AI May Soon Detect Antibiotic-Resistant Infections Early in Children

Researchers from Ariel University and Clalit Health Services have developed artificial intelligence models designed to identify, at the diagnostic stage, whether a urinary tract infection in children is caused by antibiotic-resistant bacteria. The study analyzed approximately 36,000 cases of urinary tract infections in children aged one month to 18 years, using electronic health records from January 2010 to August 2020.

The AI models assess the risk of infection by bacteria producing the ESBL enzyme, a mechanism that confers resistance to major antibiotic classes and may necessitate different treatment. Currently, treatment for urinary tract infections is often initiated before culture results, which can take two to three days, are available. The AI models utilize existing patient data from the initial medical encounter, including age, gender, socioeconomic status, comorbidities, infection source (community or hospital-acquired), prior antibiotic use, and history of previous infections.

A significant finding was the strong correlation between previous infections with resistant bacteria and the risk of recurrence; children with a prior ESBL infection were 18 times more likely to experience such an infection again. Other risk factors identified included age, gender, socioeconomic status, infection source, prior antibiotic use, and the specific bacterial strain.

The models demonstrated a high capability in ruling out resistant infections, with a negative predictive value of about 0.98 across all five models. This suggests that an AI-based system could provide physicians with a risk assessment before culture results are obtained, aiding decisions on whether to consider antibiotics targeting resistant bacteria. Early detection is crucial as ESBL-producing bacteria are linked to higher morbidity, prolonged hospital stays, and intensive care unit admissions.

Despite promising results, the researchers emphasize that this is an early research phase. The models are not intended to replace a physician's clinical judgment, which also considers the child's overall condition, physical examination, and laboratory results. Further research is needed to improve model performance, validate them in diverse populations, and assess their real-world impact on treatment outcomes. The study highlights AI's potential to enhance medical decision-making by leveraging existing health data, particularly amid the growing global challenge of antibiotic resistance.