IHE IT-Infrastructure fall 2021

IHE ITI Tech met last week and had a very productive meeting. It seems we are getting the hang of virtual meetings. I will say that they are still about 50% as productive as a face-to-face. There are times at which we are more productive, but there are times when the virtualness gets in the way. This can be due to the lack of non-verbal queues (oh, you really didn't understand what I said), or inability to stand up and draw a picture on a whiteboard, or because everyone is multi-tasking.

Finished this meeting: (will take a week or two to be actually published on https://profiles.ihe.net )

• mCSD Whitepaper 
• PIXm in IG format with Feed transaction - for Trial Implementation
• PDQm in IG format - for Trial Implementation

Continuing Work Items:

• Quality improvements to the HTML publication of the Technical Framework publications
• Maintenance (CP)
• MHD to a Federation task-142

New Work Items:

• Convert mCSD to IG Publisher task-151
• Create Implementation Guide Profile ATNA auditEvent patterns for generic FHIR RESTful transactions with pattern informing IHE profile further specializing task-144
• Convert Metadata Handbook to html publication without technical changes task-169
• Convert MHDS to html publication without technical changes (not an IG yet) task-166 

Ill cover these in other articles.

Security of #FHIR implementations concerns

Security Report: "The New Healthcare Ecosystem will depend on FHIR APis, but Are They Secure?"

Alissa Knight did some invited and funded cyberSecurity research and found some good and some bad. No-one should be surprised by that conclusion. The point we should take from this research is that 

1. EHRs are doing a good job of securing their FHIR implementations
2. FHIR is good and worthy
3. There is room for improvement in some implementations 
4. There are included recommended improvements. 

Some have take a very negative view of the research, most of this negativity is unfounded and driven by the fact that Alissa is excellent at marketing and drama. She has excellent presentation skills, excellent writing skills, excellent hacking skills, and excellent artwork. 

This research was funded by a vendor that provides security frontend for APIs, so certainly there should be some consideration here. I, however, have good reason to believe that this funding did not influence the research. Alissa has a couple of decades of proving she knows what she is doing. This is not just fake research complaining about products that didn't use the vendors solution. Yes there is one mention that a product that did use that vendor solution was not found vulnerable, yes the vendor got mention at the bottom of the report for funding it.

This was not an attempt to bring down FHIR, but rather to challenge us all to be better.  This report says nothing negative about the FHIR standard, it was focused on implementations. 

There is some bad news, some implementations were specifically disgustingly bad at securing the data, and they had lots of data (aggregators). These services had signed up to be part of this research. They were informed of the problems. The report was expressing the trend, not the specific details. Alissa addressed this during the YouTube below, and also on twitter later.

Alissa hosted a Ask Me Anything on live YouTube, the recording here

Deeper dive

The report stands alone, and I have tried to come up with better ways to say what it says, but I just can't. The things that Alissa found were very remedial cyber security issues. Editing URLs in the browser to try to get at obvious other information. Using legitimate security tokens for things it was not intended to do. ...  Alissa has so much deeper skills that she simply never needed to rely on. She goes into all of these details about as detailed as is reasonable without naming products. 

Page 16 was critical for me. I literally was stunned. Not an exaggeration. I knew that there would likely be mistakes, hence why above I said we all should not be surprised she found issues. It was clear there would be issues found. But the class of issues found are just unexplainable.  All of the issues failed to implement even the FIRST section on the FHIR Security page.

Data Aggregators were the worst offenders in my view. I know they provide a useful service, but they do it in ways that are not transparent, and they clearly don't take their responsibility seriously. They have been around for decades, so likely these same bad security implementations were just as bad on their previous proprietary "secret" APIs. The use of FHIR just makes them more obviously bad.

The last bullet on page 16 was the worse, quote:

With one patient engagement app, the API endpoint sent me all the patient and clinician records in its database, indicating it was using the mobile app to filter out just my record.

NO WAY!!??? yup, that happened

mic drop

Conclusion:

I need to bring this back to positives... FHIR is good, the EHRs are good, many apps and services are good... but some are really NOT.  

FHIR is a building block. It has so much to give to us humans (and animals). It will enable many things while it transitions from a "Standard for Trial Use" to something that is used to build CarePlans, Patient Engagement, Emergency Medicine, Disaster Management, Public Health, Artificial Intelligent clinical decision support aids, world wide COVID-19 tracking, and COVID-19 vaccine credentials...  helping patients stay healthy rather than get unhealthy in the first place ...

FHIR is built upon http RESTful concepts, Document concepts, and messaging. These platforms have security and privacy layers available to be used. Healthcare does not need to invent security, we just need to implement it properly. Mostly apply the security layer to the interactions following good policies. Have bad policy, is bad. Have bad implementations, is bad. Have no security, is bad.

All systems that have access to health data need to be designed from the beginning with Privacy Principles which include Security and Security which is about (protecting risks to Confidentiality, Integrity, and Availability).  Failing at "Access Control" makes me wonder about all the other opportunities to have forgotten Privacy and Security in the design.

Now is a good time to be reminded. We are not too far gone, we can do the right thing.

FHIR is the right standard to work with. 

Others comments

• Youtube post release discussion including Grahame and John
• SC Media article
• Grahame article
• Keith article
• Keith tweet storm
• Sean article
• Alissa twitter responses 

#FHIR Basic AuditEvent for generic RESTful actions

I have drafted a prototype Implementation Guide covering the AuditEvent profiling for generic FHIR RESTful actions.

For any FHIR REST operation there is a well-defined AuditEvent specified in this implementation guide. The appropriate AuditEvent shall be recorded by Client and Server applications that claim conformance to this implementation guide. The resulting set of AuditEvents are made available to a client authorized to retrieve them. The AuditEvent in this case is useful for the typical privacy office and security office use, but is also useful to enable a Patient facing app that can inform the patient when and how their data are used.

Basic Auditing where there is a known subject of the data involved. This profile is a formal specification of the guidance given in the FHIR Core AuditEvent under Common Scenarios

This guide does not cover all AuditEvents. It does not cover

• how accesses to data where their is no subject, such as a Provider Directory. Although this is likely similar, just without the mandated Patient entity.
• how failures are recorded. Failures are recorded with the .outcome that is not success, and is thus a very large body of possibilities. Failures are logged with best-effort and with verbose content. This makes the AuditEvent of a failure very hard to characterize, vary hard to automatically process, and possibly exposing of privacy or business secrets. These might be access control denials, which the patient would be interested in but for which it is not clear they would be due these kinds of notices. These might be infrastructural failures, which are too hard to characterize.

The AuditEvent profiles here could be used as a prototype for a more specific AuditEvent profile in a use-case specific Implementation Guide. Where a use-case specific Implementation Guide defines an AuditEvent profile, those profiles should be used rather than the Basic AuditEvent profiles found here. Both could be recorded without harm. 

Actors defined in the Implementation Guide

Data using Client

Requesting application in a REST relationship with the Server.

Note that the Client may also record the appropriate AuditEvent into the Audit-Repository. For security use-cases it is very helpful for the client to record the AuditEvent too, as this sets up a pattern of normal operation that can be watched for deviations. Deviations such as the client stopping audit logging should be investigated, a possible cause is that the client credentials have been stolen and are being used by another application than the one authorized.

Data Server

Responding server that holds the data the Client is requesting thru REST. Server records the appropriate AuditEvent into the Audit-Repository.

Audit Repository

FHIR repository holding the AuditEvents created, and provides access to the AuditEvents to Audit-Clients. The Audit-Repository would typically not allow Update or Delete of any AuditEvent previously recorded. Thus only allowing Create, and Read of AuditEvents.

Note that the Audit-Repository may be the same system as the Server.

Audit using Client

A Client that retrieves AuditEvents for some functionality. Where the functionality is not constrained or defined here. The Audit-Client queries AuditEvents for a given Patient.

Purpose

The reason for me to have written this Implementation Guide is for two specific reasons

1. To provide a structureDefinition Profile set of these basic audit even patterns. Which does test the FHIR core AuditEvent specification.
2. To provide a pattern for an Audit using Client that uses the AuditEvent(s) for various purposes. Including the purpose of providing a  Patient Engagement - Access Log

References

• Basic Audit Implementation Guide
• IHE- Mobile Access to Health Documents (MHD) - has AuditEvent profiles 
• IHE Patient Demographics Query for Mobile (PDQm) - has AuditEvent profiles
• IHE Patient Identifier Cross-reference for Mobile (PIXm) - has AuditEvent profiles
• IHE Internet User Authorization (IUA) - has profiled AuditEvent but does not have profiles

FHIR Document Digital Signatures

I was asked about Digital Signatures for FHIR documents:

I am working on _____  IG that is FHIR document based and we need a means to prove authenticity. The model is relatively simple in that a document and all of its parts represent a single thing that needs to be “signed”.

I have looked around for examples of this in IGs and in example documents and I have not found anything. I see a lot of references to CDA documents and signatures, but not much in the ay of FHIR documents. Can you point me in the right direction? Are there example FHIR IGs and documents out there. Where should I start?

Documents are good

The right way to do this is to have the signature cover the whole document, you have gotten to that point well. This is important as the signature covers all of the contents, including identity, date/time, context, etc; and also the meat of the content you are needing signed.  The point here is that a Document does not rely on references to outside data that might change, but rather a document should copy within itself everything that needs to be protected with the signature.

A FHIR-Document is not different than a CDA Document or any other kind of document. It is seen by the digital signature as simply a bucket of bits. Thus anything you see showing a digital signature on a CDA document is likely just as applicable to a FHIR-Document. 

The wrong way to do this is to believe that one can include a signature within the document (or within anything that is signed -- for example a FHIR Resource that contains a signature element). As soon as you need to exclude something in the bucket of bits, you open up the opportunity for other things to be excluded from the signature. So, always sign a whole bucket of bits, and a whole bucket of bits that is internally complete (doesn't rely on outside data).

The solution

A signature over a document is a document itself. It is a document of type XML-Signature.

There is already a specification for this from IHE – Document Digital Signature (DSG); and is what the FHIR core specification recommends. https://profiles.ihe.net/ITI/TF/Volume1/ch-37.html

Both documents would have DocumentReference that point at the bits (My preference is using a Binary, but the enclosed base64 data is an alternative).

The two documents would have a relationship. The digital signature (DocumentReference) would have a .relatesTo with the .relatesTo.target of the DocumentReference with the content; and the .relatesTo.code of ‘signs’.

Some more context on this https://healthcaresecprivacy.blogspot.com/2017/04/ihe-document-digital-signature-dsg.html

Note the concept of having everything needed (document) in one blob to be signed is very similar to what the COVID-19 credential does, but they strip things down to the bare minimum in order to fit in a reasonable QR code. They do use a JSON signature and encapsulate the content. So it is logically similar to the above, but practically it looks very different.  (Updated to be more correct)

My other articles on Digital Signatures

* Foundation of IHE - DSG profile

FHIR data in existing Nationwide Health Information Exchange

In the USA and elsewhere, there are Document Sharing based Health Information Exchanges. These exchanges have been built up over the past ten or so years. They have security models, privacy models, patient identification models, record location models, and data format models.  They also have mature testing tools, events, and have been specialized for many projects.

The solution is to leverage this existing solution, and just add FHIR. 

The exchanges in the USA include a PUSH model as well as a Query / Retrieve model. The PUSH model is used to convey information to a specified recipient. This can be done by way of the IHE-XDR protocol or the Direct Project (which uses IHE-XDM). The point I make below applies here, but I am not going to focus on PUSH models, simply for simplifying the message.

Nationwide Federated Exchanges

The Query / Retrieve models are based on the IHE-XCA, which is a federated query model that is designed to enable query of various types of patient data among a federation of partners. This model is used at the state level, and has three flavors at the national level, with interoperability between them... Thus a Federation of Federations.

Both the PUSH and the Query/Retrieve models today are dominated by the

C-CDA content specification. This is a good specification, and many organizations have gone to great lengths to build up their skills at creating the C-CDA, and also at consuming the C-CDA. Thus many want to continue to use CDA to preserve this investment. My point in this article is that this does not need to be a restriction on those that do want to move on to FHIR.

Ready for anything

The IHE-XCA implementation guide is designed to be content agnostic, focusing on a set of metadata that describes the document. These metadata are well talked about. They are designed to be minimal, yet powerful. Defining who the Patient is, Who the Author is, where was the content created, why was the content created, what are the privacy considerations, what are the relationships to other healthcare workflows and things. 

The metadata is not specific to CDA. Indeed the metadata schema was defined back when CDA was just emerging, when there were other content formats that might have become dominant. So the design of the metadata planned for a future when different formats would exist.

And, where the exact same content might be needed in two or more different formats. There is a specific type of relationship between metadata entries to indicate that the two content objects have the same context, with just a different technical encoding.

Just add FHIR

FHIR is a content format. There are Implementation Guides already published for a "Medical Summary" that is using the FHIR content format encoding. The International Patient Summary (IPS) is a FHIR Document that covers the same need as the C-CDA Medical Summary. Thus for those that want a FHIR content format, the choice would be the IPS rather than the C-CDA.  IHE has provided an Implementation Guide that takes the IPS content and explains how to communicate it within the Document Sharing infrastructures. So IPS over XCA.

Transition with no disruption

The best part is that the transition from everyone using C-CDA to everyone using IPS can be easy. There needs to be no change to Patient Identity, Security, Privacy, Records Discovery, Records Retrieval. By Just adding FHIR, the whole nationwide exchange that is functioning today just works.

Those organizations that can publish an IPS, just indicate in metadata entries that they can also provide an IPS. Those organizations that would prefer to consume an IPS, can detect those that are willing to provide an IPS. This addition of FHIR does not get in the way of those that are stuck in the C-CDA encoding. But it automatically enables those that want to speak FHIR.

The key is that there is an association between the C-CDA and the IPS content to indicate that they are about the same context, just different encodings. This association is shown above as a linkage between the FHIR content (IPS) and the CDA content (CCDA). In this diagram the C-CDA content is the master, but it does not need to be that way. An organization could focus on creating FHIR content and have a derived C-CDA for those partners that need C-CDA. This is all about making the transition smooth.

Conclusion

The existing nationwide networks provide many capabilities that have matured over time and are working, we should leverage this. The design of XCA is such that this transition is possible with no disruption, just add FHIR.

The use of FHIR in the existing networks does not mean that we should give up on designing solutions that use FHIR RESTful API. I encourage this too. This setting will need to solve for all of the capabilities that the existing nationwide networks have already solved for, so it will take some time, but it will eventually work.

My message is that the existing networks can be better, just add FHIR.

See the HIE-Whitepaper - section 2.7 Document Relationships where this is discussed.

Book: IHE Profiles for Health Information Exchange

 I am late to promoting this book, should have done it back in March when it was released. It is called a book, but is available for free download in multiple formats.

IHE Profiles for Health Information Exchange

By Keith Boone

He is the author, but he gives credit for pulling from many sources including my blog.

It goes a bit deeper than the IHE HIE Whitepaper. 

InScope podcast: #FHIR security

I was honored to be on the In Scope podcast, and excited to be paired up with Alissa Knight. We talked about FHIR security and such.  The host Mike Murry, who I worked with at GE for many years. He doesn't get to be in the pretty picture (Not my best picture, well I don't think I have better. Really sorry to Alissa for what I bring to the picture ). 

In the podcast, the combination of Alissa, Mike, and I is powerful as we all focus on different parts of security applied to FHIR: Design, Protection, Vulnerability.

The link to the episode on apple or general feed on inscope site, but you can find it on your favorite podcast app searching for "In Scope Security". They even have transcripts.

I hope this is just the first podcast on this topic. I would love to discuss more intersections between design, protection, and vulnerability. I would like to get into Mike and my passion for audit logging and analysis (hoping to catch Alissa).