Artificial Intelligence and Dashboard Tools to Collect, Anonymize, and Summarize Companion Animal Antimicrobial Use Data
openFDA - Food and Drug Administration
The objective of the proposed project is to develop a data extraction, analysis, and reporting pipeline for
sustainable companion animal antimicrobial use (AMU) surveillance. Previous surveillance efforts have been
restricted to snapshots of AMU in specific veterinary practices or at limited timepoints because AMU data is time-
consuming and difficult to collect. In addition, there have been substantial time lags between AMU data collection
and publication because each companion animal AMU surveillance study has conducted a bespoke analysis of
AMU data that requires a high-level of statistical, data visualization, and software skills. Recent advances in
natural language processing (NLP) and machine learning, particularly large-language models (LLMs), have
made efficient AMU data collection possible. However, even with new models, challenges remain to create a
scalable system for long-term data collection. The proposed project will overcome these challenges by creating
an LLM to extract AMU data that is agnostic to the veterinary electronic health record (EHR) system used and
create a dashboard to streamline AMU analysis. The first aim is to make an LLM that is scalable and
generalizable across EHRs by developing it on free text from multiple EHRs and validating it against existing
gold-standard AMU datasets collected from manual EHR review. The LLM will be available to participating
practices on a secure platform to ensure data privacy and security. Practices can process their own EHR data
and share only de-identified record-level AMU data, including antimicrobial drug(s) used, dosage, duration,
patient weight, and indication for AMU, plus the number of animal visits. The second aim creates an RShiny
dashboard, the AMU Data Visualizer, to streamline AMU data analysis and generate standardized AMU reports.
The dashboard will take the anonymous information generated by the LLM, or by other AMU data extraction
methods, and generate AMU metrics, such as prevalence and number of defined daily doses, with customizable
filters to understand AMU across all companion animal sectors (e.g., primary care vs. emergency care, young
animals vs. geriatric animals). The AMU Data Visualizer will allow anyone, regardless of their statistical software
skills, to create AMU reports, aggregated tables, and figures for their practice. The third aim is to collect
anonymous, aggregated AMU reports from participating veterinary practices to identify national trends in
companion animal AMU. Aggregated AMU tables from the AMU Data Visualizer will be combined into annual,
public AMU reports with benchmarks for different practice types and trend analyses. The proposed project is
conceptually innovative because it develops an EHR-agnostic AMU data collection method and a reproducible
analysis pipeline, and does not require practices to share EHR free text that could contain sensitive or identifiable
information. The project is technically innovative because it advances LLM data extraction by having the LLM
“show its work” for easy data verification. Overall, the LLM, AMU Data Visualizer, and aggregated reports will
reveal the full picture of companion animal AMU.
Up to $200K
health research