Prediction of multidrug-resistant bacteria (MDR) hospital-acquired infection (HAI) and colonisation: A systematic review.

Abstract

Background: Hospital-Acquired Infections (HAI) represent a public health priority in most countries worldwide. Our main objective was to systematically review the quality of the predictive modeling literature regarding multidrug-resistant gram-negative bacteria in Intensive Care Units (ICUs).

Methods: We conducted and reported a Systematic Literature Review according to the recommendations of the PRISMA statement. We analysed the quality of the articles in terms of adherence to the TRIPOD checklist.

Results: The initial search identified 1935 papers and 15 final articles were included in the review. Most studies analysed used traditional prediction models (logistic regression), and only three developed machine-learning techniques. We noted poor adherence to the main methodological issues recommended in the TRIPOD checklist to develop prediction models, such as handling missing data (20% adherence), model-building procedures (20% adherence), assessing model performance (47% adherence), and reporting performance measures (33% adherence).

Conclusions: Our review found few studies that use efficient alternatives to predict the acquisition of multidrug-resistant gram-negative bacteria in ICUs. Furthermore, we noted a lack of strategies for dealing with missing data, feature selection, and imbalanced datasets, a common problem in HAI studies.