Risk factors for mortality in patients with sepsis on extracorporeal membrane oxygenation and/or continuous renal replacement therapy: a retrospective cohort study based on MIMIC-IV database

This study aimed to identify risk factors for mortality in septic patients undergoing extracorporeal membrane oxygenation (ECMO) and/or continuous renal replacement therapy (CRRT). Data from the MIMIC-IV database were retrospectively reviewed for 24,502 septic patients treated with ECMO or CRRT between 2008 and 2019. After applying inclusion and exclusion criteria, 70 patients receiving ECMO, 513 receiving CRRT, and 22 receiving both were included in the final analysis. Univariate and multivariate stepwise Cox regression analyses were performed to identify independent risk factors for mortality. Model performance was assessed using receiver operating characteristic (ROC) curve analysis. We also provided model-agnostic explanations for each Cox regression model. For septic patients on ECMO, prothrombin time (per 1-s increase, HR 1.037, 95% CI 1.007–1.068, <i>p</i> = .015) was the key independent risk factor. For septic patients undergoing CRRT, SOFA score (per one-point increase, HR 1.100, 95% CI 1.055–1.147, <i>p</i> &lt; .001) was the most significant factor. For septic patients requiring both ECMO and CRRT, prior history of hypertension (HR 4.342, 95% CI 1.332–14.153, <i>p</i> = .015) was the sole independent risk factor. ROC analysis showed satisfactory model performance (AUC &gt; 0.75). For septic patients requiring ECMO, prothrombin time was the key independent risk factor. For those needing CRRT, SOFA score was the most significant independent risk factor. Prior history of hypertension was the primary independent risk factor for septic patients needing both CRRT and ECMO.

本研究旨在确定接受体外膜肺氧合（extracorporeal membrane oxygenation, ECMO）和/或连续性肾脏替代治疗（continuous renal replacement therapy, CRRT）的脓毒症患者死亡率的独立风险因素。研究回顾性分析了2008年至2019年间MIMIC-IV数据库中接受ECMO或CRRT治疗的24502例脓毒症患者数据；经纳入与排除标准筛选后，最终分析纳入70例接受ECMO治疗、513例接受CRRT治疗及22例同时接受两种治疗的患者。通过单变量及多变量逐步Cox回归分析确定死亡率独立风险因素，并采用受试者工作特征（receiver operating characteristic, ROC）曲线分析评估模型性能，同时为每个Cox回归模型提供模型无关解释。对于接受ECMO治疗的脓毒症患者，凝血酶原时间（prothrombin time）（每增加1秒，风险比（hazard ratio, HR）1.037，95%置信区间（confidence interval, CI）1.007–1.068，p=0.015）是关键独立风险因素；对于接受CRRT治疗的脓毒症患者，序贯器官衰竭评估（Sequential Organ Failure Assessment, SOFA）评分（每增加1分，HR 1.100，95% CI 1.055–1.147，p<0.001）是最显著的独立风险因素；对于同时需要ECMO和CRRT治疗的脓毒症患者，既往高血压病史（HR 4.342，95% CI 1.332–14.153，p=0.015）是唯一的独立风险因素。ROC曲线分析显示模型性能良好（曲线下面积（Area Under the Curve, AUC）>0.75）。对于需要ECMO治疗的脓毒症患者，凝血酶原时间是关键独立风险因素；对于需要CRRT治疗的患者，SOFA评分是最显著的独立风险因素；既往高血压病史是同时需要CRRT和ECMO治疗的脓毒症患者的主要独立风险因素。