Prognostic analysis of sepsis-related liver injury and development of a prediction model based on machine learning method
-
摘要:目的
分析脓毒症相关肝损伤(SRLI)患者的预后, 并使用8种机器学习方法建立脓毒症患者入住ICU后发生SRLI的预测模型。
方法纳入MIMIC-IV数据库中满足脓毒症诊断标准且无肝脏、胆系基础疾病的患者。根据肝酶≥5倍正常值上限(ULN)或胆红素≥2.0 mg/dL将患者分为SRLI组和非SRLI组。采用卡方检验、多因素Logistics回归分析及倾向性评分匹配法分析2组患者死亡风险。采用8种机器学习算法[Logistics回归、分类回归树(CART)、随机森林(RF)、支持向量机(SVM)、K-近邻(K-NN)、朴素贝叶斯(NBM)、极端梯度提升(XGBoost)、梯度提升树(GBDT)]构建SRLI预测模型并进行验证。
结果卡方检验(P < 0.001)、多因素Logistics回归分析(P < 0.05)、倾向性评分匹配分析后Log-rank (P < 0.05)均显示SRLI增加患者死亡风险。SRLI预测模型中, RF算法的曲线下面积(AUC)最高为0.866, 其后依次是GBDT (AUC=0.862)、Logistics回归(AUC=0.859)、SVM (AUC=0.837)、NBM (AUC=0.830)、CART (AUC=0.771)、XGBoost (AUC=0.764)、K-NN (AUC=0.722)。
结论SRLI增加患者死亡风险。RF算法构建预测模型有较高的诊断价值。
Abstract:ObjectiveTo analyze the prognosis of patients with sepsis-related liver injury (SRLI) and establish a prediction model for the occurrence of SRLI after ICU admission in sepsis patients using eight machine learning methods.
MethodsPatients who met the sepsis diagnostic criteria and had no underlying liver or biliary diseases were included from the MIMIC-IV database, and were classified into SRLI and non-SRLI groups based on liver enzymes ≥5 times the upper limit of normal (ULN) or bilirubin ≥2.0mg/dL. Chi-square test, multivariate Logistic regression analysis, and propensity score matching were used to analyze the mortality risk between the two groups. Eight machine learning algorithms[Logistic regression, classification and regression tree (CART), random forest (RF), support vector machine (SVM), K-nearest neighbors (K-NN), naive Bayes method (NBM), extreme gradient boosting (XGBoost), and gradient boosting decision tree (GBDT)]were employed to construct and validate the SRLI prediction model.
ResultsThe chi-square test (P < 0.001), multivariate Logistic regression analysis (P < 0.05), and log-rank test after propensity score matching (P < 0.05) all indicated that SRLI increased the mortality risk of patients. Among the SRLI prediction models, RF algorithm had the highest area under the curve (AUC), with its value of 0.866, followed by GBDT (AUC=0.862), Logistic regression (AUC=0.859), SVM (AUC=0.837), NBM (AUC=0.830), CART (AUC=0.771), XGBoost (AUC=0.764), and K-NN (AUC=0.722).
ConclusionSRLI increases the mortality risk of patients. The prediction model constructed using the RF algorithm has high diagnostic value.
-
-
表 1 SRLI组与非SRLI组基线资料比较[n(%)][M(QL, QU)]
因素 非SRLI组(n=6 559) SRLI组(n=2 080) t/χ2/Z P 男性 3 600(54.89) 1 254(60.29) 19.00 < 0.001 年龄/岁 70.27(58.04, 81.27) 65.83(52.88, 77.14) 46.68 < 0.001 BMI/(kg/m2) 26.87(23.30, 31.87) 27.57(23.95, 32.41) 52.82 < 0.001 GCS评分/分 14.00(9.00, 15.00) 13.00(7.00, 15.00) 48.70 < 0.001 MBP/mmHg 58.00(51.00, 64.00) 56.00(47.00, 63.00) 48.14 < 0.001 心率/(次/min) 105.00(91.00, 120.00) 112.00(97.00, 128.00) 55.82 < 0.001 呼吸频率/(次/min) 28.00(24.00, 32.50) 29.00(25.00, 34.00) 54.03 < 0.001 体温/℃ 36.44(36.11, 36.72) 36.40(35.72, 36.72) 48.31 < 0.001 SpO2/% 92.00(90.00, 95.00) 92.00(88.00, 94.00) 48.71 < 0.001 SOFA/分 3.00(2.00, 4.00) 4.00(2.00, 6.00) 59.02 < 0.001 SAPS-Ⅲ/分 51.00(38.00, 67.00) 67.00(48.00, 91.00) 60.77 < 0.001 因素 分类 非SRLI组(n=6 559) SRLI组(n=2 080) t/χ2/Z P 24 h液体量/mL 1 820.41(927.39, 3 009.11) 2 777.71(1 452.49, 4 445.51) 59.17 < 0.001 机械通气 3 215(49.02) 1 422(68.37) 237.00 < 0.001 VDI/(μg/min) 0(0, 0.10) 0.10(0, 0.50) 59.22 < 0.001 休克 1 390(21.19) 965(46.39) 505.00 < 0.001 CHF 2 101(32.03) 748(35.96) 11.00 0.001 COPD 1 776(27.08) 491(23.60) 9.70 0.002 CRD 1 654(25.22) 423(20.34) 20.00 < 0.001 肿瘤 978(14.91) 329(15.82) 0.94 0.332 糖尿病 2 105(32.09) 546(26.25) 25.00 < 0.001 Charlson共病指数 5.00(3.00, 8.00) 6.00(4.00, 8.00) 48.53 < 0.001 感染源 肺部 1 403(21.39) 454(21.83) 12.00 0.014 腹腔 373(5.69) 110(5.29) 泌尿系 1 001(15.26) 256(12.31) 皮肤软组织 59(0.90) 21(1.01) 其他 3 723(56.76) 1 239(59.57) 实验室检查 乳酸/(mmol/L) 1.70(1.20, 2.80) 3.05(1.70, 5.93) 62.28 < 0.001 pH 7.35(7.28, 7.41) 7.28(7.18, 7.37) 42.29 < 0.001 pa(O2)/mmHg 80.00(63.00, 109.00) 75.00(60.00, 98.00) 48.00 < 0.001 pa(CO2)/mmHg 43.00(37.00, 50.00) 45.00(38.00, 54.00) 53.58 < 0.001 HCT/% 30.40(25.90, 34.90) 29.40(24.30, 34.73) 48.78 < 0.001 血红蛋白/(g/dL) 10.00(8.50, 11.55) 9.70(8.00, 11.50) 49.24 < 0.001 血小板/(×109/L) 181.00(129.00, 249.00) 141.00(87.00, 203.00) 43.10 < 0.001 白细胞/(×109/L) 13.60(9.70, 18.70) 15.80(10.60, 21.40) 54.75 < 0.001 白蛋白/(g/L) 3.20(2.70, 3.60) 3.00(2.50, 3.50) 45.77 < 0.001 尿素氮/(mmol/L) 25.00(16.00, 41.00) 29.00(18.00, 46.25) 54.24 < 0.001 肌酐/(mg/dL) 1.20(0.90, 1.90) 1.50(1.00, 2.40) 56.20 < 0.001 PT/s 14.20(12.70, 16.60) 15.80(13.60, 20.73) 58.45 < 0.001 APTT/s 31.80(27.70, 41.60) 37.20(30.30, 65.13) 58.35 < 0.001 ALT-ad/(U/L) 21.00(14.00, 34.00) 70.00(28.00, 190.25) 68.57 < 0.001 AST-ad/(U/L) 29.00(20.00, 45.00) 130.00(45.00, 301.00) 70.48 < 0.001 TBIL-ad/(mg/dL) 0.50(0.30, 0.80) 0.90(0.50, 1.90) 63.39 < 0.001 SRLI: 脓毒症相关肝损伤; BMI: 体质量指数; GCS: 格拉斯哥昏迷评分; MBP: 平均动脉压; SpO2: 血氧饱和度; SOFA: 序贯器官衰竭评分; SAPS-Ⅲ: 简化急性生理功能评分; VDI: 血管活性药物强度; CHF: 慢性心功能不全; COPD: 慢性阻塞性肺疾病; CRD: 慢性肾功能不全; pH: 酸碱度; pa(O2): 动脉血氧分压; pa(CO2): 动脉血二氧化碳分压; HCT: 红细胞压积; PT: 凝血酶原时间; APTT: 活化部分凝血活酶时间; ALT-ad: 入ICU时谷丙转氨酶; AST-ad: 入ICU时谷草转氨酶; TBIL-ad: 入ICU时总胆红素。1 mmHg=0.133 kPa。 表 2 SRLI组与非SRLI组患者预后比较[n(%)][M(QL, QU)]
预后指标 非SRLI组(n=6 559) SRLI组(n=2 080) t/χ2/Z P 急性肾损伤 4 585(69.90) 1 726(82.98) 137.00 < 0.001 机械通气时间/d 0(0, 1.50) 1.20(0, 4.04) 58.99 < 0.001 血管活性药物使用时间/h 0(0, 12.92) 7.52(0, 49.67) 58.66 < 0.001 住院时间/d 7.80(4.78, 13.54) 10.10(5.17, 19.07) 55.13 < 0.001 入住ICU时间/d 2.97(1.71, 5.86) 4.33(2.16, 9.57) 56.89 < 0.001 28 d死亡 1 316(20.06) 681(32.74) 142.00 < 0.001 住院死亡 1 003(15.29) 651(31.30) 260.00 < 0.001 表 3 机器学习模型预测效能
模型 AUC 准确度 精确度 召回率 F1得分 RF 0.866 0.842 0.950 0.388 0.551 GBDT 0.862 0.843 0.816 0.483 0.607 Logistics 0.859 0.855 0.852 0.510 0.638 SVM 0.837 0.830 0.776 0.449 0.569 NBM 0.830 0.809 0.973 0.245 0.391 CART 0.771 0.814 0.694 0.463 0.555 XGBoost 0.764 0.809 0.688 0.435 0.533 K-NN 0.722 0.772 0.610 0.245 0.350 RF: 随机森林模型; GBDT: 梯度提升树模型; SVM: 支持向量机模型; NBM: 朴素贝叶斯模型; CART: 分类与回归树模型; XGBoost: 极端梯度提升模型; K-NN: K-近邻模型。 -
[1] EVANS L, RHODES A, ALHAZZANI W, et al. Surviving sepsis campaign: international guidelines for management of sepsis and septic shock 2021[J]. Intensive Care Med, 2021, 47(11): 1181-1247. doi: 10.1007/s00134-021-06506-y
[2] 邢冬梅, 隋冰冰, 王磊. 老年脓毒症患者住院期间死亡风险预测模型的建立与验证[J]. 实用临床医药杂志, 2024, 28(8): 39-44. doi: 10.7619/jcmp.20233722 [3] DHAINAUT J F, MARIN N, MIGNON A, et al. Hepatic response to sepsis: interaction between coagulation and inflammatory processes[J]. Crit Care Med, 2001, 29(7 Suppl): S42-S47.
[4] PEREZ RUIZ DE GARIBAY A, KORTGEN A, LEONHARDT J, et al. Critical care hepatology: definitions, incidence, prognosis and role of liver failure in critically ill patients[J]. Crit Care, 2022, 26(1): 289. doi: 10.1186/s13054-022-04163-1
[5] VAN DEN BROECKE A, VAN COILE L, DECRUYENAERE A, et al. Epidemiology, causes, evolution and outcome in a single-center cohort of 1116 critically ill patients with hypoxic hepatitis[J]. Ann Intensive Care, 2018, 8(1): 15. doi: 10.1186/s13613-018-0356-z
[6] KOBASHI H, TOSHIMORI J, YAMAMOTO K. Sepsis-associated liver injury: incidence, classification and the clinical significance[J]. Hepatol Res, 2013, 43(3): 255-266. doi: 10.1111/j.1872-034X.2012.01069.x
[7] LEVY M M, EVANS L E, RHODES A. The surviving sepsis campaign bundle: 2018 update[J]. Intensive Care Med, 2018, 44(6): 925-928. doi: 10.1007/s00134-018-5085-0
[8] SEYMOUR C W, LIU V X, IWASHYNA T J, et al. Assessment of clinical criteria for sepsis: for the third international consensus definitions for sepsis and septic shock (sepsis-3)[J]. JAMA, 2016, 315(8): 762-774. doi: 10.1001/jama.2016.0288
[9] HU W H, CHEN H, MA C C, et al. Identification of indications for albumin administration in septic patients with liver cirrhosis[J]. Crit Care, 2023, 27(1): 300. doi: 10.1186/s13054-023-04587-3
[10] WANG D W, YIN Y M, YAO Y M. Advances in sepsis-associated liver dysfunction[J]. Burns Trauma, 2014, 2(3): 97-105. doi: 10.4103/2321-3868.132689
[11] HORVATITS T, DROLZ A, TRAUNER M, et al. Liver injury and failure in critical illness[J]. Hepatology, 2019, 70(6): 2204-2215. doi: 10.1002/hep.30824
[12] JONSDOTTIR S, ARNARDOTTIR M B, ANDRESSON J A, et al. Prevalence, clinical characteristics and outcomes of hypoxic hepatitis in critically ill patients[J]. Scand J Gastroenterol, 2022, 57(3): 311-318. doi: 10.1080/00365521.2021.2005136
[13] 孙维维, 黄晓英, 王亚东. HELENICC评分预测脓毒症相关急性肾损伤行持续肾脏替代治疗患者早期病死率的价值[J]. 实用临床医药杂志, 2023, 27(15): 29-34. doi: 10.7619/jcmp.20231689 [14] WOZNICA E A, INGLOT M, WOZNICA R K, et al. Liver dysfunction in sepsis[J]. Adv Clin Exp Med, 2018, 27(4): 547-551. doi: 10.17219/acem/68363
[15] GINÈS P, SCHRIER R W. Renal failure in cirrhosis[J]. N Engl J Med, 2009, 361(13): 1279-1290. doi: 10.1056/NEJMra0809139
[16] XIE T H, XIN Q, CAO X R, et al. Clinical characteristics and construction of a predictive model for patients with sepsis related liver injury[J]. Clin Chim Acta, 2022, 537: 80-86. doi: 10.1016/j.cca.2022.10.004
[17] DAI J M, GUO W N, TAN Y Z, et al. Wogonin alleviates liver injury in sepsis through Nrf2-mediated NF-κB signalling suppression[J]. J Cell Mol Med, 2021, 25(12): 5782-5798. doi: 10.1111/jcmm.16604
[18] HUANG H, TOHME S, AL-KHAFAJI A B, et al. Damage-associated molecular pattern-activated neutrophil extracellular trap exacerbates sterile inflammatory liver injury[J]. Hepatology, 2015, 62(2): 600-614. doi: 10.1002/hep.27841
-
期刊类型引用(6)
1. 郝俐,郑志龙. 可溶性髓系细胞触发受体1、乳酸对难治性支原体肺炎患儿预后的影响. 实用心脑肺血管病杂志. 2025(02): 16-19 . 百度学术
2. 严秋丽,张方园,陈颖青,施金金,钮文思. 新型指标HROX指数在儿童重症肺炎早期诊治中的应用价值研究. 中国临床新医学. 2024(08): 881-885 . 百度学术
3. 史小慧,曾晓兵,康莉莉,张筱婷. 儿童重症肺炎进展风险预测模型构建. 赤峰学院学报(自然科学版). 2024(08): 13-17 . 百度学术
4. 王晓,张玉玲. 婴儿社区获得性肺炎进展为重症肺炎预警指标的研究现状. 妇儿健康导刊. 2024(19): 21-24 . 百度学术
5. 宫秀娟,赵慧霞,张小庆,张连霞. 重症社区获得性肺炎患者PIV、IL-6、HBP、PAB水平对疗效的评估价值. 天津医药. 2024(11): 1216-1220 . 百度学术
6. 韩萍萍,刘娅. 静脉滴注人免疫球蛋白辅助治疗对小儿重症肺炎的临床疗效. 深圳中西医结合杂志. 2024(20): 83-86 . 百度学术
其他类型引用(1)