audgendb
play

AudGenDB Intuitive biomedical query and reporting tool using Django - PowerPoint PPT Presentation

AudGenDB Intuitive biomedical query and reporting tool using Django Byron Ruth July 13, 2011 SciPy 2011 Austin, Texas Center for Biomedical Informatics The Childrens Hospital of Philadelphia Scope National Institute of


  1. AudGenDB Intuitive biomedical query and reporting tool using Django Byron Ruth · July 13, 2011 · SciPy 2011 · Austin, Texas Center for Biomedical Informatics · The Children’s Hospital of Philadelphia

  2. Scope • National Institute of Health (NIH) funded project • Audiologic and Genetics database of unprecedented scope and size • Started with CHOP’s data, intending to integrate other pediatric institutions’ data

  3. Data Problems • Clinical data is complex and messy • Data is scattered throughout EHR • Most clinical data is not “research-friendly” • Clinical systems are built for clinical workflows • Given a drop-down of choices, it can most likely be overridden

  4. Example Concept ID Line # Concept Value CHOP#AUDIO#0001 1 Ear;Stim;125;250;500;750;1000;1500;2000;3000;4000;6000;8000 CHOP#AUDIO#0001 2 R;AIR;;;20M;;;;55;; 75;;; CHOP#AUDIO#0001 3 L;AIR;;;15;; 25;;55;;;;; CHOP#AUDIO#0001 4 L;BONE;;; 15;; 20;;;; 60N;;; WHAT WE HAVE WHAT WE WANT

  5. Solutions • Extract-Transform-Load (ETL) • Extract patients of interest from EHR • Stage the data in a more “queryable” data model • Data curation • Push data to production

  6. Great!

  7. We have data • 37,000+ Patients • 367,000+ Procedures (including 22,000+ surgical procedures) • 165,000+ Encounters • 56,000 Typanograms • 57,000+ Audiograms • 16,000+ Auditory Brainstem Responses (ABR) • 4000+ Radiology Impressions

  8. Proposed interface.. Done!

  9. Approach

  10. Constraints • Reach as many diverse groups of people as possible • We only have two software developers... • No resources to support distributed self-installed applications (e.g. Microsoft Office)

  11. The Web! • No installation required • Rapid iterative development cycles • Virtually every device has a Web browser these days...

  12. It should.. • Be Intuitive • Be “Discoverable” • Be in a familiar language • Not Be intimidating • Not Require knowledge of the data model • Not Require knowledge of the data

  13. Typical Interface...

  14. Our Interface...

  15. Demo

  16. Harvest

  17. The Stack • ModelTree - generates and executes Django QuerySets at runtime • Avocado - wraps ModelTree with additional metadata and abstractions • Serrano - REST API exposing core parts of Avocado • Cilantro - web app with Concept “query views”

  18. ModelTree • https://github.com/cbmi/modeltree

  19. Avocado • Definition - represents a single data field • Concept - one or more definitions represented together in a domain-specific way • Domain - collection of Concepts

Recommend


More recommend