Construction And Application Of Risk Prediction Model For Catheter-associated Urinary Tract Infection Among Patients In Neuro-intensive Care Unit

Objective1.To systematically review the risk prediction models of catheter-associated urinary tract infection(CAUTI),and analyze the strengths,limitations,and risk factors of existing CAUTI models.2.To analyze the current situation and risk factors of CAUTI in neuro-intensive care unit(NICU)patients,and to construct and verify the risk prediction model of CAUTI in NICU patients.3.To analyze the in-hospital mortality of patients with CAUTI,identify the risk factors of in-hospital mortality,and construct and externally validate the risk prediction model for CAUTI patients.4.According to the CAUTI risk prediction model,the CAUTI risk early warning system for neurological critical patients was further developed for subsequent application.Methods1.PubMed,Cochrane Library,Web of Science,Scopus,Embase,CNKI,VIP,and WanFang database were systematically searched for papers on risk prediction models for CAUTI in hospitalized patients.These databases were searched up until September 30,2022.The data was extracted and the literature screened independently by two researchers.The risk of bias and applicability of the included studies were analyzed and the content was integrated using PROBAST,a tool for assessing the risk of bias of predictive models.2.Based on the risk factors identified in Part I,a database of CAUTI was constructed.This study conducted a retrospective analysis of patients who were admitted to the NICU at Subei People’s Hospital of Jiangsu Province between January 2019 and January 2020.The patients were randomly assigned to either the training or validation group,in a ratio of 7:3.Data was collected on the patients’ general condition,clinical characteristics,and laboratory examination.Univariate and multivariate analyses were used to analyze the independent risk factors of CAUTI.A nomogram risk prediction model was constructed according to the results of multivariate logistic regression.The performance of the nomogram model was assessed using both receiver operator characteristic curve(ROC)analysis and calibration curve.Additionally,a decision curve analysis was conducted to determine the clinical impact value of the nomogram.3.Using SQL statement code to obtain CAUTI target population data in the critical care database eICU and MIMIC-Ⅳ,of which eICU database was used as the modeling group and MIMIV-Ⅳ as the external validation.We extracted data on demographic characteristics,vital signs,complications,severity of illness,and laboratory indicators of patients diagnosed with CAUTI from both databases.We used five machine learning methods to screen variables and construct risk models for in-hospital mortality of CAUTI patients.Accuracy,precision,recall,F1 score,sensitivity,specificity and AUC were used to choose the best model and further construct a visual nomogram model.4.Based on the CAUTI risk prediction model of NICU patients and the in-hospital mortality risk prediction model,the CAUTI risk early warning system of neurological critical patients was further developed by C#language and Fine UI framework.Results1.A total of 14 studies on the development of CAUTI risk prediction models in hospitalized patients were included through systematic search,and a total of 21 risk prediction models were developed.The AUC of the models ranged from 0.725 to 0.970.Three studies were externally validated and two studies were applied.Duration of indwelling catheter,age,length of hospital stays,diabetes,albumin level,antibiotic used,ICU admission,renal dysfunction,and female sex were the most included predictors in the included models.All 14 studies had a high risk of bias,and applicability was unclear,mainly because missing data were not reported or handled,blinding was not reported,screening factors based on univariate analysis,and incomplete evaluation of model performance.2.The incidence of CAUTI among patients in NICU was 3.91%.The main pathogenic bacteria of CAUTI were gram-negative bacteria,and Escherichia coli was the most common pathogenic bacteria.Multivariate regression analysis showed that age≥60 years(OR:14.12,95%CI:1.57-126.96),epilepsy(OR:26.89,95%CI:4.51-178.17),length hospitalization(OR:48.71,95%CI:3.27-726.53)and low albumin levels(<35g/L)(OR:6.58,95%CI:1.49-29.06)were independent risk factors for CAUTI among NICU patients.The nomogram model showed good discrimination and calibration in the training and validation group,and the decision curve analysis also showed good clinical application value.3.We finally recruited 863 and 212 CAUTI patients from eICU and MIMIC-Ⅳ databases,respectively.The in-hospital mortality of CAUTI patients was 13.7%and 11.3%,respectively.Among the five machine learning models,the traditional logistic model had the best prediction performance(AUC=0.823),and the AUC of external validation was 0.765,which also showed good stability.The nomogram model constructed by the optimal logistic model included four predictors:age,intubation,simplified acute physiology-Ⅱ score and mean corpuscular volume.In addition,the calibration curve showed satisfactory agreement between the predicted and actual results.The decision curve covered a large threshold probability,and the model also showed good clinical application value.4.This study developed a CAUTI risk early warning system for neurological critical patients,including five parts:user login and registration,patient management,CAUTI risk prediction,system management and account management.By inputting the predictors of CAUTI,the risk probability of CAUTI can be effectively obtained.Conclusion1.There are still some limitations in CAUTI prediction models,and all models have high risk of bias and methodological defects.Future research should enhance both research design and report,carry out large sample and multi-center research,combine big data and machine learning algorithm analysis methods,develop prediction models with excellent prediction performance,simple and convenient to use,and constantly update and calibrate in the process of use to provide practical risk prediction models for clinical practice.2.A simple and feasible nomogram model was developed to predict the risk of CAUTI among NICU patients.The nomogram incorporated clinical variables at admission(age,admission diagnosis,albumin)and length of hospital stay.The nomogram can not only effectively predict the possibility of CAUTI in critically ill neurological patients,but also can be used as a tool to judge the timing of catheter removal,improve the quality of care for critically ill neurological patients and reduce the incidence of CAUTI.3.Based on a large critical care patient database,a machine learning-based nomogram model was developed to predict the in-hospital mortality of CAUTI patients,and the traditional logistic model was found to have the best predictive value.The AUC of the model in the external validation cohort was 0.765,and the accuracy was 0.906.The model has good discrimination ability,quantifies the probability of infection of patients,accurately identifies high-risk patients,and is easy to operate.It is expected to be used routinely in ICU to assist medical staff to provide appropriate treatment and care for critically ill patients,especially those with uncertain survival outcomes.4.The CAUTI risk early warning system for critically ill neurological patients developed,on the one hand,collect long-term and systematic data information of critically ill neurological patients and follow up and manage them after discharge.On the other hand,the system can predict the risk of CAUTI in NICU patients through the risk factors of CAUTI,and allows for targeted screenings of individuals with a higher probability of developing CAUTI and can guide nursing practices in order to improve the overall quality of care.