At Penn Med, using software to cut health costs

See also my column in the June 12 Sunday Inquirer. -- Roy Rosin is Chief Innovation Officer at the University of Pennsylvania's $5 billion-a-year medical complex. It's not exactly R&D: Innovation there means cutting costs -- Rosin calls it "getting to better value and patient outcomes" -- by using software to save time and steps.

Rosin told me about what he and his colleagues are building -- doctors and profs from Penn's several schools, plus Penn health innovation chief David Asch, Independence Blue Cross innovation chief Tom Olenzak, and investors like DreamIt Health boss Steve Barsch (predecessor Elliott Menschik is now at Amazon -- including "Innovation Accelerator" experiments, "solving some of the big, intractable problems in healthcare delivery:"   

ROSIN: I actually think solving the big healthcare problems requires an interdisciplinary approach. You can't just design a new intervention -- product or servcie -- without making sure you have a viable business model. 

If you're still being paid for the servcies you provide, even while moving to accountable care, and come up with something that rduces the need for those servcies -- it generally means we need to evolve how that increased value is shared. 

So we need creative business thinkers, alongside innovative clinicians.

Many of the big, costly health problems out there have to do with human decision making and how people interact with systems, so behavioral sciences and design are critical. 

It's really fun to see the impact of these new perspectives coming together. In the past few weeks alone, I've seen:
-- Design students reimagine a bassinet that keeps newborns safer in the hospital;
-- Engineering students come together with our nurses to make breast milk inventory management both safer, and much more efficient;
-- Wharton teams with MBAs, physicians, designers and engineers work with our orthopedic practices to re-invent the pre-operative process for hip-replacement surgeries, to improve the patient experience.

This isn't the form some folks in medicine fear, where these outsiders storm in saying, 'We know best and can solve all your problems for you.'...

Instead (it's) close collaboration, where we're leveraging clinicians' insights with new skills and perspectives.

How do you know a project is working? In every project, we try to quantify a benefit. I can't always measure how we are doing at the program level. But each project is chartered to move an explicit needle (such as readmission numbers, or access to appointments)...

The teams we worked with as part of our last Innovation Accelerator projects -- where we look for colleagues with both new ideas and a passion for solving high-priority problems -- we (went) 8 for 8 (showing improvement in defined metrics.)

One was for getting people discharged from the hospital to show up for their follow-up appointments, where the team designed an intervention that took the metric from less than half of patients, to 100% (for one cohort).

One was for a better model for helping women who had early pregnancy complications -- specifically miscarriages, where Dr. Courtney Schreiber and her team, not only crafted a much better patient experience, but by avoiding the Emergency Room and Operating Room, saved $500 per case.

That's important because the federal government has expanded the breadth of health insurance subsidies, but also cut the reimbursement for procedures. If the old rates gave hospitals and doctors more than the cost of services, current commercial and Medicare rates are often way below cost and force you to economize. It means a therapy that keeps patients alive but needing care can create enormous expenses...

One example of fundamentally better interntions that can be hindered by payment models can be found in the case of  diabetic retinopathy. It is a common condition and causes blindness that is easily preventable, as it develops over a fairly long time. You can either pay hundreds for treatment if caught early, or tens of thousands later for laser treatment and injections in the patients' eyes. Catching it late in the game can literally mean $16,000 per eye per year, (and) vision loss.

You've been to the eye doctor. You know about eye drops and dilation. You can't work, you aren't supposed to drive. Dilation is unpleasant. And since it requires a specialist, you also have access issues, so there can be delays.

So now some brilliant engineers have invented a camera that doesn't require dilation to take a high-quality retinal image. You don't need a caregiver to drive you to appointments if you use this, since you don't have vision issues from being dilated. And you no longer require specialists for the screening procedure.

The question, can we make this work in a real setting?


Dr. Thomasine Gorry, an Opthamologist and one of Penn’s specialists with deep expertise in this arena, brought us this concept. She’s a wonderful clinician, and a passionate innovator, in a group really intent on improving how we provide care. 

What she and her team were able to demonstrate very quickly was that we can catch a high incidence of retinopathy -- and glaucoma -- by placing these cameras in primary care and... in labs where people go to get blood work done (for) specialty practices. 

The cost of screening is lower, the patient experience is better with less discomfort, time and inconvenience and access is improved. 


So what’s the problem? In the old model, while screening costs were higher, we were reimbursed for more than our cost.  In the new, higher-value, lower-cost model – better for the patient in terms of outcome and experience, and better for the payer – teleretinal imaging reimbursement is far less than our costs. 

We’re working with our payer partners on this. But we have to figure out how make sure payments for novel interventions are fair and create incentives and financial support for delivering this care.


Does the pace of academics frustrate you, after a career at Intuit and in the private sector?

I want the health system to operator on a cadence like Silicon Valley cadence. That means iterations in days and weeks, instead of years. We are getting there.

And we've shown it's absolutely possible. In our Accelerator program, Dr. Schreiber and her team, working on the PEACE model for better care in the case of miscarriage, did 8 pilots in 90 days. 

People said, are you kidding? That's not the cadence of health care.

We had three things to prove:
Can we safely sort and route patients so those who should be in the emergency room (arrive) there or stay there?
Second, can I catch them and reroute them to the clinic (designed for an "improved experience" plus "cost savings") in time to avoid the costs associated with emergency room or operating room care?
Third, is it economically viable to do this? 

So it worked. It worked insanely well. The team working on this has now seen interest in establishing this team design, including (from) Pennsylvania Hopsital. That's what we look for -- reproducible results.

We love these fast experiments. The goal is scaled impact... 

Anna Doubenis is a family medicine practicioner and professor. She led another of the Innovation Accelerator projects focused on "super utilizers," the 1% of people who account for 30% of costs. On her initial, small-scale, high-touch model, she showed a 42% drop in Emergency Department utilization.

Shreya Kangobi is a brilliant researcher and physician we worekd with in an earlier cycle of the Accelerator program. She had developed a novel, evidence-based community health worker model. She redesigned how to effectively use community health workers. She changed who she hires, how she trains them, how she deploys them. Shreya was employing a rapid innovation model even before she met us. She has this entrepreneurial midset and an approach that involved frequent, rapid iterations based on teh deep immersion her team has with this patient population.

She finds these people she calls natural helpers. People who would bring (a sick neighbor) chicken soup before it was formally their role to help. They are in the community. They are not college kids with a social mission. 

The major determinant of health is (often) not a medical thing. (They may have symptoms of illness) because they are depressed. Because they have food insecurity. Because they are being abused. Because they have addiction issues.
The stories that come from Shreya's group, they are mind boggling. A woman whose son was incarcerated, once released, became her major concern. So helping situate him was teh most important first step in helping her begin to take care of her own health.

A man who wasn't taking his medications -- why? He had been a musician earlier in life. Shreya's group was able to reconnect him to that passion. Once he was engaged in life they were able to work on his health needs...

You see, the key wasn't patient education, or reminders to take your medicine or go to your appointments. It's, 'what's the root of this problem?'

They got to the bottom of that. It was work on human motivation. These are such complicated things.

There was a woman who was a super utilizer. A 'Frequent Flyer.' This group got her to open up. She was being abused by the landlord. He was stealing her checks. They had to get her set up with safe housing. All of a sudden her health improved.

The root cause is different from what we think. 

Dr. Anna Doubeni, the family medicine practicioner -- she does house visits -- her super-utilizer program is making progress. She is looking at multi-factorial problems. There is no way to have good health if you are homeless. You can't keep your meds cold, for example. 

Big problem was not knowing who needed attention, who had increasing problems and needs, who started showing up more frequently in the emergency room.

My team built Doubeni a simple dashboard so we could identify which patients required more attention to get them back into the family medicine practice. .

Don't insurers and Medicaid programs, with all their patient and payment data, track all that?

I don't think insurers' information is timely enough to be actionable at this time. Claims data which gives us the full picture, even if Temple or Jeff or another teaching hospital is involved, it's not realtime.

Dr. Shreya Kangovi, her program is called IMPaCT, she has this Penn Center for Community Health Workers. 500 programs have downloaded Impact. It's  becoming a national model.

And there's a DreamIt company, Keriton, they came through the Penn Apps Hackathon, which (Penn Center for Innovation COO) Laurie Actman (CORRECTED) helped set up to connect academic guys with real world advisors...

Lactation nurses in the HUP NICU needed some way to track and measure breast milk -- pumping too much, or too little, are both problems. It turns out that the inventory management of mother's milk takes thousands of hours for nurses -- who while they're doing it can't do patient care. 

Laurie Actman from Penn's Center for Innovation and I connected clinicians who had real-world problems they were trying to solve with the engineers participating in the Hackathon.

We had people from Vijay Kumar's Engineering School in the audience listening. A couple students with robotics, mechanical engineering and computer science backgrounds were interested in this problem. One of them had a personal family issue, that student said, "I get it."

Bhatnagar Vidur,  a software engineer, and his cofounder Sneha Rajana, who has hardware expertise, rapidly prototyped a solution that won second place at the competition.

They got energized about the problem and helping the nurses, new moms and badides. (So they) decided to pursue this as a company.

I recently attended a meeting where they presented their progress back to the nurses they had met with and learned from. The energy in the room was through the roof. This kid is a first-time entrepreneur. He'd been living in the ICU, he just immersed himself. And here he was, telling the nurses what he had learned from them.

He understood not only the problems, but the context in which the problems emerged. He knew where problems occurred, why the system had to be designed to be 'hands free,' and what functionality would eliminate the most wasted time. These insights are leading toward a solution that the nurses can't wait to use.

So they gave the program remote sensing. Sensors that can measure and tell you how much has been pumped and how often the woman is pumping. Bottles and labels that can be tracked with hands-free scanning, so checking in multiple bottles takes a fraction of the usual time and effort.

Automated inventory management -- that means we know when we're running low, that the right baby always receives the right milk, and that the baby never receives expired milk.

I love these guys. The Philly ecosystem is coming together to support this high-potential company. Not only did they secure $50,000 from Dream-It. They also received support from the Athena fund (for women founders and cofounders), the Wharton Innovation Fund, and First Round Capital's Dorm Room Fund. All together that gave them nearly $100,000 to get started.

Is Penn doing anything you don't see on other campuses? It's not normal in most places.

Penn has this interesting advantage: Everything is on one campus. At Harvard the medical school is in Boston, the business school is across the river, the engineers are down over there. Penn is totally different. You can walk there.

Our work with Sarah Rottenberg in the Design School's (Integrated Product Design) program means that every year talented designers tackle real health system problems.

We partner with Chris Murphy, a computer scientist at the Engineering School, as his students are also looking for important, real-world problems, and they bring programming skills that can turn ideas into prototypes quickly.

I can't say the process is very fancy. I blast emails to a few contacts: 'Who has a problem we can (attack)?' 

It doesn't work all the time. I can't say the success rate is 100%. We have done dozens of projects. We have maybe a 60-70% hit rate. 

I love this generation. They want to improve the world. It's the concept of repair the world. (Tikkun olam. Or St. Francis.) That is their attitude. They have gotten to this point: 'I don't want to work on the next Facebook. I want to solve a real problem.'

We work with Drexel grads, too. They have great training and technical skills. (He cited the example of Drexel grads at the Conshohocken-based software engineering group, Transmorgify.)

You have found a way to speed up research? I believe my rapid experimentation techniques matter. We work with teams to identify critical assumptions and design experiments that can produce early evidence, quickly, at low cost. But it's not cut-and-paste from the software industry.

The biggest change in innovation in the last 15 years have been techniques for how you learn fast at low cost, validating or invalidating core hypotheses that tell you either to keep going or change direction.

Innvations often fail due to scaling prematurely. So the key now is to get it right ifrst, even if for only one group of patients in one location, before making it big.

Silicon Valley uses a lot of quick-and-dirty techniques to help focus product development... 

One technique is called 'Fake Back Ends," where behind hte scenes you have a first version held togehter by chewing gum and Scotch tape, that allows yo to try something out at very small scale.

One example is when we went after trying to keep women at risk of post partum hypertension safe after they went home -- but (they needed) blood pressure (measurement) to do so.

Two of our superstars -- Katy Mahraj, an innovation manager, and Matt Vandertuyn, a design scientist -- worked with Drs. Sindhu Srinivas and Adi Hirshberg. They improvised a system of sending patients home with blood pressure cuffs, adn started texting them in to send in the (blood pressure) values.

It's not something that works out of the gate. So, again, you dont even know what to build at first.

Instead of investing in a fancy automated system, the back end in this case was actually Dr. Hirshberg pretending to be the system. (Which) we might eventualy build. But since it wasn't hard-coded, we could keep iterating and trying new things until we found what worked: Texting at different times, personalizing the messages, enlisting help from the patients' family or friends if they weren't responding.

After several iterations, the team had successfully and dramatically moved the needle of getting the informationa nd keeping these women safe and out of the hospital.

Now that we know what works, we'r building a scaled version of this intervention on a real technology platform. The team has won severla awards and grants...

So the doctors in Orthopedics told us, we have an access problem. Everyone said we should do same-day scheduling. But it turns out that's really hard. You have to negotiate with the physicians on changing how slots are held, and infromation has to be moved and routed differently in the call center.

In this case the practice leader did an outstanding fake back end. He grabbed a friend in marketing and put up a billboard on our Web site: SAME DAY APPOINTMENTS Now Available. They had a big clear call-to-action, with the phone number on the site. But that was his personal cell phone.

He became the fake back end. An improvised call center. (That circumvented) a great deal of the complexity of changing systems.

A physician agred to try this with him, even if only for a few patients, or a few days, so they could test demand (and) how they might operationalize it...

Everything changed in those three days. The conversion rate, from inquiry to appointment, went up, and almost half of the patients coming in were new to Penn...

It was this population of people who were convenience-oriented. It turned out we hadn't been known as the easiest place to get into.

One learning was that we didn't even really need to do same-day appointments (patients were happy to wait a day or a few days) -- but by announcing this availlability it communicated yo gould get in quickly.

The early evidence their experiment generated in three days, led to the program being implemented.

So those kinds of techniques, are really the same as what's increasing the rate of change in Silicon Valley. What can you learn in hours, that might have taken years, using the old way of working?

I'm not trying to prove anything scientifically, at first. I am experimenting anecdotally. And when you do that, all of a sudden you start to have data, evidence and insights based on reality, not an abstract concept, theoretical strategy, or a PowerPoint.

One of the early people at Google, Alberto Savoia, liked to say that ideas, by themselves, have no value. They are worthless. But once someone acts on it and you start to have some data, it's no longer worthless. 

So the question becomes, can you overcome inertia to do something? To learn something? You have to DO something to learn something. And then you can use that anecdotal evidence produced by these fast experiments to overcome the inertia.

Dr. Mitesh Patel recently joined the Penn faculty. He is affiliated both with our innovation center and CHIBE (Penn's Center for Health Incentives and Behavioral Economics). He's one of the most productive persons on the planet, doing incredible work at a hyperproductive pace.

Mitesh is now leading our new Nudge Unit, the first such group in a U.S. academic institution.

Nudges use techniques of choice architecture and other insights from cognitive sciences to help move people to making better decisions and choices. 

One of my favorite examples from his work includes an intervention where he worked with our IS team and our chief medical information officer, Dr. Bill Hanson, to make generic medications the default in our emergency room.

Use of generics not only lowers costs to drive higher value care, but also, due to affordability supports higher medication adherence. 

We saw an immediate and dramatic impact shifting towards appropriate prescriptions of generic medications based on this work.

Some of his latest work revealing intervention designs that best promote physical activity is a good example of where rapid experiments can accelerate and support but not replace gold standard research methods like randomized controlled trials, RCTs.

When I came here I looked at all the great work Kevin Volpp, David Asch, Mitesh and other faculty in CHIBE were leading, applying behavior change insights to healthcare delivery. 

I wondered if some of their insights – including the power of social support – could be used to promote physical activity. 

In the mode of fast experimentation I structured a competition where we organized teams of four to get accelerometer apps on their phones and compete in terms of daily steps taken by the team. 

Every enterprise has a wellness program. We suspected they weren’t optimally designed, especially as the incentive designs are not based on evidence and CHIBE faculty’s work was revealing that incentive designs really matter. 

When companies have walking competitions, traditional programs are designed so employees with the most steps win. 

I played around with that. We ran some fast experiments. We were able to quickly demonstrate some interesting insights in the effects ofteam structures (team interactions promoted increased activity) and incentive designs. 

Pretty quickly it started looking like ‘most steps win’ designs just made active people superactive But it also just makes sedentary people discouraged.

So what do you do to improve your real target, the more sedentary people?

Mitesh took it from there and generated new interventions using more rigorous methods... The quick and dirty initial work meant that once you get to the more in-depth, expensive, scientific testing you can avoid wasting time and money on interventions that have no anecdotal support and focus on those where there’s some reason to believe they might be worth (following up with in-depth research.)

Mitesh's work has appeared everywhere from the leading peer-reviewed medical journals to the Today Show. He does great research.