John Halamka is the very model of an early technology adopter. Blockchain, augmented reality, precision medicine, machine learning, digital identity, and credentialing—Halamka has tested its applications in healthcare and has an opinion. “My life has been devoted to the pursuit of innovation—attempting to embrace new ideas and new technologies before the path ahead is completely clear,” he writes on his blog, “Life as a Healthcare CIO” (founded way back in 2007.)
Halamka is not only Chief Information Officer for Boston’s Beth Israel Deaconess Medical Center, he is also a practicing emergency physician, chairman of the New England Healthcare Exchange Network, and the International Healthcare Innovation professor at Harvard Medical School.
Beth Israel serves 3,000 doctors, 12,000 employees, and one million patients a year.
We spoke with him in March 2018 about his advice on bringing new technologies into a large healthcare organization—and some of the projects he has worked on recently with Amazon and Google.
From Dissemination to Diffusion
What you see so often in healthcare is announcements of pilots, where three people try something for a day, and it never goes anywhere. Then someone claims victory. That’s just silly.
The real measure of success is adoption. The February issue of the journal Health Affairs is devoted entirely to innovation, and it includes an article on dissemination versus diffusion. Dissemination is when I try something with blockchain, go to the HIMMS conference, and talk about it. Diffusion is when every doctor’s office—even the two doctor practice in North Dakota—is using it.So how do you get from a pilot to dissemination to diffusion?
Having a secret skunkworks outside of production and operations never, ever works. The cool guys in the skunkworks say, “We’ve got a flying car!” And everyone in the salt mines says, “We’re on old skateboards here.”
So we’ve tried to embed innovation in operations. It’s similar to Google’s idea of 20 percent [time for employees to work on their own projects.] You have this total alignment of business owner demand with innovation. It’s not IT trying something, like a 3D, holographic iPad. It’s because a business owner says, “We have a crisis right now—bed capacity, quality, readmissions—and then IT says, “I see that crisis, and I have an idea. Let’s try that idea and see if it helps your crisis.” Then if it does, the business owner says, “Let’s drive it everywhere.”
At Beth Israel Deaconess, our most expensive real estate is the OR. It’s $100-a-minute to keep it going. They’re expensive to build and maintain. The crisis the business owner had was that demand exceeds supply. And they said, “What can you do?”
We said, “How do you allocate OR time?” And they said [they allocate two hours for an appendectomy, whether it is being done by] an intern versus Dr. Famous, and whether the patient is an 18-year old or an 80-year old. We said, “Let’s have Amazon study our last two million operations. We’ll feed it three variables: who was the patient, who was the doctor, and what was the procedure?”
Beth Israel Deaconess has maybe 500 people who operate. So we started with 25, and said, “What if we simply re-do the schedule of 25 surgeons, and Amazon does it rather than a human? We were able to create 30 percent increased capacity in the OR.” That’s because you’re given 25 minutes, [rather than two hours,] for an 18-year old patient [being operated on by] Dr. Famous, who has done thousands of these procedures.
As a result, the administration says, “You’ve been able to free up 30 percent of OR capacity for 25 doctors. Let’s roll that out to 500, tomorrow.” The roll-out to everybody is happening now, and we have had the 25 have been in full production for about six months.
A Pilot That Didn’t Fly
[We ran a pilot test with Google Glass in 2014, which involved about 20 doctors.] Google Glass is the worst engineered device ever. It’s the Edsel of IT innovation. [At the outset, we] said, “We believe the idea of a wearable technology is inherently good.
Emergency physicians just have a hard time doing CPR while holding an iPad. [So] we went into production with Google Glass for emergency physicians. We told the patients we were doing a trial, and [the devices we used] were orange, so the patient could see it. We had incredible doctor satisfaction, getting hands-free [access to] patient data, and incredible patient satisfaction. Instead of sitting and staring at a keyboard, the doctor was staring at you.
It failed for three reasons. The battery life [of the Glass headsets] was an hour and a half, but doctors in the ER work an 8- or a 12-hour shift. So were you supposed to carry a car battery on your back?
Problem two is it used a Texas Instruments processor so underpowered that as soon as you tried to push it to do something, like voice capture [of a doctor’s dictation], it overheated and melted the plastic. We had Glasses melting on physicians’ faces.
The third was the TI processor was so old, there were no Android updates available for it. [That is especially important] around security. It was a medical device with no patching available.
We would have used it in a continuous way in full production if the engineering of the device had been sound. We do look forward to somebody—maybe not Google—creating that next-gen version of a wearable camera, voice control, and network connectivity which has 12-hour battery life and fully secure. As soon as we get that, we’ll try again.
We’re fine with failing, but we want to fail because technology isn’t ready for prime time—not because of change management or budget issues.
Buying Time from Collaborators
How did we learn not to use a skunkworks approach? What we found in the past is that you get shadow IT. Shadow IT means it’s not coordinated, and it doesn’t have the rigor of controls and evaluation and prioritization. In my 30s, I was the rogue. In the 1990s, I ran a skunkworks called the Center for Quality and Value. It was meant to be this really edgy, innovative place. But even if you innovated, it was hard to sustain those innovations, because they weren’t part of the fabric of the day-to-day operations.
Today, inside my organization, the IT organization, we have the Center for IT Exploration. It has two full-time people, but then [also collaborates with] about twenty part-time people [around the company.] Those twenty part-timers, some of them are in tech, but a number of them are clinicians who code, and can do interesting data analytics or machine learning activities.
If you carve out a protected day or two a week [for those part-time people] to work on innovation projects—and if it’s a meritocracy—people tend to put in nights, weekends, and free time, because [the work is] so exciting.
[We don’t ask the part-timers to volunteer their time, though.] I buy doctors’ time. [Do I have a vast budget to do that?] My budget is 1.9 percent of the budget of the organization. So I go out and seek donations. We’re a nonprofit; I’ll go out to our trustees or people in industry and say, “Would you contribute $100,000 to fund a series of innovations that are going to improve patient experience?” Increasingly, in the innovation economy, fundraising from non-traditional sources is part of the job.
I open source pretty much everything we do. It’s available free to anyone.
Governance is unbelievably important. If every time there’s a new bright shiny object, people say “Squirrel!” and chase it, you’ll wind up with a cloud-hosted blockchain with machine learning for APIs. There has to be a process where you bring stakeholders together, and say, “What should we do?” There are always a million things you could do, but what is the highest priority? What are our metrics for success? If we’re going to fail, how do we fail fast? You’ve got limited time and resources.
Machine Learning Applications
I have five Amazon employees in my organization, and we have twelve Amazon machine learning apps in production, and 36 in the pipeline…
Amazon is looking for the opportunities in healthcare. And unlike IBM Watson replacing doctors—not—Amazon is saying there’s a bunch of prosaic stuff that goes on every day that could benefit from cloud-hosted utilities. What is the most common technology in a doctor’s office today? A fax machine. If you were to have a surgery at Beth Israel Deaconess, how do we get your surgical consent? Your doctor faxes it to us.
We built a machine learning service to read faxes. Our faxes are read by Amazon’s [machine learning service.] Amazon says, “Oh, this is a surgical consent. Let me insert it into his record as a digital document, and check the box that surgery can begin.” [You can also apply machine learning to questions like, which patient] is going to need the ICU, and who is going to be discharged when?
My chairman of radiology told me that when he was a fellow, he was reading MRIs that had 20 images. Today, they have 300 images. Can a mere human look at 300 pictures in a single case and be able to find the needle in the haystack? It’s just hard. But what if you had a machine learning algorithm that said, “I’ve sifted through the 300, and here’s 20 the human should look at?” He told me that that would bring the joy back to practicing for him.
Accepting Reasonable Risk
Maybe there’s a point where you’re an early adopter, and it’s kind of risky. I think if you want to be an innovator, you have to have some tolerance of reasonable risk. If you wait until all the risk goes away, it’s too late.
[When I tell people that] all my production systems run on Amazon Web Services, [they say,] “How could you do that?” My answer is, “How many employees does Amazon have paying attention to resiliency and security?” I have five. They have more.
What’s the Catalyst for Innovation?
You do need a catalyst [to make innovation happen in a large healthcare organization], whether it’s a frustration or problem or economic urgency. I’ve certainly heard lots of pitches where what is being pitched is a solution in search of a problem.
You’re trying to connect your FitBit to Twitter to create a social network for weight loss? That [is likely] innovation for innovation’s sake. There’s no urgency.
[In the past decade or so,] our urgency was driven by regulatory compliance: HIPAA, the Affordable Care Act, meaningful use. Those all happened simultaneously. … Today, our urgency is about the business imperatives; it’s not the federal government telling us what we should work on next. It’s how do we survive changes in market conditions or reimbursement.
*This case study and many more can be found in our special report: Healthcare Innovation — Case Studies from the Leading Edge.