Stream: analytics on FHIR
Topic: DevDays
Brian Postlethwaite (Jun 11 2019 at 16:18):
Session now on at DevDays in Baker!
Dan Connolly (Jun 11 2019 at 16:19):
/me waves
Dan Connolly (Jun 11 2019 at 16:20):
"Give us the normalized schema". "I disappoint you; this schema does not exist." because reasons...
Dan Connolly (Jun 11 2019 at 16:21):
the SQL on FHIR talk that compiled the FHIR structures to .proto was pretty cool. And the cerner Bunsen thing that exposes Spark SQL is pretty cool too.
Dan Connolly (Jun 11 2019 at 16:21):
ah... HAPI JPA is key-value id -> blob
Grahame Grieve (Jun 11 2019 at 16:24):
what doesn't exist is a single standard SQL schema - because there are so many ways people want to do that, for different performance and architecture reasons
Dan Connolly (Jun 11 2019 at 16:25):
the fact that google and cerner are able to consume the FHIR spec details and generate .proto or spark schemas tells me there are nice (if latent) types in the FHIR spec details.
Grahame Grieve (Jun 11 2019 at 16:27):
yes indeed. we do schema, and you can. But we don't do a single standard SQL schema
Dan Connolly (Jun 11 2019 at 16:27):
where are these FHIR spec details that people consume? when I search for FHIR structure definition, I get http://www.hl7.org/fhir/structuredefinition.html , i.e. metadata, not data.
Jason Walonoski (Jun 11 2019 at 16:30):
http://hl7.org/fhir/downloads.html
Jason Walonoski (Jun 11 2019 at 16:30):
http://hl7.org/fhir/definitions.xml.zip
Jason Walonoski (Jun 11 2019 at 16:30):
or http://hl7.org/fhir/definitions.json.zip
Jason Walonoski (Jun 11 2019 at 16:31):
There are also npm packages you can use
Dan Connolly (Jun 11 2019 at 16:31):
thanks! (sorry for the "please do my homework for me" question)
Dan Connolly (Jun 11 2019 at 16:37):
JSONpath in Oracle?! Anybody used it?
Dan Connolly (Jun 11 2019 at 16:38):
/me dreams of migrating our shop away from Oracle to postgres
Dan Connolly (Jun 11 2019 at 16:39):
ah... the Bunsen thing does a standard-ish SQL on FHIR. Did he just say the dev went from cerner to google? So is it no longer maintained?
Grahame Grieve (Jun 11 2019 at 16:40):
@Ryan Brush might want to comment directly
Dan Connolly (Jun 11 2019 at 16:41):
"choice is done wrong in FHIR" ;)
Grahame Grieve (Jun 11 2019 at 16:43):
...some particular person thinks so, but the community as a whole did not agree
Dan Connolly (Jun 11 2019 at 16:43):
fhir-fuel / fhir-storand-and-analysis-track
Dan Connolly (Jun 11 2019 at 16:44):
"CMS example" <- where does that link go?!!?
Dan Connolly (Jun 11 2019 at 16:45):
every time he says FHIRBASE I hear firebase
Dan Connolly (Jun 11 2019 at 16:46):
Q: FHIRBASE with HAPI or [missed]
A: I saw a fork of HAPI with FHIRBASE (details, anyone?)
Dan Connolly (Jun 11 2019 at 16:51):
oh... the CMS example doesn't seem to be related to the sort of CMS data I'm used to: RIF files.
Dan Connolly (Jun 11 2019 at 16:51):
Dan Connolly (Jun 11 2019 at 16:56):
what were the 3 transforms they do? choice, first class extensions, and ... ???
Ryan Brush (Jun 11 2019 at 17:43):
Hey everyone! I'll try to answer some of the questions that popped up here and am happy to chat further at the conference as well. (I was in another session this morning so missed some context so hopefully the following will make sense.)
As mentioned above, I won't be highly involved in the Bunsen project any longer, but it is being maintained by folks at Cerner. Posting bugs to http://github.com/cerner/bunsen is probably the best place for discussions there.
As for SQL schemas based on FHIR, I think our best approach is to think of that as views over FHIR data that is easily used by SQL-based analytic tools. Trying to represent all of FHIR as a lossless SQL schema creates a lot of complexity that makes queries pretty difficult. That said, I think it is possible to have well-defined but lossy views of FHIR data expressed in SQL, as described at https://github.com/FHIR/sql-on-fhir/blob/master/sql-on-fhir.md. This topic didn't get as much attention in the last few months for unrelated reasons but I think can pick up steam again if there is general interest.
nicola (RIO/SS) (Jun 11 2019 at 17:45):
Here is a link to my presentation - https://niquola.github.io/fhir-devdays-2019-slides/#/, and link to SQL on FHIR draft by @Ryan Brush - https://github.com/FHIR/sql-on-fhir/blob/master/sql-on-fhir.md
Paul Church (Jun 11 2019 at 18:46):
There is a lossless mode on GCP export to BigQuery - the schema is derived from the data since contained resources make the true schema infinite. It's as difficult to write queries against as everyone expected. Its only selling point is that it is lossless - I have yet to hear of anyone finding it more useful than the SQL-on-FHIR projection.
Dan Connolly (Jun 12 2019 at 18:12):
codegen talk continues this theme of structured defintion to, for example, graphql
Dan Connolly (Jun 12 2019 at 18:13):
TIL: https://www.aegis.net/touchstone.html Touchstone
Build. Validate. Share.
Dan Connolly (Jun 12 2019 at 18:14):
https://bit.ly/2KLohJc https://asymmetrik.com/solutions/healthcare/code-generation/?referrer=healthcare
Last updated: Apr 12 2022 at 19:14 UTC
