基于年龄和合并症的脓毒症患者预后风险模型评估:一项大规模多中心回顾性研究中机器学习和传统方法的启示
这项回顾性研究纳入四家中国医学中心的3287例脓毒症患者(内部验证队列2974例,外部验证队列313例),评估年龄校正Charlson合并症指数(ACCI)及其简化版qACCI对死亡的预测价值。机器学习模型将ACCI识别为院内死亡的关键预测因子,ACCI与院内、30天和ICU死亡风险呈线性相关;中ACCI组HR 2.18(95%CI 1.77-2.70)、高ACCI组HR 3.72(95%CI 2.99-4.65),均以低ACCI组为参照。ACCI预测院内死亡的AUC为0.788,在胃肠道感染(0.831 vs. 0.794)和中枢神经系统感染(0.803 vs. 0.739)中优于SOFA;qACCI在内部和外部队列的AUC分别为0.734和0.758。作者认为ACCI和qACCI可作为临床医生早期识别高危患者的可靠工具。
要点
- ACCI预测脓毒症患者院内死亡的AUC为0.788,在胃肠道感染(0.831 vs. 0.794)和中枢神经系统感染(0.803 vs. 0.739)中优于SOFA。
- 与低ACCI组相比,中ACCI组死亡风险HR 2.18(95%CI 1.77-2.70),高ACCI组HR 3.72(95%CI 2.99-4.65)。
- ACCI与院内死亡、30天死亡和ICU死亡风险均呈线性相关,敏感性分析显示各亚组结果一致。
- 简化版qACCI在内部验证队列AUC为0.734,外部验证队列AUC为0.758,具有中等预测效能。
适用人群与场景
- 四家中国医学中心的脓毒症患者,共3287例:内部验证队列2974例(平均年龄67.96岁,男性37.5%),外部验证队列313例(平均年龄67.90岁,男性48.2%)。
研究设计与关键结果
- 研究类型:大规模、多中心、回顾性研究,纳入四家中国医学中心的3287例脓毒症患者。
- 机器学习模型将ACCI识别为院内死亡的关键预测因子。
- ACCI与院内死亡、30天死亡和ICU死亡风险呈线性相关;敏感性分析显示各亚组结果一致。
- 与低ACCI组(HR 1,参照)相比,中ACCI组死亡风险HR 2.18(95%CI 1.77-2.70),高ACCI组HR 3.72(95%CI 2.99-4.65)。
- ACCI预测院内死亡的AUC为0.788,在胃肠道感染中优于SOFA(0.831 vs. 0.794),在中枢神经系统感染中优于SOFA(0.803 vs. 0.739)。
- qACCI在内部验证队列AUC为0.734,在外部验证队列AUC为0.758。
证据类型、等级与局限
- 证据类型:多中心回顾性研究。
- 局限:原文未报告局限性。
对床旁实践的改变
- 作者认为,ACCI和qACCI作为年龄与合并症的综合指标,为临床医生早期识别高危脓毒症患者提供了有价值且可靠的工具。
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
Background: Age and comorbidity significantly impact the prognosis of septic patients and inform treatment decisions. To provide clinicians with effective tools for identifying high-risk patients, this study assesses the predictive value of the age-adjusted Charlson Comorbidity Index (ACCI) and its simplified version, the quick ACCI (qACCI), for mortality in septic patients. Methods: This retrospective study included septic patients from four Chinese medical centers. The internal validation cohort comprised patients from Xinhua Hospital, Ruijin Hospital, and Huashan Hospital, while participants from Renji Hospital served as the external validation cohort. Machine learning models identified ACCI's feature importance. Restricted cubic spline regression and subgroup analysis assess the correlation between ACCI and mortality risk. The qACCI, derived from the ACCI components, was also evaluated for predictive reliability. Results: A total of 3,287 septic patients were included: 2,974 in the internal cohort (mean age 67.96 years; 37.5% male) and 313 in the external cohort (mean age 67.90 years; 48.2% male). Machine learning models identified ACCI as a key predictor of in-hospital mortality. A linear correlation was confirmed between ACCI and risks of in-hospital, 30-day, and ICU mortality. Sensitivity analysis revealed consistent results across subgroups, demonstrating significantly higher mortality risks in the moderate- (hazard ratio [HR] 2.18, 95% CI 1.77-2.70) and high-ACCI (HR 3.72, 95% CI 2.99-4.65) groups compared to the low-ACCI group (HR 1, reference). The ACCI achieved an AUC of 0.788 for in-hospital mortality, outperforming the SOFA in gastrointestinal (0.831 vs. 0.794) and central nervous system infections (0.803 vs. 0.739). The qACCI showed moderate predictive performance in both the internal (AUC, 0.734) and external (AUC, 0.758) cohorts. Conclusions: As composite indicators of age and comorbidity, ACCI and qACCI provide valuable and reliable tools for clinicians to identify high-risk patients early.
参考来源
- Liu G, Shang Z, Ning N, et al. EVALUATION OF PROGNOSTIC RISK MODELS BASED ON AGE AND COMORBIDITY IN SEPTIC PATIENTS: INSIGHTS FROM MACHINE LEARNING AND TRADITIONAL METHODS IN A LARGE-SCALE, MULTICENTER, RETROSPECTIVE STUDY. Shock. 2025. PMID: 39965627. doi:10.1097/SHK.0000000000002562
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。