Interpretative reading of the antibiogram--a semi-naïve Bayesian approach

Area of Science:

• Medical Microbiology
• Computational Biology
• Infectious Diseases

Background:

• Antibiograms (ABG) provide in vitro antimicrobial susceptibility test results for pathogens.
• Institutional cross-ABGs represent conditional probabilities of antimicrobial susceptibility between drug pairs.
• Current methods struggle to optimally integrate isolate-specific and institutional data for susceptibility prediction.

Purpose of the Study:

• To explore the interpretative reading of isolate antibiograms to enhance institutional antibiogram prior probabilities.
• To develop and evaluate Bayesian approaches for predicting antimicrobial susceptibility.
• To improve the accuracy of predicting antimicrobial resistance and susceptibility patterns.

Main Methods:

• Utilized a database of 3347 clinically significant blood isolates from an Israeli university hospital.
• Calculated average institutional antibiograms and cross-antibiograms for 14 pathogen groups.
• Employed naïve and semi-naïve Bayesian methods, including a novel min2max2 approach, evaluating performance using normalized Brier distance and 5-fold cross-validation.

Main Results:

• The naïve Bayes method reduced the normalized Brier distance from 37.7% to 28.2%.
• The semi-naïve min2max2 method achieved the lowest normalized Brier distance of 25.3%.
• The min2max2 method leverages cross-resistance and cross-susceptibility data from cross-antibiograms.

Conclusions:

• A practical method for predicting antimicrobial susceptibility can be developed using a semi-naïve Bayesian approach.
• The proposed min2max2 method demonstrates a significant advantage in reducing prediction error (Brier distance).
• This approach offers a valuable tool for enhancing antimicrobial stewardship and clinical decision-making.