How to open an industry: data points from Strata Rx

O'Reilly conference brings together health care and data

O’Reilly’s first conference devoted to health care, Strata Rx, wrapped up earlier this week. Despite competing with at least three other conferences being held on the same week around the country on various aspects of health care and technology, we drew a crowd that filled the ballroom during keynotes and spent the breaks networking more hungrily than they attacked the (healthy) food provided throughout.

Springing from O’Reilly’s Strata series about the use of data to change business and society, Strata Rx explored many other directions in health care, as a peek at the schedule will show. The keynotes were filmed and will soon appear online. The unique perspectives offered by expert speakers is evident, but what’s hard is making sense of the two days as a whole.

In this article I’ll try to show the underlying threads that tied together the many sessions about data analytics, electronic records, disruption in the health care industry, 21st-century genetics research, patient empowerment, and other themes. The essential message from the leading practitioners at Strata Rx is ultimately that no one in health care (doctors, administrators, researchers, regulators, patients) can practice their discipline in isolation any more. We are all going to have to work together.

We can’t wait for insights from others, expecting researchers to hand us ideal treatment plans or doctors to make oracular judgments. The systems are all interconnected now. And if we want healthy people, not to mention sustainable health care costs, we will have to play our roles in these systems with nuance and sophistication.

But I’ll get to this insight by steps. Let’s look at some major themes of Strata Rx.


The first thing to acknowledge is that the data scientists have provided us with very powerful tools indeed to organize the world. We see this in the way retailers make recommendations for our next purchases, the effort politicians put into wooing us, and everything else. Health care has lagged behind, but a few nubile practitioners are putting their data to good use. Increasingly, such practitioners will push others out of the business on the grounds of both cost and quality.

Some projects aired at Strata Rx that show the power of data include:

  • Tina Brown-Stevenson of UnitedHealth Group, the country’s largest insurer, found changes in metabolism that could predict the onset of diabetes two years in advance. Imagine how much behavior change you could teach a patient in two years. UnitedHealth Group also discovered that the cost of treating these patients went up even before the onset of diabetes.
  • Bharat Rao, part of a team at Siemens, asked doctors to predict the survival rates of patients after cancer treatments and found that the doctors’ predictions turned out no better than a flip of the coin. On the other hand, accumulating data and comparing patients with different profiles, Siemens came up with a system that predicted outcomes pretty well. Although this is not reliable enough yet to be used in planning treatment, they hope they will be able to provide that support someday.
  • Niall Brennan, at the Centers for Medicare and Medicaid Services (CMS), drilled down into the economics of hospital readmissions for the senior citizens covered by Medicare. A power law applies here: only 4% of the 31 million Medicare beneficiaries have readmissions (7 million don’t become inpatients in the first place), and only 0.5% have three or more readmissions. But CMS spends an average of $36,000 on each of that 0.5%. This is a quality-of-life issue as well: it’s not helping people to keep readmitting them when the hospital stays clearly aren’t curing them. Concentrated efforts to reduce readmissions among the 0.5% will improve care and help Medicare’s bottom line, which is so much in the news these days. I was a little disappointed in Brennan’s talk because I was hoping he could show how analytics could fix the problem, not just measure it. But CMS’s deft manipulation of complex data was still impressive.

If we accept the old adage that power brings responsibility, we have to make sure that big data is used for patients and not against them (or against doctors). Keynote speaker Fred Trotter walked a fine line in releasing a potentially virulent strain of data into the conference: he distributed nationwide data he had wrested from the Department of Health and Human Services about referrals between doctors. When this goes online under an open Creative Commons license, anyone can check who is getting referrals (and therefore may be a more desirable specialist to visit) and who knows the desirable specialists. He describes the referral data as a “social graph” for doctors. Combined with other data, one could expose things such as which doctors refer patients to hospitals with high infection rates.

Trotter, who has written a book with David Uhlman for O’Reilly on the health care industry, challenged the audience to develop “a doctor rating algorithm that patients find useful and doctors find fair.”

John Wilbanks brings responsibility to the health care field by designing a system he calls Portable Legal Consent for educating patients and allowing them to choose the use of their data. Wilbanks gave a keynote explaining the role of data sharing in creating a standards-based health research system.

Wilbanks claimed that health care is such an inefficient system that monopolies tend to emerge — in delivery, in record-keeping, and in research — in order to take advantage of economies of scale. He wants to shift from monopolistic institutions to universal standards, and allow more innovation that way by both incumbents and new entrants. Wilbanks works on Portable Legal Consent for Sage Bionetworks, whose president Stephen Friend also delivered a keynote about open research.


Cheek by jowl with power lies humility, which everyone in the health care field has to experience when dealing with recalcitrant patient cases or systems. I will probably experience mine after readers comment on this article.

Humility was promulgated by Jamie Heywood of PatientsLikeMe in a rousing keynote that revealed numerous research studies where data analytics on PatientsLikeMe members disproved the results of the carefully controlled clinical studies.

For centuries, clinical studies have been the gold standard of health research, but they are being questioned more and more — and not just because the pharmaceutical companies and academic researchers who conduct them have conflicts of interest. As Bharat Rao pointed out, clinical trials try to choose subjects who are healthy in every way except for the condition being researched. They are not representative of the larger population who will ultimately use the treatment. And a clinical study can never use a population as large as a group such as PatientsLikeMe.

Still, it’s shocking to hear (as a speaker pointed out at the Ignite! talks on the first evening of Strata Rx) that 80% of studies in medical journals are irreproducible. Heywood pushed the limits with his keynote, suggesting we might do just as well to throw out all the results of clinical research up to now and start over.

But self-selecting populations such as PatientsLikeMe and 23andMe aren’t necessarily representative either. Bastian Greshake (@gedankenstuecke on Twitter), who runs one of these patient recruitment sites called openSNP, pointed out to me that they are short of certain types of people, particularly those who are more reluctant to share information. He can prove this by genetic tests on the people who have signed up!

The data itself humbles us by its sheer size. It has become a modern pastime to awe one another by recounting the number of search requests handled by Google each second or the petabytes of information collected by radio telescopes. Health care offers no end of such machismo statistics, particularly when we consider medical images and genetic data. At Strata Rx, David Ewing Duncan played the game best, although he is busy generating data about himself and encouraging others to do so.

Good algorithms can extract insights even from dirty data, but quantity is no guarantee of value. Our ignorance of what works in treating people is matched by our lack of good data. Medical records are untrustworthy: they are inconsistent because so many people enter data at different times, and often just plain wrong.

Perhaps the worst problem is that a lot data isn’t entered even though the rules state that it should be. The clinician forgets to ask the important question (such as about smoking or substance abuse), or if she asks the question she doesn’t write down the answer, or if she writes down the answer she does it in some free-text abbreviation that’s hard to find programmatically.

Thus, one researcher said most work with health data involves cleaning and harmonizing it, leaving only a small amount of time at the end for interesting research.

Whatever is left of good data must be weakened further, if data is to be shared, by removing facts that could re-identify patients. So if you have a great idea for a consumer app based on health data, go ahead and develop it — just be careful what you promise.

We must be humble about the pace of our achievements. As forums on electronic records and health information exchange showed at Strata Rx, most sites are just beginning to collect data that can be valuable for research and treatment — much less to share it with others and use it to effect change. According to Andrew Kress of IMS Health, only 14% of prescriptions in the US are made electronically, even though e-prescribing has been around for a couple decades.

And how much can doctors accomplish at all? Given many of the age’s most serious challenges — obesity, substance abuse, sedentary lifestyles — we need a public health approach rather than a medical approach. As Alexandra Drane of Eliza Corporation pointed out, marketers do better at persuading people to eat frozen dinners than we do persuading them to eat broccoli.

We need to update our appeals, and the beginning is to recognize that the average person doesn’t define himself by his weight or glucose levels. He does, however, talk about job stress and money problems and being a caregiver at home. Health professionals must tie these issues at the top of the patient’s mind to the behaviors that can improve his life — and his health along with it.


I found the most fertile and far-reaching theme at Strata Rx to be integration. Data must come from many places, and a successful researcher must use everything he can get. A few simple examples of integration include:

  • Accountable Care Organizations (ACOs), the new vision of Centers for Medicare and Medicaid Services, which can succeed only by getting many doctors and institutions to work together and keep their patients inside the system by demonstrating high quality. Integrating their electronic health record systems isn’t required, but it’s a natural step toward maintaining the high level of care as the patient moves between doctors and facilities. But John Kansky of the Indiana Health Information Exchange reported that patients in an ACO will still receive 30-45% of their care outside it. This means an ACO will have to adopt standards and exchange data seamlessly with other providers.
  • Consumer sharing sites, such as PatientsLikeMe and 23andMe, whose founder Anne Wojcicki delivered a Strata Rx keynote. Such sites not only allow a single patient to understand herself better by comparing her background to others, but create an enormous database that researchers can draw on. There are certainly privacy risks associated with publicly sharing one’s data, but lots of people clearly are willing to take the risk in the interest of pushing the boundaries of science. More sophisticated iterations of these patient sites may help people perform some of the self-censorship that data experts do for them when de-identifying clinical data.
  • Open research, which is being seen increasingly as the salvation of pharmaceutical companies who are having trouble developing new drugs and getting them through successful trials. This movement is the provenance of Sage Bionetworks. If researchers can be persuaded or compelled to open their data sets, the costs of research can decrease. If companies publish all their data on trials — including trials they would rather not have the public know about — we can avoid going down unproductive research paths.

All these current trends highlight integration. But Strata Rx showed that the health care industry industry must go much deeper — and must give up many of its most comfortable practices to do so. Its very workflows must be integrated in all the following areas — and more:

Clinical care and accounting

Old way: a doctor diagnoses each patient who comes before her and prescribes a treatment, theoretically without considerations of costs (although in reality such considerations can sway a decision). The billing office then handles payment. Clerical staff handle the patient’s next appointment as he walks out the door.

New way: Scheduling becomes a data-driven clinical decision. Part of a treatment plan is to follow up by phoning the patient and checking to make sure meds are properly taken or other recommendations followed. Furthermore, longitudinal data about patient visits are checked to determine whether treatment is effective and what should be changed.

Clinical research

Old way: Academic clinicians derive suggested experiments from their knowledge of the human body and its genes. After winning funding, they recruit patients who follow directions and are informed of results at the end of the trial.

New way: Patient advocacy groups haggle with researchers to design trials that meet patient needs. Researchers hide less about the experiment from patients, giving them feedback that help them participate with more commitment and gusto.

Patient-doctor relationship

Old way: The patient approaches the doctor with a complaint, or a routine checkup reveals a problem requiring further research. After the poking, prodding, and imaging that give the doctor material for a diagnosis, the doctor gives the patient a treatment plan and sends her home. Records stay with the doctor, who generates all the information in them and treats them as intellectual property

New way: Health is a life-long endeavor whose main guardian is the patient herself, along with her family and other cohorts. The doctor must not only suggest a treatment but motivate the patient to stick with it. Patients keep track of their own records and gain the knowledge to second-guess their doctors. Patients also collect their own data, both by hand and through devices in the home, so the doctor is just as dependent on the patient for data as the patient is dependent on the doctor. And patients are researching where they can get the best care for the lowest relative cost.

Public health reporting

Old way: Government agencies give health providers lists of incidents and facts to report regularly. The agencies then combine the data into their own reports, which are released to the public and which providers can draw on or ignore.

New way: Public agencies collect data from a variety of sources and use it to influence provider behavior, such as identifying hot spots. The providers are constantly comparing their performance against the norm and using feedback to adjust their practices.

In short, one-way flows are being converted into collaborations all over the place. And all these interactions are greased by data, which makes it possible to assert claims and check on their validity. This is what health care needs to enter the 21st century, and why Strata Rx is a significant contribution to the field.

tags: , , , , , , , , , ,