AI Helping Doctors Do Better

When it comes to AI, I wasn’t always a believer. As an acolyte of Rita Charon and narrative medicine (1) during my medical training, I would routinely sit with a patient, holding their hand and hearing their entire life story. It seemed unlikely that I would end up working with robots. I always thought AI was about replacing humans, and in the healthcare setting – sometimes cold and inhumane – this seemed like a step in the wrong direction.

But I’ve been converted. For the past few months, I’ve been working with people dedicated to leveraging AI to help humans do better – to help humans be better humans. Our team recently came across the podcast episode How AI can help us be more human which reaffirmed our view on AI in  healthcare – that it can be a tool to help humans be better as clinicians.

At PredictionHealth we’ve seen the results of adding AI into the clinical workflow and how it improves patient care. The tool actually allows clinicians to re-experience why they went into medicine – to connect with their patients. With the PredictionHealth app open on their computer our clinicians can focus on their patient for the entire visit while not worrying about their documentation because AI and US-based remote scribes do it for them.This results in the doctor spending more time focusing on the patient and less time on the computer.



Another perhaps less obvious byproduct is that patients are hearing more about their health and disease processes as clinicians discuss more of what needs to be included in the documentation in front of the patient instead of dictating it later. Other authors have discussed the benefits of verbalizing some of the complexities of the patient visits out loud (2), and we’re seeing this play out in the ways physicians are holding their conversations when they know the AI documentation system is at work. 

There is additional information that could help clinicians improve. For example, how much actual audio time are they (the clinician) taking up vs the patient in each visit? How much time is spent on counseling vs. other parts of the encounter? How does the tone change throughout the visit?

In addition to improving the interaction during the visit, there is also the suggestion of improved documentation. For example, one of our providers saw her notes become more detailed than she would have accomplished on her own:

“Before I opted for PredictionHealth, my notes had been stripped down to the bare bones needed to avoid litigation and ensure proper billing and payment. I wasn’t accounting for the detailed conversations I had with patients or the joint decision-making employed in nearly every visit. PredictionHealth allows me to easily recall what we’ve talked about so we don’t cover the same ground every visit. More detailed notes have also been helping my office staff get pre-certification for procedures and referrals much more easily.”

Another provider recently added, “I always knew I was doing a lot of complex work with patients during the appointments.  Thanks to your team, I don’t have to worry about the changes in E&M coding that went into effect this year because your faithful documentation of the patient encounters already shows the work that justifies my billing code.”

Harnessing conversational data could even influence the language of medicine, getting back to documentation that is comprehensive and intelligible, not just utilitarian. Documentation that is both easier and higher quality is a recipe for alleviating provider burnout associated with this clinical chore.

We do need to be careful, though, because AI still has its limitations. Even while I was writing this post in praise of AI, I came across this article by one of Microsoft’s AI researchers, Kate Crawford, who warns of the dangers of AI which every industry must bear in mind (3). One of the biggest surprises for me, and something Crawford references, is the continued high human cost for businesses to build AI technologies. It is still humans who are responsible for building and maintaining the data models upon which these systems function. So while we are developing the technologies we must simultaneously develop processes to make them equitable and sustainable so that they can make our human work better while not jeopardizing humankind.

Even with the potential pitfalls, I’m still enthusiastic about the worthy goal – AI that helps doctors spend more time with their patients, less time on the computer, produce more comprehensive documentation, provide more face-to-face patient education, and maybe even more benefits we’ve yet to discover.

1. Narrative Medicine: A Model for Empathy, Reflection, Profession, and Trust | JAMA
2. Seven Reasons to Dictate in the Presence of Your Patients | FPM
3. Microsoft’s Kate Crawford: ‘AI is Neither Artificial nor Intelligent.’ | The Guardian