"Epic fail?" Atul Gawande on the EHR

From an excellent New Yorker long read.

The article has an accompanying audio transcript that runs 59.07. Read/listen to all of it.

“Something’s gone terribly wrong. Doctors are among the most technology-avid people in society; computerization has simplified tasks in many industries. Yet somehow we’ve reached a point where people in the medical profession actively, viscerally, volubly hate their computers.”

Take it from the top. Atul Gawande, MD:

On a sunny afternoon in May, 2015, I joined a dozen other surgeons at a downtown Boston office building to begin sixteen hours of mandatory computer training. We sat in three rows, each of us parked behind a desktop computer. In one month, our daily routines would come to depend upon mastery of Epic, the new medical software system on the screens in front of us. The upgrade from our home-built software would cost the hospital system where we worked, Partners HealthCare, a staggering $1.6 billion, but it aimed to keep us technologically up to date.

More than ninety per cent of American hospitals have been computerized during the past decade, and more than half of Americans have their health information in the Epic system. Seventy thousand employees of Partners HealthCare—spread across twelve hospitals and hundreds of clinics in New England—were going to have to adopt the new software. I was in the first wave of implementation, along with eighteen thousand other doctors, nurses, pharmacists, lab techs, administrators, and the like.

The surgeons at the training session ranged in age from thirty to seventy, I estimated—about sixty per cent male, and one hundred per cent irritated at having to be there instead of seeing patients. Our trainer looked younger than any of us, maybe a few years out of college, with an early-Justin Bieber wave cut, a blue button-down shirt, and chinos. Gazing out at his sullen audience, he seemed unperturbed. I learned during the next few sessions that each instructor had developed his or her own way of dealing with the hostile rabble. One was encouraging and parental, another unsmiling and efficient. Justin Bieber took the driver’s-ed approach: You don’t want to be here; I don’t want to be here; let’s just make the best of it.

I did fine with the initial exercises, like looking up patients’ names and emergency contacts. When it came to viewing test results, though, things got complicated. There was a column of thirteen tabs on the left side of my screen, crowded with nearly identical terms: “chart review,” “results review,” “review flowsheet.” We hadn’t even started learning how to enter information, and the fields revealed by each tab came with their own tools and nuances.

But I wasn’t worried. I’d spent my life absorbing changes in computer technology, and I knew that if I pushed through the learning curve I’d eventually be doing some pretty cool things. In 1978, when I was an eighth grader in Ohio, I built my own one-kilobyte computer from a mail-order kit, learned to program in basic, and was soon playing the arcade game Pong on our black-and-white television set. The next year, I got a Commodore 64 from RadioShack and became the first kid in my school to turn in a computer-printed essay (and, shortly thereafter, the first to ask for an extension “because the computer ate my homework”). As my Epic training began, I expected my patience to be rewarded in the same way.

My hospital had, over the years, computerized many records and processes, but the new system would give us one platform for doing almost everything health professionals needed—recording and communicating our medical observations, sending prescriptions to a patient’s pharmacy, ordering tests and scans, viewing results, scheduling surgery, sending insurance bills. With Epic, paper lab-order slips, vital-signs charts, and hospital-ward records would disappear. We’d be greener, faster, better.

But three years later I’ve come to feel that a system that promised to increase my mastery over my work has, instead, increased my work’s mastery over me. I’m not the only one…

Fascinating. While I'm now 5 years out of the daily EHR trenches professionally, my world has been "all Epic all the time" ever since, in my roles as a patient (2015 prostate cancer dx & tx, 2018 SAVR px) and caregiver to my now-late younger daughter (2017-2018 pancreatic cancer dx & tx). All the major players here in the Bay Area -- Kaiser, Sutter, Muir, Stanford, UCSF -- are on Epic. I continue to be an acute observer of the EHR workflows I witness at every encounter, and I frequently query my clinicians about their experiences using Epic.

Nearly all I have seen during our many patient encounters across the past few years has been that of clinicians at all license levels whipping around the Epic EHR at lightning speed. Yes, they also all grouse about what they see as nuisance diversionary billing and compliance documentation, but the clinical care aspects of the EHR are about as efficient as you could hope for. That there are hundreds to thousands of clinical variables to be recorded and tracked is simply a core reality of medical care -- not the fault of the EHR.

Paper is not better.

Dr. Gawande:

“… the computer, by virtue of its brittle nature, seems to require that it come first. Brittleness is the inability of a system to cope with surprises, and, as we apply computers to situations that are ever more interconnected and layered, our systems are confounded by ever more surprises. By contrast, the systems theorist David Woods notes, human beings are designed to handle surprises. We’re resilient; we evolved to handle the shifting variety of a world where events routinely fall outside the boundaries of expectation. As a result, it’s the people inside organizations, not the machines, who must improvise in the face of unanticipated events.”

I am reminded of my prior post "Are structured data the enemy of health care quality?" Also, see my "Update on our favorite whipping boy, the EHR."

I might note that the bulk of the litany of complaints set forth in the Gawande article (and those of many others) are hardly news to those of us who have been involved in the EHR wars. I've been listening to these gripes since I came to the health IT space in 2005 with the onset of the QIO 8SOW "DOQ-IT" program.

Responding to the immediately foregoing Gawande observation inescapably leads me to, among other resources, this glorious book I recently finished.

"AI" to the rescue? Skeptics remain legion (including eminent AI pioneer Judea Pearl).

Big Data and causal inference together play a crucial role in the emerging area of personalized medicine. Here, we seek to make inferences from the past behavior of a set of individuals who are similar in as many characteristics as possible to the individual in question. Causal inference permits us to screen off the irrelevant characteristics and to recruit these individuals from diverse studies, while Big Data allows us to gather enough information about them.

It’s easy to understand why some people would see data mining as the finish rather than the first step. It promises a solution using available technology. It saves us, as well as future machines, the work of having to consider and articulate substantive assumptions about how the world operates. In some fields our knowledge may be in such an embryonic state that we have no clue how to begin drawing a model of the world. But Big Data will not solve this problem. The most important part of the answer must come from such a model, whether sketched by us or hypothesized and fine-tuned by machines…

Pearl, Judea. The Book of Why: The New Science of Cause and Effect (p. 352). Basic Books. Kindle Edition.

Now, while none of that speaks to the chronic, clinically enervating "productivity treadmill" concerns so adroitly addressed by Dr. Gawande, it is nonetheless relevant more broadly.

Regarding Dr. Gawande's "brittleness" characterization of computers, Dr. Pearl:

The goal of strong AI is to produce machines with humanlike intelligence, able to converse with and guide humans. Deep learning has instead given us machines with truly impressive abilities but no intelligence. The difference is profound and lies in the absence of a model of reality.

Just as they did thirty years ago, machine learning programs (including those with deep neural networks) operate almost entirely in an associational mode. They are driven by a stream of observations to which they attempt to fit a function, in much the same way that a statistician tries to fit a line to a collection of points. Deep neural networks have added many more layers to the complexity of the fitted function, but raw data still drives the fitting process. They continue to improve in accuracy as more data are fitted, but they do not benefit from the “super-evolutionary speedup.” If, for example, the programmers of a driverless car want it to react differently to new situations, they have to add those new reactions explicitly. The machine will not figure out for itself that a pedestrian with a bottle of whiskey in hand is likely to respond differently to a honking horn. This lack of flexibility and adaptability [emphasis mine -BG] is inevitable in any system that works at the first level of the Ladder of Causation… [Pearl, pp. 30-31]

"Lack of flexibility and adaptability" -- i.e., "brittleness."

You'd have to study the entire Pearl book to fully appreciate that AI has quite a way to go before it significantly enables digital workflow "adaptability" borne of "causal reasoning" capacity, particularly in the complex data health IT space. Nonetheless, there remains a lot that can be accomplished below the "strong AI" level to get us closer to more digitally "frictionless" clinical workflow usability. 

BTW, apropos, see also my prior posts concerning "The Digital Doctor."

Stay tuned.
_____________

More to come...