切换至 "中华医学电子期刊资源库"

中华普外科手术学杂志(电子版) ›› 2026, Vol. 20 ›› Issue (06) : 576 -581. doi: 10.3877/cma.j.issn.1674-3946.2026.06.015

论著

联合临床特征与血清学指标的机器学习模型预测急性胰腺炎持续性器官衰竭
李东, 董烈峰†(), 赵文月, 宋振晴   
  1. 250000 济南,山东大学附属山东省立第三医院消化内科
  • 收稿日期:2026-05-14 出版日期:2026-12-26
  • 通信作者: 董烈峰

A machine learning model based on combined clinical features and serological indicators for predicting persistent organ failure in acute pancreatitis

Dong Li, Liefeng Dong†(), Wenyue Zhao, Zhenqing Song   

  1. Department of Gastroenterology, Shandong Provincial Third Hospital Affiliated to Shandong University, Jinan Shandong Province 250000, China
  • Received:2026-05-14 Published:2026-12-26
  • Corresponding author: Liefeng Dong
  • Supported by:
    Shandong Provincial Medical and Health Science and Technology Project(202403030224)
引用本文:

李东, 董烈峰, 赵文月, 宋振晴. 联合临床特征与血清学指标的机器学习模型预测急性胰腺炎持续性器官衰竭[J/OL]. 中华普外科手术学杂志(电子版), 2026, 20(06): 576-581.

Dong Li, Liefeng Dong, Wenyue Zhao, Zhenqing Song. A machine learning model based on combined clinical features and serological indicators for predicting persistent organ failure in acute pancreatitis[J/OL]. Chinese Journal of Operative Procedures of General Surgery(Electronic Edition), 2026, 20(06): 576-581.

目的

 研究基于临床特征-血清学指标的机器学习模型构建急性胰腺炎(AP)继发持续器官功能衰竭(POF)的预测模型。

方法

回顾性收集2019年10月至2024年12月350例AP患者病例资料,按7∶3分为训练集(n=245)和内部验证集(n=105),另纳入新城分院同期82例患者作为测试集。符合正态分布的计量资料采用(

±s)描述,行独立样本t检验;计数资料以[例(%)]描述,行χ2检验;经LASSO回归筛选变量,构建逻辑回归(LR)、随机森林(RF)、支持向量机(SVM)模型,以受试者工作特征(ROC)曲线、曲线下面积(AUC)优选模型;通过测试集验证模型泛化能力。

结果

LASSO回归筛选出的AP继发POF相关变量包括体重指数、急性生理与慢性健康系统Ⅱ(APACHEⅡ)评分、随机血糖、白细胞介素6(IL-6)、C反应蛋白(CRP)、乳酸脱氢酶(LDH)、中性粒细胞计数/淋巴细胞计数(NLR)。基于临床特征构建的LR,RF,SVM模型的AUC分别为0.700、0.862、0.707;基于临床特征-血清学指标构建的LR,RF,SVM模型的AUC分别为0.805、0.904、0.840。临床特征-血清学指标在3种机器学习模型中的AUC均优于单独的临床特征(P<0.05)。内部验证集中基于临床特征-血清学指标的RF模型预测AP继发POF的AUC为0.899。测试集数据绘制ROC曲线,结果显示,基于临床特征-血清学指标的RF模型预测AP继发POF的AUC为0.896,灵敏度为80.7%,特异度为87.8%,准确度为85.7%,精确率为73.5%,F1为0.831。

结论

基于临床特征(体重指数、APACHEⅡ评分、随机血糖)与血清学指标(IL-6、CRP、LDH、NLR)联合的RF模型预测AP继发POF性能优良、泛化能力可靠,可用于早期识别高危患者与风险分层。

Objective

To construct a predictive model for acute pancreatitis (AP) secondary to persistent organ failure (POF) based on clinical characteristics and serological indicators using machine learning.

Methods

A total of 350 AP patients from October 2019 to December 2024 were retrospectively collected and divided into a training set (n=245) and an internal validation set (n=105) in a 7∶3 ratio. Another 82 patients from Xincheng Hospital during the same period were included as the test set. Quantitative data with normal distribution were described as (

±s) and analyzed using independent sample t tests; count data were described as [cases (%)] and analyzed using the χ2 test. Variables were selected using LASSO regression and models of logistic regression (LR), random forest (RF), and support vector machine (SVM) were constructed to optimize the model using the area under the receiver operating characteristic curve (AUC). The generalization ability of the models was verified using the test set.

Results

The AP secondary POF-related variables selected by LASSO regression included body mass index, Acute Physiology and Chronic Health Evaluation Ⅱ (APACHE Ⅱ) score, random blood glucose, interleukin 6 (IL-6), C-reactive protein (CRP), lactate dehydrogenase (LDH), and neutrophil count/lymphocyte count (NLR). The AUC of the LR, RF, and SVM models based on clinical characteristics were 0.700, 0.862, and 0.707, respectively; the AUC of the LR, RF, and SVM models based on clinical characteristics and serological indicators were 0.805, 0.904, and 0.840, respectively. The AUC of the serological indicators-based models in the three machine learning models were all better than those of the clinical characteristics alone (P < 0.05). The AUC of the RF model based on clinical characteristics and serological indicators in the internal validation set for predicting AP secondary POF was 0.899. The ROC curve was plotted using the test set data, and the results showed that the AUC value of the RF model based on clinical characteristics and serological indicators for predicting AP secondary POF was 0.896, with a sensitivity of 80.7%, specificity of 87.8%, accuracy of 85.7%, precision of 73.5%, and F1 of 0.831.

Conclusion

The RF model based on the combination of clinical characteristics (body mass index, APACHE II score, random blood glucose) and serological indicators has excellent performance and reliable generalization ability for predicting AP secondary POF, and can be used for early identification of high-risk patients and risk stratification.

表1 训练集、内部验证集及测试集患者资料比较
表2 训练集PSM匹配后POF组与Non-POF组核心指标比较(
±s)
图1 LASSO回归筛选急性胰腺炎继发持续器官功能衰竭相关变量 注:APACHEⅡ评分为急性生理与慢性健康系统Ⅱ评分;CRP为C反应蛋白;NLR为中性粒细胞计数/淋巴细胞计数;IL-6为白细胞介素6;LDH为乳酸脱氢酶
图2 7个核心变量的Pearson相关性热图 注:NLR为中性粒细胞计数/淋巴细胞计数;IL-6为白细胞介素6;CRP为C反应蛋白;APACHEⅡ评分为急性生理与慢性健康系统Ⅱ评分;LDH为乳酸脱氢酶
图3 优化指标RF模型中核心变量的重要性排序(Gini系数) 注:RF为随机森林;NLR为中性粒细胞计数/淋巴细胞计数;APACHEⅡ评分为急性生理与慢性健康系统Ⅱ评分;IL-6为白细胞介素6;LDH为乳酸脱氢酶;CRP为C反应蛋白,其中NLR为模型中权重最高的预测指标
图5 基于临床特征-血清学指标的受试者工作曲线
表3 3种模型对AP继发POF的预测效能
图6 联合RF模型的校准曲线在训练集和内部验证集上的校准曲线 注:RF为随机森林;训练集Brier=0.064;内部验证集Brier=0.077
图7 内部验证集中基于临床特征-血清学指标的受试者工作曲线 注:RF为随机森林
图8 测试集中基于临床特征-血清学指标的受试者工作曲线 注:RF为随机森林
[1]
Tang J,Chen T,Ni W,et al. Dynamic nomogram for persistent organ failure in acute biliary pancreatitis: Development and validation in a retrospective study[J]. Dig Liver Dis,2022,54(6): 805-811.
[2]
Mahapatra SJ,Garg PK. Organ Failure and Prediction of Severity in Acute Pancreatitis[J]. Gastroenterol Clin North Am,2025,54(1): 1-19.
[3]
中华医学会急诊医学分会,上海市医学会急诊专科分会. 急性胰腺炎急诊诊治专家共识[J]. 中华急诊医学杂志,2024,33(4): 470-479.
[4]
Pandey R,Pathak R,Jha A,et al. Comparison of Acute Physiology and Chronic Health Evaluation Ⅱ,Bedside Index for Severity in Acute Pancreatitis and Modified Computed Tomography Severity Index Scores in Predicting the Outcome in Acute Pancreatitis in a Tertiary Care Centre in Nepal[J]. J Nepal Health Res Counc,2023,21(2): 203-206.
[5]
Capurso G,Ponz de Leon Pisani R,Lauri G,et al. Clinical usefulness of scoring systems to predict severe acute pancreatitis: A systematic review and meta-analysis with pre and post-test probability assessment[J]. United European Gastroenterol J,2023,11(9): 825-836.
[6]
中华医学会急诊分会,京津冀急诊急救联盟,北京医学会急诊分会,等. 急性胰腺炎急诊诊断及治疗专家共识[J]. 临床肝胆病杂志,2021,37(5): 1034-1041.
[7]
Leppäniemi A,Tolonen M,Tarasconi A,et al. 2023 WSES guidelines for the management of severe acute pancreatitis[J]. World J Emerg Surg,2023,18(1): 38.
[8]
Machicado JD,Gougol A,Tan X,et al. Mortality in acute pancreatitis with persistent organ failure is determined by the number,type,and sequence of organ systems affected[J]. United European Gastroenterol J,2021,9(2): 139-149.
[9]
刘德仁,刘加宁,苏峰. 入院血清肝素结合蛋白水平与急性胰腺炎持续器官衰竭的相关研究[J]. 现代消化及介入诊疗,2021,26(10): 1225-1229.
[10]
Zhang Y,Peng W,Zheng X. The prognostic value of the combined neutrophil-to-lymphocyte ratio(NLR)and neutrophil-to-platelet ratio(NPR)in sepsis[J]. Sci Rep,2024,14(1): 15075-15084.
[11]
陈系东,王绍闯,赵何伟,等. 高龄BCLC B期肝癌患者常规TACE术后急性肝功能恶化的危险因素研究[J/OL]. 中华普外科手术学杂志(电子版),2025,19(6): 689-692.
[12]
Zhou J,Yang R,Zhu W,et al. NLR inflammasome pathways: key targets for pathogenesis and therapy of metabolic diseases[J]. Front Immunol,2026,17(1): 1823724-1823749.
[13]
王迎芝,张先进,杨士彦. 基于Logistic回归分析急性胰腺炎并发持续性器官功能衰竭的因素[J]. 肝胆外科杂志,2022,30(5): 365-369.
[14]
中华医学会外科学分会胰腺外科学组. 中国急性胰腺炎诊治指南(2021)[J]. 中华消化外科杂志,2021,20(7): 730-739.
[15]
张洁,罗小霞,余鸿. 系统性免疫炎症指数对急性胰腺炎患者并发器官功能损伤的预测价值[J/OL]. 中华普外科手术学杂志(电子版),2025,19(1): 68-71.
[16]
Gu K,Wang Q. Establishment and Validation of a dynamic nomogram for persistent organ failure in acute biliary pancreatitis: a retrospective study[J]. J Inflamm Res,2024,17(1): 8513-8530.
[17]
Guo P,Liu L,Yang X,et al. Irisin improves BBB dysfunction in SAP rats by inhibiting MMP-9 via the ERK/NF-κB signaling pathway[J]. Cell Signal,2022,93(1): 110300.
[18]
罗竹. 基于机器学习的随机森林模型预测重症急性胰腺炎的价值[D]. 温州医科大学,2022.
[1] 卞文霞, 黄东亚. ICU急性胰腺炎患者早期血清阴离子间隙与谵妄发生风险的关联研究[J/OL]. 中华危重症医学杂志(电子版), 2026, 19(04): 315-321.
[2] 刘传于, 胡建国, 王荔, 许鑫玥, 刁帅, 杨茜, 胡丽娜. 特殊类型宫颈腺癌患者的多中心回顾性临床分析[J/OL]. 中华妇幼临床医学杂志(电子版), 2026, 22(02): 128-139.
[3] 刘美, 杨永辉, 朱桂云, 马伟立, 韩云霄. CXC趋化因子受体3及其配体表达水平与肺结核患者临床特征及病变程度的相关性[J/OL]. 中华实验和临床感染病杂志(电子版), 2026, 20(03): 155-166.
[4] 王维凤, 查桂芳, 袁茂月, 牟一鑫, 杨青. 567例百日咳患儿临床特征[J/OL]. 中华实验和临床感染病杂志(电子版), 2026, 20(03): 182-187.
[5] 赵欣, 李文强, 杨莹. 宏基因组二代测序用于非结核分枝杆菌肺病的诊断价值及临床特征分析[J/OL]. 中华肺部疾病杂志(电子版), 2026, 19(04): 632-637.
[6] 王家瑶, 叶开婷, 祝清清, 费黎明, 董佳慧. 54例重症鹦鹉热衣原体肺炎临床特征及预后回顾性分析[J/OL]. 中华肺部疾病杂志(电子版), 2026, 19(03): 444-450.
[7] 汤亚南, 刘晓亮, 马秀英, 赵璐珊, 王丽颖, 周鹏翔, 盖晓燕, 邢燕. 577例儿童社区获得性肺炎临床特征及诊疗现状回顾性分析[J/OL]. 中华肺部疾病杂志(电子版), 2026, 19(02): 303-308.
[8] 叶金宝, 梁声强, 马建新, 林小强, 张海森. 小野寺预后营养指数对肝癌复发预测价值[J/OL]. 中华肝脏外科手术学电子杂志, 2026, 15(03): 372-378.
[9] 王浩, 张伟杰, 员百慧, 宋知远, 于艳龙. 甘油三酯葡萄糖指数预测急性胰腺炎严重程度Meta分析[J/OL]. 中华肝脏外科手术学电子杂志, 2026, 15(03): 405-412.
[10] 杨宁, 张立翱, 李仁君, 金水. 非遗传性结直肠息肉病患者的临床特征及息肉复发的影响因素分析[J/OL]. 中华结直肠疾病电子杂志, 2026, 15(02): 168-174.
[11] 卢毅, 王国兴, 谢苗荣. “急救理念”在重症急性胰腺炎救治中的作用和地位[J/OL]. 中华重症医学电子杂志, 2026, 12(02): 134-140.
[12] 方振东, 黄伟鹏, 高伊帆, 虞洪, 沈波, 郭丰. 腹腔镜腹膜后胰腺坏死清创术治疗感染性胰腺坏死的预后研究[J/OL]. 中华重症医学电子杂志, 2026, 12(02): 141-146.
[13] 黄东亚, 侯超群, 李强. 急性胰腺炎患者红细胞分布宽度与白蛋白比值对院内死亡风险的影响[J/OL]. 中华重症医学电子杂志, 2026, 12(02): 147-153.
[14] 任伟洁, 毛妮, 胡宁丹, 沈锋. KLF4调控TXNIP介导线粒体氧化应激激活AR42J细胞NLRP3炎症小体的机制研究[J/OL]. 中华消化病与影像杂志(电子版), 2026, 16(04): 294-301.
[15] 韩一梅, 冯仕川, 陈志娟. 高脂血症性急性胰腺炎复发的危险因素及其列线图预测模型构建[J/OL]. 中华消化病与影像杂志(电子版), 2026, 16(03): 222-228.
阅读次数
全文


摘要


AI
小
编
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?