Unless you have a vested interest in health care, you can completely be forgiven for missing this one. No worries. It’s a touch obscure, but very relevant to me personally.
So what did you miss? Today HDS is announcing their new Hitachi Clinical Repository (HCR) product. At which point everyone reading goes “their who what now?” They’re too busy marveling at the amazing Watson, and that’s all good – Watson’s going to come into play here in a bit. First, let’s talk about what Hitachi Clinical Repository actually is.
HCR works with Hitachi’s other software products to create a massive metadata repository based on all sorts of sources within health care; prescription history, doctor’s notes, medical imaging, the works. According to Hitachi it will work with any software and source out there. As an example in the presentation folks at HIMSS11 will see, Hitachi shows HCR sourcing data from:
Lab IS as HTML, Radiology IS as DICOM and HL7, Pharmacy as XML, Nursing as PDF, ADT Registration as XML, Accounts Payable as XLS, Bloodbank as HTML, CPOE as HL7, Emergency as HL7, and Microbiology as CDA. (As a note; these are standards as part of Electronic Records.)
Klinikum Wels-Grieskirchen GmbH in Austria has already deployed HCR, and seems pleased as could be with it. They describe it as enabling the “ability to search for information across all deployed systems, irrespective of vendor or location”. (Elmar Flamme, CIO, Klinikum-Wels-Grieskirchen)
That’s a big deal in health care, where often multiple specialists will write conflicting prescriptions, and it doesn’t get caught by the pharmacy till you go to fill that prescription, because everyone’s on a different system. The argument for HDS’ HCR is that with it, you no longer have this problem. All of the information is accessible through a single system, independent of what’s behind it. Health Care IT can deploy this without the political wars involved in moving other department’s systems. They can continue to use their imaging solution of preference, while HCR provides physicians with all the information in one place.
But what about Watson? I said I was going to bring up the current wunderkind of IT, the “gentlemachine” who’s made IBM sexy to work for again. And so I shall.
Think of HDS’ HCR as the First Step. HCR consolidates all data interfaces into a single access point; a single pane of glass if you will. (Sorry; I hate that phrase as much as you do, believe me.) This is a key part, the single point of access. If we take this single point of access, thousands of data points, a library’s worth of medical textbooks and feed all this to Watson, what do we get? Chances are good, we get diagnosis. Will it be right? Maybe, maybe not. This isn’t a step in eliminating humans from health care, but rather, a step in eliminating human error from health care.
Systems like Watson need vast amounts of data points. HCR delivers it. But humans like you and I can’t necessarily manage so many data points effectively. Much less always remember that medication A and medication B can’t go together with ailment C but can with ailment D. Intelligent decision making engines like Watson can, though. Machines are beginning to understand the nuances of language, enabling them to interpret a physician’s notes provided sufficient supporting data.
If a physician tells Watson that the patient has condition A, Watson can go to HCR for data, then suggest medications with 100% certainty every time of avoiding drug interactions, known allergic reactions, and potentially harmful side effects.
But this brings us to the other problem which HCR seeks to solve. Medical data quadruples every two years according to the NIH. For a system like Watson to work with medical data, you need to store a lot of it. A single patient can easily represent gigabytes to terabytes of data when imaging is factored in – these aren’t snapshots with a digital camera, after all.
Not only that, but this vast sea of data is incredibly disparate. Radiology stores images, nurses store notes, doctors store notes and diagnoses, pharmacies store prescriptions, and labs store test data. HCR creates a metadata repository that normalizes the data as much as possible down to it’s most granular level, per patient.
So when the lab sends blood results, it will for example, store the metadata as Insulin Level, PSA Level, Potassium Level, High Level. Then under PSA Level, it will have Integer, Text, and Date. The data on the lab’s systems is still there, unchanged, untouched. Instead HCR compiles metadata from it and presents a single accessible interface.
Without this normalization of data, a system like Watson would have to speak to hundreds or thousands of completely different data sources, interpret vastly differing data, and try to assemble it’s own repository. It’s incredibly CPU expensive, time consuming, and more likely to result in errors. Yes, IBM did that for Jeopardy – the multiple disparate data sources – but Watson in that sense is very much a one-off. A commercialized Watson targeting health care would need to be efficient in it’s use and intake of data. HCR enables that; a Watson system could look at the patient’s known ailments, then look at the Insulin level Integer and Date. Based on changes of the Insulin level over time, the system can then accurately say whether or not there is clinical data pointing to diabetes, or recommend dosage adjustments, or even more impressively – accurately determine a specific patient’s metabolizing of insulin.
As Ken Jennings said it best: “I for one welcome our new computer overlords.”
And as usual, you can find out more about Hitachi Data Systems’ HCR at their website.