Skip links

Predicting Antibiotic Resistance By Applying Machine Learning Techniques

Even before the COVID-19 pandemic, antimicrobial resistance (AMR) posed a significant global public health challenge. According to estimates from the World Health Organization, the European region alone witnesses over 670.000 infections each year caused by bacteria resistant to one or more antibiotics, resulting in approximately 33.000 fatalities. Antimicrobial resistance is a natural phenomenon, driven primarily by genetic changes in microorganisms. However, the misuse and overuse of antimicrobials accelerate the development and selection of drug-resistant pathogens. The most alarming AMR rates are concentrated in southern and eastern parts of Europe, with Italy and Greece carrying the highest estimated burden of antibiotic-resistant bacterial infections.

Currently, the gold standard for detecting AMR relies on classical culture-based antimicrobial susceptibility testing. This method, while reliable, takes around 24 hours for microorganism identification, and an additional 48 hours to report results. This time-consuming process can impact patient outcomes, either by delaying treatment or by forcing clinicians to initiate empiric antibiotic therapy based on educated guesses and local guidelines because immediate intervention is crucial.

Artificial intelligence (AI) and machine learning (ML) techniques have gained increasing significance in the field of medicine, demonstrating promising performances that mark a significant step towards personalized and precision medicine. There are different ways in which ML can be applied to AMR. AI has been employed to design new antibiotics or predict synergistic drug combinations. These models are trained on a dataset of known compound interactions and represent a powerful tool for designing new drug combinations and treatment strategies, thereby optimizing therapeutic outcomes. Researchers have harnessed ML models for the surveillance of AMR and to review antibiotic prescription appropriateness. A study conducted in 2020 on 715 patients admitted to the ED at Tan Tock Seng Hospital, the second-largest adult hospital in Singapore, shows that combining the results from their three AI models, 58.3% of study participants would not need antibiotics.

Kelyon has teamed up with experts and researchers from the University of Salerno, IRCCS Istituto Dermopatico dell’Immacolata (IDI), the Istituto Superiore di Sanità (ISS), and the National Council of Research (CNR) to address the AMR challenge. The research team is analyzing large datasets of antibiotic susceptibility testing to develop new AI models to predict antibiotic resistance. The collaboration started in 2021 with the activation of a doctoral position co-supervised by Kelyon and the University of Salerno. The preliminary results of the research activities were presented at the 10th annual congress of the European Society for Translational Medicine, held on October 6-8, 2023, and at the 18th edition of the Bioinformatics and Computational Biology conference, held on December 5-7, 2023.

It is undoubtedly true that AI will enhance our ability to support healthcare decision-making. Literature in the field of AI/ML applied to microbiology demonstrates that there is a growing interest in finding new solutions to address antimicrobial resistance. Dashboards and data visualization tools, if well integrated into a user-friendly software interface, are extremely helpful to researchers, clinicians, and other stakeholders to assess antibiotic resistance trends, define new treatment policies, and evaluate the effectiveness of antimicrobial stewardship programs or the potential consequences of inaction.

The preliminary results obtained by the research team indicate that machine learning models applied to antimicrobial susceptibility data have the potential to provide valuable predictions about antibiotic resistance profiles, supporting clinicians in selecting the most appropriate therapy on time. These models can be embedded into a data-driven clinical decision support system (CDSS) and be integrated into the clinical workflow, leading to a more personalized antibiotic prescription, and giving clinicians another tool to provide better care and mitigate the burden of antimicrobial resistance.

This website uses cookies to improve your web experience.