ICU患者肺部真菌感染预测模型的建立和预测效能评价  

Establishment of prediction model for pulmonary fungal infection in ICU patients and evaluation of prediction efficiency

作　　者：高兴娟[1] 杨兴唐 刘守珠[1] 程磊[1] GAO Xingjuan;YANG Xingtang;LIU Shouzhu;CHENG Lei(Department of Laboratory Medicine,Qinghai Red Cross Hcxspital,Xining,810001,China)

机构地区：[1]青海红十字医院检验科,西宁810001

出　　处：《临床血液学杂志》2021年第6期403-406,411,共5页Journal of Clinical Hematology

摘　　要：目的:基于ICU肺部真菌感染患者影响因素,建立预测模型,并评价模型预测效能。方法:分析2018年2月—2020年6月ICU住院患者198例,收集所有患者临床资料,包括真菌类别、临床症状、年龄、性别、激素使用时间、广谱抗生素种类、抗生素使用时间、基础疾病、机械通气、血清白蛋白、肾功能、留置管、右心衰竭、深静脉置管、粒细胞高低,以是否为继发性肺部真菌感染分为感染组(102例)、非感染组(96例),进行单因素、多因素分析,并建立风险预测模型。结果:共检出白假丝酵母菌46例,占45.10%。单因素分析结果显示,感染组、非感染组在激素使用时间、广谱抗菌药物使用时间、机械通气、血清白蛋白、深静脉置管、糖尿病方面差异均有统计学意义(t=7.098~40.030,χ^(2)=7.540~19.251,均P<0.05)。预测模型:P=1/{1+EXP[-(2.177+0.166×激素使用时间+0.829×机械通气+0.067×抗菌药物使用时间-1.150×血清白蛋白+0.439×深静脉置管+0.610×糖尿病)]},预测模型受试者工作曲线面积(AUC)=0.857(95%CI:0.801~0.913)与随机面积0.500比较差异有统计学意义。以预测概率0.511(Youden指数最大)作为切割点,预测模型的灵敏度、特异性、准确率分别为82.35%、83.33%、82.83%,当切割点为0.500时,预测模型的灵敏度、特异性、准确率分别为83.33%、78.13%、80.81%。结论:以激素使用时间、广谱抗菌药物使用时间、机械通气、血清白蛋白、深静脉置管、糖尿病建立的风险预测模型可为ICU患者肺部真菌感染的早期诊断提供参考。Objective: To establish a predictive model and evaluate the predictive efficacy of the model based on the influencing factors of patients with pulmonary fungal infection in ICU. Methods: A total of 198 inpatients in the ICU of our hospital from February 2018 to June 2020 were analyzed, and clinical data of all patients were collected, including fungal type, clinical symptoms, age, gender, hormone use time, broad-spectrum antibiotics type, antibiotic use time, basic diseases, mechanical ventilation, serum albumin, renal function, indwelling catheter, right heart failure, deep venous catheterization and level of granulocytes. The patients were divided into infection group(102 cases) and non-infected group(96 cases) according to whether it was secondary pulmonary fungal infection, and conducted single-factor and multi-factor analysis, and established a risk prediction model. Results: A total of 46 cases of Candida albicans were detected, accounting for 45.10%. Univariate analysis results showed that the infection group and the non-infectious group had statistically significant differences in hormone use time, broad-spectrum antibiotic use time, mechanical ventilation, serum albumin, deep venous catheterization, and diabetes(t=7.098-40.030, χ^(2)=7.540-19.251, all P<0.05). Prediction model was P=1/{1+EXP[-(2.177+0.166× hormone use time+0.829×mechanical ventilation+0.067×antibacterial drug use time-1.150×serum albumin+0.439×deep venous catheter+0.610×diabetes) ]}. The difference between the predictive model receiver operating curve area(AUC)=0.857(95%CI: 0.801-0.913) and the random area 0.500 was statistically significant. Taking the prediction probability of 0.511(the largest Youden index) as the cutting point, the sensitivity, specificity, and accuracy of the prediction model were 82.35%, 83.33%, and 82.83%, respectively. When the cutting point was 0.500, the sensitivity, specificity and accuracy of the prediction model were 83.33%, 78.13%, and 80.81%, respectively. Conclusion: The risk prediction model estab

关 键 词：真菌感染 预测模型 危险因素 病原体 临床特点 

分 类 号：R512.5[医药卫生—内科学]