Abstract: Background　Klebsiella pneumoniae is a common nosocomial infection pathogen，and carbapenem antibiotics is "the last line of defense" to treat its infection.In recent years，hospital infections caused by carbapenem-resistant Klebsiella pneumoniae（CRKP） are increasing due to the overuse of antibiotics，especially carbapenems.The detection rate of CRKP is also growing each year with the progress of medical technology.After infection，there are only a few sensitive antibiotics that can be selected clinically，yielding unoptimistic anti-infection effects，which in turn leads to a rising mortality rate.Objective　To identify the risk factors of mortality in patients of CRKP infection，and use them to establish a mortality risk prediction model，then evaluate the model's prognostic value.Methods　Clinical data of 199 patients with CRKP infection were collected from January 2017 to April 2019 from the Second Affiliated Hospital of Xi'an Jiaotong University，including 138〔modeling group consisting of 104 survivals and 34 deaths（classified at discharge）〕 receiving inpatient treatment from January 2017 to December 2018，and 61（validating group） from January to April 2019.General data，prevalence of underlying diseases，pre-hospital treatment，complications during hospitalization，invasive treatment，biochemical indices and temperature after 3-day initial treatment，post-admission treatment，and other indices were compared between the survival and death subgroups.Multivariate Logistic regression models were applied to investigate the risk factors of mortality in patients with CRKP infection，and based on the derived related coefficients，a mortality risk prediction model for CRKP infection was established，and ROC curve of the model was drawn to predict the mortality in modeling and validating group.Results　The mortality rate was 24.6%（34/138） in the modeling group and 24.6%（15/61） in the validating group.Multivariate Logistic regression analysis showed that multiple organ dysfunction syndrome during hospitalization，APACHE Ⅱ score ≥14 points after initial treatment，and use of vasoactive drugs were closely related to the mortality after CRKP infection（P<0.05）.The regression equation was that multiple organ dysfunction syndrome during hospitalization ×2.4 + （APACHE Ⅱ score ≥14 points after initial treatment）×1.5 + use of vasoactive drugs × 2.0.The mortality risk prediction model in CRKP infection was applied in modeling group with the value of H-L test 0.866，in validating group with 0.807.The area under the curve（AUC）of the model in modeling group was 0.851〔95%CI（0.774，0.928）〕，with optimal cut-off value of 2.75，sensitivity of 84.4%，specificity of 76.0% and Youden index of 0.584，and the AUC of which in validating group was 0.966〔95%CI（0.894，1.000）〕，with optimal cut-off value of 2.95，sensitivity of 100.0%，specificity of 87.0% and Youden index of 0.870.Conclusion　Multiple organ dysfunction syndrome during hospitalization，APACHE Ⅱ score ≥14 after initial treatment and use of vasoactive drugs are independent risk factors of mortality in CRKP infected patients.The mortality risk prediction model in CRKP infection fitted out based on the above-mentioned indices has a good prognostic value.

Key words: Klebsiella infections, Carbapenem-resistant Klebsiella pneumoniae, Infection, Mortality, Root cause analysis, Predictive value

摘要： 背景　肺炎克雷伯菌是院内感染常见致病菌，碳青霉烯类抗生素是治疗其感染的“最后一道防线”。近年来由于抗生素尤其是耐碳青霉烯类的过量暴露，碳青霉烯类耐药肺炎克雷伯菌（CRKP）引起的院内感染增加，其检出率亦随医疗技术的进步而逐年增加。CRKP感染后，临床可供选择的敏感抗生素屈指可数，且抗感染治疗效果不佳，死亡率也随之上升。目的　探究CRKP感染患者发生死亡的危险因素，建立CRKP感染死亡风险预测模型，并评估该模型对患者预后的预测价值。方法　收集2017年1月—2019年4月西安交通大学第二附属医院收治的199例CRKP感染患者的临床资料。将2017年1月—2018年12月收治的患者作为建模组（n=138），2019年1—4月收治的患者作为验证组（n=61）。将建模组的患者依据离院时状态分为存活亚组、死亡亚组，分别为104例和34例。比较存活亚组和死亡亚组患者一般资料、基础疾病、入院前治疗情况、住院期间并发症、有创治疗情况、初始治疗（3 d）后血生化指标及体温、入院后治疗情况等。采用多因素Logistic回归分析探究CRKP感染患者发生死亡的影响因素，并依据多因素Logistic回归分析结果的相关系数建立CRKP感染死亡风险预测模型，分别绘制CRKP感染死亡风险预测模型预测建模组及验证组患者发生死亡的受试者工作特征（ROC）曲线。结果　建模组患者死亡率为24.6%（34/138），验证组患者死亡率为24.6%（15/61）。多因素Logistic回归分析结果显示，住院期间并发多器官功能障碍综合征（MODS）、初始治疗后急性生理学与慢性健康状况评分系统Ⅱ（APACHEⅡ）评分≥14分、使用血管活性药物是CRKP感染患者发生死亡的影响因素（P<0.05）。CRKP感染死亡风险预测模型回归方程=住院期间并发MODS×2.4+（初始治疗后APACHⅡ评分≥14分）×1.5+使用血管活性药物×2.0。将CRKP感染死亡风险预测模型应用于建模组患者，H-L检验P值为0.866；应用于验证组患者，H-L检验P值为0.807。CRKP感染死亡风险预测模型预测建模组患者发生死亡的ROC曲线下面积（AUC）为0.851〔95%CI（0.774，0.928）〕，最佳截断值为2.75，灵敏度为84.4%，特异度为76.0%，约登指数为0.584。CRKP感染死亡风险预测模型预测验证组患者发生死亡的AUC为0.966〔95%CI（0.894，1.000）〕，最佳截断值为2.95，灵敏度为100%，特异度为87.0%，约登指数为0.870。结论　住院期间并发MODS、初始治疗后APACHEⅡ评分≥14分、使用血管活性药物是CRKP感染患者发生死亡的独立危险因素，依据上述指标拟合出的CRKP感染死亡风险预测模型对CRKP感染患者预后具有较好的预测价值。

关键词: 克雷伯菌感染, 碳青霉烯类耐药肺炎克雷伯菌, 感染, 死亡, 影响因素分析, 预测价值