A Novel Approach to Antimicrobial Resistance: Machine Learning Predictions for Carbapenem-Resistant Klebsiella in ICUs

Overview

The aim of this study to predict carbapenem resistant Klebsiella spp. earlier in our patients monitored in our Intensive Care Unit in the future, using artificial intelligence. Patients with bloodstream infection and pneumonia caused by Klebsiella spp. will be comparatively examined in two groups, as sensitive and resistant. Resistance will be attempted to be predicted with deep machine learning.

Antimicrobial resistance is a globally increasing threat and has serious consequences on both public health and the economy. In an infection that may develop with a resistant microorganism, therapeutic options are limited, hence early and effective treatment that can be initiated by predicting resistance will make a difference in patient prognosis. Today, artificial intelligence and machine learning are changing our medical practice. When the literature is reviewed, there are studies suggesting that machine learning can predict antimicrobial resistance.Risk factors for carbapenem-resistant Klebsiella spp. have been previously identified. These previously identified risk factors will be evaluated retrospectively in our own patients and an algorithm related to the prediction of resistance will be developed with the help of machine learning. Our goal is to predict bacterial resistance earlier in our patients monitored in our Intensive Care Unit in the future, using artificial intelligence, and to facilitate our patients' access to early and effective treatment options. Secondarily, it is also aimed to provide economic benefits by preventing unnecessary antibiotic use. Access to patients' data will be obtained retrospectively through the hospital automation system. Publications in the literature will be examined, and the risk factors causing the development of infection with carbapenem-resistant Klebsiella spp. will be evaluated. Patients with carbapenem resistance and sensitivity will be compared in two separate subgroups. The obtained features will be classified using various decision trees and neural algorithms separately. The data obtained will be statistically compared in the distinction of resistance and sensitivity. Statistical evaluation was done with IBM SPSS 29.0 (IBM Corp., Armonk, NY, USA). Demographic data, descriptive statistics, Categorical variables will be expressed in terms of frequency (percentage). Categorical variables will be expressed with the chi-square test. The performance of Machine Learning algorithms will be evaluated by ROC analysis, AUC, classification accuracy, sensitivity, and specificity values will be calculated.