ontodesigns

These thoughts were originally posted as a comment (against this discussion) on www.thehealthcareblog.com. Unfortunately, the comment never appeared on the site. Not sure why. The discussion was a couple of months ago now but the thoughts are still relevant so seems like a good idea to post them here. Note: This is an edit of the original comment text.

——-

i’ve just spent 3 years working on the UK NHS national EHR. It has been hard and pretty difficult. Is it a success? At the moment, not really, but at least a basic level of functionality has been put in place and there is potential for it to grow. This assumes that the significant non-technical challenges can be overcome. I won’t repeat these because they chime with everything that has been said before. In terms of health standards, the UK NHS EHR builds on HL7 and SNOMED-CT.

I have some concerns related to the quality of HL7 and the scope of SNOMED.

HL7 V2 is relatively old now and is complex to use, and leaves a lot undefined which makes data exchange difficult. As far as I can tell, IHE is doing some good work to tighten up the constraints and get everyone on the same page (Btw, should we be talking more about the interesting work the IHE is doing, e.g. XDS, etc.) Still the HL7 V2 standard is pretty old and needs to be replaced. HL7 V3 (mostly used on the UK NHS project) is the new version, but is really a bit of a dog’s dinner (excuse the technical term - and for more info see below) and shouldn’t really go anywhere, but probably will because there isn’t anything else.

SNOMED appears to be a good piece of work, especially given its scope. My concern here is particularly that scope. SNOMED essentially tries to be the framework for pretty much the entire vocabulary of healthcare (correct in the comments, if you think I’m wrong). This is a massive scope. Does this worry anyone else? The IHTSDO seems like a pretty smart organisation, which is good, but given how much they are trying to control, I don’t feel comfortable.

So, my questions in this area are related to moving forward.

Ultimately, do we know if HL7 is fit for purpose? HL7 V2 seems to do the job, but given its age, should we really be building new systems with it? What are the consequences of continuing to use it? Certainly feels like it could be relatively expensive to implement and costly to maintain? HL7 V3 really wasn’t/isn’t great for the UK NHS. Just too complex given that the same could be achieved using new technologies with much less. It might get us through, but it does feel like a worrying standard to build new healthcare system on top of. If HL7 does need replacing, who is going to do it? What frameworks are future standards going to developed with? Should we continue with the current HL7 org? No answers to these questions, but surely they are valid questions to ask?

In terms of SNOMED and its very broad scope, is this really in our interest? It might be, and I’m open here, but I do worry about how the HIT industry ensures it isn’t overly coupled to IHTSDO and that the single approach doesn’t slow progress/adaption/evolution down in a way that isn’t helpful. I must admit that I think I’d like to see an approach which provides greater de-coupling. I’ve yet to be convinced that a single org needs to control it all. Maybe we should be considering separating some of the concerns so not a single org for everything? Btw, has anyone successfully implemented SNOMED post-coordination yet, so again, while parts of the standard may work well, there are significant issues with other parts. Will we have to wait for IHTSDO to fix these, or would completely new approaches be valuable? Not sure what the right approach should be here, but feels like a lot of eggs in one basket.

All of the above, make me wonder whether we need a new approach to healthcare standards in some areas. Should the new generation of Health 2.0 applications (for example) be getting together to agree interop approaches that build on the examples of Web 2.0, etc. I know there are issues here, for example, the WS-* standards are hardly a super success. Still, I think there is something in the modern open internet approach/standards/design paradigms that could really benefit HIT.

Btw, CCR is an interesting standard that I quite like. It limits its scope and doesn’t try to do too much. Should we be considering more “micro” standards like CCR and the narrower separation of concerns that such an approach implies? Might be nice not to go through ASTM though. If not, which standards orgs would make more sense and what “micro” standards would be really high priority?

This is another interesting EHR related presentation. This time from 2004 and Dr. Paul Tang who is Chief Medication Information officer for the Palo Alto Medical Foundation (PAMF). He covers the challenges of patient safety and how EHRs/EMRs can be shown to materially make a difference. He also discusses patient access to their records and the opportunities this creates.

Overall I really like the presentation and highly recommend that anyone interested in EMRs watches it even though it is long. The Q&A is pretty interesting at the end and does bring up some of the challenges of EMR implementations, including touching on ICD-9 coding (for more information, see e-Patient Dave’s recent issues). I will post more on coding shortly. I think there some real real challenges here at the moment.

Also, worth noting that the PAMF was working together with Epic on some of this work, although I couldn’t quite work out the nature of the relationship. This does make the presentation even more interesting given KP’s apparent success with My health manager.

In this presentation Jerry Lin (a product manager at Google covering Google Health) talks about some of the challenges that Google are facing with Google Health.

At the beginning he talks about placing the patient at the centre of the health information flow. This seems a reasonable position, particularly if you believe that patients need to be a the centre of their care (which I do).

He then proceeds to talk about two particular challenges (of so many!), user interface and data management.

The UI challenge seems to really mix up what should be shared (privacy) with the UI for sharing. Not sure that this is purely a UI challenge but I can understand why they have been combined. No easy answers here, I think everyone is learning at the moment. The NHS CFH org has recently updated the consent model to be used in the UK. I agree with the changes, and think they are aligned with to continuing to ensure the patient stays at the centre of managing their care.

The data challenge refers to coding, which is really still a super significant challenge. The video only spends a few minutes on a couple of issues here, but it does give a clue to how significant the challenges are. He raises, what should Google do if they see text and a code that they don’t believe match? This is a general issue with including presentation and coded details at the same time. Again, no easy answers, and risks need to be traded-off against benefits. I’m not a expert in coding, but it is an area that I’d like to explore further, so expect some more blog entries. Jerry discusses a couple of other data issues, but watch the video for those.

There is an interesting Q&A at the end of the presentation that I also recommend watching.

I’m posting this interview on InfoQ with Ian Robinson, as SOA was a discussion point in the future e-health architectures discussions mentioned here.

Ian is a ex-colleague of mine from ThoughtWorks. We worked together on some interesting ideas surrounding service evolution and versioning. See the end of this blog post.

I really enjoy listening to Ian and believe he is one of those people who will always bring interesting, rational and pragmatic points of view to any discussion. Not all of Ian’s answers are clear in this interview, but there is certainly enough straight talking to justify listening, especially for those considering SOA architectures.

I attended this (hopefully details updated soon!) one day event on the potential evolution of e-health systems architectures by 2020. The day was very well run by staff from Leeds University, and specifically the YCHI team. Very interesting it was too. I’ve been meaning to start blogging again and this will be a nice springboard.

There were a number of subjects covered ranging from the emerging business models for (e-)heathcare, to the NHS Spine, to specific health related ‘cloud’ technology projects. The attendees were from a broad range of the UK healthcare IT stakeholders.

Over the next weeks and months, I intend to explore some of the challenges, and maybe thoughts of solutions, in this blog. While my (few) previous posts have been generally about large systems design, the scope covered will open up to cover large e-health systems design.

Expect more frequent posts!

If you haven’t been following Steve Vinoski’s blog then you might want to check it out here. I’ve been listening to what Steve has been saying about Erlang for a while and it has caught my eye. I’ve since bought Joe Armstrong’s book on Erlang, and while only read the first couple of chapters, looks very interesting so far.

Anyway, back to the point of the blog entry … Firstly, you just have to agree with Joe’s discussion below.

Joe Armstrong’s Blog - The road we didn’t go down

Secondly, one of my pet dislikes in software development is technology that sells ‘illusion’. Essentially, something that appears to solve a problem but that actually doesn’t. It isn’t RPC that’s the problem, but the selling of RPC as ‘just like call procedure call but remote’. In my mind, this is just illusion. A local procedure call is nothing like a remote procedure call, because of the latency and failure patterns. I’m happy with abstraction, but not illusion.

Pat Helland is back at Microsoft and blogging again. Great news! I’m behind the times because he rejoined Microsoft a little while back and actually restarted his MSDN blog on the 14th May 2007. Just about the time I started this blog and in fact stated that he was at Amazon. Oh well.

He is in good form, challenging conventional wisdom (whatever that is!) and writing papers that outline alternative views that are truely useful. See the link below.

Pat Helland’s Blog: http://blogs.msdn.com/pathelland/

Another blog entry … Is it an age thing, but time flies by so quickly?

One of the things that I like about REST is the fact that it takes a different perspective on things. At an early age in our computing careers we are taught about functions / methods / procedures / operations and functional decomposition. This thinking then tends to dominate our approach to problem solving. REST suggests an alternative. Maybe we should be thinking about a simple consistent set of operations on resources (usually data, identified a URI). These simple operations may then trigger a more complex set of interactions, driven internally by events, which then result in a changed service state. What I like about this, is that the interaction model (protocol) for the service interface can be kept fairly simple and changes in the service’s state are usually easier to visualise. Service interfaces that are based around complex protocols of request / responses seem to lead to service behaviour with complex underlying state models that are difficult to use.

I’m not sure why this is but could it be cognitive? When I do something, it is the resulting state that I think about, rather than the activities involved in completing the action. In fact, the activities are pretty quickly forgotten and the resulting state really frames my mental model. Using an example, I don’t think about the activities involved in depositing money in a bank account. I think about the state after the transaction (e.g. money deposited). This state may then trigger internal events, leading to further externally visible state changes (e.g. increased balance). I do wonder whether our ‘procedural’ outlook masks what we really care about, which is ‘in what state did we leave the system’.

Does this make sense? Anyone agree? Disagree?

Anyway, for an in-depth view on REST, I’ve always like Duncan Cragg’s blog. Duncan explains REST with a passion, see his blog here.

Large SOA designs are difficult and there doesn’t seem to be much really good advice on how to build them. To get this blog started, I thought it might be useful to provide links to three papers that are influencing my thinking around service architectures. Two are by Pat Helland (who is now at Amazon), and one is by an ex-colleague at ThoughtWorks.

Pat Hellands first paper is called ‘Data on the Outside, Data on the Inside. It covers a number of key service data concepts such as immutability, data versioning and reference data.

http://msdn2.microsoft.com/en-us/library/ms954587.aspx

The second Pat Helland paper is about distributed transactions and how these aren’t appropriate in near-infinitely scalable applications. He then continues to outline some approaches to building such applications without using distributed transactions.

http://www-db.cs.wisc.edu/cidr/cidr2007/papers/cidr07p15.pdf

The final paper was written by Ian Robinson at ThoughtWorks. Ian and myself (and Ian Cartwright) were working together on a ThoughtWork’s project and the paper summarises some of the things that we were thinking about. Understanding the versioning of services and backward compatibility are critical to any Service architecture, especially in enterprises where the cost of changing service consuming applications is a real problem.

http://www.martinfowler.com/articles/consumerDrivenContracts.html

If you know of any similarly good papers on SOA design, then it would be great hear about them. Please feel free to email me at robins at ontodesigns.com and I’ll put together a follow up blog entry.

I’ve been mean to do this for such a long time … it will take a little while to get the blog looking as I would like … but hopefully more time will be spent on content than style …