A Bayesian approach to modeling antimicrobial multidrug resistance.

Area of Science:

• Microbiology
• Computational Biology
• Statistical Modeling

Background:

• Multidrug resistance (MDR) poses a significant threat to public health and effective bacterial infection treatment.
• Current MDR identification methods are often belated, relying on large proportions of resistant isolates.
• Detecting correlations in minimum inhibitory concentration (MIC) and binary susceptibility can enable proactive MDR identification.

Purpose of the Study:

• To develop a proactive method for identifying multidrug-resistant bacteria.
• To establish a Bayesian framework for estimating joint antibiotic resistance levels.
• To infer correlations in latent MIC and binary susceptibility data.

Main Methods:

• Utilized a Bayesian framework with a Gaussian mixture model to estimate joint resistance levels for multiple antibiotic classes.
• Incorporated a latent MIC variable to handle censored data and a latent class variable for MIC components.
• Applied the model to Salmonella heidelberg isolates from the National Antimicrobial Resistance Monitoring System (NARMS).

Main Results:

• The proposed Bayesian model accurately and robustly estimates correlations in antibiotic resistance compared to current methods.
• Identified significant joint resistance patterns, specifically between Amoxicillin-clavulanic acid & Cephalothin, and Ampicillin & Cephalothin in Salmonella heidelberg.
• Demonstrated the model's capability to infer correlations from both quantitative MIC and binary susceptibility data.

Conclusions:

• The developed Bayesian framework provides a timely tool for early detection of multidrug resistance (MDR).
• Estimated large correlations serve as an early warning signal for clinical intervention against MDR bacteria.
• This proactive approach enhances the management of bacterial infections exacerbated by multidrug resistance.