目的: 探究应激性血糖升高与重症监护病房(intensive care unit, ICU) 患者28 d全因死亡风险之间的关系, 并比较不同应激性血糖升高指标的预测效能。方法: 以重症医学(Medical Information Mart for Intensive Care Ⅳ, MIMIC-Ⅳ) 数据库中符合纳入、排除标准的ICU患者为研究对象，将应激性血糖升高指标按照百分位数分为Q1(0~25%)、Q2(>25%~75%)、Q3(>75%~100%)组，以是否发生ICU内死亡及在ICU内接受治疗的时间为结局变量，以人口学特征、实验室指标、合并症等为协变量，利用Cox回归及限制性立方样条探究应激性血糖升高和ICU患者28 d全因死亡风险之间的关联; 采用受试者工作特征(receiver operation characteristic, ROC)曲线下面积(area under curve, AUC)评价不同应激性血糖升高指标的预测效能，应激性血糖升高指标包括应激性血糖升高比值(stress hyperglycemia ratio, SHR) 1、SHR2、血糖间隙(glucose gap, GG); 进一步将应激性血糖升高指标纳入牛津急性疾病严重程度评分(Oxford acute severity of illness score, OASIS)，探究其改善评分的预测效能，即采用AUC评估评分区分度，AUC越大表明评分区分度越好，采用Brier score评价评分校准度，Brier score越小，表明评分校准度越好。结果: 共纳入5 249例ICU患者，其中发生ICU内死亡的患者占7.56%。调整混杂因素后的Cox回归分析结果表明，SHR1、SHR2和GG的最高组Q3与最低组Q1相比，ICU患者28 d全因死亡HR(95%CI)分别为1.545(1.077~2.217)、1.602(1.142~2.249)和1.442(1.001~2.061)，且随着应激性血糖升高指标的增加，ICU患者死亡风险也逐渐增加(P_trend < 0.05)。限制性立方样条分析表明，SHR和28 d全因死亡风险之间呈线性关系(P>0.05)。SHR2和GG的AUC显著高于SHR1: AUC_SHR2=0.691(95%CI: 0.661~0.720)，AUC_GG=0.685(95%CI: 0.655~0.714)，AUC_SHR1=0.680(95%CI: 0.650~0.709)，P < 0.05。将SHR2纳入OASIS评分中，能显著提高评分的区分度和校准度：AUC_OASIS=0.820(95%CI: 0.791~0.848)，AUC_OASIS+SHR2=0.832(95%CI: 0.804~0.859)，P < 0.05; Brier score_OASIS=0.071，Brier score_OASIS+SHR2= 0.069。结论: 应激性血糖升高与ICU患者28 d全因死亡风险密切相关，可为重症监护患者的临床管理和决策提供参考。

Objective: To investigate the relationship between stress glucose elevation and the risk of 28 d all-cause mortality in intensive care unit (ICU) patients, and to compare the predictive efficacy of different stress glucose elevation indicators. Methods: ICU patients who met the inclusion and exclusion criteria in the Medical Information Mart for Intensive Care Ⅳ (MIMIC-Ⅳ) database were used as the study subjects, and the stress glucose elevation indicators were divided into Q1 (0-25%), Q2 (>25%- 75%), and Q3 (>75%-100%) groups, with whether death occurred in the ICU and the duration of treatment in the ICU as outcome variables, and demographic characteristics, laboratory indicators, and comorbidities as covariates, Cox regression and restricted cubic splines were used to explore the association between stress glucose elevation and the risk of 28 d all-cause death in ICU patients; and subject work characteristics [receiver operating characteristic (ROC) and the area under curve (AUC)] were used to evaluate the predictive efficacy of different stress glucose elevation indicators, The stress hyperglycemia indexes included: stress hyperglycemia ratio (SHR1, SHR2), glucose gap (GG); and the stress hyperglycemia index was further incorporated into the Oxford acute severity of illness score (OASIS) to investigate the predictive efficacy of the improved scores: the AUC was used to assess the score discrimination, and the larger the AUC indicated, the better score discrimination. The Brier score was used to evaluate the calibration of the score, and a smaller Brier score indicated a better calibration of the score. Results: A total of 5 249 ICU patients were included, of whom 7.56% occurred in ICU death. Cox regression analysis after adjusting for confounders showed that the HR (95%CI) for 28 d all-cause mortality in the ICU patients was 1.545 (1.077-2.217), 1.602 (1.142-2.249) and 1.442 (1.001-2.061) for the highest group Q3 compared with the lowest group Q1 for SHR1, SHR2 and GG, respectively, and The risk of death in the ICU patients increased progressively with increasing indicators of stressful blood glucose elevation (P_trend < 0.05). Restricted cubic spline analysis showed a linear relationship between SHR and the 28 d all-cause mortality risk (P>0.05). the AUC of SHR2 and GG was significantly higher than that of SHR1: AUC_SHR2=0.691 (95%CI: 0.661-0.720), AUC_GG=0.685 (95%CI: 0.655-0.714), and AUC_SHR1=0.680 (95%CI: 0.650-0.709), P < 0.05. The inclusion of SHR2 in the OASIS scores significantly improved the discrimination and calibration of the scores: AUC_OASIS=0.820 (95%CI: 0.791-0.848), AUC_OASIS+SHR2=0.832 (95%CI: 0.804-0.859), P < 0.05; Brier score_OASIS=0.071, Brier score_OASIS+SHR2=0.069. Conclusion: Stressful glucose elevation is strongly associated with 28 d all-cause mortality risk in ICU patients and may inform clinical management and decision making in intensive care patients.