SIIM Hackathon Project
We will be working on the tooling around improving the utility of CDE Sets at the SIIM Hackathon this year! We already have a toolchain that lets us nearly automate the process of generating CDE Set definitions, including some techniques using LLMs for assistance (post on this soon!)
At the Hackathon, we will be extending this work to include the ability to automatically (or at least semi-automatically) add connections between CDE Sets (and Elements and the individual values of Elements) and some important ontologies:
This process will mean that we'll have a repository of CDE definitions richly interconnected with other ontologies, both for outward and inward connections. This will help lower one of the major barriers to CDE adoption and make our use of CDEs much more powerful. For example, this can mean that when we create FHIR Observations using CDE codes, we can also include the relevant SNOMED CT codes (and RadLex, and Anatomic Locations, and... 😄), which are more readily understood/accepted by downstream systems such as EMRs.
Approach
In essence, the goal is to automate semantic search against the target ontologies for all levels of a CDE definition—Set, Element, and element value.
- We've already begun gathering content and have processed into a NoSQL database with collections for each ontologies.
- We're enabling a full-text search against the ontologies.
- We're also using foundation models to generate vector representations of each node in the target ontologies, which we will also index.
- We will then create tools (Python and TypeScript) that can take a CDE Set definition and automate the process of performing a hybrid semantic search against the ontologies, bringing back a set of candidate linkages.
- We can then ask an LLM to assess the candidate linkages and format the results for expert review.
We've already begun work on the toolset and hope to push it forward more quickly by the SIIM conference. We hope to see you all there!