切换至 "中华医学电子期刊资源库"

中华危重症医学杂志(电子版) ›› 2025, Vol. 18 ›› Issue (04) : 274 -281. doi: 10.3877/cma.j.issn.1674-6880.2025.04.002

论著

基于机器学习算法构建药物中毒患者ICU住院时间延长的预测模型
戴辉水1,2, 吕嵩1, 张劲松2, 巴根2, 石齐芳2,()   
  1. 1239400 安徽滁州,明光市人民医院重症医学科
    2210029 南京,南京医科大学第一附属医院急诊与危重症医学科
  • 收稿日期:2025-01-07 出版日期:2025-08-31
  • 通信作者: 石齐芳
  • 基金资助:
    国家自然科学基金项目(82172184)

Prediction model for prolonged ICU stay in drug poisoning patients based on machine learning algorithms

Huishui Dai1,2, Song Lyu1, Jinsong Zhang2, Gen Ba2, Qifang Shi2,()   

  1. 1Department of Intensive Care Medicine, Mingguang People's Hospital, Chuzhou 239400, China
    2Department of Emergency and Critical Care Medicine, the First Affiliated Hospital with Nanjing Medical University, Nanjing 210029, China
  • Received:2025-01-07 Published:2025-08-31
  • Corresponding author: Qifang Shi
引用本文:

戴辉水, 吕嵩, 张劲松, 巴根, 石齐芳. 基于机器学习算法构建药物中毒患者ICU住院时间延长的预测模型[J/OL]. 中华危重症医学杂志(电子版), 2025, 18(04): 274-281.

Huishui Dai, Song Lyu, Jinsong Zhang, Gen Ba, Qifang Shi. Prediction model for prolonged ICU stay in drug poisoning patients based on machine learning algorithms[J/OL]. Chinese Journal of Critical Care Medicine(Electronic Edition), 2025, 18(04): 274-281.

目的

基于机器学习算法构建药物中毒患者ICU住院时间延长的预测模型。

方法

纳入美国重症监护医学信息数据库Ⅳ(MIMIC-Ⅳ)与电子重症监护病房合作研究数据库(eICU-CRD)中的药物中毒患者,按ICU住院时间分为非延长组(≤ 48 h)和延长组(> 48 h)。按7 ∶ 3将MIMIC-Ⅳ中1 342例患者分为训练集(939例)与测试集(403例),eICU-CRD作为外部测试集(2 144例)。在训练集中,通过单因素分析和最小绝对收缩和选择算子(LASSO)回归联合筛选变量,利用6种机器学习算法(逻辑回归、极端梯度提升、轻量级梯度提升机、随机森林、决策树、支持向量机)建模。同时,采用受试者工作特征(ROC)曲线、霍斯默-莱梅肖(H-L)检验、Brier评分及决策曲线分析(DCA)在内部和外部测试集中评估模型性能。

结果

训练集中共筛选出7个关键变量,分别为脑血管疾病、肝脏疾病、吸入性肺炎、脓毒症、呼吸频率、序贯器官衰竭估计(SOFA)评分和机械通气。逻辑回归模型在训练集中表现最佳[曲线下面积(AUC)= 0.767,95%置信区间(CI)(0.667,0.868),P < 0.001)],其在内部测试集中的AUC为0.762[95%CI(0.712,0.811),P < 0.001],在外部测试集中的AUC为0.732[95%CI(0.708,0.756),P < 0.001]。且逻辑回归模型在内部及外部测试中均具良好校准度和净收益。

结论

本研究构建的逻辑回归模型由7个预测因素组成,包括脑血管疾病、肝脏疾病、吸入性肺炎、脓毒症、呼吸频率、SOFA评分和机械通气,可有效预测药物中毒患者ICU住院时间延长风险,有助于临床早期识别和干预。

Objective

To develop a prediction model for the prolonged ICU stay in patients with drug poisoning based on machine learning algorithms.

Methods

Drug poisoning patients included in the Medical Information Mart for Intensive Care Ⅳ (MIMIC-Ⅳ) and Electronic Intensive Care Unit-Collaborative Research Database (eICU-CRD) were divided into a non-prolonged group (≤ 48 hours) and a prolonged group (> 48 hours) according to the length of ICU stay. A total of 1 342 patients in MIMIC-Ⅳ were divided into a training dataset (939 cases) and a test dataset (403 cases) at a ratio of 7 ∶ 3, with eICU-CRD serving as an external test dataset (2 144 cases). In the training dataset, variables were jointly screened through single-factor analysis and least absolute shrinkage and selection operator (LASSO) regression. Six machine learning algorithms (logistic regression, extreme gradient boosting, light gradient boosting machine, random forest, decision tree, and support vector machine) were used for modeling. Meanwhile, the receiver operating characteristic (ROC) curve, Hosmer-Lemeshow (H-L) test, Brier score, and decision curve analysis (DCA) were used to evaluate the model performance in both internal and external test datasets.

Results

Seven critical predictors were screened out in the training dataset: cerebrovascular disease, liver disease, aspiration pneumonia, sepsis, respiratory rate, sequential organ failure assessment (SOFA) score, and mechanical ventilation. The logistic regression model performed best in the training dataset [area under the curve (AUC) = 0.767, 95% confidence interval (CI) (0.667, 0.868), P < 0.001]. Its AUC was 0.762 [95%CI (0.712, 0.811), P < 0.001] in the internal test dataset, and 0.732 [95%CI (0.708, 0.756), P < 0.001] in the external test dataset. Moreover, the logistic regression model showed good calibration and net returns in both internal and external test datasets.

Conclusions

The logistic regression model constructed in this study consists of seven predictive factors, including cerebrovascular disease, liver disease, aspiration pneumonia, sepsis, respiratory rate, SOFA score, and mechanical ventilation. It can effectively predict the risk of prolonged ICU stay in drug poisoning patients, which is helpful for early clinical identification and intervention.

表1 训练集药物中毒患者ICU住院时间延长组与非延长组基线资料的比较
因素 非延长组(n = 593) 延长组(n = 346) Z/χ2 P 因素 非延长组(n = 593) 延长组(n = 346) Z / χ2 P
年龄[岁,MP25P75)] 45(31,58) 45(30,58) 0.435 0.663 评分系统[分,MP25P75)]      
男性[例(%)] 319(54) 165(48) 3.022 0.082 GCS评分 15(15,15) 15(15,15) 1.141 0.110
白种人[例(%)] 341(58) 201(58) 0.012 0.914 CCI评分 1(0,3) 1(0,4) 3.436 < 0.001
基础疾病[例(%)]       SOFA评分 2(1,4) 4(2,7) 9.942 < 0.001
糖尿病 85(14) 54(16) 0.664 0.189 入ICU后首次实验室检查[MP25P75)]    
高血压 164(28) 106(31) 0.807 0.369 白细胞(× 109 / L) 9.1(6.6,11.5) 10.3(7.8,13.9) 5.459 < 0.001
肥胖 39(7) 29(8) 0.808 0.369 血红蛋白(g / L) 121(108,131) 122(110,135) 1.660 1.660
房颤 34(6) 25(7) 0.592 0.442 红细胞压积(%) 36(33,39) 36(33,40) 1.440 0.147
CHF 51(9) 39(11) 1.504 0.220 血小板(× 109 / L) 215(175,269) 210(166,261) 1.659 0.097
心肌梗死 29(5) 27(8) 2.807 0.094 RDW(%) 14(13,14) 14(13,14) 0.517 0.605
脑血管疾病 22(4) 31(9) 10.343 0.001 肌酐(μmol / L) 80(62,97) 80(62,141) 3.439 < 0.001
COPD 100(17) 63(18) 0.190 0.663 尿素氮(mmol / L) 4.6(3.2,6.8) 5.4(3.6,8.9) 4.685 < 0.001
LQTS 24(4) 13(4) 0.002 0.963 血钾(mmol / L) 3.9(3.6,4.3) 4.0(3.5,4.5) 1.251 0.210
肝脏疾病 86(15) 100(29) 27.622 < 0.001 血钠(mmol / L) 140(138,142) 140(137,142) 1.258 0.206
肾脏疾病 45(8) 28(8) 0.023 0.879 AG(mmol / L) 14(12,16) 15(12,18) 3.597 < 0.001
入ICU的首次生命体征[MP25P75)]     碳酸氢盐(mmol / L) 23(21,25) 22(19,25) 2.945 0.003
体温(℃) 36.7(36.4,37.0) 36.8(36.4,37.2) 2.679 0.007 血糖(mmol / L) 6.2(5.2,7.6) 6.7(5.5,8.4) 3.941 < 0.001
心率(次/ min) 85(73,100) 92(77,106) 4.059 < 0.001 入ICU后24 h内的治疗措施[例(%)]    
RR(次/ min) 18(15,21) 20(17,24) 6.268 < 0.001 机械通气 166(28) 200(58) 80.389 < 0.001
收缩压(mmHg) 127(113,141) 125(113,141) 0.310 0.756 肾脏替代治疗 11(2) 12(3) 1.753 0.186
SpO2(%) 98(96,100) 99(96,100) 0.242 0.803 血管活性药物 64(11) 92(27) 38.228 < 0.001
药物中毒信息[例(%)]       抗生素 116(20) 158(46) 70.789 < 0.001
故意中毒 215(36) 130(38) 0.111 0.739          
合并酒精中毒 117(20) 67(19) 0.003 0.959          
入ICU后24 h内并发症[例(%)]                
吸入性肺炎 45(8) 87(25) 54.300 < 0.001          
脓毒症 84(14) 171(49) 135.528 < 0.001          
图1 训练集药物中毒患者ICU住院时间延长风险的LASSO回归分析注:LASSO.最小绝对收缩和选择算子;λ.正则化参数;a图为LASSO回归系数路径图;b图为LASSO回归交叉验证最佳参数λ的选择过程图
图2 6种机器学习模型对训练集药物中毒患者ICU住院时间延长风险的评估注:a图为各模型的受试者工作特征曲线;b图为各模型的决策曲线分析
图3 基于逻辑回归模型预测药物中毒患者ICU住院时间延长风险的列线图注:SOFA.序贯器官衰竭估计
图4 逻辑回归模型对急性药物中毒患者ICU住院时间延长风险内部测试的评估注:a图为混淆矩阵,展示真实类别与预测类别的分布,颜色深浅表示样本数量;b图为受试者工作特征曲线;c图为校准曲线;d图为决策曲线分析
图5 逻辑回归模型对急性药物中毒患者ICU住院时间延长风险外部测试的评估注:a图为混淆矩阵,展示真实类别与预测类别的分布,颜色深浅表示样本数量;b图为受试者工作特征曲线;c图为校准曲线;d图为决策曲线分析
1
Lewer D, Brothers TD, Gasparrini A, et al. Seasonal, weekly and other cyclical patterns in deaths due to drug poisoning in England and Wales[J]. Addiction, 2023, 118 (8): 1596-1601.
2
Gummin DD, Mowry JB, Beuhler MC, et al. 2023 annual report of the national poison data systemR (NPDS) from America's poison centersR: 41st annual report[J]. Clin Toxicol (Phila), 2024, 62 (12): 793-1027.
3
高艳霞,孙同文. 中国急性中毒十年研究回顾与展望[J]. 中华急诊医学杂志2023,32(3):282-287.
4
Deng X, Jin Y, Wang Y, et al. Characteristics of poisoning cases admitted to outpatient and emergency department - China, 2019[J]. China CDC Wkly, 2023, 5 (47): 1052-1057.
5
Reisinger AC, Schneider N, Schreiber N, et al. Critical care management of acute intoxications, dynamics and changes over time: a cohort study[J]. Intern Emerg Med, 2024, 19 (7): 2015-2024.
6
Shi Q, Zhang J. Clinical prediction models for intensive care unit admission in patients with acute poisoning: is it time for a comprehensive evaluation of their utility?[J]. Toxicol Res (Camb), 2024, 13 (2): tfae031.
7
Peres IT, Hamacher S, Oliveira FLC, et al. What factors predict length of stay in the intensive care unit? Systematic review and meta-analysis[J]. J Crit Care, 2020, 60: 183-194.
8
Peres IT, Hamacher S, Cyrino Oliveira FL, et al. Data-driven methodology to predict the ICU length of stay: a multicentre study of 99,492 admissions in 109 Brazilian units[J]. Anaesth Crit Care Pain Med, 2022, 41 (6): 101142.
9
Shi Q, Dai H, Ba G, et al. Development and internal validation of a predictive model for prolonged intensive care unit stays in patients with psychotropic drug poisoning[J]. Heart Lung, 2024, 68: 350-358.
10
戴辉水,石齐芳,巴根,等. 急性药物中毒性脑病患者ICU住院时间预测模型的构建[J]. 中国工业医学杂志2024,37(2):133-137.
11
Johnson AEW, Bulgarelli L, Shen L, et al. MIMIC-IV, a freely accessible electronic health record dataset[J]. Sci Data, 2023, 10 (1): 1.
12
Pollard TJ, Johnson AEW, Raffa JD, et al. The eICU Collaborative Research Database, a freely available multi-center database for critical care research[J]. Sci Data, 2018, 5: 180178.
13
Da B, Chen H, Wu W, et al. Development and validation of a machine learning-based model to predict survival in patients with cirrhosis after transjugular intrahepatic portosystemic shunt[J]. EClinicalMedicine, 2024, 79: 103001.
14
Shao L, Wang Z, Xie X, et al. Development and external validation of a machine learning-based fall prediction model for nursing home residents: a prospective cohort study[J]. J Am Med Dir Assoc, 2024, 25 (9): 105169.
15
Nattino G, Pennell ML, Lemeshow S. Assessing the goodness of fit of logistic regression models in large samples: a modification of the Hosmer-Lemeshow test[J]. Biometrics, 2020, 76 (2): 549-560.
16
Yang W, Jiang J, Schnellinger EM, et al. Modified Brier score for evaluating prediction accuracy for binary outcomes[J]. Stat Methods Med Res, 2022, 31 (12): 2287-2296.
17
庄燕,戴林峰,张海东,等. 脓毒症患者早期生存影响因素及Cox风险预测模型构建[J/OL]. 中华危重症医学杂志(电子版)2024,17(5):372-378.
18
Zwaag SM, van den Hengel-Koot IS, Baker S, et al. The INTOXICATE study: methodology and preliminary results of a prospective observational study[J]. Crit Care, 2024, 28 (1): 316.
19
Hurtado D, Quintero JA, Rodriguez YA, et al. Principal causes of acute poisoning in an emergency service: experience between 2014 and 2021 at a University Hospital in Southwestern Colombia[J]. Sci Rep, 2024, 14 (1): 3544.
20
Naim G, Lacoste-Palasset T, M'Rad A, et al. Factors associated with prolonged intensive care stay among self-poisoned patients[J]. Clin Toxicol (Phila), 2022, 60(9): 997-1005.
21
Megahed FM, Chen YJ, Megahed A, et al. The class imbalance problem[J]. Nat Methods, 2021, 18 (11): 1270-1272.
22
Liisanantti JH, Ohtonen P, Kiviniemi O, et al. Risk factors for prolonged intensive care unit stay and hospital mortality in acute drug-poisoned patients: an evaluation of the physiologic and laboratory parameters on admission[J]. J Crit Care, 2011, 26 (2): 160-165.
23
Ichikura K, Okumura Y, Takeuchi T. Associations of adverse clinical course and ingested substances among patients with deliberate drug poisoning: a cohort study from an intensive care unit in Japan[J]. PLoS One, 2016, 11 (8): e0161996.
24
Song YX, Yang XD, Luo YG, et al. Comparison of logistic regression and machine learning methods for predicting postoperative delirium in elderly patients: a retrospective study[J]. CNS Neurosci Ther, 2023, 29 (1): 158-167.
25
Song X, Liu X, Liu F, et al. Comparison of machine learning and logistic regression models in predicting acute kidney injury: a systematic review and meta-analysis[J]. Int J Med Inform, 2021, 151: 104484.
26
石齐芳,张劲松. 急性中毒临床预测模型需要关注的几个问题[J]. 中华急诊医学杂志2024,33(11):1479-1481.
[1] 明昊, 宋宏萍, 白誉誉, 党晓智, 赵阳, 程燕妮, 王琪, 肖迎聪. 自动乳腺超声联合临床病理特征预测Luminal B 型乳腺癌术后复发风险的临床研究[J/OL]. 中华医学超声杂志(电子版), 2025, 22(04): 337-347.
[2] 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.
[3] 张丽花, 胡耀华, 周春献, 张晓亭, 胡韶军. 恶性肿瘤患者住院化疗期间医院感染状况及列线图预测模型构建[J/OL]. 中华实验和临床感染病杂志(电子版), 2025, 19(04): 205-213.
[4] 李炜, 王倩, 张娥, 张光辉. 基于血清降钙素原、可溶性白细胞介素-2受体及临床特征建立慢性化脓性中耳炎术后复发风险预测模型及效能评价[J/OL]. 中华实验和临床感染病杂志(电子版), 2025, 19(04): 223-229.
[5] 孙圣梅, 习一清, 安宁. 人表皮生长因子受体2阳性型乳腺癌新辅助治疗响应的基因预测模型[J/OL]. 中华普通外科学文献(电子版), 2025, 19(05): 332-339.
[6] 赵晨皓, 张序东, 杨浚沫, 周何. 血清肿瘤标志物对结直肠癌患者术后复发的预测效能研究进展[J/OL]. 中华普外科手术学杂志(电子版), 2025, 19(04): 467-470.
[7] 关雪, 于颖, 李京, 刘莹, 崔亚娟, 刘名鹤. 肝移植受者术后下肢深静脉血栓形成风险预测模型的建立[J/OL]. 中华移植杂志(电子版), 2025, 19(03): 138-144.
[8] 张志昱, 张凡, 周奇, 欧阳骏, 林宇鑫, 张学锋. 术前炎性指标对肾透明细胞癌WHO/ISUP病理分级的预测价值[J/OL]. 中华腔镜泌尿外科杂志(电子版), 2025, 19(03): 323-328.
[9] 汪锐, 陈自武, 杨朴强, 田静, 陈莹, 林成, 汪伟. 基于血清标志物机器学习模型对慢性阻塞性肺疾病急性加重期机械通气风险的预测分析[J/OL]. 中华肺部疾病杂志(电子版), 2025, 18(04): 615-619.
[10] 林水荣, 宋子敏, 于玺, 李绍强, 华赟鹏, 沈顺利. 术前抗病毒治疗对HBV相关肝癌肝切除术后肝衰竭影响[J/OL]. 中华肝脏外科手术学电子杂志, 2025, 14(05): 700-706.
[11] 唐善华, 赖展鸿, 刘海晴, 王小振, 王恺, 周杰. 基于XGBoost算法构建肝癌肝切除术后肝衰竭早期识别预测模型[J/OL]. 中华肝脏外科手术学电子杂志, 2025, 14(05): 725-731.
[12] 赵俊宇, 林航宇, 李会灵, 王显飞, 游川. 肝癌肝切除术后大量腹水预测模型的建立与验证[J/OL]. 中华肝脏外科手术学电子杂志, 2025, 14(05): 740-747.
[13] 李媛媛, 李荣山. 机器学习:肾脏疾病研究与诊疗的新前沿[J/OL]. 中华肾病研究电子杂志, 2025, 14(04): 181-187.
[14] 王柯云, 孙雅佳, 李甜, 张钰哲, 郑颖, 张伟光, 王倩, 董哲毅. 糖尿病肾脏疾病早期发生风险预测模型的研究进展[J/OL]. 中华肾病研究电子杂志, 2025, 14(04): 218-225.
[15] 夏炎, 朱帅帅, 张岩松. 基于生物信息学和机器学习探究前庭神经鞘瘤生物标志物在免疫微环境中的相关功能[J/OL]. 中华脑科疾病与康复杂志(电子版), 2025, 15(03): 171-179.
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?