关键词: 耐碳青霉烯肠杆菌目细菌, mcr基因, 继发血流感染, 风险预测模型, 机器学习, 重症监护病房

Abstract:

Objective To investigate the characteristics of the mcr gene of carbapenem-resistant Enterobacteriaceae（CRE）in anal swabs of intensive care unit（ICU）patients and its relation with secondary bloodstream infection（BSI），and to construct a risk prediction model for secondary BSI in ICU patients. Methods Totally，700 patients who were admitted to the ICU of Yellow River Conservancy Commission Yellow River Central Hospital from March 2020 to March 2023 were enrolled as modeling group. The patients in the modeling group were classified into infection group（85 cases）and non-infection group（615 cases）based on whether they had secondary BSI. Totally，300 ICU patients who met the same inclusion and exclusion criteria in Yellow River Conservancy Commission Yellow River Central Hospital from April 2023 to August 2024 were enrolled as validation group. The patients in the validation group were classified into infection group（36 cases）and non-infection group（264 cases）. Univariate and stepwise Logistic regression analysis was used to analyze the influencing factors of secondary BSI in ICU patients. Logistic regression，decision classification regression tree（CRT）and backpropagation neural network（BPNN）algorithms were used to construct a risk prediction model for secondary BSI in patients，and the efficiency was compared using receiver operating characteristic（ROC）curve for ICU patient. Results Age，intra-abdominal infection，ICU stay time，blood culture Pitt bacteremia score（PBS），mcr-9 in the strain and neutrophil deficiency were all factors that affected secondary BSI in ICU patients（P<0.05）. CRT had the optimal efficiency in predicting whether ICU patients would have secondary BSI. Factors used for classification were ICU stay time，strain carrying mcr-9，PBS，patient age，where ICU stay time is the root node. The area under curve（AUC）was 0.984，the sensitivity was 85.90%，and the specificity was 98.90%. The external validation results showed that the predictive efficiency of the 3 models was good，with AUC>0.900（P>0.05）. Conclusions There is a relation between the mcr-9 gene and secondary BSI in ICU patients. The constructed prediction models for secondary BSI in ICU patients based on machine learning have good predictive efficiency.

Key words: Carbapenem-resistant Enterobacteriaceae, mcr gene, Secondary bloodstream infection, Risk prediction model, Machine learning, Intensive care unit