Increasing antimicrobial resistance (AMR) poses a substantial threat to global health and economies, which has led many countries and regions to develop AMR National Action Plans (NAPs). However, inadequate logistical capacity, funding, and essential information can hinder NAP policymaking, especially in low-to-middle-income countries (LMICs). Therefore, major gaps exist between aspirations and actions, such as fully operationalized environmental AMR surveillance programs in NAPs. To help bridge knowledge gaps, we compiled a multilingual database that contains policy guidance from 146 countries composed of NAPs, internal reports, and other guidance documents on AMR mitigations, including environmental considerations. Leveraging this database, we developed an AMR-Policy GPT, a large language model with advanced retrieval-augmented generation capabilities. This prototype model can search and summarize evidence from plans, metadata, and technical knowledge and provide traceable references from global document databases. It was then manually validated to show its proficiency in accurately managing diverse inquiries while minimizing misinformation. Overall, the AMR-Policy GPT offers a prototype that, with the deepening of its database and further road testing, has the potential to support inclusive, evidence-informed AMR policy guidance to support governments, research, and public agencies. A conversational version of our prototype is available at www.liuhuibot.com/amrpolicy.
Keywords: LMICs; antimicrobial resistance; artificial intelligence; large language model; one health; policymaking; retrieval-augmented generation.