Predictive Modeling of Emerging Antibiotic Resistance Trends

Antibiotic resistance is constantly evolving, requiring frequent reevaluation of resistance patterns to guide treatment of bacterial infections. Antibiograms, aggregate antimicrobial susceptibility reports, are critical for evaluating the likelihood of antibiotic effectiveness. However, these antibiograms provide outdated resistance knowledge. Thus, this research employs predictive modeling of historic antibiograms to forecast the current year’s resistance rates. Utilizing subsets of the expansive 15-year Massachusetts statewide antibiogram dataset, we demonstrate the effectiveness of using our model selector PYPER with regression-based models to forecast current antimicrobial susceptibility. A PYPER variant is effective since it leverages the fact that different antibiotic-bacteria-location combinations have different antimicrobial susceptibility trends over time. In addition, we discuss relative weighting of the regression-variant models, the impact of location granularity, and the ability to forecast multiple years into the future.