The Minnesota Department of Health (MDH) has been recognized as a 2025 AWS Champions Award winner for pioneering a cutting-edge application of generative AI to streamline medical record abstraction for COVID-19 surveillance. By using large language models (LLMs) in a powerful data system, MDH turned a time-consuming, manual public health task into a fast, automated process. This has sped up research and provided deeper insights into public health trends

Project Overview
To improve COVID-19 monitoring efforts, MDH developed a retrieval-augmented generation (RAG) pipeline that ingests unstructured PDF medical records, converts them to text, and indexes the data to enable intelligent search and structured data extraction. This system combines the power of generative AI with rule-based variables to automatically generate clinical data abstracts for use by MDH’s Epidemiology staff.

The solution was initially piloted for COVID-19 case surveillance, a critical need during the pandemic when traditional interview-based data collection became inconsistent due to overwhelming case volumes. The new pipeline supports rapid, consistent, and scalable abstraction of medical records.

Challenges Addressed: Data Bottlenecks in Pandemic Response
Prior to implementing this solution, MDH, like many public health agencies, faced major limitations in managing and interpreting COVID-19 patient data:

• Interview-based data collection: Initially relied on interviews to characterize infections — an approach that quickly became unsustainable as case counts surged.
• Manual medical record abstraction: Required staff to individually review and summarize patient records, consuming hundreds of hours and prone to inconsistencies.
• Data access and timeliness: Delays in converting unstructured documents into usable datasets limited the department’s ability to make real-time public health decisions.

These challenges constrained MDH’s ability to produce standardized, high-quality epidemiological data at the speed necessary during a public health emergency.

Results: Faster Abstraction, Greater Insight, and a Scalable Model
MDH has realized multiple transformational benefits:

• Significant time savings: Automation drastically reduces the number of labor hours required for medical record review, allowing epidemiologists to focus on data analysis instead of data entry.
• Improved accuracy and consistency: Using generative AI in conjunction with deterministic checks yields more standardized and reliable data than manual methods alone.
• Actionable epidemiological insights: Enables real-time aggregation of structured data to track the clinical course, outcomes, and risk factors associated with COVID-19 hospitalizations.
• A scalable and replicable framework: The project laid the foundation for broader applications in public health beyond COVID-19, supporting surveillance for other infectious diseases and chronic conditions.

MDH’s innovative approach turned a traditionally retrospective, manual task into a near real-time analytical capability. This not only advances state-level response, but also national surveillance through programs like COVID-NET, a CDC initiative involving 14 sites nationwide.

A New Era of Public Health Data Intelligence
The Minnesota Department of Health has demonstrated how cloud technologies can reshape public health practice, turning previously unmanageable data volumes into timely, actionable insights.