Predicting the distribution patterns of antibiotic-resistant microorganisms in the context of Jordanian cases using machine learning techniques

Abstract

Antimicrobial resistance (AMR) is identified as the fourth leading cause of mortality in Jordan. However, there is a scarcity of data addressing the demographics and clinical characteristics associated with AMR against commonly used antibiotics in Western Jordan. To address this knowledge gap, a retrospective analysis was undertaken on the microbiology records of AMR at Al-Hussein/Salt Hospital in Jordan West from October 2020 to December 2022 and included 2893 reports. Two machine learning (ML) models, specifically categorization regression trees (CARTs) and random forests (RFs) were trained using microbiology reports and then utilized to forecast the AMR for different categories of antibiotics. The most commonly isolated microorganisms were Escherichia coli (53.3%), Klebsiella pneumoniae, and Staphylococcus aureus. Bacterial strains belonging to the Enterococcus faecium, Staphylococcus aureus, Klebsiella pneumoniae, Acinetobacter baumannii, Pseudomonas aeruginosa, and Enterobacter species category demonstrated elevated levels of resistance. The RF model demonstrated superior accuracy compared to the CART, exhibiting a range of 0.64–0.99. This finding suggests a significant level of dependability in the predictive capability of the RF models in forecasting AMR patterns. AMR is susceptible to the impact of demographic factors such as age, sex, and bacterial species. This study emphasized the significance of monitoring AMR to facilitate the administration of appropriate antibiotic therapy.