Combating antimicrobial resistance (AMR) has prompted the implementation of numerous policy and antibiotic stewardship efforts. However, evaluating their impact remains challenging, limiting our ability to understand how these interventions may have delayed or prevented transmission—and, in turn, hindering their scalability and adaptation.
To address this gap, we propose to develop a counterfactual machine learning framework to simulate how resistance might have evolved under different policy scenarios, providing evidence-based guidance for future AMR interventions. By analyzing how past policy changes—or the absence of change—have contributed to current resistance patterns, we can generate “what-if” scenarios across regions and pathogens. This approach will help estimate the causal impact of antimicrobial and drug-use policies on resistance trends using historical AMR data.
Our primary objective is to predict counterfactual resistance trends—scenarios that did not occur but could have—based on patterns of drug usage across different regions. To achieve this, we will focus on Gram negative infections and leverage pharmaceutical datasets, such as SMART Surveillance Heatmaps, Pfizer-ATLAS, Venus Remedies, Shionogi, and Venatorx-Gears, using demographic information, isolate data, and resistance profiles. Additional external datasets may also be incorporated as needed.
The primary deliverables will include side-by-side visualizations of observed versus counterfactual resistance trends under various hypothetical policy scenarios, effect-size estimates quantifying the projected impact of specific interventions on antimicrobial resistance over time, such as the percentage reduction in resistance expected over a given period, and an interactive dashboard. This dashboard will enable policymakers, hospital administrators, and global health agencies to explore multiple intervention scenarios by adjusting key levers such as timing, magnitude, and scope of policy changes and immediately observe the projected impact on AMR.
By quantifying how past stewardship shaped resistance patterns and simulating alternate histories, this project offers insights for future AMR interventions. Our counterfactual approach advances methodological innovation in AMR surveillance and informs global policy.