Study On Risk Factors And Prediction Model Of Multi-drug Resistant Bacteria Infection In ICU Patients

Objective To clarify the current situation of ICU patients with multidrug-resistant bacteria(MDRO)infection.The risk factors of MDRO infection were explored by taking MDR-AB and MRSA as examples,and the risk prediction model was established to predict the occurrence of MDRO infection,which may support for the reasonable prevention and control strategy of MDRO infection.Methods Patients who were admitted to ICU(hospital stay > 48h)in a tertiary general hospital in Ningxia from July 1,2018 to December 31,2020 were selected as the research objects in this case-control study.MDR-AB and MRSA infection were identified as case group by drug sensitivity test and drug-resistant bacteria monitoring.The control group was determined by the nearest neighbor matching algorithm of propensity score(PSM),and the age,gender,hospital area and year of hospitalization were used as matching variables.The risk factors of MDR-AB and MRSA infection were screened by single factor analysis.The risk prediction models of MDR-AB and MRSA infection were constructed by Logistic regression,artificial neural network and decision tree model.The area under ROC and prediction accuracy were used to select the optimal prediction model.Results 1.A total of 688 ICU MDRO patients were found in 4502 ICU hospitalized patients from July 1,2018 to December 31,2020.The detection rate was 15.24%,with a total detection rate of 18.50%(688/3718).Among them,the detection rates of MDRO in 2019 and2020 were 21.36%(304/1423)and 17.82%(298/1672)respectively.The results showed that gram negative bacteria accounted for 84.74% of MDRO,and the top three bacteria were MDR-AB,E.coli and KPN.MRSA accounted for a large proportion of Gram-positive bacteria,and the detection rate was 20.99%.Most of MDRO samples were from sputum,accounting for 68.02%,followed by blood and urine,accounting for 9.88% and 8.43%respectively.2.Univariate analysis identified the candidate variables of risk factors and substituted them into Logistic regression model,neural network model and decision tree model.The results showed that the duration of antibiotic use(OR=2.248,95% CI: 1.436～3.520)and combination(OR=1.918,95%CI: 1.485～2.476),GCS score(OR=2.359,95%CI:1.601～3.476),length of stay(OR= 2.248,95% CI: 1.436～3.520),times of hospitalization(OR= 3.284,95% CI: 2.041～5.285)and central venous catheterization(OR = 2.002,95% CI:1.325～3.024)were independent risk factors for MDR-AB.The length of hospital stay(OR=8.607,95%CI: 3.567～20.768),number of hospital stay(OR=2.265,95% CI:1.014～5.060),respiratory failure(OR=2.553,95%CI: 1.147～5.683)and antibiotic use time(decision tree model),blood transfusion and product history(decision tree model)were closely related to MRSA infection.3.In prediction model of MDR-AB,the area under ROC curve of Logistic regression model,neural network model and decision tree model showed no significant difference(P>0.05).Combined with the prediction accuracy,the prediction ability of neural network model(84.0%)was better than that of Logistic regression model(72.2%)and decision tree model(71.6%).In prediction model of MRSA,the area under curve(AUC)combined with the accuracy of prediction can be obtained.The neural network model(0.814,89.5%)is better than the Logistic regression model(0.742,84.0%),and the Logistic regression model is better than the decision tree model(0.660,80.0%),and the difference is statistically significant.Conclusion 1.During the study period,the total detection rate of MDRO was 18.50% in a tertiary hospital in Ningxia.The detection rates of MDRO in 2019 and 2020 were 21.36%and 17.82% respectively.The detection rate of MDRO in research hospitals showed a downward trend and was slightly lower than that in domestic hospitals of the same level.2.Combined with the three prediction models,the independent risk factors of MDR-AB were the duration and combination of antibiotics,GCS score,length of hospital stay,times of hospital stay and central venous catheterization.The duration of hospital stay,times of hospital stay,suffering from respiratory failure,time of antibiotic use,history of blood transfusion and blood products were closely related to MRSA infection.Medical staff should use drugs rationally and screen high-risk groups actively.3.Combined with the prediction accuracy and the area under the ROC curve,the prediction ability of neural network model is better than Logistic regression model and decision tree model.However,in clinical practice,it is still necessary to further explore the applicability of each model in combination with the actual situation,so as to obtain the highest prediction value in practice.