AAFP's Innovation Fellow Studies Tech, Digital Scribes

For those dissatisfied with the current state of health IT, here's an opportunity to do something about it. Each year, the AAFP offers a Primary Care Innovation Fellowship for family physicians and advanced students interested in

• enhancing usability of health IT,
  
• improving health IT support for primary care, and
  
• applying industrial and systems engineering to family medicine.

The deadline to apply for the fellowship is May 17.

AAFP News recently discussed the program with current fellow and University of Michigan Medical School student Erkin Otles, M.S.E., whose research dovetails with a project approved by the Academy's Board of Directors last fall to drive innovation in health IT, reduce physician burden and support care delivery by leveraging artificial intelligence (AI) and machine learning.

Otles, who also is pursuing a doctorate in industrial and operations engineering, is studying the use of voice recognition and AI technologies to develop a digital scribe.

Q: Would you tell us more about what you are working on during your fellowship?

A: There is a great deal of interest in the medical and tech communities to explore the use of AI assistants -- like Siri and Alexa -- to build documentation tools for physicians. Everyone has significantly different ideas about what these tools would look like and how the patient and physician would interact with them, but there are a lot of people interested in this space.

My specific interest is understanding whether we can utilize existing natural language processing tools to build a documentation assistant -- a digital scribe, or dScribe. dScribes would be able to listen in on a patient-physician encounter and automatically write the majority of the encounter note, like a scribe. The ultimate goal is to free physicians from the burden of note documentation (or dictation).

Q: How has it been beneficial for you to have the Academy involved in your work?

A: Working with the AAFP has been immensely helpful. Though there's a lot of interest on the part of tech companies and physicians, the academic players have been slow to join. This work is a little different than what traditional researchers are interested in, and there are myriad challenges in attempting this type of cross-disciplinary research.

Having the AAFP say, "We see this as valuable work for us and for all the family physicians we represent," has been fantastic. Potential collaborators and mentors have been incredibly responsive because this project is something the AAFP values. I've been able to meet and work with several different tech companies because they value the AAFP's role in advancing family medicine. Additionally, the technical support for my work has been amazing. I've been able to tap into experts at the AAFP -- from physician mentors to research librarians to experts in survey design.

Q: What is the state of the technology you are studying, and how far away are we from things like a self-documenting EHR?

A: If you asked 100 experts in the field, you could get 100 different answers; fully automatic documentation in the EHR is the holy grail. Some of the grander ideas being discussed -- like the EHR becoming an invisible, fully conversational and omnipotent clinical assistant -- could be a decade or more away. But there are other things that could happen significantly sooner.

What will likely happen during the next couple of years is the increasing use of AI assistants that are good at specific tasks, like capturing conversation information in the note automatically. So, the evolution we will likely see is a series of components, or skills, gradually being added to the clinical information technology space. Some of these may be integrated with your EHR, others might not be.

Companies are working on many of these tasks, and there are even products available right now using those technologies. Are they 100 percent perfect? No. But are they getting better? Yes, because they are fine-tuning their algorithms and collecting more data. The biggest barrier to success with artificial intelligence products, including natural language processing and machine learning, is that you have to have a good data set; it has to be fairly representative of the real world, and you need a ton of it. Some people seem to think AI is magic, but underlying that magic are algorithms that try to find patterns in data, and to do that well takes a lot of data.

Q: What have you learned so far?

A: One of the biggest things I've learned is that there's this massive mismatch between the amount of interest in these technologies and amount of academic medical research done. Most of the technical components are known quantities in that they've been studied by computer scientists and informaticists. However, they've been studied in contexts very different from the clinic, and that's where I think we have a big gap in our understanding.

Before I started the fellowship, I thought I would be spending the majority of my time coding, but that actually has been a small portion of my overall work.

Taking these tools into the clinic is not simply a technical problem. In order to really serve the needs of physicians and patients, we need to figure out how these tools will alleviate their cognitive workload and not add burden or noise. Right now, we have the opportunity to act proactively and shape this technology. If we get the right minds together from medicine, computer science and engineering, I believe we can avoid many of the struggles we've seen with EHR adoption.

Q: Why would you encourage others to apply for this fellowship?

A: There are many reasons, but the biggest thing for me was that it allowed me to take on a project I wanted to do with great resources. Between clerkships and my Ph.D. coursework, having this fellowship has allowed me to set aside protected time every week to explore this fascinating area.