Antimicrobial minimum inhibitory concentrations can be imputed from phenotypic data using a random forest approach.

Area of Science:

• Veterinary microbiology
• Public health
• Computational biology

Background:

• Antimicrobial resistance (AMR) poses a significant public health threat, necessitating cross-sectoral monitoring.
• Data challenges like missing values and protocol changes hinder accurate AMR trend analysis.
• Machine learning offers a potential solution for imputing missing antimicrobial susceptibility data.

Purpose of the Study:

• To develop and evaluate machine learning models for imputing missing minimum inhibitory concentrations (MICs).
• To assess the accuracy of these models using internal and external datasets.
• To enhance the evaluation of AMR trends and inform public health policies.

Main Methods:

• Random forest models were trained using cattle-associated Escherichia coli data from the National Antimicrobial Resistance Monitoring System.
• Models predicted MICs for 10 antimicrobials based on isolate metadata and other MICs.
• Performance was validated on held-out test data and external datasets from chickens and humans.

Main Results:

• Models achieved over 80% accuracy for all 10 antimicrobials and over 90% for 5 antimicrobials on test data.
• Six models demonstrated consistent performance across test, human, and chicken datasets.
• Four models showed similar accuracy on human data but reduced accuracy on chicken data.

Conclusions:

• The developed machine learning models accurately predict MIC values, suitable for imputing missing data in AMR surveillance.
• Accurate imputation can improve AMR trend evaluation, inform stewardship, and streamline susceptibility testing.
• These models offer a promising tool for enhancing AMR monitoring and public health strategies.