Hello Community! Please find the recording from our AI conversation at this link, or in the embedded video below: https://iu.mediaspace.kaltura.com/media/t/1_hoei5nqb
Our whiteboard with the key points shared is public on the OpenMRS Wiki here: https://openmrs.atlassian.net/wiki/spaces/projects/whiteboard/389939204?atlOrigin=eyJpIjoiMDkzNjY2Y2FjNTdhNDM1ODhhYjMwZDBjNGU1NDIwYzQiLCJwIjoiYyJ9
Next Meetings
- Terry Mochire will organize a follow-on general community session in 2-3 weeks.
- Detailed presentation and demo of Content & Mappings Automation project coming to community in a few weeks!
What was discussed
The key ideas shared were:
1. Automated Concept Mapping help
- @michaelbontyes from @Madiro and @paynejd from @OpenConceptLab shared their project to help implementers / form-builders to rapidly match form content with existing Concepts / Terminology codes. Saves days to weeks of refining content and terminology picking.
- Next steps:
- Demo coming in March/April 2025!
- Get involved and ask questions on Slack at: ocl channel
- Michael & Vero to work on ticketing a community Epic.
- Wiki for more information: https://openmrs.atlassian.net/wiki/x/JIB4Fg
- Next steps:
2. Generate SMS Discharge Instructions
- Dr. Terry Mochire from @Intellisoft shared how they are using GPT’s API to generate discharge messages to patients after their OPD visit.
- Next steps:
- Looking to pilot test soon.
- Talk post & docs coming.
3. Chart Search (LLM-supported)
- @burke from @Regenstrief shared a vision of embedding a Chart Search feature in the patient chart, and then having an LLM GenAI model summarize the findings for the clinician. Wiki page on this: https://openmrs.atlassian.net/wiki/x/D4BAFg?atlOrigin=eyJpIjoiOGI5YWFkMjU1ZTAwNGQxYWJkYzhiNTFlYjg0ZWJmNDYiLCJwIjoiYyJ9
- Jing Chang & @bashir from Google Health / Open Health Stack team shared they are interested in partnering on an open-weight FHIR-based Search; locally hosted. “Ask a FHIR-store questions” approach.
- Next steps: OHS and Regenstrief to discuss further; Intellisoft interested as well.
4. Query Support: For Clinic Managers, or M&E Officers, or Report Generation
- Clinic Query Support: A UI connected to an LLM, where a user (like a clinician or clinic manager) could ask questions of their EMR, and in the background the LLM would convert their plain english questions into OpenMRS-friendly SQL queries, then query the OpenMRS DB; e.g. “How many patients are sick with 1,2,3…”
- Or, could first focus on helping M&E Officers: an LLM helping data managers with crafting SQL. Why focus on M&E staff? Because: End-user-generated SQL via LLM (other than very simple examples) sounds like a bridge too far at the moment. Knowing the types of questions clinicians ask and what it takes to answer these properly usually requires a data manager & skills beyond today’s LLMs.
- Report Generation: Not waste a lot of time writing complicated report queries.
- @bennyange from @EMR4All shared helpful detail about the practical investigations EMR4All has been pursuing with using LLMs in this regard:
- Natural Language to SQL for Cohort Builder – AI to convert plain English queries into SQL to simplify data retrieval for Clinical Decision Support – AI to analyzes patient records for insights and recommendations.
- EMR4All tested Ollama (Smollama2) allowing offline AI execution without relying on OpenAI APIs. We also tested DeepSeek More accurate but slower due to hardware constraints.
- Next steps: EMR4All team will explore agent-based AI for OpenMRS automation for Medical Data Processing & Summarization to automate extraction of key insights from patient notes. They will keep us posted on how this goes.
5. Writing Clinical Visit Notes
- Not discussed in detail, but this was raised as an example use case worth considering.
6. Use as data source for Drug-Drug Interactions
@muppasanipraneeth19 proposed the following ideas shared publicly on GitHub that would involve using existing LLMs to inform the content needed to help catch drug-drug interactions in Prescription Workflows: GitHub - Muppasanipraneeth/aashayams