为85%
1
Wang F, Fan J, Zhang Z, et al. The global burden of liver disease: the major impact of China [J]. Hepatology (Baltimore, Md.), 2014, 60(6):2099-2108.
2
Angeli P, Garcia-Tsao G, Nadim MK, et al. News in pathophysiology,definition and classification of hepatorenal syndrome: A step beyond the International Club of Ascites (ICA) consensus document [J]. J Hepatol, 2019, 71(4): 811-822.
3
Piano S, Rosi S, Maresio G, et al. Evaluation of the Acute Kidney Injury Network criteria in hospitalized patients with cirrhosis and ascites [J]. J Hepatol, 2013, 59(3): 482-489.
4
Durand F, Olson JC, Nadim MK. Renal dysfunction and cirrhosis [J].Curr Opin Crit Care, 2017, 23(6): 457-462.
5
Desai AP, Knapp SM, Orman ES, et al. Changing epidemiology and outcomes of acute kidney injury in hospitalized patients with cirrhosisa US population-based study [J]. J Hepatol, 2020, 73(5): 1092-1099.
6
王朋, 李泽宇, 蒋亚婷, 等. MELD评分、ALBI评分联合β2-微球蛋白对肝硬化合并急性肾损伤的预测价值 [J]. 临床肝胆病杂志,2023, 39(12): 2839-2844.
7
潮燕, 陈聪. 基于Nomogram建立肝硬化并发急性肾损伤的风险评估模型 [J]. 中南医学科学杂志, 2023, 51(5): 771-774.
8
Bewick V, Cheek L, Ball J. Statistics review 14: Logistic regression [J].Crit Care, 2005, 9(1): 112-118.
9
Siriborvornratanakul T. Human behavior in image-based road health inspection systems despite the emerging autoML [J]. J Big Data, 2022,9(1): 96.
10
Alsharef A, Aggarwal K, Kumar M, et al. Review of ML and autoML solutions to forecast time-series data [J]. Arch Comput Methods Eng,2022, 29(7): 5297-5311.
11
王甘红, 陈健, 沈支佳, 等. 基于自动化机器学习建立结肠镜肠道准备失败风险预测模型及评价 [J]. 中国内镜杂志, 2024, 30(5):36-47.
12
Shi Y, Lin J, Zhu J, et al. Predicting the recurrence of common bile duct stones after ERCP treatment with automated machine learning algorithms [J]. Dig Dis Sci, 2023, 68(7): 2866-2877.
13
中华医学会消化病学分会. 中国肝硬化临床诊治共识意见 [J]. 中华消化杂志, 2023, 43(4): 227-247.
14
Flamm SL, Wong F, Ahn J, et al. AGA clinical practice update on the evaluation and management of acute kidney injury in patients with cirrhosis: expert review [J]. Clin Gastroenterol Hepatol, 2022, 20(12):2707-2716.
15
Chen Y, Liu X, Gao L, et al. Using the H2O automatic machine learning algorithms to identify predictors of web-based medical record nonuse among patients in a data-rich environment: mixed methods study [J]. JMIR Med Inform, 2023, 11: e41576.
16
Nohara Y, Matsumoto K, Soejima H, et al. Explanation of machine learning models using shapley additive explanation and application for real data in hospital [J]. Comput Methods Programs Biomed, 2022,214: 106584.
17
Fahmy AS, Csecs I, Arafati A, et al. An explainable machine learning approach reveals prognostic significance of right ventricular dysfunction in nonischemic cardiomyopathy [J]. JACC Cardiovasc Imaging, 2022, 15(5): 766-779.
18
Tariq R, Hadi Y, Chahal K, et al. Incidence, mortality and predictors of acute kidney injury in patients with cirrhosis: a systematic review and meta-analysis [J]. J Clin Transl Hepatol, 2020, 8(2): 135-142.
19
Ning Y, Zou X, Xu J, et al. Impact of acute kidney injury on the risk of mortality in patients with cirrhosis: a systematic review and metaanalysis [J]. Ren Fail, 2022, 44(1): 1-14.
20
赵静涵, 陈玉龙, 张琛. 超声参数与UPJO致肾积水患儿肾功能的相关性分析 [J/OL]. 中华腔镜泌尿外科杂志(电子版), 2023, 17(4):372-376.
21
Mindikoglu AL, Dowling TC, Magder LS, et al. Estimation of glomerular filtration rate in patients with cirrhosis by using new and conventional filtration markers and dimethylarginines [J]. Clin Gastroenterol Hepatol, 2016, 14(4): 624-632.e2.
22
Sawada Y, Shiraki M, Iwasa M, et al. The effects of diuretic use and the presence of ascites on muscle cramps in patients with cirrhosis: a nationwide study [J]. J Gastroenterol, 2020, 55(9): 868-876.
23
Montano-Loza AJ, Meza-Junco J, Prado CM, et al. Muscle wasting is associated with mortality in patients with cirrhosis [J]. Clin Gastroenterol Hepatol, 2012, 10(2): 166-73, 173.e1.
24
Wang R, Hu H, Hu S, et al. β2-microglobulin is an independent indicator of acute kidney injury and outcomes in patients with intracerebral hemorrhage [J]. Medicine (Baltimore), 2020, 99(8):e19212.
25
陈杰桓, 许志荣, 刘颖培, 等. 超声造影对急性肾损伤治疗后肾血流灌注水平的评价 [J/OL]. 中华腔镜泌尿外科杂志(电子版),2023, 17(1):58-62.
26
褚雪倩, 周炜, 黄萱, 等. 乙肝肝硬化腹水患者发生急性肾损伤的影响因素及预测模型 [J]. 临床肾脏病杂志, 2024, 24(8): 623-628.
27
Murphree DH, Quest DJ, Allen RM, et al. Deploying predictive models in a healthcare environment - an open source approach [J].Annu Int Conf IEEE Eng Med Biol Soc, 2018, 2018: 6112-6116.
[1] 李培真, 刘海亮, 李大伟, 贾昊, 张泽瑾, 刘力维, 申传安. 重度烧伤患者发生早期急性肾损伤危险因素分析及预测模型建立[J/OL]. 中华损伤与修复杂志(电子版), 2025, 20(03): 199-205.
[2] 钱小梅, 罗洪, 李智慧, 周代君, 李东. 76例乙型肝炎肝硬化并发原发性肝癌的高危因素Logistic分析[J/OL]. 中华普外科手术学杂志(电子版), 2025, 19(03): 251-253.
[3] 石洋, 郭辉, 李坚. 门静脉高压症TIPS操作规范及其手术相关并发症处理[J/OL]. 中华肝脏外科手术学电子杂志, 2025, 14(03): 372-378.
[4] 罗沙, 邱乐乐, 于祥, 冯哲, 蔡广研, 黄静. 无创容量评估新技术在急性肾损伤患者连续性肾脏替代治疗中的应用研究[J/OL]. 中华肾病研究电子杂志, 2025, 14(02): 61-67.
[5] 李菲, 郭晓夏, 郑悦, 郑爔, 李鑫成, 李文雄. 他汀类药物对甘油三酯葡萄糖指数增高的脓毒症相关急性肾损伤患者预后的影响[J/OL]. 中华肾病研究电子杂志, 2025, 14(02): 68-76.
[6] 张佳馨, 刘洋, 吴贞, 程庆砾, 敖强国. 高钠血症对老年院内急性肾损伤患者短期预后的影响[J/OL]. 中华肾病研究电子杂志, 2025, 14(02): 83-90.
[7] 张可颖, 冀雨薇, 付章宁, 张益帆, 王晓晨, 杨滟, 陈香美, 蔡广研, 洪权. 人参皂苷Rb1 预处理间充质干细胞的转录组分析及急性肾损伤治疗关键基因挖掘[J/OL]. 中华肾病研究电子杂志, 2025, 14(01): 26-33.
[8] 何许巍, 刘洋, 程庆砾, 敖强国. 急性肾损伤早期生物标志物即时检测的研究进展[J/OL]. 中华肾病研究电子杂志, 2025, 14(01): 48-52.
[9] 辛宇, 王常松, 于凯江. 重症肾脏:2024年度进展与展望[J/OL]. 中华重症医学电子杂志, 2025, 11(01): 17-21.
[10] 王冉, 常炜, 杨毅, 徐静媛. 脓毒症休克液体复苏的晶体液种类对急性肾损伤影响的研究进展[J/OL]. 中华重症医学电子杂志, 2025, 11(01): 100-104.
[11] 王玉琳, 王中华, 刘子文, 吕梦鑫, 于源滋, 李涛, 胡锦华, 张小茜. 肝硬化消化道出血患者经颈静脉肝内门体分流术前后肠道微生态的宏基因组学分析[J/OL]. 中华消化病与影像杂志(电子版), 2025, 15(02): 135-143.
[12] 陈镇, 余天垒, 刘晓超, 尹萌. 肝硬化门静脉高压伴上消化道急性出血患者的血细胞参数、肾功能及营养评分变化及其与预后的相关性[J/OL]. 中华消化病与影像杂志(电子版), 2025, 15(02): 149-154.
[13] 张志硕, 王紫仪, 马宁. MSCT 肝脏体积测量评估肝硬化患者肝脏功能代偿状态的可行性研究[J/OL]. 中华消化病与影像杂志(电子版), 2025, 15(02): 155-161.
[14] 颜世锐, 熊辉. 感染性心内膜炎合并急性肾损伤患者的危险因素探索及死亡风险预测[J/OL]. 中华临床医师杂志(电子版), 2024, 18(07): 618-624.
[15] 原江东, 周丽珍, 段鹏程, 许婕璇, 黎江, 卢慕荣. 国际标准化比值与脓毒症相关急性肾损伤患者死亡率的关系[J/OL]. 中华临床实验室管理电子杂志, 2025, 13(01): 11-16.
阅读次数
全文


摘要


AI


AI小编
你好!我是《中华医学电子期刊资源库》AI小编,有什么可以帮您的吗?