A Lesson In Clinical Decision Support: We Cannot Defeat Human Nature

      Our UCSF Clinical Informatics group met a few months ago with several representatives from a major Health IT vendor. The vendor, we’ll call them RxLabs, is a provider of pharmaceutical related knowledge in many domains, including decision support tools for the EHR. Our conversation centered around how to better customize medication alerts. We talked about the popular topic of “alert fatigue,” and how to improve EHR decision support tools to improve their impact, rather than just being white noise annoying clinicians.
      The vendor was walking us through a slide-deck about their hypotheses and data about EHR medication alerts and we were having a vibrant discussion about how to improve provider adherence with decision support. We saw slide after slide about how to make pop-ups smarter and about trying to get more buy-in from providers with paying attention to alerts. After all, why would a provider trying to take care of her patient ignore an alert that is trying to help provide an important message? It must be sloppiness or laziness on the part of providers!
      Ten minutes in to this conversation about drug alerts, up pops the following:
Windows 7 Display Alert
      I’ll give you a second to guess what happened next.
      Without a moment’s hesitation or thought, the presenter clicked the little X in the upper right corner. Our conversation went on. More slides. More data about medication alerts in the EHR. Ten minutes later, guess what happened?
      Up came the same pop-up Windows alert. The presenter again, hastily, without paying attention, and perhaps giving a small huff of displeasure, clicked the little X in the upper right corner. More slides, ten more minutes, same thing. You get the idea.
      This happened three times, with each passing pop-up, the presenter becoming slightly more annoyed. The fourth time the pop-up appeared, my colleague Russ Cucina, the Associate CMIO at UCSF, paused the presenter to have us all read the pop-up alert message. We took ten seconds together to learn that selecting any of the three choices rather than clicking the “x” would have satisfied the alert and kept it from coming back.
      The room broke out into laughter. We all understood our own hypocrisy. We cannot defeat human nature.
Advertisements

“But who is going to pay for it?!” — New Medicare Billing Codes for 2015 Include Remote Chronic Disease Management

“But who is going to pay for it?!”
This has been the common refrain for years. The world of diabetes care experienced this dilemma relatively early-on, as some of the earliest digital health tools were in the diabetes field. When home glucose monitoring became easier and more ubiquitous, and then continuous, people with diabetes were all of a sudden collecting loads of data at home that might dramatically impact their care… and then waiting 3 months to come in to the office to discuss that data. I am asked this question all the time about the startup company I advise, Tidepool, because Tidepool facilitates better and easier remote diabetes care.
It is not just diabetes. In general, there has been more hype and excitement over digital health than impact in clinical practice. A significant reason is the mismatch between payment models and digital health use cases. We still largely live in a fee-for-service world, where we are paid to provide care during a “face to face” office visit and everything is measured by having a “billable encounter.” Most digital health tools, by bringing platforms, apps, sensors, devices, and analytics onto mobile and onto the consumer at home or at work, facilitate care happening outside of my exam room. This does not generate a “billable encounter” and there is no “face to face” office visit.
I don’t think I’m revealing anything new here by saying that it has been beyond a tough sell getting the healthcare system to implement digital health innovations in a fee-for-service environment. How enticing is it to anybody to do a lot of work for free? Doctors do it, but begrudgingly and in small batches.
As for Tidepool, we’ve known that it would be a tough sell initially, but had faith that payment models would change and that we would be ready when they did. I’ve written before about how I’d like to see my future practice operate once payment models changed. And now they are continuing to do so.
Medicare now looks to be slowly facilitating change to align payment models with exciting new technologies. As many media outlets are reporting (see CNN Money, iHealthBeat, Modern Healthcare, mHealthNews), CMS has announced that it will add new telemedicine billing codes starting January 1, 2015 (the CMS document is here). Doctors will be able to start billing Medicare using the 99490 and 99091 CPT codes for providing non-face-to-face, remote care, for patients with chronic conditions. Medicare has never in the past paid for the provision of these services.
A huge caveat that, in my opinion will continue to stymie progress, is that Medicare will still require patients to be in rural areas for these payments.
But, this remains a step forward toward the holy grail of aligning payment models and incentives with new digital health technologies. Paying doctors to provide remote, non-face-to-face care for patients with chronic diseases is the right thing to do for patients and for the healthcare system. Digital health innovations that would sputter under current payment models may take flight once remote care is reimbursed.

What I Learned At Epic UGM… And Other Random Thoughts

Epic’s User Group Meeting (UGM) is a Healthcare Conference
The Epic EHR is so ingrained in healthcare now that the UGM conference is really a healthcare conference, not an IT conference. This was a conference where more than 10,000 healthcare professionals met to share best practices about how to run a healthcare organization and deliver care, and oh by the way, the tool you’re using is this software called Epic.

There were clearly dominant themes this year among the priorities of the healthcare organizations in attendance:
1— Population health and ACOs
2— Patient-______: patient-engagement, patient-centeredness, patient reported outcomes, patient collected data, patient portal, etc
3— Health information exchange
4— Capture and use of discrete data by physicians
5— E-visits and video visits to improve access (and maybe end the long reign of the office visit)
6— Algorithms and analytics, especially with combining of multiple data sources
7— Personalized medicine using genomic data and home-collected data alongside traditional clinical data

Epic Should Do SaaS
If I were Epic, I would develop a SaaS (Software as a Service) version (call it “EpicLite”) and cannibalize my own business from the bottom up. Epic is making some fantastic improvements to their software, but a major complaint you hear around the lunch tables at UGM is that no organization has the resources to implement all of Epic’s features and functions. Epic has made their software endlessly customizable in an attempt to please customers who asked them for such customization. But the end result is that we all bog ourselves down. I’d like to see Epic push back a bit more against what we all tell them we want, be bolder, and push out software to us all that just works out of the box. They can start with the “EpicLite” version and sell it to organizations less complex than the very large customers they most frequently serve now. Follow the 80/20 rule, pick the things that work best, and give it to people. I promise that we will complain, but then we will deal with it and save a lot of money and effort. They could then slowly move up-market with this SaaS version to sell it to the more complex and large customers in true Clayton Christiansen-esque disruptive innovation to disrupt their own core business. To analogize based on one of Christiansen’s examples, they won’t want to be selling mainframes in ten years when everyone wants PCs.

Open.Epic and Apple HealthKit Integration
I’ve heard a lot of skepticism about this effort over the past year because Epic has always had the reputation of being a very closed system, but Open.Epic should change that perception. I think that this is going to be a big deal. I believe that a major reason for the lack of success of many digital health apps is that they are silos and built in standalone fashion. Let’s face it: the EHR is the hub of clinical workflows and no matter how cool and important your app is, it is still just an add-on. Apps cannot be successful if they don’t fit into clinical workflows. Therefore, to be successful, the workflow of using an app needs to blend in with the use of the EHR. Epic publishing APIs through Open.Epic for people to connect apps in is a game-changer and will enable an entirely new generation of apps that bolt on alongside the EHR.

Similarly, I think the Epic and Apple Healthkit integration will catalyze many of the currently stagnant use cases for sensor and device data, as it will now easily feed into the clinical environment.

Random Thoughts and Impressive Numbers
I found myself wondering what Epic would be like if it were in Silicon Valley instead of Wisconsin. I don’t think it would be very Epic-like. You probably wouldn’t see them announcing next year’s product releases in the form of a musical.

It is hard to tell how much the healthcare system is shaping Epic’s software development plan versus the other way around. I’m sure it is some of both.

Epic is incredibly successful at energizing its customers and getting them to evangelize and espouse the virtues of their product for them. And we all pay to fly out to Wisconsin to do it! When I walked by it, their usability testing lab had a more than half-hour wait to get in and a line down the hallway.

54% of the US and 2.5% of the global population have an EpicCare chart. There were 5,000,000 Epic<—>Epic information exchanges in Aug 2014.

The Most Important Digital Health App of 2013: Now THIS is a Learning Healthcare System

The Most Important Digital Health App of 2013: Bugs and Drugs

In a year that saw consumer-facing digital health app after consumer-facing digital health app, the app that impressed me most was actually clinician-facing, not consumer-facing.  In 2012, the digital health apps that stood out to me most were Kinsa, a smartphone-connected thermometer enabling real-time community maps of infectious disease, and GeckoCap, a wireless sensor-in-a-smartcap for asthma inhalers enabling parents to track their kids’ asthma.  (Of course, for fairness sake, I’m leaving out Tidepool, the open platform for type 1 diabetes for which I’m medical advisor, and about which I am incredibly enthusiastic.)

When seeing new digital health devices and apps, I usually have one of three reactions, either: a) “Nope, next!”; b) “This has potential, I want to hear more about it”; or c) “I need to immediately call everyone I know and tell them about what I just saw”.  This year, reaction C came from the AthenaHealth/ePocrates Bugs and Drugs app.  This app makes me feel optimistic about real progress happening in healthcare.  This app makes me feel like the promise of the Learning Healthcare System is either upon us, or truly just around the corner.

If you’ve not seen this app yet, stop reading this article for a moment (come back to finish it, of course!) and go download it from the App Store.  The Bugs and Drugs app is a real-time, aggregated, cloud antibiogram.

What’s an antibiogram?  

Here is an example of the 2011 UCSF adult antibiogram.  First, a quick explanation for the non-clinician.  To test a patient for urine or bloodstream infections, clinicians order cultures to see if bacteria will grow (literally) out of the respective collection sites from a patient.  If bacteria grows from a culture and the patient is thus deemed infected, tests are done to see which bacteria is the specific cause.  Additional tests are then done to see which antibiotics will be effective at killing this particular bacteria strain.  This is known as sensitivity or susceptibility data.  This information can make the difference between giving a patient an ineffective antibiotic and an effective one.  Without it, we as clinicians are guessing about which bacterial strain we think the patient might have and which antibiotic to use.  We base this on our knowledge about which bacteria are most commonly pathogenic and which antibiotics are designed to kill which bacteria.  We also use available past data about cultured bacteria and antibiotic susceptibilities.  This last piece of data comes from antibiograms.  Many hospitals regularly publish an antibiogram, a handout that aggregates all of the culture and susceptibility data from each culture site (e.g. blood or urine) from the past year.  It shows the relative frequency of the occurrence of each bacterial strain and the frequency of each particular bacteria being sensitive or resistant to each common antibiotic.  For example, in the example UCSF antibiogram linked to above, there were 810 E. coli isolates (the most common bacteria isolated), and 85% of these were susceptible to ceftriaxone, a common antibiotic.  You might find that in another hospital in another region of the country, say North Carolina, that the sensitivity rate of E. coli to ceftriaxone is 35%.  Thus in the first hospital, the treating doctor would be likely to use ceftriaxone to treat the next patient with an E coli urinary tract infection, whereas in Texas, the doctor would certainly want to choose something else, knowing that ceftriaxone is unlikely to be effective.

So, this information can truly be life or death information.  It also contributes greatly to the concept of antibiotic stewardship and appropriate use of antibiotics to maintain their effectiveness for future use.  Traditionally, antibiograms are published regularly with an aggregation of the previous year’s data for each particular hospital.  But, that is static data, a collection of one year at a time.  It is also data bound within the physical or virtual walls of each healthcare organization or medical center.

Bugs and Drugs: An Antibiogram for the Learning Healthcare System

The Bugs and Drugs app has taken this concept and moved it into the cloud era.  The app capitalizes on the fact that AthenaHealth, as a cloud EHR provider, is able to aggregate all of the clinical data from their EHR, in real-time.  They have aggregated together all of the bacterial culture and antibiotic susceptibility data from all of their users and display it in real time in this app.  You are a doctor in Wichita and your patient has a urinary tract infection?  Pull open the Bugs and Drugs app and you can actually see what the most common bacteria are in the Wichita area right now that are causing urinary tract infections.  You can see which antibiotics are effective against those bacteria in the Wichita area right now.  This data is not from last year, it is from the last few weeks.  This data is not just from your hospital’s lab, it is from all of the hospitals’ labs in the area.

The catch of course is that this still lacks true health information exchange.  While the data does cross boundaries between health systems, it does not cross EHR vendor boundaries, coming only from AthenaHealth locations.  So, in the example above, you would not be getting data from every location in Wichita, just those that use AthenaHealth.

However, the really important thing about this app is that it shows on a nuts-and-bolts clinical level what we can do with aggregated real-time clinical data when it is put into a useful format in the hands of a clinician.  This information can influence care right now, for the patient sitting right in front of you.  This is the realization of the possibilities of the Learning Healthcare System, moving valuable information much more efficiently into the hands of the treating physician.  I predict (and hope) that we’ll see many more innovations like this in the coming year.

Feedback Loops and Teachable Moments: The Future Diabetes Care Paradigm

The current paradigm of office visits every three months for PWDs (people with diabetes) is not the right model (nor is it for other similar chronic conditions).  The management of diabetes requires a patient to make dozens of daily self-management decisions.  “How much insulin should I give for this slice of pizza?  Do I need to eat a snack to prevent my blood sugar from going low before I go for a jog?”  Diabetes related questions and issues do not occur on an every-three month basis in synch with this current model for office visits.  They are predictably unpredictable.  Accordingly, to best serve our patients, our system must be flexible and nimble.

In the current model, I see a PWD in my office and let’s say, for example, that we decide together to make a change to his insulin to carbohydrate dosing ratio.  He then leaves my office and we wait three months to reconvene and see if that dosing plan change is working or not.  It’s not that it takes three months to decide.  We could probably know within a week or two if the change is working.  It’s just that healthcare isn’t set up that way.  Our entire world now, in every industry and facet of life, is about data, analytics, and metrics.  Other industries have learned that rapid feedback loops are effective.  Adjusting a PWD’s insulin to carbohydrate dosing ratio should be no different.  By the time he comes back to my office three months later, the opportunity for learning may already have been lost.  Neither one of us has gotten timely and relevant feedback about our decisions.  We may have lost the opportunity for a teachable moment.  Healthcare needs to develop a new model where these feedback loops are much tighter and much faster, actually capitalizing on opportunities for teachable moments.  (Sidebar: One doctor who realized this years ago was Dr. Jordan Shlain, who founded HealthLoop)  Research studies show that PWDs are more successful and confident with managing their diabetes when they feel like they have the backup and support of their clinical providers looking over their shoulders to make sure things are going ok.  If we were to design the system from scratch to accomplish these goals, we probably would not have built it to rest on the concept of office visits every three months.

So, what should be the future model of a Diabetes and Endocrinology clinical practice?  Here’s what I imagine my practice looking like in the (hopefully near) future.  Instead of having 16 office visit slots per day of 30 minutes each, I imagine myself seeing 5 patients a day for 45-60 minutes each, allowing us to take our time working together in person and truly addressing the needs and goals of the patient.  These longer visits are essential for a patient new to my practice, a patient with a complicated or unknown diagnosis, a patient with complications or a major change in their disease state, or for discussing major changes in therapeutic course or strategy.  The rest of my day will be spent using a dashboard to do remote population management, looking for trouble spots among my patient population and focusing in on those, and doing telemedicine, connecting with patients through video-chats to make more minor adjustments and to do brief “check ins.”  Ten minutes spent with a patient at the point where there is a teachable moment like a low blood sugar from walking the dog might be more effective than a standard 30 minute office visit every three months.  We’ll have to test this hypothesis, of course, but we must try it.

This is why I’m brimming with so much enthusiasm and excitement about working with the non-profit, Tidepool, who is building an open data platform and a new generation of software applications for the management of type 1 diabetes.  Tidepool will provide us with the technology infrastructure to reach this vision of more frequent feedback loops and teachable moments.  I’m also very excited about the work that my UCSF colleagues, Drs. Ralph Gonzales and Nat Gleason, are doing to pilot the use of telephone visits and e-visits with patients in place of office visits.  Their work is paving the way toward demonstrating efficacy of e-visits, helping to achieve payer reimbursement so that such a model can take root.

UCSF Lean Launchpad: The right way to redesign healthcare

I recently had the fortunate opportunity to be part of the inaugural UCSF Lean Launchpad course, formed by Erik Lium and Stephanie Marrus at UCSF, founded by Steve Blank, and taught by Steve and our digital health cohort instructor, Abhas Gupta.  This was a very intense and demanding ten week class that was not about reading and memorizing and taking tests, but about going out and talking to people; “getting out of the building,” as Steve famously says.  The fundamental insight that led to the offering of this course was that scientific and clinical innovation in healthcare does not happen in a vacuum.  While everyone knows how important it is to test and validate scientific hypotheses, it turns out that it is just as important to test and validate your business hypotheses.  Moreover, these should happen in parallel.  This business model hypothesis testing cannot be outsourced after your scientific validation is completed.  This business hypothesis testing cannot be done by sitting in your office and bouncing ideas off colleagues.  Just as we demand data to prove scientific hypotheses, we need data to prove business hypotheses.  Otherwise we’re just guessing.

The Business Model Canvas and Lean Launchpad provide the framework for innovators to literally get out of the building and talk to dozens of customers, partners, and others to help validate, or more often, invalidate, their hypotheses.  Without doing this, talented people will often waste literally years of effort pursuing a product that nobody really wants to use and that nobody will pay for.

This is not news to the world of entrepreneurs at large, who have heard these ideas from Steve, Eric Ries, and others for years.  However, I think this is still a novel concept in the life sciences and healthcare.  Without validating product-market fit, revenue strategy, channels, and the other parts of the business model canvas, healthcare innovators are hurting their chances at disseminating their products to reach broad audiences.  To fully realize the efficiencies of translational medicine, healthcare has to buck the belief that science and commercialization happen sequentially rather than in parallel.  One caveat: There’s obviously something still to be said for early basic science, where one can explore basic mechanisms without having the constraints of having to worry about commercialization.  But for anybody who is working on the more translational end of the innovation spectrum (i.e. the entire digital health industry), doing this is mandatory.

It was amazing to see the changes in strategy among the teams in our class as the weeks went by.  Making Friends started out planning to build a game to help socialize children with autism, but realized along the way that parents and special needs schools were much more interested in having a dashboard to communicate and track the childrens’ progress.  Tidepool, for whom I’m a medical advisor, started out thinking that our early customers would be tech-savvy 20-somethings with type 1 diabetes, but quickly learned that the most interested customers would be parents of children with type 1 diabetes (see the video about our process here).  The Lean Launchpad class was filled with similar stories — we all found that most of our initial guesses were flat out wrong once we went out and talked to people.  As Steve always notes, one smart person is not as smart as the collective wisdom of hundreds of customers.

Following these lessons will be crucial to future successful innovations in healthcare and I sincerely hope that this curriculum spreads throughout the healthcare community.  We in healthcare have to have the courage to get out of the building and test our assumptions early instead of blindly plowing forward.  We should apply the same rigor to our business plans and dissemination strategy as we do to our science.  We should shed the attitude that, “if we build it, they will come.”

A hearty thank you goes out to all of those who designed this curriculum and ran this class.

Redesigning HealthCare: More Thoughtful, More Caring

I went in to the dermatologist last week for an annual skin check and, instead of a humiliating, cold, and uncomfortable paper gown, this cotton spa robe was instead waiting for me to change into.  My experience of whether I was working with an empathetic and caring physician was shaped before she even set foot in the exam room.  Small touches like this robe can make a dramatic difference in the patient experience.  This does not mean that “luxury” can or should replace high-level medical care.  However, thoughtful touches like this robe can enhance and augment high quality medical care to make it even better, and we should not ignore these opportunities to make our patients feel more comfortable.

Spa Robe and Sonos at Doctor's Office

 

 

 

%d bloggers like this: