Detection of extended-spectrum β-lactamase-producing bacteria isolated directly from urine by infrared spectroscopy and machine learning.

Resistant bacteria have become one of the leading health threats in the last decades. Extended-spectrum β-lactamase (ESBL) producing bacteria, including Escherichia (E.) coli and Klebsiella (K.) pneumoniae (the most frequent ones), are a significant class out of all resistant infecting bacteria. Due to the widespread and ongoing development of ESBL-producing (ESBL+) resistant bacteria, many routinely used antibiotics are no longer effective against them. However, an early and reliable ESBL+ bacteria detection method will improve the efficiency of treatment and limit their spread. In this work, we investigated the capability of infrared (IR) spectroscopy based machine learning tools [principal component analysis (PCA) and Random Forest (RF) classifier] for the rapid detection of ESBL+ bacteria isolated directly from patients' urine. For that, we examined 1881 E. coli samples (416 ESBL+ and 1465 ESBL-) and 609 K. pneumoniae samples (237 ESBL+ and 372 ESBL-). All samples were isolated directly from the urine of midstream patients. This study revealed that within 40 min of receiving the patient urine it is possible to determine the infecting bacterium as E. coli or K. pneumoniae with 95% success rate while it was possible to determine the ESBL+E. coli and ESBL+K. pneumoniae with 83% and 78% accuracy rates, respectively.