CDS Hooks vs HL7 v2 triggered interfaces

[Boris Key]

Hi Kevin and everyone else!

After spending some hours reading about all neatness of FHIR and CDS hooks, I still have a question that I am not clear on conceptually.

Based on my research, most of the interop these days is done over HL7 v2 and heavily relying on a push model - source system will push HL7 v2 message to other systems via a middleman (integration engine like Cloverleaf) to other systems. Other systems can then react to these events. 

I think most FHIR implementations will follow a push model instead (Cerner, for example) but FHIR spec itself can support both

https://www.hl7.org/fhir/pushpull.html

Now my question. Are CDS Hooks intended ONLY for CDS systems and a target CDS system must render a card back and displayed in EHR or CDS hooks are also intended to be used much like old HL7 push model, so a workflow event like patient admission, will trigger a FHIR message, sent to all the subscribers, and will be similar to ADT message?

I was in a few meeting with architects and it seemed to me they were thinking to use CDS hooks for that purpose.

Thanks,

Boris

[Kevin Shekleton]

Hi Boris,

Today, CDS Hooks is targeted towards a model where an application (eg, the EHR) invokes a remote CDS Service for a particular hook. The CDS Service performs its decision support and returns back any CDS in the form of "cards". All of this communication is done via HTTPS and via a JSON API using FHIR data.

I know of use cases where CDS Hooks is being used so that the EHR informs the CDS Service of the hook but no cards are returned to the user. In this case, the CDS Service is simply using the EHR to be notified of hooks/events in the workflow. Our design today around CDS Hooks, while allowing for this use case, is really centered on user-driven workflows.

I speak about CDS Hooks often and if you haven't seen this slide deck yet, take a look at it as it may help fill in some gaps or answer some of your questions. The latest deck that I've presented can be found here:

https://speakerdeck.com/kpshek/upmc-remote-decision-support-with-cds-hooks

As an aside, the Cerner FHIR implementation is not a "push" model. Instead, SMART applications or backend systems "pull" data by invoking HTTPS requests against the Cerner FHIR server. Of the EHR FHIR implementations I'm aware of, this is the model in use today.

Best,

Kevin

[Boris Key]

thanks for a quick response, Kevin! Your slide desk is very helpful, thanks for sharing it.

sorry, I meant to say that the Cerner FHIR implementation was a pull model :) I've been playing with the sandbox api, really cool stuff!