Machine learning to predict antimicrobial resistance: future applications in clinical practice?

Area of Science:

• Medical Informatics
• Computational Biology
• Infectious Diseases

Background:

• Antimicrobial resistance (AMR) is a growing global health threat.
• Machine learning (ML) is emerging as a powerful tool for predicting AMR patterns.

Purpose of the Study:

• To review the existing literature on the application of ML for predicting AMR.
• To provide physicians with an overview of ML-based AMR prediction methods.

Main Methods:

• Comprehensive literature search across major scientific databases (MEDLINE/PubMed, EMBASE, Google Scholar, ACM, IEEE) up to December 2023.
• Inclusion of 36 studies focusing on ML for AMR prediction, analyzing data sources, applications, and performance metrics.

Main Results:

• Most studies (89%) utilized hospital data, predominantly in high-resource settings (92%).
• Key applications included predicting drug resistance in infected patients (67%) and ML-assisted antibiotic prescription (22%).
• Common inputs were demographics, prior antibiotic susceptibility testing, and antibiotic exposure, with a focus on Gram-negative bacteria resistance prediction (92%). Performance metrics (AUROC) ranged from 0.56 to 0.93.

Conclusions:

• ML holds significant potential to aid in AMR prediction and clinical decision-making.
• The field is rapidly expanding, but further research is required for the design, implementation, and evaluation of ML decision support systems in clinical settings.