基于SMOTE算法的脓毒症并发急性呼吸窘迫综合征早期预警模型
该回顾性病例对照研究纳入济南市人民医院2016年10月至2022年10月收治的566例脓毒症患者,其中163例住院期间发生ARDS、403例未发生。多因素Logistic回归显示年龄(OR=3.449)、肺部感染为首发病因(OR=2.309)、胰腺及胰周感染(OR=1.937)、脓毒性休克(OR=3.381)、SOFA评分(OR=9.311)为独立影响因素。基于上述因素构建的预警模型AUC为0.882(95%CI 0.851-0.914),敏感度79.8%、特异度83.4%;经SMOTE算法改进后模型AUC为0.890(95%CI 0.867-0.913),敏感度85.3%、特异度79.1%。作者认为临床可据此评估脓毒症患者发生ARDS的概率,以便早期干预、改善预后。
要点
- 566例脓毒症患者中163例住院期间发生ARDS,SOFA评分是最强的独立危险因素(OR=9.311,95%CI 5.831-14.867,P=0.000)。
- 脓毒性休克(OR=3.381,95%CI 1.890-6.047,P=0.000)和年龄(OR=3.449,95%CI 2.197-5.414,P=0.000)同样显著增加ARDS风险。
- 肺部感染为首发病因(OR=2.309,95%CI 1.427-3.737,P=0.001)与胰腺及胰周感染(OR=1.937,95%CI 1.236-3.035,P=0.004)也是独立危险因素。
- 原始预警模型AUC 0.882(95%CI 0.851-0.914),敏感度79.8%、特异度83.4%;SMOTE改进模型AUC 0.890(95%CI 0.867-0.913),敏感度85.3%、特异度79.1%。
- 作者认为临床可基于该预警模型评估脓毒症患者发生ARDS的概率,以便早期干预、改善预后。
适用人群与场景
- 2016年10月至2022年10月济南市人民医院收治的脓毒症患者566例,按住院期间是否发生ARDS分为ARDS组163例与非ARDS组403例。
研究设计与关键结果
- 回顾性病例对照研究,纳入566例脓毒症患者,其中163例住院期间发生ARDS、403例未发生。
- 单因素分析显示两组在年龄、体重指数(BMI)、恶性肿瘤、输血史、胰腺及胰周感染、胃肠道感染、肺部感染为首发病因、急性生理学与慢性健康状况评分Ⅱ(APACHEⅡ)、序贯器官衰竭评分(SOFA)、白蛋白(Alb)、血尿素氮(BUN)、机械通气治疗、脓毒性休克及ICU住院时间方面差异有统计学意义。
- 二元多因素Logistic回归显示年龄(OR=3.449,95%CI 2.197-5.414,P=0.000)、肺部感染为首发病因(OR=2.309,95%CI 1.427-3.737,P=0.001)、胰腺及胰周感染(OR=1.937,95%CI 1.236-3.035,P=0.004)、脓毒性休克(OR=3.381,95%CI 1.890-6.047,P=0.000)、SOFA评分(OR=9.311,95%CI 5.831-14.867,P=0.000)为脓毒症患者住院期间发生ARDS的独立影响因素。
- 基于上述危险因素建立预警模型:P1=-4.558+1.238×年龄+0.837×肺部感染为首发病因+0.661×胰腺及胰周感染+1.218×脓毒性休克+2.231×SOFA评分;ROC曲线分析显示该模型预测脓毒症患者住院期间发生ARDS的AUC为0.882(95%CI 0.851-0.914),敏感度79.8%、特异度83.4%。
- 基于SMOTE算法改进数据集后重建预警模型:P2=-3.279+1.288×年龄+0.763×肺部感染为首发病因+0.635×胰腺及胰周感染+1.068×脓毒性休克+2.201×SOFA评分;ROC曲线分析显示该模型AUC为0.890(95%CI 0.867-0.913),敏感度85.3%、特异度79.1%。
证据类型、等级与局限
- 证据类型:回顾性病例对照研究。
- 局限:原文未报告局限性。
对床旁实践的改变
- 作者认为临床可基于年龄、肺部感染为首发病因、胰腺及胰周感染、脓毒性休克和SOFA评分构建的预警模型评估脓毒症患者住院期间发生ARDS的概率,以便早期干预、改善预后。
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 explore the independent risk factors of acute respiratory distress syndrome (ARDS) in patients with sepsis, establish an early warning model, and verify the predictive value of the model based on synthetic minority oversampling technique (SMOTE) algorithm.
METHODS: A retrospective case-control study was conducted. 566 patients with sepsis who were admitted to Jinan People's Hospital from October 2016 to October 2022 were enrolled. General information, underlying diseases, infection sites, initial cause, severity scores, blood and arterial blood gas analysis indicators at admission, treatment measures, complications, and prognosis indicators of patients were collected. The patients were grouped according to whether ARDS occurred during hospitalization, and the clinical data between the two groups were observed and compared. Univariate and binary multivariate Logistic regression analysis were used to select the independent risk factors of ARDS during hospitalization in septic patients, and regression equation was established to construct the early warning model. Simultaneously, the dataset was improved using the SMOTE algorithm to build an enhanced warning model. Receiver operator characteristic curve (ROC curve) was drawn to validate the prediction efficiency of the model.
RESULTS: 566 patients with sepsis were included in the final analysis, of which 163 developed ARDS during hospitalization and 403 did not. Univariate analysis showed that there were statistically significant differences in age, body mass index (BMI), malignant tumor, blood transfusion history, pancreas and peripancreatic infection, gastrointestinal tract infection, pulmonary infection as the initial etiology, acute physiology and chronic health evaluation II (APACHE II) score, sequential organ failure assessment (SOFA) score, albumin (Alb), blood urea nitrogen (BUN), mechanical ventilation therapy, septic shock and length of intensive care unit (ICU) stay between the two groups. Binary multivariate Logistic regression analysis showed that age [odds ratio (OR) = 3.449, 95% confidence interval (95%CI) was 2.197-5.414, P = 0.000], pulmonary infection as the initial etiology (OR = 2.309, 95%CI was 1.427-3.737, P = 0.001), pancreas and peripancreatic infection (OR = 1.937, 95%CI was 1.236-3.035, P = 0.004), septic shock (OR = 3.381, 95%CI was 1.890-6.047, P = 0.000), SOFA score (OR = 9.311, 95%CI was 5.831-14.867, P = 0.000) were independent influencing factors of ARDS during hospitalization in septic patients. An early warning model was established based on the above risk factors: P1 = -4.558+1.238×age+0.837×pulmonary infection as the initial etiology+0.661×pancreas and peripancreatic infection+1.218×septic shock+2.231×SOFA score. ROC curve analysis showed that the area under the ROC curve (AUC) of the model for ARDS during hospitalization in septic patients was 0.882 (95%CI was 0.851-0.914) with sensitivity of 79.8% and specificity of 83.4%. The dataset was improved based on the SMOTE algorithm, and the early warning model was rebuilt: P2 = -3.279+1.288×age+0.763×pulmonary infection as the initial etiology+0.635×pancreas and peripancreatic infection+1.068×septic shock+2.201×SOFA score. ROC curve analysis showed that the AUC of the model for ARDS during hospitalization in septic patients was 0.890 (95%CI was 0.867-0.913) with sensitivity of 85.3% and specificity of 79.1%. This result further confirmed that the early warning model constructed by the independent risk factors mentioned above had high predictive performance.
CONCLUSIONS: Risk factors for the occurrence of ARDS during hospitalization in patients with sepsis include age, pulmonary infection as the initial etiology, pancreatic and peripancreatic infection, septic shock, and SOFA score. Clinically, the probability of ARDS in patients with sepsis can be assessed based on the warning model established using these risk factors, allowing for early intervention and improvement of prognosis.
参考来源
- Duan H, Li X, Yang X, et al. [An early warning model for sepsis complicated with acute respiratory distress syndrome based on synthetic minority oversampling technique algorithm]. Zhonghua Wei Zhong Bing Ji Jiu Yi Xue. 2024. PMID: 38813628. doi:10.3760/cma.j.cn121430-20230925-00821
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。