**Views of Biomedical Data **
One way to think about “common data models” is in terms of focus, making the metaphor of focal planes an appropriate one:
aa1 a1
ab7 ab75 ab76 c9 c90
c93 c935 m77 m75 n25 n275
e432966
cbc- 8/44587920
Prot 2
Hb Plt
Alb
Pathology
In the diagram above, we move from “raw data” on the left to increasingly abstracted and curated data on the right. Where we focus depends on the business we are in. To deliver care, physicians need efficient access to records in any form. To do research, we need quality-checked, discrete data.
This note is an attempt to capture the direction of our discussion at the face-to-face meeting in Baltimore. There is a variety of “common data models” to choose from, and partisanship aside, each has its advantages and disadvantages, its ideal domain applications and its mismatches. The questions are: (a) Is this a problem? Why not “let a thousand flowers bloom”? (b) Do we need to agree on a preferred model for CTSA / IDTF / CD2H / “your favorite project” activity? (c) What to do about the areas in which the commonest CDMs do not agree on concepts, value sets, and the like?
It appeared at one point that FHIR might be championed as the common data supermodel to end all CDMs. What seemed to swing this view and shift the FHIR plane to the other end of the abstraction chain was the argument that most of our (research) data still has to come from practice, so that an approach that makes the process of extracting data for research from data from practice would be the most advantageous to the research community.
Where do we go from here?
Tony Solomonides 07-17-2019
transactional EHR data, e.g. in Chronicles
a coat of many colors made up of common data models
**close to the source to pick out what we need? **
**or is it really a common data supermodel? **
first level of abstraction, e.g. normalization, say Clarity
second level of abstraction, e.g. integration with other sources, e.g. data warehouse
de-identified research data warehouse