Machine Learning Techniques to Identify Antimicrobial Resistance in the Intensive Care Unit

Area of Science:

• Infectious Diseases
• Medical Informatics
• Computational Biology

Background:

• Antimicrobial resistance (AMR) poses a significant global health threat, with increasing social and economic burdens.
• Rapid identification of bacterial resistance is critical for patients in intensive care units (ICUs).
• Conventional antibiotic resistance tests take 24-48 hours post-culture, delaying crucial treatment decisions.

Purpose of the Study:

• To apply machine learning (ML) techniques for predicting bacterial resistance to antimicrobials.
• To identify potential relationships between patient data, culture results, and antimicrobial resistance patterns.
• To provide a faster alternative to traditional resistance testing in critical care settings.

Main Methods:

• Utilized machine learning techniques to analyze clinical, demographic, culture, and antibiogram data.
• Employed correspondence analysis to visualize relationships between bacteria and antimicrobial families.
• Evaluated ML model performance based on antimicrobial family and identified trends in resistance.

Main Results:

• ML models demonstrated effectiveness in identifying antimicrobial resistance, with performance varying by antimicrobial family.
• Non-linear relationships were identified, offering insights into resistance mechanisms.
• Correspondence analysis revealed graphical relationships between bacteria and antimicrobial classes.
• A discernible shift in antimicrobial resistance trends was observed.

Conclusions:

• Machine learning offers a promising approach for the rapid prediction of antimicrobial resistance in ICUs.
• This methodology can significantly reduce the time to identify resistance, enabling quicker therapeutic interventions.
• Understanding the complex relationships between bacteria, antimicrobials, and patient factors is key to combating AMR.