Predicting antimicrobial resistance of bacterial pathogens using time series analysis

Area of Science:

• Veterinary Medicine
• Microbiology
• Data Science

Background:

• Antimicrobial resistance (AMR) poses significant global health and economic threats.
• Effective AMR surveillance in food animal production is crucial but challenged by costly and time-consuming detection methods like Minimum Inhibitory Concentration (MIC) testing.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting the future AMR burden in bacterial pathogens.
• To provide a more efficient and accurate tool for routine AMR surveillance in food animal production.

Main Methods:

• Collected pathogen and antimicrobial data from over 600 US farms (2010-2021) to create AMR time series data.
• Applied machine learning, specifically Seasonal Auto-Regressive Integrated Moving Average (SARIMA), to predict AMR trends.
• Compared SARIMA performance against five baseline models, including ARMA and ARIMA.

Main Results:

• The SARIMA model demonstrated superior performance in predicting AMR trends compared to baseline models.
• The study successfully generated predictive AMR time series data for five key bacterial pathogens: *Escherichia coli, Streptococcus suis, Salmonella sp., Pasteurella multocida*, and *Bordetella bronchiseptica*.

Conclusions:

• Machine learning, particularly SARIMA, offers a powerful tool for predicting AMR burden in food animal pathogens.
• This predictive capability can enhance AMR surveillance strategies, potentially reducing costs and improving response times.
• The methodology can be extended to predict AMR for other bacterial pathogens beyond those studied.