Machine learning in predicting antimicrobial resistance: a systematic review and meta-analysis

Area of Science:

• Medical Informatics
• Computational Biology
• Epidemiology

Background:

• Antimicrobial resistance (AMR) poses a significant global health challenge.
• Accurate and prompt identification of AMR is crucial for improving patient outcomes and optimizing antibiotic stewardship.
• Machine learning (ML) offers potential tools for predicting AMR patterns.

Approach:

• A systematic literature review was conducted using major databases (PubMed, Web of Science, Embase, IEEE) up to September 2021.
• Studies employing ML algorithms or risk scores for AMR prediction were included, with 25 studies utilizing ML.
• Common ML algorithms included logistic regression, decision trees, and random forests.

Key Points:

• ML models demonstrated a pooled area under the curve (AUC) of 0.82 for AMR prediction.
• Specific resistance patterns like extended-spectrum β-lactamases, MRSA, and carbapenem resistance were frequently studied.
• ML prediction showed higher specificity (87%) compared to risk scores (37%), though sensitivity was comparable.

Conclusions:

• ML holds promise as a technology for predicting antimicrobial resistance.
• Current limitations include retrospective study designs, non-standardized data processing, and a lack of real-world validation.
• Further research and validation are needed to integrate ML models into clinical practice effectively.