White House Champions of Change in Precision Medicine: Duke's Commitment

Today I had an opportunity today to attend a White House event highlighting work in the field of precision medicine. I was joined by my colleague Geoff Ginsburg who directs the Duke Center for Personalized and Precision Medicine (video replay here, with Geoff's comments at 2:08:30). As part of the event, Duke was highlighted for our commitment to precision medicine. This was for two reasons:

• The innovative MeTree platform created by Geoff, Lori Orlando and the rest of the MeTree team. This platform provides a way for patients to work together with their doctors to improve the recording of their detailed family health history - information that is critical to the success of precision medicine.
• The fact that the MeTree platform will be integrated into our Epic-based EHR using SMART on FHIR. Duke is an "Implementer" of the Argonaut Project, and was the first health system with an Epic-based EHR to run unmodified SMART apps directly (see our HIMSS demonstration video here).

As I've said elsewhere, we're on the verge of a renaissance in healthcare technology modularity and interoperability, led by open and familiar standards. The same principles that have led to the successful mobile app ecosystem are being applied to healthcare, with the result that many more innovators will have access to tools that allow them to build EHR-compatible apps. The problems in healthcare are enormous, and the more brilliant minds we have focused on these problems, the more likely it is that we'll find compelling solutions, and soon.

Geoff shared a few comments with the group at the event today, which I've shared here with his permission:

Good afternoon. My name is Geoff Ginsburg and I direct the Duke University Center for Applied Genomics and Precision Medicine. 

It is an honor for our work to be recognized at this Champions for Change for Precision Medicine event. I want to acknowledge up front the support of the Duke Health System, of my colleague Ricky Bloomfield (Director of Mobile Technology Strategy and hospitalist at Duke) who is in the audience and of Lori Orlando (a health services researcher and internist at Duke) who could not be here today but who has been key to the development of this idea and making it real. 

Today we are announcing the development of a platform that will make it easier for the patient to provide information about their family history to their provider and for providers to access family history and risk information to better care for their patients
 -- information that is critical to precision medicine. 

Several years ago we recognized that family history is fundamental to optimizing effective clinical approaches to personalized and precision medicine. However, our research showed that seldom, if at all, was a patient’s family history captured and adequately documented by providers. Furthermore, when histories were taken, there were challenges in interpreting the risk information in a multigenerational family history.

How many of you in the audience have had a truly detailed family history taken by your doctor and learned something from the results?

To address this challenge we created a patient facing, web-based, evidence based software platform to capture family health history – called MeTree. Patients talk to their families and loved ones about what illnesses family members have had and their age of onset and enter the information via the web into our software platform. The information is used to calculate risks for developing disease and the results are reported back both to the provider and to the patient creating an effective provider-patient interaction about their hereditary health risks and what to do about them. 

Two years ago we were fortunate to be funded by National Human Genome Research Institute to expand the reach of this platform to five different health systems across the country representing a variety of care environments and demographic groups. 

Now thousands of patients are learning about their family history of disease and using that information with their providers to get appropriate screening, genetic counseling and testing and taking actions to enhance disease prevention. 

Duke is pioneering the use of open, vendor-neutral standards espoused by the Argonaut Project. These standards will be in place by year’s end to enable this platform to be integrated into multiple patient portals and EHRs, which will allow will near universal accessibility of family history information to patients and providers seamlessly --  improving shared decision making related to prevention of inherited disease. 

With commitment comes responsibility. We will now get to work to honor this commitment to the President and to our patients. It is our hope that this work will help facilitate many more innovations from health systems and technology companies in the future so that we can realize our shared goal of higher quality and more cost-effective healthcare in our country.

Duke's on FHIR (but it's ok)!

Current EHRs are among the most complex pieces of software ever written. They serve a critical role to help standardize clinical workflow, facilitate billing, and integrate simplistic forms of clinical decision support, yet tremendous effort is required to customize EHRs to meet the needs of diverse hospital systems.

We felt there had to be a better way.

Starting in 2012 we started investigating ways to create a framework built on top of our Epic-based EHR that would allow us to access the EHR in a standard way from any device or platform. By the summer of 2013 we had a functional proof of concept that allowed us to access patient demographics, problem lists, medications and more from a simple Android app. Duke Apps Supporting Healthcare, referred to internally as DASH, was born!

Around this same time, we started learning more about a similar effort underway at Boston Children’s Hospital called SMART (Substitutable Medical Apps and Reusable Technologist) that incorporated a new REST-based open API called FHIR (Fast Healthcare Interoperability Resources, supported by HL7). Since our goal was general ease of use and interoperability, it made sense to join this effort, originally funded by the ONC via the SHARP program. There were already several proof-of-concept apps written to be SMART on FHIR compliant, so from a practical standpoint, this made sense.

We updated our code to be compliant with both SMART and FHIR and as of January 2015 we became the first Epic-based hospital system to run unmodified SMART apps within our EHR (in our proof of concept environment).

We were thrilled to present this work at the HIMSS 2015 national conference, and you can view a sample the demonstration here (UI bits blurred at Epic's request):

In order to have a compelling demo, we wanted to show several types of apps running in multiple environments. We chose:

• Growth Chart, an open-source pediatric growth chart app with an award-winning interface
• Meducation RS, a closed-source app by Polyglot that presents a patient's medication list in simple language translated into 21 languages
• Duke PillBox, a skill-based interactive learning tool to help patients teach themselves how to take their medications as part of the discharge process. This was developed by MedAppTech for a Duke research initiative.

We demonstrated each of these apps running in both the Epic desktop environment as well as the Epic mobile apps for iOS. Once the infrastructure was in place, it took less than 5 minutes to add these SMART apps to our mobile EHR. Playing with FHIR has never been so fun (or easy)!

Finally, we also demonstrated a native iOS app, Pediatric Growth Charts, that we launched with the patient context from the Epic iOS app.

So, we demonstrated an open source app, a closed source app, and an internally-developed app all functional within desktop and mobile EHR environments, plus a native iOS app to boot. And we're just getting warmed up!

Probably the most frequent question I get is: How did you do it?

Answer: Very carefully.

Ok, so the real answer is that this was a development effort in the truest sense of the phrase. Prior to our switch to Epic, Duke had a home-growth EHR, which means we have some seriously talented developers here who know how to write production-grade EHR code. In this case, we were able to use many of the web services already provided by Epic and simply add a FHIR wrapper. Some data elements required a bit more work. To tie it all together, we wrote a rate-limiting and authorization server in Node.js and installed MITREid Connect to handle authentication.

Is this the path we'd recommend for everyone? No, probably not. Fortunately, most hospital systems won't have to write their own SMART on FHIR implementations. Through the Argonaut Project, many major EHR vendors and academic medical centers have committed to FHIR as well as the OAuth profiles needed to make it all work. This includes Epic. So hang tight, and your EHR vendor of choice will likely soon provide you the tools you need to kindle the flame.

Duke is involved in Argonaut as an "Implementer," which means we have real-world experience implementing apps that use the standard, and have provided feedback to improve both SMART and FHIR.

As I've said before, there probably hasn't been a more exciting time to work in healthcare technology. Between novel patient engagement tools such as HealthKit and Google Fit, cutting-edge research platforms such as ResearchKit, and now truly modern APIs that will usher in a new generation of substitutable EHR apps, there's plenty to keep us busy, and most importantly, to help patients take better care of themselves and providers take better care of patients.

And the best part? Things are just starting to heat up ...

The Apple Watch: Counting Calories Counts

I've now spent a few weeks with Apple Watch, and while there's a lot to talk about, one feature in particular has "surprised and delighted" me, and it's not the one I was expecting.

I exercise regularly, but I wasn't expecting Apple Watch to help me much with respect to my routine. That's not because I didn't think it would be accurate or easy to use, but simply because Apple Watch wasn't made for me: I'm a swimmer.

Granted, it's been shown that Apple Watch can withstand a good swim, but it's still not designed to track swimming (I use my Pebble with the swim.com app for that), and Apple specifically discourages it.

So rather than lament Apple Watch's neglect of swimmers the world over, I went for a walk. A few of them, actually (dog needs exercise, too). And what I saw surprised me. Check out the following two workouts that Apple Watch saved to the Health app. Notice anything interesting?

I did a double-take when I saw these results, but then it dawned on me what was going on, and it's super cool. If you noticed, the first workout was a 0.93 mile walk that lasted 17 minutes, during which I burned 66 calories. The second workout was only 0.91 miles, lasted only 14 minutes, yet I burned 72 calories!

So how could a shorter walk (both in distance and duration) lead to more calories burned? While you mull that over, take a few minutes to head over to ABC News to view a 5-minute video about Apple's secret fitness lab. I'll wait.

Wasn't that cool? Apple is amassing what is probably the world's largest and most complete set of physiologic data during exercise (over 18,000 hours and counting), all while volunteers wear Apple Watch. This testing takes multiple factors into account, including activity level (accelerometer), heart rate, temperature, and probably also factors in height and weight when available, although I don't have access to their algorithms to know for sure. The volunteers in the fitness lab also wear specialized (and very expensive) gear designed to accurately measure caloric expenditure. These data can then be used to create accurate data models to fit almost every profile, leading to an extremely accurate estimate of calories burned.

In other words, Apple is validating the watch to be the most accurate consumer-oriented calorie-counting machine ever created.

Of course, this explains the "discrepancy" in my workout numbers. In the second workout, I had to walk much quicker in order to cover 0.91 miles three minutes faster than the 0.93 miles I walked the day before. Apple Watch knew I was working harder because it was tracking my heart rate every 5 seconds for the duration of the workout, and my heart rate reflected the increased activity.

Lesson: to burn calories, nothing can replace good old-fashioned, heart-pumpin' aerobic exercise!

Want to know the coolest part? Measuring heart rate is just the beginning! As I've mentioned before, now that we have a computer in constant contact with our skin, the market for transcutaneous sensors is going to explode, and the more data we collect, the more accurate these measurements will be.

Just hurry it up with the swim tracking, ok, Apple?

ResearchKit is upon us!

Today, Apple has unveiled the open source code to their ResearchKit framework. You can find it here:

http://researchkit.github.io/

I was pleased to see the code hosted on GitHub, which I've found to be exceedingly user-friendly and which we already use here at Duke Medicine.

Also, not only has Apple released the code for ResearchKit itself, but they've also released the code for all 5 of the ResearchKit apps developed to date, as well as the back end code used for those apps, called AppCore. That's the power of open source, folks!

We've been discussing ResearchKit internally since it was announced in March. Anyone waiting for the code to be released before starting on their ResearchKit apps is already behind. Why? I'd estimate that over 90% of the effort required to create a ResearchKit app comes before a single line of code is written. The code itself is simple and straightforward, making it easy to create a consent workflow and make use of active tasks.

But the real work comes in designing the study itself. What do you want to study? Why? What is your target population? Who gets access to the data? How many versions of the consent do you need? Do you want it localized? US? International? Has it been approved by the IRB? The list goes on.

In my opinion, this is the future of research for a certain category of studies that require access to large numbers of individuals for more refined data, or that need access to a very specific and hard-to-reach patient population, such as one with a rare disease.

I've already spent too long on this blog post. Our developers are jumping into the framework as we speak, ready to begin our study implementation. Time to dig in!

Meaningful Use of Patient-Generated Health Data

This week the NPRM for Meaningful Use 3 was made available in "unpublished" form on the Federal Register site. It seems that one of of the primary aims for MU 3 is to streamline the set of objectives applicable to eligible providers (EPs), eligible hospitals (EHs) and critical access hospitals (CAHs).

The new item most interesting to me is Objective 6: Coordination of Care through Patient Engagement (starts on page 103 of the linked document).

This proposed objective aims to "Use communications functions of certified EHR technology to engage with patients or their authorized representatives about the patient's care" and employs three measures:

Measure 1: >25% of all unique patients "actively engage with the electronic health record made accessible by the provider" either by 1) viewing, downloading or transmitting to a third party their health information; or 2) "access[ing] their health information through the use of an ONC-certified API that can be used by third-party applications or devices."

I've previously discussed the Argonaut Project Implementation Program and its relation to the SMART on FHIR project. The FHIR APIs and added functionality of the SMART project (OAuth, OpenID) will dramatically lower the barrier for third-parties to easily add functionality and significant value to current EHRs. While these APIs are already enjoying broad support even before they are complete, seeing this emphasized in the MU 3 NPRM is a testament to their importance.

Measure 2: For >35% of all unique patients, a secure message should be sent using electronic messaging function of CEHRT to the patient, or in response to a secure message sent by the patient.

It's critically important that we encourage direct engagement and interaction between patients and providers, and this measure intends to do just that.

Measure 3: "Patient-generated health data or data from a non-clinical setting is incorporated into the certified EHR technology for more than 15 percent of all unique patients."

This is exciting. While patient-generated data can come in many forms, including manual entry by patients, this measure will only be achievable if we employ technologies that reduce or remove such barriers. Apple's HealthKit is by far the easiest-to-deploy tool to facilitate this data handoff currently, and it's available right now. We're hopeful an Android-equivalent will be available soon for patients with those devices (Google Fit doesn't yet ... fit that purpose).

Neither SMART on FHIR nor HealthKit are yet widely deployed or adopted, but these technologies will undoubtedly be critical to ushering in the learning health system, and it's great to see APIs and patient-generated data being emphasized in the latest NPRM.

MU 3, welcome to the 21st century!

2015-03-24 UPDATE: Just to make sure this is clear, MU 3 is still draft at this stage, and the content is subject to change. Also, attestation for objective 6 will require meeting only 2 of 3 of the measures listed above.

ResearchKit - More Details

ResearchKit is exciting, as I've already noted. Apple has posted some additional information in the form of a Technical Overview document. This document publicly sheds a little more light on what ResearchKit will enable.

Apple describes three "modules" available within ResearchKit, including Surveys, Informed Consent and Active Tasks.

Surveys

ResearchKit provides a standardized interface to quickly build surveys. These modules are already localized. This would work for the majority of current research use cases.

Informed Consent

One of the most significant inclusions is native informed consent capability, as you may have seen in the keynote:

Credit: Apple

This is something many researchers have been scrambling to buy, create or produce since it's not currently a common feature of modern EHRs. The fact that this is now available out of the box and in open source form is compelling. The informed consent mechanism is flexible and takes into account the use of waivers, use of institution-specific ethics language, and also provides the ability to insert comprehension tests ensure the patient has capacity to sign. The consent framework then generates a PDF for upload or email.

Active Tasks

As demonstrated in the video today during the keynote, ResearchKit provides the ability to capture patient interactions which Apple calls "Active Tasks." The Active Task modules currently available in ResearchKit include:

• Motor Activities
• Gait, tapping
• Fitness
• 6-minute walk
• Cognition
• Spatial memory
• Voice
• Phonation

All of these are accomplished using hardware built-in to the phone itself.

Open Source

Probably the most significant innovation here, however, is the fact that this is all (or will be soon) open source. For this to succeed and expand, it's vitally important that academic medical centers keep this in mind when developing their apps, and that they choose to share the source of their own apps so other researchers can build on their foundation. This collaboration will be the key to accelerating the transition to a Learning Health System.

ResearchKit and the Future of Healthcare

Apple landed a pretty big bombshell today with the announcement of ResearchKit. And probably the biggest part of the announcement was this:

Yes, ResearchKit is fully open source. Why is that important? Because it means that this technology will eventually be available on any platform. Yup, including Android. And Windows. And whatever else comes in the future.

And it also means that integration between ResearchKit and other emerging healthcare technologies will be possible, including the SMART on FHIR platform, which is leading the way to modernize healthcare interoperability.

Given Apple's commitment to privacy, I have no doubt that the healthcare industry will quickly adopt this new platform, as evidenced by the high-quality institutions already participating.

As I've mentioned previously, the Learning Health System represents the vision of a world in which we are constantly learning from a stream of high-quality data, using that data to quickly make even better decisions, even at the point of care. ResearchKit will help us reach that goal even more quickly, possibly even before the 10-year goal as set forth by the ONC.

Let's get started!