Development and External Validation of an Evidence-based Risk Prediction Model for Multidrug-resistant Bacterial Infections in ICU Patients

Abstract

Background Previous research has found that multi-drug resistant (MDR) bacteria can be transmitted between ICU patients, and the infections caused by MDR bacteria may negatively affect the efficacy of current treatment. As the speed of diagnostic testing to identify MDR bacteria is relatively slow in clinical practice, the research regarding the prediction of MDR bacteria infections has developed. Objective To develop an evidence-based risk prediction model for MDR bacterial infections in ICU patients, and to verify it using the real-world clinical data collected retrospectively. Methods Potential risk factors for MDR bacterial infections in ICU patients were identified by a meta-analysis of studies regarding MDR bacterial infections in ICU patients included in databases of PubMed, EMBase, the Cochrane Library, CNKI, Wanfang, China Science and Technology Journal Database, Ace Base of CMA during January 2012 to June 2020 using Stata/SE 12.0 software, and were used to develop a risk prediction model by transforming effect size to the standardized regression (β) coefficient. Next the model was fully established and externally verified using the clinical data of adult ICU patients (n=3 908) recruited from Shanghai General Hospital from January 2018 to June 2021. ROC analysis was used to describe the predictive accuracy of the prediction model. Results Seventeen potential risk factors of MDR bacterial infections in ICU patients were identified through the meta-analysis of 31 included studies. The MDR bacterial infection risk prediction model incorporating these 17 factors with corresponding β value as coefficient (derived from converting the risk effect size of each factor) was developed: Logit (P) =-2.476 3 +0.086X1〔gender (male) 〕+0.191X2 (history of hospitalization) +0.392X3 (being transferred from another hospital) +1.723X4 (length of ICU stay) +0.315X5 (other infections) +0.385X6 (chronic obstructive pulmonary disease) +0.131X7 (diabetes) +0.536X8 (renal disease) +0.285X9 (renal failure) +0.565X10 (dialysis) +0.148X11 (mechanical ventilation) +0.742X12 (central venous catheter) +0.336X13 (urinary catheter) +3.483X14 (types of used antimicrobial drugs) +0.174X15 (history of antimicrobial use) +0.975X16 (history of carbapenems use) +1.151X17 (history of aminoglycosides use) . External verification of the model revealed that the model had 64.36% sensitivity, 80.39% specificity, and 0.447 4 Youden index, and an AUC of 0.724. Conclusion Our model has been proved to have good performance in predicting the MDR bacterial infection risk in ICU patients, as well as relatively good applicability, scientificity, and practicability. The development regimen may be used as a reference for developing a risk prediction model for other diseases.

Keywords

• intensive care units|multi-drug resistant bacteria|logistic models|prediction model|meta analysis|evidence-based theory