利用纵向生物标志物进行脓毒症相关急性肾损伤动态死亡率预测的联合建模：回顾性队列研究
Joint Modeling for Dynamic Mortality Prediction in Sepsis-Associated Acute Kidney Injury Using Longitudinal Biomarkers: A Retrospective Cohort Study
摘要
目的：本研究采用联合模型框架，评估了血流及肾功能纵向生物标志物对脓毒症相关急性肾损伤（SA-AKI）患者动态死亡风险预测的预后价值。
方法：研究共纳入11809例SA-AKI患者。基线特征定义为首次入住ICU时的数据。在28天的随访期内，收集了血小板计数和血清肌酐的纵向测量值，共计80582个观测值用于分析。
结果：研究建立了一个联合模型，用于评估SA-AKI患者的ICU死亡风险。其中，纵向子模型纳入了血小板计数和估算肾小球滤过率（eGFR）;生产子模型则包含了年龄、APACHEIII评分、平均红细胞体积（MCV）、红细胞分布宽度（RDW）、氧分压（PO?）、乳酸、氯化物、国际标准化比值（INR）和部分凝血活酶时间（PTT）。该模型显示出良好的区分能力：训练集中，其在48小时、72小时和96小时的受试者工作特征曲线下面积（AUC）值分别为0.85、0.78和0.71；验证集中对应的AUC值分别为0.81、0.76和0.69。
结论：该联合模型通过整合血小板计数和eGFR的纵向变化轨迹，能够动态预测患者结局。它捕捉了生物标志物当前及历史的波动，并展现出稳健的随时间变化的预测性能。
Abstract
BACKGROUND: We employed a joint modeling framework to evaluate the prognostic value of longitudinal blood and renal function biomarkers for dynamic mortality prediction in sepsis-associated acute kidney injury (SA-AKI) patients.
METHODS: A total of 11,809 patients diagnosed with sepsis-associated acute kidney injury (SA-AKI) were included. Baseline characteristics were defined at the time of first ICU admission. Longitudinal measurements of platelet count and serum creatinine collected during the 28-day follow-up period yielded 80,582 observations for analysis.
RESULTS: We developed a joint model to estimate ICU mortality risk in patients with sepsis-associated acute kidney injury (SA-AKI). The longitudinal sub-model incorporated platelet count and estimated glomerular filtration rate (eGFR), while the survival sub-model included age, APACHE III, mean corpuscular volume (MCV), red cell distribution width (RDW), partial pressure of oxygen (PO?), lactate, chloride, international normalized ratio (INR), and partial thromboplastin time (PTT). The model demonstrated discriminative performance with area under the receiver operating characteristic curve (AUC) values of 0.85 at 48 hours, 0.78 at 72 hours, and 0.71 at 96 hours in the training set, and corresponding values of 0.81, 0.76, and 0.69 in the validation set.
CONCLUSIONS: The joint model (JM) dynamically predicts outcomes by integrating longitudinal trajectories of platelet count and eGFR (estimated glomerular filtration rate), capturing both current and historical biomarker fluctuations, the model demonstrated robust time-dependent predictive performance.关键词
脓毒症/联合模型/动态预测/生物标志物轨迹Key words
Sepsis/ Joint model/ Dynamic prediction/ Biomarker trajectories引用本文复制引用
刘欢,汤在祥.利用纵向生物标志物进行脓毒症相关急性肾损伤动态死亡率预测的联合建模：回顾性队列研究[EB/OL].(2026-04-07)[2026-04-09].http://www.paper.edu.cn/releasepaper/content/202604-57.学科分类
医学研究方法/基础医学/生物科学研究方法、生物科学研究技术
评论