The Future of Medicine: AI Detects Resistant Infections at the Diagnostic Stage
Could artificial intelligence help doctors determine at the start of treatment if a child has an antibiotic-resistant infection? A study by researchers from Ariel University and Clalit Health Services found that machine learning models using existing medical record data can accurately assess the risk of infections caused by ESBL-producing bacteria.
Could artificial intelligence help doctors know at the very beginning of treatment if a child with a urinary tract infection has been infected with an antibiotic-resistant bacterium? A study by researchers from Ariel University and Clalit Health Services examined machine learning models designed to do exactly that, using information already existing in the child's medical record at the time of initial diagnosis. The models are designed to assess the risk that the infection was caused by bacteria producing the ESBL enzyme – a mechanism that gives bacteria resistance to major groups of antibiotics and may require treatment different from the standard. Today, when a child is diagnosed with a urinary tract infection, treatment is often given even before the culture result is received, which may take at least two to three days.
About 36,000 cases of urinary tract infections were examined
In the study, data from electronic medical records of Clalit Health Services were analyzed, including about 36,000 events of proven urinary tract infections in children from one month to 18 years of age. The data were collected between January 2010 and August 2020. The researchers examined information that was already available at the first medical encounter, including the child's age and sex, socioeconomic status, background diseases, whether the infection was community-acquired or hospital-acquired, previous use of antibiotics, and history of previous infections. Based on the data, five different machine learning models were developed to assess the likelihood that it was an infection caused by an ESBL-producing bacterium.
Risk jumped 18-fold in children who had already been infected in the past
One of the most prominent findings in the study was the strong link between a previous infection with a resistant bacterium and the risk of recurrence: children who had already suffered from an ESBL infection in the past were at an 18-fold higher risk of suffering from this type of infection in another event. Other factors found to be associated with the risk were the child's age and sex, socioeconomic status, the place where the infection was acquired, previous use of antibiotics, and the type of bacterium that caused the infection.
Goal: to know earlier which antibiotic to give
One of the main advantages of the models was their high ability to rule out a resistant infection. In all five models, a negative predictive value of about 0.98 was recorded, meaning that when the model estimated that the infection was not caused by an ESBL bacterium, in the vast majority of cases the assessment was correct. The possible implication is that a system based on artificial intelligence could in the future provide the doctor with a risk assessment even before the culture results arrive, and thus assist in the decision of whether there is a need to consider antibiotic treatment adapted to resistant bacteria. The need for early identification is particularly significant because infections caused by ESBL-producing bacteria have been linked to higher rates of morbidity, prolonged hospitalizations, and even hospitalization in intensive care units.
Not yet a replacement for a doctor
Alongside the encouraging results, the researchers emphasize that this is a research stage. The models are not intended to replace the doctor's judgment, which is also based on the child's clinical condition, physical examination, and laboratory test results. According to the researchers, further studies are needed to improve the models' performance, validate them in other populations, and especially to examine whether their use in the real world actually improves treatment outcomes. The study points to additional potential for integrating artificial intelligence into medicine: using information already in the medical record to assist in faster and more accurate decision-making, precisely at a time when antibiotic resistance is becoming one of the main challenges for health systems worldwide.
