文献速递重症综合管理与镇静镇痛复苏、休克与脓毒症

脓毒症患者28天死亡危险因素分析及预测模型的构建与验证

[Analysis of 28 day-mortality risk factors in sepsis patients and construction and validation of predictive model].
PubMed 文献 · Zhonghua Wei Zhong Bing Ji Jiu Yi Xue · 2024-01-01
一句话结论基于血管活性药物使用、APACHE II评分、合并MODS、中性粒细胞计数和氧合指数的列线图模型对脓毒症患者28天死亡有良好预测能力(训练集AUC 0.899,验证集AUC 0.909)。

该回顾性队列研究纳入2017年1月至2022年12月就诊于中国人民解放军联勤保障部队第九四〇医院ICU的281例脓毒症患者,按7:3分为训练集(197例)和验证集(84例)。采用Lasso回归及单因素、多因素Logistic回归筛选出5个与28天死亡相关的独立预测因素:使用血管活性药物、APACHE II评分、合并MODS、中性粒细胞计数和氧合指数,并据此构建列线图模型。模型在训练集和验证集的AUC分别为0.899和0.909,C-index分别为0.900和0.920,校准曲线及Hosmer-Lemeshoe检验显示校准良好,DCA和CIC显示临床效用良好。该模型有助于ICU脓毒症患者28天死亡风险评估。

要点

  • 281例脓毒症患者中82例死亡,28天死亡率为29.18%;训练集死亡54例(27.41%),验证集死亡28例(33.33%)。
  • 多因素Logistic回归显示合并MODS的OR=17.298(95%CI 5.517-76.985,P<0.001),使用血管活性药物的OR=5.924(95%CI 1.244-44.571,P=0.043)。
  • APACHE II评分OR=1.051(95%CI 1.000-1.107,P=0.050),中性粒细胞计数OR=0.934(95%CI 0.879-0.988,P=0.022),氧合指数OR=0.994(95%CI 0.988-0.998,P=0.017)。
  • 列线图模型训练集AUC为0.899(95%CI 0.856-0.943),验证集AUC为0.909(95%CI 0.845-0.972),C-index分别为0.900和0.920。
  • Hosmer-Lemeshoe检验均P>0.05提示校准良好,DCA和CIC显示模型临床效用良好。

适用人群与场景

  • 2017年1月至2022年12月就诊于中国人民解放军联勤保障部队第九四〇医院ICU的脓毒症患者281例,按7:3分为训练集197例和验证集84例。

研究设计与关键结果

  • 回顾性队列研究,纳入281例脓毒症患者,其中82例死亡,28天死亡率为29.18%;训练集死亡54例(27.41%),验证集死亡28例(33.33%)。
  • Lasso回归及单因素、多因素Logistic回归筛选出5个独立预测因素:使用血管活性药物(OR=5.924,95%CI 1.244-44.571,P=0.043)、APACHE II评分(OR=1.051,95%CI 1.000-1.107,P=0.050)、合并MODS(OR=17.298,95%CI 5.517-76.985,P<0.001)、中性粒细胞计数(OR=0.934,95%CI 0.879-0.988,P=0.022)和氧合指数(OR=0.994,95%CI 0.988-0.998,P=0.017)。
  • 基于上述5个因素构建列线图模型,训练集AUC为0.899(95%CI 0.856-0.943),验证集AUC为0.909(95%CI 0.845-0.972);C-index训练集为0.900,验证集为0.920,区分度良好。
  • Hosmer-Lemeshoe检验均P>0.05,提示校准良好;DCA和CIC图显示模型临床效用良好。

证据类型、等级与局限

  • 证据类型:回顾性队列研究。
  • 局限:原文未报告局限性。

对床旁实践的改变

  • 作者认为,基于血管活性药物使用、APACHE II评分、合并MODS、中性粒细胞计数和氧合指数5个指标构建的列线图模型对脓毒症患者28天死亡有良好预测能力。

提炼说明

  • 本条依据论文摘要整理;摘要未涉及「诊断标准与关键阈值」「处置流程与药物剂量」「红旗征与不能漏的情况」「指南或共识出处」,故未列出这些栏目;全文请点「阅读原文」。
英文原文(摘要)· 点击展开

OBJECTIVE: To construct and validate a nomogram model for predicting the risk of 28-day mortality in sepsis patients.

METHODS: A retrospective cohort study was conducted. 281 sepsis patients admitted to the department of intensive care unit (ICU) of the 940th Hospital of the Joint Logistics Support Force of PLA from January 2017 to December 2022 were selected as the research subjects. The patients were divided into a training set (197 cases) and a validation set (84 cases) according to a 7 : 3 ratio. The general information, clinical treatment measures and laboratory examination results within 24 hours after admission to ICU were collected. Patients were divided into survival group and death group based on 28-day outcomes. The differences in various data were compared between the two groups. The optimal predictive variables were selected using Lasso regression, and univariate and multivariate Logistic regression analyses were performed to identify factors influencing the mortality of sepsis patients and to establish a nomogram model. Receiver operator characteristic curve (ROC curve), calibration curve, decision curve analysis (DCA), and clinical impact curve (CIC) were used to evaluate the nomogram model.

RESULTS: Out of 281 cases of sepsis, 82 cases died with a mortality of 29.18%. The number of patients who died in the training and validation sets was 54 and 28, with a mortality of 27.41% and 33.33% respectively. Lasso regression, univariate and multivariate Logistic regression analysis screened for 5 independent predictors associated with 28-day mortality. There were use of vasoactive drugs [odds ratio (OR) = 5.924, 95% confidence interval (95%CI) was 1.244-44.571, P = 0.043], acute physiology and chronic health evaluation II (APACHE II: OR = 1.051, 95%CI was 1.000-1.107, P = 0.050), combined with multiple organ dysfunction syndrome (MODS: OR = 17.298, 95%CI was 5.517-76.985, P < 0.001), neutrophil count (NEU: OR = 0.934, 95%CI was 0.879-0.988, P = 0.022) and oxygenation index (PaO2/FiO2: OR = 0.994, 95%CI was 0.988-0.998, P = 0.017). A nomogram model was constructed using the independent predictive factors mentioned above, ROC curve analysis showed that the AUC of the nomogram model was 0.899 (95%CI was 0.856-0.943) and 0.909 (95%CI was 0.845-0.972) for the training and validation sets respectively. The C-index was 0.900 and 0.920 for the training and validation sets respectively, with good discrimination. The Hosmer-Lemeshoe tests both showed P > 0.05, indicating good calibration. Both DCA and CIC plots demonstrate the model's good clinical utility.

CONCLUSIONS: The use of vasoactive, APACHE II score, comorbid MODS, NEU and PaO2/FiO2 are independent risk factors for 28-day mortality in patients with sepsis. The nomogram model based on these 5 indicators has a good predictive ability for the occurrence of mortality in sepsis patients.

来源:Zhonghua Wei Zhong Bing Ji Jiu Yi Xue,查看原文。版权归原作者所有。

参考来源

  1. Shao H, Wang Y, Zhang H, et al. [Analysis of 28 day-mortality risk factors in sepsis patients and construction and validation of predictive model]. Zhonghua Wei Zhong Bing Ji Jiu Yi Xue. 2024. PMID: 38845493. doi:10.3760/cma.j.cn121430-20231109-00961

供执业医师抢救与诊疗参考,不替代临床判断;剂量请结合体重、肝肾功能和本院药品规格复核,以指南原文为准。