切换至 "中华医学电子期刊资源库"

第五届中国出版政府奖音像电子网络出版物奖提名奖

中国科技核心期刊

中国科学引文数据库(CSCD)来源期刊

中华重症医学电子杂志 ›› 2025, Vol. 11 ›› Issue (03) : 267 -277. doi: 10.3877/cma.j.issn.2096-1537.2025.03.009

重症医学研究

基于可解释机器学习构建成人心脏瓣膜术后转入ICU行机械通气治疗患者发生PMV的预测模型
邓猛1, 张星星2,(), 李晓青1, 俞云1, 王文春1, 李海亮1, 黄海云1, 孟翔飞1, 马文1, 潘陈伟1   
  1. 1 210009 南京,江苏省重症医学重点实验室 东南大学附属中大医院重症医学科
    2 210000 南京,中国药科大学附属浦口中医院护理部
  • 收稿日期:2024-09-24 出版日期:2025-08-28
  • 通信作者: 张星星

Development of an interpretable machine learning model for predicting prolonged mechanical ventilation in adults after cardiac valve surgery

Meng Deng1, Xingxing Zhang2,(), Xiaoqing Li1, Yun Yu1, Wenchun Wang1, Hailiang Li1, Haiyun Huang1, Xiangfei Meng1, Wen Ma1, Chenwei Pan1   

  1. 1 Department of Critical Care Medicine, Jiangsu Provincial Key Laboratory of Critical Care Medicine, Zhongda Hospital, School of Medicine, Southeast University, Nanjing 210009, China
    2 Department of Nursing, Pukou Chinese Medicine Hospital Affiliated to China Pharmaceutical University, Nanjing 210000, China
  • Received:2024-09-24 Published:2025-08-28
  • Corresponding author: Xingxing Zhang
引用本文:

邓猛, 张星星, 李晓青, 俞云, 王文春, 李海亮, 黄海云, 孟翔飞, 马文, 潘陈伟. 基于可解释机器学习构建成人心脏瓣膜术后转入ICU行机械通气治疗患者发生PMV的预测模型[J/OL]. 中华重症医学电子杂志, 2025, 11(03): 267-277.

Meng Deng, Xingxing Zhang, Xiaoqing Li, Yun Yu, Wenchun Wang, Hailiang Li, Haiyun Huang, Xiangfei Meng, Wen Ma, Chenwei Pan. Development of an interpretable machine learning model for predicting prolonged mechanical ventilation in adults after cardiac valve surgery[J/OL]. Chinese Journal of Critical Care & Intensive Care Medicine(Electronic Edition), 2025, 11(03): 267-277.

目的

利用机器学习模型预测成人心脏瓣膜术后转入ICU接受机械通气治疗的患者发生延长机械通气(PMV)的风险。

方法

回顾性收集东南大学附属中大医院2023年7月1日至2024年6月30日期间,心脏瓣膜术后转入ICU并接受机械通气治疗的患者的一般资料及临床资料。通过LASSO回归进行特征筛选,并将数据按照7∶3的比例随机分为训练集和验证集,并进一步通过多因素Logistic回归排除混杂因素。通过比较不同机器学习模型的受试者工作特征曲线(ROC)的曲线下面积(AUC)、准确率、敏感度和特异度F1值,Logistic回归模型被确定为最佳预测模型。使用沙普利加性解释(SHapley additive explanation,SHAP)方法对模型的结果进行解释。

结果

在纳入的711例患者中,有143例发生了PMV,发生率为20.1%。Logistic回归模型在预测PMV方面表现出较好的AUC(训练集:0.903,95%CI:0.869~0.938;验证集:0.899,95%CI:0.854~0.943)。SHAP分析表明,模型中变量重要性从高到低依次为:术后6 h乳酸值≥4 mmol/L、新发房颤、体外循环时间≥180 min、手术类型、平均肺动脉压力(MPAP)≥35 mmHg(1 mmHg=0.133 kPa)、射血分数(EF)≤0.35、氧分压<90 mmHg、术前肾功能衰竭、二次瓣膜置换、术后6 h引流量≥600 ml及急性生理学和慢性健康状况评价(APACHEⅡ)≥21分。

结论

本研究构建的模型在预测成人心脏瓣膜术后转入ICU接受机械通气治疗患者发生PMV方面表现出良好的性能。此外,SHAP方法有效地解释了模型结果,为重症医务人员提供了早期快速识别心脏瓣膜手术患者发生PMV风险的依据,有助于缩短患者的机械通气时间。

Objective

To develop and validate an interpretable machine learning model for predicting the risk of prolonged mechanical ventilation (PMV) in adult patients requiring mechanical ventilation after cardiac valve surgery in ICU.

Methods

A retrospective cohort study of patients admitted to ICU after cardiac valve surgery who received mechanical ventilation at a tertiary hospital between July 1, 2023 and June 30, 2024, was conducted. Feature selection was performed using Least Absolute Shrinkage and Selection Operator (LASSO) regression. The dataset was randomly split into a training set and a validation set in a ratio of 7∶3. Multivariate logistic regression was used to adjust for potential confounders. Several ML models were evaluated and compared based on the area under the ROC curve (AUC), accuracy, sensitivity, specificity and F1 score. The optimal model was interpreted using SHapley Additive exPlanations (SHAP).

Results

Among the 711 patients included, 143(20.1%) developed PMV. The logistic regression model demonstrated excellent discriminative ability, with an AUC for 0.903 (95%CI: 0.869-0.938) in the training set and 0.899 (95%CI: 0.854-0.943) in the validation set. SHAP analysis identified the following top predictors of PMV (in descending order of importance): lactate level ≥ 4 mmol/L at 6 hours post-operation, new-onset atrial fibrillation, cardiopulmonary bypass time ≥ 180 minutes, type of surgery, mean pulmonary artery pressure (MPAP) ≥ 35 mmHg, ejection fraction (EF) ≤ 0.35, partial pressure of oxygen < 90 mmHg, preoperative renal failure, re-do valve replacement, chest tube drainage ≥ 600 ml at 6 hours post-operation, and APACHE Ⅱ score≥ 21.

Conclusion

We developed an interpretable prediction model that performs well in identifying adult patients at high risk for PMV following cardiac valve surgery. The use of SHAP enhances the clinical utility of the model by providing insights into key risk factors, thereby aiding ICU clinicians in early risk stratification and potentially facilitating interventions to shorten the duration of mechanical ventilation.

表1 训练集和验证集患者基线特征比较[例(%)]
基本情况 总数(711例) 验证集(214例) 训练集(497例) χ2 P
年龄 0.271 0.601
18~59岁 249(35.02) 78(36.45) 171(34.41)
≥60岁 462(64.98) 136(63.55) 326(65.59)
性别 1.739 0.187
309(43.46) 101(47.20) 208(41.85)
402(56.54) 113(52.80) 289(58.15)
心功能分级 1.613 0.205
Ⅰ~Ⅲ级 628(88.33) 194(90.65) 434(87.32)
Ⅳ级 83(11.67) 20(9.35) 63(12.68)
手术类型 0.022 0.884
≥2种术式 316(44.44) 96(44.86) 220(44.27)
单纯瓣膜手术 395(55.56) 118(55.14) 277(55.73)
APACHEⅡ 0.091 0.765
<21分 648(91.14) 194(90.65) 454(91.35)
≥21分 63(8.86) 20(9.35) 43(8.65)
高血压 0.936 0.333
343(48.17) 109(50.93) 234(47.08)
368(51.83) 105(49.07) 263(52.92)
二次瓣膜手术 0.227 0.630
666(93.67) 199(92.99) 467(93.96)
45(6.33) 15(7.01) 30(6.04)
肾功能衰竭 0.024 0.889
679(95.49) 204(95.33) 475(95.57)
32(4.51) 10(4.67) 22(4.44)
体外循环时间 2.367 0.124
<180 min 678(95.35) 208(97.20) 470(94.56)
≥180 min 33(4.65) 6(2.80) 27(5.44)
输注库存血 0.842 0.358
525(73.80) 153(71.50) 372(74.80)
186(26.20) 61(28.50) 125(25.20)
6 h引流量 0.914 0.341
<600 ml 665(93.53) 203(94.86) 462(92.94)
≥600 ml 46(6.47) 11(5.14) 35(7.06)
P/F 1.944 0.164
<200 mmHg 81(11.39) 19(8.88) 62(12.48)
≥200 mmHg 630(88.61) 195(91.12) 435(87.52)
PO2 2.902 0.089
<90 mmHg 78(10.97) 17(7.94) 61(12.28)
≥90 mmHg 633(89.03) 197(92.06) 436(87.72)
EF值 1.956 0.161
<0.35 55(7.74) 12(5.61) 43(8.66)
≥0.35 656(92.26) 202(94.39) 454(91.34)
MPAP 0.328 0.566
<35 mmHg 483(67.93) 142(66.36) 341(68.61)
≥35 mmHg 228(32.07) 72(33.64) 156(31.39)
新发房颤 0.078 0.784
664(93.38) 199(92.99) 465(93.56)
47(6.62) 15(7.01) 32(6.44)
术后6 h Lac 0.293 0.587
<4 mmol/L 539(75.80) 165(77.10) 374(75.25)
≥4 mmol/L 172(24.20) 49(22.90) 123(24.75)
血清肌酐 0.890 0.346
<177 μmol/L 685(96.34) 204(95.33) 481(96.78)
≥177 μmol/L 26(3.66) 10(4.67) 16(3.22)
白蛋白 2.380 0.123
<35 g/L 373(52.47) 103(48.13) 270(54.32)
≥35 g/L 338(47.53) 111(51.87) 227(45.68)
PMV 1.061 0.304
568(79.89) 176(82.24) 392(78.87)
143(20.11) 38(17.76) 105(21.13)
肺部感染 3.161 0.076
690(97.05) 204(95.33) 486(97.79)
21(2.95) 10(4.67) 11(2.21)
COPD 0.030 0.866
697(98.03) 209(97.66) 488(98.19)
14(1.97) 5(2.34) 9(1.81)
胸腔积液 0.341 0.562
676(95.08) 205(95.79) 471(94.77)
35(4.92) 9(4.21) 26(5.23)
吸烟史 0.804 0.371
566(79.61) 175(81.78) 391(78.68)
145(20.39) 39(18.22) 106(21.32)
饮酒史 0.019 0.901
593(83.40) 179(83.64) 414(83.29)
118(16.60) 35(16.36) 83(16.71)
合并脑梗 0.343 0.559
584(82.13) 179(83.64) 405(81.49)
127(17.87) 35(16.36) 92(18.51)
围手术期IABP 0.071 0.785
705(99.16) 213(99.53) 492(98.99)
6(0.84) 1(0.47) 5(1.01)
图1 LASSO回归选择特征因子。图a:使用10折交叉验证在选定的值处绘制垂直线,其中最优的值产生18个非零系数;图b:在套索模型中,从log(λ)序列中提取了27个特征的系数轮廓。垂直虚线以最小均方误差(λ=0.006)和最小距离的标准误差绘制
图2 多因素Logistic回归森林图(P<0.05) 注:APACHEⅡ为急性生理与慢性健康状况评分;EF为射血分数;MPAP为平均肺动脉压力;Lac为血气乳酸值
图3 分类多模型的综合分析。图a为训练集多模型ROC曲线;图b为验证集多模型ROC曲线;图c为Logistic回归分析校正曲线;图d为Logistic回归分析决策曲线 注:PMV为延长机械通气;ROC为受试者工作特征曲线
表2 验证集数据模型性能对比表
图4 SHAP解释模型。图a为在SHAP中的特征属性,每一行代表一个特征,横坐标是SHAP值,其中紫点代表较低的特征值,黄点代表较高的特征值;图b为特征重要性排名矩阵图,显示每个协变量在最终预测模型开发中的重要性,由SHAP表示;图c为非PMV患者SHAP得分;图d为PMV患者SHAP得分 注:MPAP为平均肺动脉压力;EF为射血分数;APACHEⅡ为急性生理与慢性健康状况评分;PO2为动脉血氧分压;Lac为血气乳酸值;PMV为延长机械通气
1
Baskett RJ, Exner DV, Hirsch GM, et al. Mitral insufficiency and morbidity and mortality in left ventricular dysfunction [J]. Can J Cardiol, 2007, 23(10): 797-800.
2
Hirasawa K, Izumo M, Akashi YJ. Stress echocardiography in valvular heart disease [J]. Front Cardiovasc Med, 2023, 10: 1233924.
3
Udzik J, Sienkiewicz S, Biskupski A, et al. Cardiac complications following cardiac surgery procedures [J]. J Clin Med, 2020, 9(10): 3347.
4
Kotfis K, Szylińska A, Listewnik M, et al. Balancing intubation time with postoperative risk in cardiac surgery patients-a retrospective cohort analysis [J]. Ther Clin Risk Manag, 2018, 14: 2203-2212.
5
O'Brien SM, Feng L, He X, et al. The Society of Thoracic Surgeons 2018 adult cardiac surgery risk models: part 2-statistical methods and results [J]. Ann Thorac Surg, 2018, 105(5): 1419-1428.
6
Jing X, Zhu Z, Fan H, et al. Impact of delay extubation on the reintubation rate in patients after cervical spine surgery: a retrospective cohort study [J]. J Orthop Surg Res, 2023, 18(1): 557.
7
Liu J, Zhang Y, Zhang W, et al. Prediction of vasoactive-inotropic score on prolonged mechanical ventilation in adult congenital heart disease patients after surgical treatment combined with coronary artery bypass grafting [J]. Braz J Cardiovasc Surg, 2024, 39(3): e20230218.
8
Ad N, Holmes SD, Patel J, et al. Comparison of EuroSCORE Ⅱ, Original EuroSCORE, and the Society of Thoracic Surgeons Risk Score in cardiac surgery patients [J]. Ann Thorac Surg, 2016, 102(2): 573-579.
9
王子玉, 傅强, 王树英, 等. STS风险评估系统对单中心非体外循环冠状动脉旁路移植术后延长机械通气的预测价值分析 [J]. 中国急救医学, 2019, 39(7): 642-648.
10
Wang TK, Li AY, Ramanathan T, et al. Comparison of four risk scores for contemporary isolated coronary artery bypass grafting [J]. Heart Lung Circ, 2014, 23(5): 469-474.
11
Jiang C, Xiu Y, Qiao K, et al. Prediction of lymph node metastasis in patients with breast invasive micropapillary carcinoma based on machine learning and SHapley Additive exPlanations framework [J]. Front Oncol, 2022, 12: 981059.
12
Shortliffe EH, Sepúlveda MJ. Clinical decision support in the era of artificial intelligence [J]. JAMA, 2018, 320(21): 2199-2200.
13
Vayena E, Blasimme A, Cohen IG. Machine learning in medicine: addressing ethical challenges [J]. PLoS Med, 2018, 15(11): e1002689.
14
Lei T, Guo J, Wang P, et al. Establishment and validation of predictive model of tophus in gout patients [J]. J Clin Med, 2023, 12(5): 1755.
15
Lundberg SM, Erion G, Chen H, et al. From local explanations to global understanding with explainable AI for trees [J]. Nat Mach Intell, 2020, 2(1): 56-67.
16
Sauerbrei W, Royston P, Binder H. Selection of important variables and determination of functional form for continuous predictors in multivariable model building [J]. Stat Med, 2007, 26(30): 5512-5528.
17
Pourhoseingholi MA, Baghestani AR, Vahedi M. How to control confounding effects by statistical analysis [J]. Gastroenterol Hepatol Bed Bench, 2012, 5(2): 79-83.
18
Sankar A, Rotstein AJ, Teja B, et al. Prolonged mechanical ventilation after cardiac surgery: substudy of the Transfusion Requirements in Cardiac Surgery Ⅲ trial [J]. Can J Anaesth, 2022, 69(12): 1493-1506.
19
王甦. 急性Stanford A型主动脉夹层术后高乳酸血症的预测模型及预后影响研究 [D]. 武汉: 华中科技大学, 2022.
20
Ferreruela M, Raurich JM, Ayestarán I, et al. Hyperlactatemia in ICU patients: incidence, causes and associated mortality [J]. J Crit Care, 2017, 42: 200-205.
21
Chiara O, Giomarelli PP, Biagioli B, et al. Hypermetabolic response after hypothermic cardiopulmonary bypass [J]. Crit Care Med, 1987, 15(11): 995-1000.
22
Öztürk S. Effect of lactate levels on extubation time in coronary artery bypass grafting surgery [J]. Cardiovasc J Afr, 2023, 34: 1-5.
23
张永苹, 雷黎明, 凌云, 等. 156例体外循环心脏术后严重高乳酸血症患者的护理 [J]. 护理学报, 2023, 30(4): 66-68.
24
张永苹, 宋亚敏, 凌云, 等. 64例全胸腔镜不停跳下行心脏三尖瓣膜手术后患者的护理 [J]. 护理学报, 2021, 28(10): 70-71.
25
中华护理学会糖尿病护理专业委员会. 住院成人高血糖患者血糖监测医护协议处方共识 [J]. 中华护理杂志, 2019, 54(8): 1142-1147.
26
余强芳, 彭渝. 心脏术后房颤危险因素与护理研究进展 [J]. 护理实践与研究, 2012, 9(4): 136-137.
27
陶红, 沈骁, 章淬. 心脏术后新发心房颤动的研究进展 [J]. 心血管病学进展, 2024, 45(3): 234-237.
28
Ali Dabbagh, Fardad Esmailian, Sary Aranki. 成人心脏外科术后重症监护学 (第二版) [M]. 济南: 山东科学技术出版社, 2021: 310-313.
29
Caldonazo T, Kirov H, Rahouma M, et al. Atrial fibrillation after cardiac surgery: a systematic review and meta-analysis [J]. J Thorac Cardiovasc Surg, 2023, 165(1): 94-103.e24.
30
Alexander S, Teshome M, Patel H, et al. The diagnostic and prognostic utility of risk factors defined by the AHA/ACCF on the evaluation of cardiac disease in liver transplantation candidates [J]. BMC Cardiovasc Disord, 2019, 19(1): 102.
31
Sandau KE, Funk M, Auerbach A, et al. Update to practice standards for electrocardiographic monitoring in hospital settings: a scientific statement from the American Heart Association [J]. Circulation, 2017, 136(19): e273-e344.
32
Ge M, Wang Z, Chen T, et al. Risk factors for and outcomes of prolonged mechanical ventilation in patients received DeBakey type I aortic dissection repairment [J]. J Thorac Dis, 2021, 13(2): 735-742.
33
白一帆, 王崇, 谈梦伟, 等. 成人双瓣膜置换术后延迟拔管的危险因素 [J]. 中华胸心血管外科杂志, 2012, 28(11): 651-654.
34
Badagliacca R, Reali M, Poscia R, et al. Right intraventricular dyssynchrony in idiopathic, heritable, and anorexigen-induced pulmonary arterial hypertension: clinical impact and reversibility [J]. JACC Cardiovasc Imaging, 2015, 8(6): 642-652.
35
吴伟春, 徐楠, 牛丽莉, 等. 不同心功能状态下特发性肺动脉高压患者心室同步性的超声研究 [J]. 中国循环杂志, 2017, 32(1): 81-84.
[1] 张小玉, 郭伟, 吴秀丽, 汪蕊, 汪莹. ICU专科护士培训大纲的构建和应用[J/OL]. 中华临床医师杂志(电子版), 2023, 17(03): 330-334.
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?