| BackgroundThermal burns are severe injuries.Burn patients are at high risk of developing bacterial infections during their stay in the Intensive Care Unit(ICU).Due to the loss of protective skin barrier and the decline of immune function,the pathogen is easy to colonize,and infection becomes the main cause of death in burn patients.If infection is not intervened in time,sepsis,systemic inflammatory response syndrome and multiple organ failure may develop in severe cases,which are life-threatening.According to statistics,more than 75%of burn deaths are directly attributed to infection,and multi-drug resistant bacteria(MDRO)increase the mortality rate of burn-related sepsis patients from 42%to 86%.At present,the primary method to control infection is to use antibacterial drugs,but the increasingly severe situation of bacterial resistance makes the selection of antibacterial drugs full of challenges,and the selection of effective antibacterial drugs is essential.At present,the investigation of pathogens in burn ICU mainly focuses on southwest China and coastal areas.There are few epidemiological surveys of pathogens in burn ICU in northwest China,and there are differences in the prevalence of pathogens in burn ICU in different areas.In order to make empirical medication well documented,it is necessary to understand the types and drug resistance of pathogens of nosocomial burn ICU infection.In addition,burn ICUs are the hardest hit areas of MDRO,and early intervention in timely identification of high-risk patients with MDRO infection can improve the prognosis of severe burns.Therefore,this study retrospectively analyzed the distribution characteristics and drug resistance of infectious pathogens in burn patients admitted to the burn ICU of the hospital,clarified the current status of burn ICU infection,and used multidrug resistant Acinetobacter baumannii(MDR-AB)and methicillin-resistant Staphylococcus aureus(MRSA)as examples to construct a risk prediction model using logistic regression,neural network and decision tree to predict the occurrence of MDRO infection and provide data support for the development of effective prevention and control strategies.MethodsThe first part of this study used a retrospective analysis,and hospitalized patients who visited the burn ICU of a tertiary care hospital in Xi’an from January 1,2015 to December 31,2021 were selected as the study subjects.Burn ICU inpatient information was collected from this hospital information system(HIS).The survey included demographic characteristics,medical conditions,diagnostic results,drug use,invasive procedures and infection-related laboratory tests:types of submitted specimens,detected pathogens and drug susceptibility test results.To analyze the distribution of infectious pathogens and drug resistance of main pathogens in burn ICU patients.The second part of the study design used a case-control study,selected burn ICU patients(length of stay>7 days)as the study subjects,and identified MDR-AB and MRSA infections as the case group by drug susceptibility testing and drug resistant bacteria surveillance.In order to eliminate other possible interfering factors,the closest matching algorithm of Propensity Score Matching(PSM)was used to match the case group and the control group,and information on burn area,surgical procedures,and drug use was collected from the patients,and univariate analysis was used to preliminarily screen MDR-AB and MRSA infection risk factors.In order to predict the risk of MDR-AB and MRSA infection more accurately,logistic regression,artificial neural network and decision tree model were used to establish the prediction model.In the screening of prediction model,the area under receiver operating characteristic curve(ROC)curve and prediction accuracy are used as evaluation indexes to select the optimal prediction model.Results1.a total of 15,304 microbial samples were collected from 746 patients from 2015 to 2021,including 10,070 positive samples,with a positive rate of 65.8%,and a total of 2,022 pathogens were isolated.Among them,1265 strains were from wound secretion,accounting for 57.4%;228 strains were from blood,accounting for 10.4%;186 strains were from catheter,accounting for 8.5%;411 strains were from sputum,accounting for 18.7%;58 strains were from urine,accounting for 2.7%;32 strains were from tissue,accounting for 1.4%;1338 strains were gram-negative bacteria,accounting for 60.7%;616 strains were gram-positive bacteria,accounting for 28.0%;248 strains were fungi,accounting for 11.3%;the top 10 pathogens detected were Acinetobacter baumannii,Staphylococcus aureus,Pseudomonas aeruginosa,Klebsiella pneumoniae,Proteus mirabilis,Enterobacter cloacae,Candida albicans,Escherichia coli,Enterococcus faecium,Hemolytic streptococcus,and Candida tropicalis.The proportion of multidrug-resistant Acinetobacter baumannii(80.6%in 2015 and 90.4%in 2021)increased significantly.Multidrug-resistant P.aeruginosa declined substantially from 83.2%in 2015 to 24.6%in 2021.The multidrug-resistant rate of K.pneumoniae was 72.5%in 2015 and 81.5%in 2021.MRSA isolated accounted for the highest 95.3%of S.aureus isolated.The top four pathogens in composition ratio:Staphylococcus aureus,Acinetobacter baumannii,Pseudomonas aeruginosa,and Klebsiella pneumoniae,except Pseudomonas aeruginosa,had a significant increase in resistance to antibiotics.2.Univariate analysis was performed to identify possible risk factors,and logistic regression,neural network and decision tree models were substituted.The results showed that large burn area(OR=2.425,95%CI:1.777～3.308),inhalation injury(OR=1.374,95%CI:1.293～4.698),length of hospital stay ≥ 30 days(OR=2.786,95%CI:1.675～4.493),delayed resuscitation(OR=1.021,95%CI:1.012～1.155),blood transfusion and its products(OR=1.213,95%CI:1.057～1.654),antibiotic use>7 days(OR=2.236,95%CI:1.566～3.163),antibiotic combination(OR=1.166,95%CI:1.068～3.079),tracheotomy(OR=1.221,95%CI:1.134～2.207)were risk factors for MDR-AB infection(P<0.05).95%CI:1.152～1.748),history of diabetes(OR=1.374,95%CI:1.193～2.698),and delayed resuscitation(OR=1.121,Fig.95%CI:1.012～1.455)and deep venous catheterization(OR=2.236,95%CI:1.366～3.863)were risk factors for MRSA infection(P<0.05),and the number of operations>2(OR=0.592,95%CI:0.387～0.904)was a protective factor for MRSA infection.3.In this study,three different models were developed for MDR-AB and MRSA infection risk prediction.For MDR-AB infection,the differences between the three models were not found to be significant by comparing the area under the ROC curve of the models.After considering the prediction accuracy,the neural network model has the best prediction ability,and its prediction accuracy is 74.4%,which is slightly higher than that of logistic regression model and decision tree model.For MRSA prediction models,in addition to comparing the area under the ROC curve,prediction accuracy was combined for comparison.The results showed that the neural network model had the best prediction effect,with an area under the ROC curve of 0.820 and a prediction accuracy of 75.1%.The logistic regression model had an area under the ROC curve of 0.812 and a prediction accuracy of 74.7%,which performed better than the decision tree model(area under the ROC curve of 0.754 and a prediction accuracy of 71.7%),and the difference was statistically significant.The results of this study showed that the neural network model had a high predictive ability in MDR-AB and MRSA infection risk prediction,while the logistic regression model also performed well and could be used as an alternative,and the decision tree model had a slightly poor predictive effect.Conclusion1.during the study period,the main pathogens circulating in the burn ICU of this hospital were Acinetobacter baumannii,Staphylococcus aureus,Pseudomonas aeruginosa,and Klebsiella pneumoniae,and the multidrug resistance ratios of Acinetobacter baumannii,Staphylococcus aureus,and Klebsiella pneumoniae gradually increased in addition to Pseudomonas aeruginosa.2.Combined with three predictive models,burn ICU patients with large burn area,inhalation injury,length of hospital stay≥30 days,delayed resuscitation,blood transfusion and its products,antibiotic use>7 days,antibiotic combination,tracheotomy are independent risk factors for MDR-AB infection,burn ICU patients with burn area,number of operations,deep vein catheterization,history of diabetes,delayed resuscitation,antibiotic use time are closely related to MRSA infection,and number of operations>2 are protective factors for MDR-AB and MRSA infection.Medical staff should actively screen high-risk groups and strengthen the rational use of antibiotics.3.In MDRO infection prediction,neural network models performed better than logistic regression models and decision tree models in terms of overall accuracy of prediction and area under the ROC curve,but the applicable conditions of each model need to be further explored in clinical practice to obtain the highest predictive value in practice. |
PDF Full Text Request