FHIR Chat · San Antonio Connectathon Track

Stream: analytics on FHIR

Topic: San Antonio Connectathon Track

Brian Wright (Dec 16 2018 at 19:38):

I don't see a "Storage and Analytics" (or similar) track currently on San Antonio connectathon Wiki. I was wondering if there are plans currently underway for a track? Thank you.

Ryan Brush (Dec 16 2018 at 19:51):

I was pretty involved with this in Baltimore and would love to see it continue, but unfortunately am not able to make it to San Antonio. If someone is interested in leading the track in San Antonio I'd support it (remotely) any way I can. I do hope to see this continue at future connectathons...just can't make this one.

nicola (RIO/SS) (Dec 17 2018 at 08:41):

Probably me :)

Brian Wright (Dec 17 2018 at 20:05):

I would be very interested in attending. Is there a set of potential topics/scenarios?

nicola (RIO/SS) (Dec 19 2018 at 05:46):

I think, same as last connectathon - http://wiki.hl7.org/index.php?title=201809_FHIR_Storage_and_Analytics

Craig McClendon (Dec 20 2018 at 15:32):

@nicola (RIO/SS) - Does this look like it's going to happen? Myself or a colleague would likely participate here.

nicola (RIO/SS) (Dec 20 2018 at 15:37):

yes, it is going to happen

David Hay (Dec 20 2018 at 18:51):

let me know when the track description is up - I'll add to the wiki - and to conman...

David Hay (Dec 20 2018 at 18:52):

btw - you might want to move discussion to the connectathon track so people know what you're up to...

nicola (RIO/SS) (Dec 21 2018 at 16:21):

@David Hay here is initial description - http://wiki.hl7.org/index.php?title=201901_FHIR_Storage_and_Analytics.

@craig mcclendon and @Brian Wright if you have specific topics for this track - please contribute!

Brian Wright (Dec 23 2018 at 17:53):

I am flexible, but here are some topics based on my own specific interests.

#1 Develop analytic model in R using synthetic data, Bunsen, Spark.

#2 Evaluate options for executing model from #1 in an environment where your model development data may not be complete for model execution (for instance: model is developed on bulk data extract, but model execution is in near real-time streaming environment).

#3 Discuss possibilities/needs for standards based cross vendor capability to execute arbitrary analytics workloads without requiring transfer of data. For example, if organization A contracts organization B for analytics work, can organization A offer a standards based analytics runtime environment for organization B to run model training and testing jobs that would not require transfer of data to organization B? Could the same capabilities be utilized at scale for real-time model execution?

#4 Investigate use of techniques described in this article:
https://www.nature.com/articles/s41746-018-0029-1

Ryan Brush (Dec 24 2018 at 17:39):

Great topics! Now I'm even more disappointed that I can't make it to San Antonio, but I'll make a few comments on these here:

1. We haven't yet written R bindings for Bunsen -- being focused on the Python/PySpark stack -- but this is definitely something we would like to do.

2. It seems like the standards-based options are either invoking the FHIR resources needed by the model (possibly using CDS Hooks as an injection point), or something like a FHIR subscription to aggregate the changes. This leaves a lot of work on the model deployment to connect things together. How much of this pattern (if any) should be defined and directly supported by FHIR standards isn't obvious, but would be a great topic to explore.

3. I think the "SQL on FHIR" discussion earlier on this thread could support that...by mapping all FHIR resources to analytic SQL operations, consumers could submit their analytic queries and receive the corresponding responses. (Plugging that into real-time model execution will be more involved depending on the latency requirements.)

4. I know some of the authors of the original paper have been involved in Connectathons, but I'm not sure if they'll make it to this one, so I'll just defer to that group. :)

nicola (RIO/SS) (Dec 27 2018 at 20:51):

Shell we book room for break out session?

Brian Wright (Jan 01 2019 at 15:06):

Sorry about the late reply, I have been on vacation and not thinking too much about FHIR. :)

For the breakout, if it would be helpful for context, I can give an overview of a project that what we are doing with FHIR subscriptions, streaming analytics, business rule and analytic model execution, persisting results to FHIR, and exposing in a SMART on FHIR application. I can cover some of the pain points and questions we have about fitting into existing FHIR standards. The potential for use of CDS Hooks for arbitrary (non EMR workflow initiated) execution is a particular area of interest.