Editor's note: Dr. Smith, a family medicine physician, is chief faculty practices officer for UCSF Health. Over the past 3–4 years, the health system has implemented a robust program using medical scribes in the outpatient setting. We spoke with her about her experience implementing this program, including the benefits and some of the potential patient safety ramifications.

Dr. Robert Wachter: What motivated you to think about implementing a scribe program at UCSF?

Dr. Susan Smith: The reasons that we started it were primarily to try to address physician burnout. The Net Promoter Score had been done for the first time in 2014 or 2015, and many of the comments about the work experience related to how much time it takes to document in the electronic health record (EHR). It's the documentation and in-basket work that put people over the edge. They want some strategy to get them out of spending all their time interfacing with the computer.

RW: This is a multifaceted problem. Scribes seem to be one solution, but what other things did you consider or have we begun to implement?

SS: What we did at UCSF was think about where is the most burdensome part of the documentation for the physicians and ask faculty what they thought. In our inpatient practice, they usually have residents and fellows who can help with documentation, but in ambulatory they have to do the documentation on their own. Scribes are probably the most common solution at the moment.

For the overall problem of in-basket work, there are several different approaches. The first is to make sure that the work going directly to the physician really can only be done by the physician. What screening and what staff can you put in place to offload work that doesn't need to be done by a physician? Sometimes on UCSF MyChart (the UCSF patient portal), patients will ask about appointments or some question that does not need a physician response. But if the message routing is set up so that it goes directly to the physician, the physician has to answer that or pass it on, neither of which are good uses of physician time. So our first effort to reduce in-basket work was to triage messages more effectively. We have not yet looked at how scribes might be used for the in-basket work, but I think that might come. For both the documentation portion and the in-basket, the use of artificial intelligence (AI) would obviously be fantastic. Maybe in 5 to 10 years, AI might help us manage both the translation from a clinician–patient conversation to a note, which would be the documentation piece, and even some of the in-basket messaging.

RW: How much do scribes cost, and what are the financial tradeoffs in bringing them in?

SS: Scribes in the United States are typically paid a fairly low wage, usually $13 to $18 an hour. The scribe companies that hire them also charge us overhead. So it costs more like $27 an hour for US-based scribes. The University of California only allows us to use US-based scribes. So those are our limitations. For organizations that can do so, a much less expensive option is overseas scribes. To figure out how to help offset those costs, we thought about what's the desirable level of RVU [relative value units] productivity for faculty. We didn't want to push people to be so productive that speed would negatively impact the quality of care we provide. So if you're producing at or above medium productivity for your specialty, we decided to provide scribe support as a cost to the health system. However, if your productivity is below median, we took the cost of the scribes and running the program, figured out what it costs per faculty member, and decided that if you could add 1.5 RVUs per half-day session, that would be enough to help support the program.

RW: For people who don't think in RVUs, how does that work out?

SS: It's typically about one extra patient per half-day session, depending on your specialty. If you're doing largely long consultative visits, it can be a shorter follow-up visit. If you're seeing 10 patients or 12 patients a half day, you might need to add 2 more short visits.

RW: What has it felt like for the docs to have a scribe in the office? What's the learning curve for the scribes, and what's the ergonomics like? How does it all work from the physician standpoint?

SS: Probably the first thing to say is that the physicians overwhelmingly like having a scribe. Once we started having a scribe for physicians, the idea that we were going to take away the scribe was not tolerable. That was not going to happen.

RW: So if you build a program like this, you have to assume that you will never be able to take it away?

SS: That is exactly right. As for the actual mechanics of it, we have a manager for the program who is very active in working both with the scribe companies and with the physicians. When we start a physician with a new scribe, he makes all the introductions. He describes the policies that we have for scribes and documents that both the scribe and the physician using the scribe have been made aware of these policies and expectations. We usually have an experienced "training scribe" working with the new scribe initially. That training scribe is there for the first couple of weeks, sometimes a little bit longer, as the new person comes up to speed. For the new scribe to be fully up to speed takes another month or two, depending on the scribe and the physician, how well they interact together and can learn each other's styles. One of the problems with US-based scribes is most of them are premed students or students interested in health care professions, so there tends to be a fairly high turnover rate. So you have to start that process over again. I don't know the exact turnover rate, but it's enough that it's somewhat troublesome and has led to the use of remote scribes for some of our workforce. The benefit of the premed students is that they are eager to learn.

RW: In general, are the scribes assigned to an individual doctor and stay with that doctor as long as they're working here, or are they floating between different docs?

SS: No, you have to train with a specific physician and they stick with that physician.

RW: It's interesting, you could imagine a world where today you have Scribe A and tomorrow is Scribe B. Why is it so important that they be linked to each other?

SS: It's that translation from a conversation to a note, which is quite different than taking dictation. You have to be able to figure out what key elements need to go into the note. How does this particular physician think about their notes, what they want to highlight, and how they want it formatted? How they put together the problems at the end and their assessment. If we all used a similar template, even if we can do that within a particular practice, then there can be more flexibility in who is scribing. But, particularly at academic medical centers, there tend to be very specialized physicians. They don't have a common template, even within a practice. That makes it more difficult for a scribe who's naïve to the language of medicine to be able to easily translate that conversation into a note.

RW: When you offered scribes to the docs who were eligible for them, did anybody turn it down?

SS: We have about an 80% acceptance rate. Sometimes people don't want to add the extra patient and they're afraid that they won't be able to do that. Sometimes they say they're not comfortable having a scribe in the room with them. And some say that they just don't feel like they need it. They have spent a lot of time working with the EHR to be able to use their documentation shortcuts ("dot phrases" in Epic); their notes are templated, and they're pretty quick at it and feel like it wouldn't help them a lot.

RW: In terms of the accuracy of the notes, are the doctors always reading the note and signing off on it before it becomes an official part of the medical record?

SS: Yes, that's correct. It's part of the policy and what the physicians sign and agree to. They have to attest to the note itself. The scribe can enter and pend an order, but the physician needs to sign off on it. Same with the after-visit summaries that we hand the patient, the physicians have to review those.

RW: What is your impression of the accuracy of the scribes? When the docs go in and look at the notes, are they generally feeling like I can just sign this thing? Or do they feel like they have to do a lot of work?

SS: I think they do very little work, although I cannot say that I have the data to back that up. The company that we work with on the majority of our scribes does their own QA [quality assurance], and our manager regularly reviews their results. I think the physicians would be complaining if they had to do a lot of edits. Occasionally, we get a scribe that doesn't work out, and we get a lot of complaints. But it's not a common problem.

RW: It sounds popular among the physicians. Any idea of how the patients accept this?

SS: What we're looking for primarily is that patients don't feel negatively about the scribes. We studied it in 2016, back when we were starting the program. The differences between the patients with scribes and the overall patient group is not significant. If you look at the literature, most of it is fairly neutral. A few studies show that patients feel positively about scribes—usually because the physician is able to have much more eye contact during the visit and doesn't have to pay attention to the computer. Our physicians feel that and cite that as one of the reasons that they like having a scribe.

RW: It's interesting. I was recently talking to a physician who has a scribe, and he described an enormous part of the benefit was that he has a colleague in the room, and after the patient leaves they can reflect on the visit. We were talking about what happens if and when AI takes this over, and it is a digital scribe rather than a person. He said that would be a real loss. Medicine has become more lonely and more isolating, so, assuming the scribe is a person you bond with, that has been a helpful unanticipated side effect.

SS: I've heard that as well, and obviously that would go away with AI. I don't know how that will play out, though most likely there will be a mix of scribe models, at least for some period of time. Although the cost of scribes may push us towards AI, there maybe be ways to figure out how to afford to maintain that relationship. Or maybe we get more freed up time using AI so that we can recreate interactions with other physicians. We used to have more of those interactions with physicians in a practice. For example, we would debrief at the end of the day and go over the difficult patients, which we cannot do so much anymore because everybody's so busy charting.

RW: Yeah, that certainly is the hope. Sort of marbled through the AI arguments are that it will free people up to be more empathic and more collegial and collaborative, and the question is whether that's true or it just speeds up the treadmill in a different way.

SS: Exactly.

RW: Have any other quality or safety concerns arisen in running with the program over a few years?

SS: There's just the occasional scribe that doesn't work out. There are some physicians who are easier to work with than others for the scribe, and vice versa. So the interpersonal interaction can be an issue, but that's not so much a safety issue. It can be a quality issue if they cannot work out a good working relationship. But we have not had any quality or safety issues surface.

RW: How much of a change in practice style is necessary to successfully work with a scribe? Do physicians need to be talking through their physical exam when they do it?

SS: That's a great question. For sure, the physical examination has to be changed. Because the best practice is that while you're doing the physical, you're speaking out loud and describing the normals and abnormals. This can be a significant change, but many of our physicians work with trainees on a regular basis, so they are already used to verbalizing many parts of the encounter—including the physical exam. Because the patient can hear, the physician may need to give an explanation about what the findings mean, which the physician might not have otherwise have done. Maybe they have to say this is an abnormality, but they'll explain it at the conclusion of the exam. "Right now we're just going to do the exam." So you have to do a little more talking and explaining to both the patient and the scribe. Again, most of that change is going to happen in the beginning month or two as you start using a scribe. So you can tell what it is that they can gather from the conversation, how you need to reword or reorganize the way you ask questions such that the scribe understands where to put the information, and how to organize it in a way that is better than just a transcription of the conversation. A lot of that also depends on the style of the physician. Some people are very directive and templated in their thinking and how they go through a series of questions. And other people are not linear thinkers; they may jump around a little bit more. That's going to make it more difficult for the scribe and a little bit more difficult to put that note together.

RW: You manage both primary care docs and some hyperspecialized physicians. Any sense of if this works better with one group or another, or it can work well with either and it depends more on physician practice style and personality?

SS: It's working well across the board. The range of specialties that we cover is pretty much everything. So it's very important that we have a system of distribution of the scribes that feels fair and transparent to all of the physicians. We based our criteria on the number of visits that a provider has per week, started at the top of the list, and now are going down the list. So we're widely distributed across a variety of departments. From the perspective of having it work, it seems to work pretty much everywhere. From the perspective of artificial intelligence, which we are just beginning to work with, that is different. The more templated the notes are, the easier it is to use artificial intelligence. Templates are easier to use in some specialties.

RW: What's an example of one where it would work well and one that it would not tend to work quite so easily?

SS: Orthopedics tends to be the example practice that everybody working in AI is trying first. That exam seems to be specific and detailed, pretty much templated. Where it doesn't work well are the specialties that have more in that note, more details and both the history and the decision making. Almost any of the oncology practices, the transplant practices, complex medical specialties, it's just going to take longer because the notes will be tend to be more complicated.

RW: For organizations that are thinking about implementing a program, describe how you triaged a limited number of scribes—both because of the economics and just the raw availability of the people. Do you have any recommendations to a program if they were trying to come up with that prioritization algorithm?

SS: That's a great question. Each organization may be quite different about this and how they're implementing it, whether it's by who can afford it or what's fair and transparent or who needs it the most. We've thought about a variety of different strategies. We thought about who could afford it, and that didn't seem like the fair way to go because some practices and departments that have large documentation needs would not have been able to afford it. We thought about whether we could measure the complexity of dictation and documentation to allocate scribes. The concern with that method was that it would incentivize longer and more complex notes, which was not the point. The point was to decrease the burden, not to increase the length of notes. And it's so easy to copy and paste and bring forward long notes that it didn't seem like a good metric. So we ended up offering scribes to the physicians with the highest numbers of visits per session and sessions per week.

RW: When you were in clinical practice, could you have envisioned scribes emerging as a thing?

SS: I don't think I would have foreseen it 10 years ago. But the documentation burden is so different than it was before the EHR became ubiquitous. That change has been the huge driver for the use of the scribes. Initially, we hoped that templates in EHRs would help with the documentation burden, but over time, even seasoned users are still spending too much time charting. I think the older physicians have certainly felt that. Younger physicians who have grown up typing everything since they were in grade school are probably more agile at the EHR and they're better at it, but it still takes up a tremendous amount of time. The value added is just not there. It's not what you train to do as you go to medical school. So the need for the solution has become increasingly evident.

RW: Do you have a sense of how much we as a health system are spending on this each year and what the tradeoffs are in terms of productivity if we are getting "something back" from people being more productive?

SS: The total budget for it is probably in the $2 to $3 million range.

RW: And UCSF Health is a $4 billion health system or thereabouts.

SS: Yes, so it's a relatively small investment. You know $2 or $3 million out of the faculty practice budget is a small percentage, but finding those several million dollars is not easy. At this point, we probably ask about two-thirds of our physicians to add the extra RVUs, and we're getting two-thirds of that. So we have a little bit of an offset. I haven't calculated the dollar amount that we make back. But it's probably under $1 million that we make back by the added RVUs. So it's an investment and it's a tradeoff.

RW: You mentioned that a large part of the motivation at UCSF Health was the finding on physician satisfaction surveys like the Net Promoter Score that there are a lot of unhappy, grumpy docs. Why does an organization like this care about that (other than we're nice people and we want people to be happy)? What's the business case to say it's worth $2 million of institutional money because of these results?

SS: We haven't spent quite enough time quantifying the financial reasons for doing this. So this is just speculative, not based on data. I think that physicians who are happier at their work are going to be more productive in their work and likely to see that extra patient, whether we're asking for it or not. When we survey physicians before and after they have a scribe, we have found an improvement in the response to questions like "When I leave work, I usually feel proud and productive," or "I feel burdened by the amount of documentation required for each patient." One of my goals at UCSF is to make clinic fun. I don't think we're there yet. But a scribe is just really helpful. It changes the dynamic of being in a room all day with patients, one after another after another, followed by hours of charting; it's a very hard job. My underlying principle would be that if it's more fun, people are going to enjoy themselves more and do a better job. And they will do a better job for our patients, which is what we ultimately care about.