Stream: implementers
Topic: Represent service running AI Model as a Device?
Brian Kaney (Jun 19 2021 at 16:57):
We have an AI model that runs a risk prediction. We are going to represent the result in FHIR as RiskAssessment. I was considering modeling the AI model as a Device (e.g. RiskAssessment.performer). Is this a terrible idea?
Josh Mandel (Jun 20 2021 at 19:45):
I think it's not a terrible idea! Would be nice to have some conventions about model versioning, named parameter sets, etc.
Brian Kaney (Jun 21 2021 at 19:46):
I was going to use Device.identifier to specify the API endpoint as the identifier.system and the identifier.value would be the version of the API. Other ideas?
Brian Kaney (Jun 21 2021 at 19:47):
I want to be able to know what version the API was when a RiskAssessment was created, this satisfies this minimum requirement.
Jose Costa Teixeira (Jun 22 2021 at 07:38):
I'd say one Device is the actual system/endpoint, another is the API software that is running on it.
Jose Costa Teixeira (Jun 22 2021 at 07:38):
So, identifier.system does not feel right, because that is the identification system, not the system where the thing is implemented
Josh Mandel (Jun 22 2021 at 21:57):
I would have looked at http://build.fhir.org/device-definitions.html#Device.version for versioning information
Brian Kaney (Jun 25 2021 at 21:53):
Okay, thanks! I am going to try Device.version and Device.url.
Last updated: Apr 12 2022 at 19:14 UTC
