Convert your FHIR JSON -> XML and back here. The CDA Book is sometimes listed for Kindle here and it is also SHIPPING from Amazon! See here for Errata.

Monday, July 26, 2010

The Impertinence of Automation on CCR and CCD Generation

One of the key statements of the CCR and CCD specifications is found in the very first sentence of the CCR standard (emphasis is mine):
1. Scope
1.1 The Continuity of Care Record (CCR) is a core data set of the most relevant administrative, demographic, and clinical information facts about a patient’s healthcare, covering one or more healthcare encounters.  It provides a means for one healthcare practitioner, system, or setting to aggregate all of the pertinent data about a patient and forward it to another practitioner, system, or setting to support the continuity of car
This statement answers the oft-asked question:  In a CCD/CCR, which observations, medications, et cetera, should be sent?  The answer is "the most relevant" or "pertient" ones.  The notion of relevance, and the need for it in the standard is best shown in John Halamka's post on Standards for Personal Health Records.  What providers don't want is a pile of "electronic paper" to wade through.  The terms relevant and pertinent were terms that were chosen with great care and deliberation by the members of the ASTM Heatlhcare Informatics committee for that very reason.

I get asked the question about what to send quite a bit in the HL7 class on the Continuity of Care Document (which I'll be teaching again in Cambridge, MA in October at the HL7 Working Group meeting), and in other venues.  The answer I give is always the same.  Relevance is a clinical decision that right now needs to be made by a trained care provider, and is based on context.  What is relevant to one care provider is not necessarily relevant to another.  For example, while your dentist may want to know that you've experienced a heart attack, details such as which side of the heart, or the exact ejection fraction ratio, may not be as relevant to them as they would be to your cardiologist or GP.

The question of relevance depends not only on who you are communicating with, but also what is known at the time of communication.  An observation that may not be relevant today could be of importance later once new medical knowledge is discovered.  Obviously, you can only determine relevance based on what you know right now.  Trying to cover ones anterior anatomy by transmitting data that "might be relevant in the future, but isn't known to be relevant now" is just another way of trying to avoid making a clinical decision.

A lot of people are struggling with how to determine relevance without the immediate judgment of a care provider.  They want to know what the definition of relevance is.  There are no simple answers here.  This is a clinical decision support problem, and while there are many different guidelines, there are no hard and fast rules.  Even a rule such as "Send the first, send the worst, and send the last" migh err in sending either too much, or too little.  The worst blood pressure reading for a patient might be during surgery, when it expectedly dropped as a result of the procedure.  Is that the "right" blood pressure result to send? Probably not.  Which is worse?  A white blood cell count that is very high or very low?  It depends, and it may well be that what you want is the trend over the last week because the patient is recuperating from an infection after chemotherapy.  Again, context is important, and I suspect that any attempt to set a rule of thumb will fail.

What I suggest developers pay attention to is the user interface that providers need to select the most relevant clinical observations to send.  Once you have some experience with that, and with what providers actually do with it, you can start thinking about how to provide more clinical decision support for providers.

Another question I get asked quite a bit is when to send a CCD/CCR, and is usually in the context of long inpatient stays.  My answer for that is fairly simple, at transfers of care.  I've also been asked whether a transfer from emergency to inpatient counts as a transfer of care.  This is also pretty simple, in the US it is a policy question, and the relevant policy makers say yes.


P.S.  IANAL, but I strongly suggest that algorithms determining the most relevant data to show a healthcare provider are properly in the domain of FDA regulated medical devices.


  1. Posting this for Virginia -- KWB

    Hi Keith,

    Thanks for blogging on this. I would like to address the specific use case of an activity rich or lengthy inpatient stay - clearly there would be results that would not be considered relevant to the receiving provider in the transition of care and their presence would obstruct the clinician's view of the important information unless the receiver had some really "smart" EHR. In fact, the way CCR and CCD work as far as I understand the regs, the receiver just needs to be able to display it - render it and that would be a really noisy "summary". I promise you I was greatly troubled by this automation of CCRs and CCDs at first and have strongly argued the side that you are arguing for quite a while - it is the perfect - the correct - the pure - solution. However, as has been proven to me time and time again, there are many clever solutions in HIT which do things in impure ways that make me uncomfortable yet yield amazing outcomes of supporting caregivers and enabling better patient care - often safer than if we'd done it the "correct" way. Lets look at today's climate. We have many motivated providers interested in Meaningful incentive dollars looking to adopt and obtain meaningful use out of HIT with standards employed. To do this, their EHR technology must be able to support the CCR or CCD standard with associated vocabularies and implementation specifications. According to you, the EHR must also provide the ability for the clinician to accept what's relevant and reject whats not from each clinical summary. But in my opinion, EHRs are not going to change overnight - nor are the customers going to be able to implement the latest EHR version overnight so if this feature is not there, there will be an obstacle to meeting meaningful use in time for incentives, etc. There is a much better chance, that, for the time being, glue technology which can be certified EHR technology - its legal as far as I can tell - will be able to produce these CCDs or CCRs to the level of specification required in the standards and implementation specs and using the appropriate vocabularies. This glue technology will be backend - take what it gets out of the EHR or any other useful feed and create the CCR/CCD. Lots of guys are out there doing this already - you know that - which is why I expect this solution. Plus, even if the EHRs were equipped with this feature with the right standards and the ability to select relevance and were installed and implemented in time, there is the not so minor task of asking clinicians to mark what's relevant. I think thats a GREAT idea but changes like that are hard. For inpatient clinical summaries I see a serious flaw in that they doctors already are required to do discharge summaries so there is redundancy. Clinicians have real jobs to do and there are lots of HIT changes we are asking them to support to meet meaningful use. So then the biggest challenge I see is to get adoption of the select relevance feature allowing meaningful use to be realized.

    --- continued in next comment ---

  2. --- continued from previous comment ---

    For these reasons, I believe an auto-generated CCR or CCD from a separate EHR Component will be a practical 2011 solution for many. If I'm right, wouldn't it be wise to provide some guidance? Here is one possible idea based on the suggestions I have heard on the related thread on the HL7 list server: - send the first occurrence of any type of result - send the last 24-48 hours of results prior to discharge - include a comment/surgeon general type warning: Warning: the following is an automatic summarization of results from the inpatient stay. It contains the first occurrence of all results and the last 24 hours of results during the patient's stay - addition results may exist - contact the sender for more information.
    Face it, the use of automation as far as I can tell is legal and fits within the meaningful use and standards regs OK. Some would even say its innovative. Without some sort of rule of relevance you will get huge send it all summaries that are rendered on receiving EHRs as linear lists or different senders defining relevance algorithms leading to less predictable results. So I argue that guidance on this is the safer bet.

    Thanks for listening. Please note that the opinions here are my own and not those of my employer (NewYork-Presbyterian Hospital).

    Yours Truly,

    Virginia Lorenzi