Using Machine Learning Algorithms to Predict Antimicrobial Resistance and Assist Empirical Treatment

Area of Science:

• Infectious Diseases
• Medical Informatics
• Computational Biology

Background:

• Multi-drug-resistant (MDR) infections pose a significant global health threat, necessitating rapid identification of pathogens and their resistance profiles.
• Current antibiotic resistance testing methods typically exceed 24 hours post-sample collection, delaying crucial treatment decisions.
• Early and accurate antibiotic susceptibility data is vital for effective patient management and combating antimicrobial resistance.

Purpose of the Study:

• To evaluate the efficacy of machine learning (ML) algorithms in predicting antibiotic susceptibility.
• To explore the potential of ML for informing empirical antibiotic treatment selection based on readily available patient and culture data.
• To reduce the time to actionable antibiotic resistance information for clinical decision-making.

Main Methods:

• Five distinct machine learning algorithms were implemented and assessed.
• Models utilized patient demographic data, Gram stain results, and culture findings as input features.
• Antibiotic susceptibility test results served as the ground truth for model training and validation.

Main Results:

• Machine learning algorithms demonstrated the ability to predict antibiotic susceptibility profiles.
• The predictive performance of the tested ML models indicates their potential utility in clinical settings.
• The study successfully identified ML as a viable tool for antimicrobial susceptibility prediction.

Conclusions:

• Machine learning offers a promising approach to expedite antibiotic susceptibility predictions.
• ML-driven insights can significantly assist clinicians in selecting appropriate empirical antibiotic therapies.
• Implementing ML in antimicrobial susceptibility testing can help mitigate the impact of MDR infections.