Stone age EHR

Today was my first #bigdatamed conference data at Stanford Medical School, which is hosting a three day conference on Big Data in Biomedicine. (I wasn't able to come Wednesday but watched two sessions on the live webcast).

I first learned of the conference last year from Atul Butte (@ajbutte), who I met when he presented at the 2012 Open Science Summit. My impression from Atul (and watching the webcast last year) is this is a bunch of computational biologists who’ve replaced their wet labs with databases (or nowadays, cloud computing accounts), in search of the next great lead to be found on their computer screen.

Certainly the first two panels yesterday fit that pattern (the first moderated by Butte). So did the after-lunch keynote by former UCSD professor Phil Bourne, creator of the PDB (protein database): a few months ago, Bourne joined NIH as its first-ever associate director for data science, reporting directly to NIH Director Francis Collins.

Translating from Science to Practice

But today the conversation broadened (as one slide put it) from the "science of medicine (biomedical research)" to the "practice of medicine (healthcare)". In other words, from faculty to the clinicians, and from universities (few industry scientists were present) to hospitals and clinics.

Some of the differences were as expected. Drug discovery researchers are at the bleeding edge of the science, and then after 5 or 10 or 15 years of drug development (animal models, clinical trials, regulatory filings, manufacturing, marketing etc.) the product finally shows up in the hands of doctors. Similarly, researchers are hoping to add to their journal publications while providers are trying to improve clinical outcomes — and increasingly under pressure to do so at higher efficiency (of both time their time and the amount spent on tests and treatments).

For clinicians, HIPAA privacy rules limit dramatically what and how data can be used and shared. Researchers have institutional review boards, but also face HIPAA restrictions. The NIH helpfully makes available a brief (16-page) note on researchers should interpret the interaction of IRB and HIPAA privacy constraints. (As it turns out, both clinicians and non-clinical researchers at the conference complained that HIPAA places unrealistic limits on combining data from differing sources to render an assessment of a given patient's health).

Proprietary vs. Open Platforms
At some point, it was inevitable that participants would discuss where the patient’s clinical data resides. Ten years ago, it was in paper charts, but now the ACA has strong incentives and penalties to store it in an electronic health record or EHR. (The administration’s healthcare IT czar says don’t call it an “electronic medical record”).

It was also inevitable that someone would ask: if we are compiling personal genomic data for patients, how will that data be made available for the clinical benefit of that patient? By one estimate, a patient’s EHR runs less than 100 megabytes while whole genome data (I’m told) runs into the gigabytes. As David Watson (ex Kaiser CTO, now at Oracle) said on today’s opening panel, medical images (such as MRI scans) are stored external to the EHR; will that happen with genomic data?

More seriously, how will such data be phased into operational systems? On the same panel, Jim Davies (CTO for England’s 100K genome project) suggested that existing EHRs would need an abstraction layer that would allow new data types to be added on, i.e. the way that apps, plug-in modules and extensions are added to other modern software systems.

However, today the EHR vendors (except for VistA) as proprietary as mainframe platform companies of the 1960s. Even Kaiser — which in 2010 had the largest private EHR implementation to date — is highly dependent on a proprietary vendor (Epic).

Proprietary control of the platform means high switching costs and other proprietary control of the customer, and so (I predict) this is something that none will relinquish unless forced to. We have a technical solution, but not a market solution. And the ACA penalties for EHR non-compliance mean that no provider can credibly defer or set aside EHR adoption until one provides the necessary openness.

So we know where we need to go, but it’s not clear how we get there. Two Harvard researchers — Zak Kohane and Ken Mandl — have proposed a way forward, and the following year won $15 million from HHS to implement their Smart Platforms project.

However, the plan seems to think that either vendors will see openness as being in their own interests or that customers will organize to demand openness. As someone who’s studied IT openness for 15 years, I can say that openness is almost always instituted by the weakest player (e.g. a late entrant), and right now I don’t see an obvious candidate in the EHR market.

WIthout such openness, health care providers are stuck with healthcare IT systems without third party add-ons. This is not just pre-app store, but pre-IBM PC, pre-Apple II, vertically integrated platforms with little if any choice to extend or change their systems. In other words, EHR systems are stuck in the stone age (1960s) of the digital computer era, with little prospect for improvement.