Analytics and AI on FHIR
summary
Claims that combining SQL on FHIR with targeted flattening lets analytics and ML read nested FHIR directly without parallel ETL, demonstrated by projecting FHIR into BigQuery and querying with Looker and CQL. A live Patient Flow Explorer dashboard is built on semi-flattened tables, deployed in a day or two against synthetic data representing trillions of records and remains performant. The team shows handling customers with dozens of EMRs, de-identification steps, and tradeoffs between full flattening and use-case-driven projections.
AI-written summary · about
chapters (11)
- 0:00Introduction and speaker setup
- 1:43Talk overview and goals
- 4:53Healthcare data challenges with FHIR
- 7:23Hierarchy of ML needs for FHIR data
- 9:50Common approaches to analytics on FHIR
- 10:52Flattening nested FHIR structures
- 18:37CQL for cohort and measure definitions
- 24:56Lessons learned from deployments
- 29:51Demo: Patient flow analytics dashboard
- 32:38Q&A session part 1
- 41:49Q&A session part 2
related talks
19mAnalyzing FHIR Data Using ML and Analytical Tools
35mScalable Data Science with FHIR
33mBridging large-scale FHIR to AI in production
21mClinical and Claims Data Transformation utilizing FHIR Bulk Data
20mExchange of analytics data using FHIR resources
30mFHIR in the W.H.O.L.E.: Graphing Clinical Data for Smarter AI Systems
22mAI and ML for Centralized Patient Data Repositories
37mA novel way to manage and represent unstructured clinical data