Patients matter most, but technology matters a lot

This was originally published on O’Reilly Media’s Strata blog, June 4, 2013.

Computing practices that used to be religated to experimental outposts are now taking up residence at the center of the health care field. From natural language processing to machine learning to predictive modeling, you see people promising at the health data forum (Health Datapalooza IV) to do it in production environments.

MedCPU checks deviations from recommended practices in real time. If a doctor orders a dose that seems unreasonably high for the particular patient, MedCPU may notice it and generate an immediate alert. The system can combine structured data and unstructured data (that is, the free text entered by the provider into the record) to come up with its judgments. Sonia Ben-Yehuda, the founder of MedCPU, says they have bypassed the commmon problem of trying to deal with multiple, incompatible EHR systems by reading what goes on directly from the EHR screen. No need for time-consuming integration!
Silicon Valley Biosystems (SVBio) collects and collates genetic data from various public sites, dealing with sets that are inconsistent in many ways: they comes from a variety of public repositories, reflect different models of the human genome as it has evolved, and suffer from gaps because researchers withhold key pieces for competitive reasons. However, by discovering overlaps among the data sets, SVBio can turn up crucial new information, such as a genetic variation associated to a disease. The algorithms that SVBio uses to find overlaps among data evolve as it feeds back results.
County Health Rankings & Roadmaps provides public health information on a county-by-county basis so that localities can compare their health rankings with others that have similar demographics. For instance, why do two communities in different parts of the country with similar populations and poverty rates have different rates of disease? It may turn out that one county has higher levels of education, and these are correlated with lower incidences of illness. Perhaps the other county can do more to keep its population in school. (Of course, if a county has a high educational level along with high poverty, it probably needs to find ways to generate more jobs, but that’s a separate issue.)

As with any conference, the second day of the Datapalooza had fewer people—over 2000 people registered overall—and less energy than the first. But the impressive parade of key figures in the health care field continued to light up the stage:

Day 1: Bryan Sivak, CTO of the Department of Health and Human Services; Kathleen Sebelius, Secretary of HHS; Jeremy Hunt, Secretary f State for Health in the United Kingdom; Atul Gawande, noted New Yorker author; Jonathan Bush, co-founder of athenahealth
Day 2: Todd Park, CTO of the United States; Farzad Mostashari, National Coordinator for Health Information Technology; Marilyn Tavener, Administrator of Centers for Medicare and Medicaid Services

Numerous other excellent speakers and deeply knowledgeable panelists joined these luminaries. And we must not forget the hilarious game show where doctors and insurance managers competed to guess what Medicare recipients care about, based on CMS surveys, in hospitals and health plans.

The references to the CMS surveys, which were conducted in a rigorously validated fashion, were more than fun. The surveys are invaluable for helping health institutions learn what their patients want and what matters. For instance, during the game show, the doctors guessed that a patient would consider average length of stay very important when selecting a hospital. It turns out that this level of sophistication is way beyond the ken of the average Medicare patient. Length of stay doesn’t turn up as a concern at all in their survey responses.

But this doesn’t mean patients are stupid or that their concerns shouldn’t drive health care. More on patient engagement later in this article.

Discontinuity of Care Documents

In I bemoaned the lag in the development of standards, among other things needed to advance the use of patient data.

The state of standards in health care can be compared to the web. Widespread adherance to the HTTP standard (as well as underlying protocols such as TCP, of course) makes it easy for a programmer to download a web page. Every modern language has a simple HTTP API. But once you get the page, understanding the data is more difficult. Usually, you’re reduced to the old "scraping" model, which requires you to visually inspect the HTML, derive some general principles behind the page’s organization, and hope that it won’t change. Web programmers recognize the difficulties and have stepped up with various mini-standards, including microformats and a handful of new tags in HTML5.

We can see a need for standards all over the health care field: the sharing of research data, the recording of device output, the exchange of patient records, and more.

I discussed research data in my coverage of Sage Congress, a summit on data sharing that brings together leaders in genetics and pharmaceutical research. I pointed out that research results, like web pages, are loosely structured, and mentioned a researcher who had trouble comparing two experiments that measured everything the same way except for medication doses.

Numerous standards have been developed to allow devices to send central systems their data. But I think the spread of consumer electronics for Quantified Self, along with the patient-centered medical home, will put strain on the standards, on the systems collecting data, and most of all on the electronic health records. Coordinating devices adds an extra layer of sophistication to the simple task of accepting data from each device separately.

But the nub of patient care, in terms of data, is the individual health record, and that’s where the big challenges lie. My article yesterday contained a modest critique of the Office of the National Coordinator, vaguely claiming they have been "slow to impose order." I managed to annoy a friend of mine at the ONC, and promised to explore the question further today. After all, didn’t the ONC promote two protocols for data exchange, CONNECT and Direct? Haven’t they been guiding the health care field toward practices that streamline secure and private exchange, such as DirectTrust and the recent implementation guidelines to assure security and interoperability over Direct?

And didn’t the Stage 2 rules Meaningful Use remove much of the mush around standards, decreeing the use of Direct for exchange and the Consolidated CDA as a format?

All those advances are worth citing. But there is much to be done.

Let’s take a few moments to look over the history of the continuity of care document (CCD), which used to be a couple pages sent with each discharged patient to his or her next institution (a rehab clinic, nursing home, or whatever) and is now in electronic form. A consortium standardized the format of this document in a sleek XML format called the CCR, standing for "continuity of care record."

The CCR became an instant hit in the IT community, adopted by both Microsoft HealthVault and the short-lived Google Health—but with incompatible changes in each case. This early lack of consistency suggests that perhaps the CCR was not as wonderful as its designers hoped. An expert in health care records has also told me that the CCR was not rich and flexible enough to accommodate the information needed in the field.

But these weaknesses were not why the HL7 standards body decided to create their own document format in direct competition with the CCR. Rather, HL7 didn’t like the CCR because it didn’t implement HL7’s Clinical Document Architecture (CDA), a complex XML structure that HL7 had designed as a basis for its various specifications. So HL7 designed its own Continuity of Care Document (CCD) format.

The CCR committee had struck off in its own direction for good reasons. HL7 standards had evolved in an ungainly manner over the decades, and the new XML documents such as the CDA were thinly wrapped versions of earlier idiosyncratic formats. I have been told that HL7 XML documents often failed XML validation suites. The need for the HL7 CCD format to adhere to the basic CDA format introduced complexity. A lack of specificity allowed vendors to wander off in incompatible directions and produce documents that required individual programmer attention to be harmonized.

The CCR really couldn’t form the basis for a national standard, but many of us were still disappointed that the HL7 CCA was adopted as the law of the land in Meaningful Use Stage 2. I’ve put that past me now, because I realize that the whole debate has the slightly musty feel of an outdated technology, like the debate of the mid-2000s between the office document formats, ODF versus OOXML.

The continuity of care document is a relic of an age when doctors thought they could just give the next doctor down the line a diagnosis and a medication listing, and they were done. Modern accountable care requires much more. Doctors must share comprehensive access to all the patient visits, test results, and treatment outcomes.

We can’t think of data exchange as a discrete event at the seam between one treatment setting and another, but as a continuous activity spanning the patient’s life. We are on the verge of a data deluge that will render the continuity of care document nearly irrelevant. Worrying about its format is like going to a barbecue and arguing over the brand of ketchup. Instead, we must make sure standards apply to the data that really matters. The chance to start anew is what excites me about Blue Button, and why I complained in yesterday’s article about the loose approach to maintaining it as an interoperable standard.

I think Bryan Sivak and Farzad Mostashari were speaking partly of standardized records when they said the ONC efforts were still "scratching the surface" of the data gap in health care today. The interface between electronic patient records, research, and medical devices needs work. And so I stand by my mild statement about the ONC.

Although I am not a fan of Linked Data, the World Wide Web Consortium’s solution to information muddiness, I think it may be useful in the health care field because the Linked Data tools, in all their complexity, are distinctly simpler than the complexity of health data, and therefore are worth the application of good minds. Linked Data requires a lot of analysis to straighten out the chaos of everyday life, but this too is a useful exercise because of the value of health data. Two sessions on Linked Data were offered today at the Datapalooza.

We are all patient advocates

I saw only two persons identified in panels as patient advocates, and met only a few others who identified themselves that way at the Datapalooza. But the modern zeal for giving patients more respect and control ran through the onference. It came up in Todd Park’s keynote and among other speakers, and motivated many of the apps on display.

One well-attended session covered basic privacy issues. We heard some of the themes that I have reported on before, such as how doctors claim HIPAA as an excuse for denying patients access to their own data, whereas HIPAA actually requires the doctor to allow access.

Perhaps the most impressive example at the conference of a large corporation paying sincere attention to patient empowerment was Pfizer Link, a site using Blue Button to give patients in clinical trials access to their own data. In the world of clinical research, this is a radical proposition. Craig Lipset, Head of Clinical Innovation for Pfizer, explained that only 5% of all patients participate in clinical trials. The industry has long been aware that the lack of volunteers adds significant delays to almost all studies. Pfizer is hoping that opening up patient access to data, along with news and relevant information on trials, will increase participation.

MedWatcher is an example of harnassing informed patient input to improve health care. After downloading the app, a person can report symptoms and any adverse effects he feels from a medication. MedWatcher combines and de-identifies this information along with reports from social media such as Twitter and Facebook, with the goal of revealing risks in medication use.

This app had a personal resonance for me because a relative of mine had recently collapsed and spent 24 hours undergoing electrocardiograms and other intensive tests. The collapse was eventually attributed to a new drug he had started taking, and I wonder whether he could have recognized the symptoms of fainting earlier and avoided the extremely expensive trip to the emergency room. The incident is a reminder: all of us are not only patient advocates—we are also patients.

Author’s home page
Other articles in chronological order
Index to other articles