I lived and worked through the transition in medicine from completely paper-based documentation to completely digital-based – using an electronic medical record (EMR). There is no question, the EMR system is much better. Access to information, communication, ordering tests, tracking results, and documenting visits are all much easier with an EMR.
But modern doctors and other health care providers are not exactly working in an EMR utopia (as was often hyped in the early days). A 2017 study, for example, found that primary care doctors work on average 11 hours per day – more than half of which is spent working with the EMR:
“Clerical and administrative tasks including documentation, order entry, billing and coding, and system security accounted for nearly one-half of the total EHR time (157 minutes, 44.2%). Inbox management accounted for another 85 minutes (23.7%).”
A recent commentary points out that time spent in the EMR correlates with physician burnout. They recommend that we do not waste the opportunity of potentially using artificial intelligence (AI) to address this issue. I share this hope, but decades of experience prevents me from being optimistic.
The problem with the EMR in health care eating up precious professional time and contributing to work burnout is a complex problem, but I do not think it is an inevitable consequence of using an EMR. The problem is with execution, which has two main components. One is the software itself, and the other is institutional practices. Let me share some of my personal experiences as an example (which the data suggests is probably typical).
I work at a large institution with an industry standard EMR (Epic). The software is powerful and it works. But the user interface, even being charitable, is horrible. I won’t give a detailed lesson on how to craft an optimal user interface by giving copious examples from the Epic EMR – those resources are out there. Suffice to say the EMR is far less than optimal. It takes far too many clicks, and requires far too much cognitive load to accomplish tasks.
If I had to distill down the problem to one underlying issue it’s this – the system is designed to make the end user work for it, rather than thoughtfully working for the end user. This is partly a workflow issue, but in many cases when a documentation task needs to be done the solution was to just have the user click another button, rather than searching for a way for the system itself to accomplish the task. At one point I was entering in the patient’s diagnosis no less than five times, for example. This has since improved, but only because some of those tasks were shifted to other people in the workflow – not to the system itself.
The other major source of inefficiency is how the EMR is used by the institution. The EMR system just seems to generate every increasing tasks to be completed by the end-user. This is the “in-box” problem. My in-box is constantly (and I do literally mean constantly) filling with tasks to be completed. Many of these are useful, and just part of patient care. But many are not useful or necessary – they are administrative busywork. There are things that I don’t even know what they are, but I have to click on them to make them go away. And I am constantly informed of things that are not my job. I am a specialist, not the primary care provider, so I don’t need to be notified each time a patient gets a flu vaccine at an outside pharmacy. In fact, the PMD should not be either – why can’t that just check a box in the chart that’s available for the doctor if and when they need it?
Each complaint may seem like it’s not that big a deal, but it results in the death of a thousand cuts. All these tiny wastes of time add up. If you are seeing 15-20 patients a day, and have a patient population of hundreds of people, then the cumulative effect eats up hours a day.
This also negatively impacts patient care. Important information gets lost in a blizzard of administrative nonsense. To keep that from happening you have to pay attention to every little thing that hits your in-box, which adds to cognitive load. Also, doctors compensate somewhat (which they have to in order to survive) by generous use of the copy-paste function. But this can lead to inaccurate documentation. So they are often told not to do this – accomplish your task in the most time-consuming and labor intensive method possible, because that is what is required to ensure quality.
Further – time spent in the EMR on the day of the visit is all billable. So much of this wasted time also adds to the cost of healthcare.
Can AI ride in to the rescue? Potentially. The authors linked above argue that we need to focus AI development on specific problems. Just doing things with AI because we can may not work (they give the Segue as an example – a solution looking for a problem that ultimately failed). But if we instead focus the development of AI systems in the medical workspace on existing problems, it can be transformative. AI can make documentation faster, sift through information quickly, and automate many of the tasks that now fall on beleaguered providers.
But here is why, even though I am hopeful, I am not optimistic. I think AI can accomplish these tasks, and more, but there is a good chance it will fail to do so for the exact same reasons that the EMR has become such a burden to providers. One big reason is that building an optimal EMR requires dual expertise – you need to have sufficient knowledge of how computer systems work, how they can work, what they can and cannot do, and how to optimize the user experience. But you also need to know how health care providers function, what they need, and what really matters. And the latter is not one thing – it is hundreds of different providers in different clinical contexts. So the system needs to be versatile and adaptive.
What happens, unfortunately, is that you have programmers who don’t know what doctors need and doctors who don’t know what programmers can do. I had hoped that as the whole EMR industry matured eventually they would figure things out, but by then systems had become entrenched and bloated.
So yes, we have an opportunity to get these two halves of the equation together, and to design AI systems that directly address the most pressing problems created by the last information revolution. We’ll see what actually happens.
Can AI Reduce Doctor Burnout?