“We see 40 million Americans a day using ChatGPT for health and wellness-related questions,” says Kate Rouch from OpenAI

Dr. McDonough:
Welcome to The Convergence on ReachMD, where innovators and physicians explore the technology transforming medicine. I'm Dr. Brian McDonough, and today, we'll be examining how your patients are using artificial intelligence to answer health and wellness questions. We'll look at the upside and the downside.

Joining me to talk about this is Kate Rouch, a breast cancer survivor who turned to AI to learn about her diagnosis and treatment, but she has more than a personal story to tell. She's now the Chief Marketing Officer at OpenAI, where she's uncovering the latest trends in AI use.

Kate, welcome to the program.

Kate Rouch:
Thanks so much for having me, Brian.

Dr. McDonough:
First thing I'd like to ask you is if you could share your personal story. We're all interested in that.

Kate Rouch:
Yeah, absolutely. So, about one month into my job with OpenAI, I was diagnosed with a serious breast cancer out of nowhere. I am the mom of two very young children and was navigating, like so many people around the country and the world, a foreign and very overwhelming health system.

I had incredible doctors and was so lucky to be at UCSF and getting absolutely incredible care. But, as you know, a complex diagnosis means many different specialties, needing to learn quickly about the specific type of cancer that I had, symptom management, and even things like, how do I talk in an age-appropriate way to my very young children about what's going on as I go through treatment? How can I help myself navigate insurance and denials that came up as part of treatment?

So, as a person, I relied on this product to empower myself and help navigate an overwhelming situation, and really to get the most out of the incredible specialists. And that's part of the reason I feel so passionately about being here and talking about the potential for ChatGPT to help people in their everyday health journeys.

Dr. McDonough:
It's very interesting. I hope you're doing well. Are things improving?

Kate Rouch:
Thank you. Yes. Cancer free.

Dr. McDonough:
That's fantastic. How did AI improve your relationship with your care team? Was there any pushback? Did you let them know what you were doing?

Kate Rouch:
UCSF physically is literally on the same street as OpenAI, so there's quite a lot of cross-pollination between the institutions. We actually have an enterprise-level rollout of our health system product, OpenAI for Healthcare, which is a HIPAA-compliant version that a lot of large hospital systems have rolled out. So, it's an institution that really understands the potential and benefit of AI from a clinical level, helping doctors obviously do things like manage paperwork all the way down to frontier science.

And so it was a very open conversation with my doctors as I did things like reviewing the clinical literature to understand my specific cancer better and being a more informed participant in talking to them about various options in my care. So they were all very supportive.

Dr. McDonough:

Kate, you answered a question that I have through your story, but I'll still ask it anyway—why are people turning into AI to answer health questions? I think you couldn't have said it any better, but from your perspective, are there certain top health questions people are asking AI?

Kate Rouch:
Yeah. We see 40 million Americans a day using ChatGPT for health and wellness-related questions. We really see this spike after hours, when hospitals and doctors may not be immediately open and accessible.

And most of the questions that we see people asking fall into three buckets. The first is doctor prep. So, "What are the best questions to ask my doctor," or my specialist in my specific case, to really make the best possible use of that brief time with your doctor or nurse? The second is insurance and paperwork. So, "Why did my insurance deny this claim? What does this letter mean? Which health plan is best for me given my specific situation?" And the third is everyday health and nutrition. So, take eczema as an example. "Does this fabric have any eczema triggers?" and things like that to actually help people manage and empower themselves in their daily lives. And that bucket could be things like, "I've connected Apple Health as part of this. Help me really understand my sleep patterns and what to do to maximize more steps," and things like that.

Dr. McDonough:
For those just tuning in, you're listening to The Convergence on ReachMD. I'm Dr. Brian McDonough, and I'm speaking with Kate Rouch, who's the CMO at OpenAI.

Kate, we have a very large physician audience, and I have a few questions I'm certain they would like to ask—I'll ask on their behalf. And the first is, how can AI companies ensure patient security and privacy when using AI to navigate their personal health information?

Kate Rouch:
Yes, of course. It's such an important question, and privacy is obviously top of mind for us and was top of mind for us when we designed this product. ChatGPT Health builds on the already strong privacy protections in regular ChatGPT. It's a dedicated space that really can be seen as a vault. There's a firewall. It does not go over into the rest of your ChatGPT experience. The data is not used by OpenAI to train or for any other purposes besides providing you answers. It really does stay in a private, dedicated, encrypted place within the app.

Dr. McDonough:
How does ChatGPT answer health-related questions? Essentially, how does it do it? But also, what sources, importantly, does it use?

Kate Rouch:
We use a variety of different sources that are extremely high quality—peer-reviewed literature, et cetera—and then we work with clinicians. So ChatGPT Health was developed with, I think, approximately 300 clinicians around the world to make sure the actual responses that ChatGPT is giving based on the large body of peer-reviewed literature that we use to train the models is actually giving good, sound health advice in the real world.

We have evaluation benchmarks that we use to understand how well our models are responding to a wide variety of health questions, and we're seeing that the most recent models have improved, I think, over eight times from an already very strong baseline. So, this is something that we're really working on very extensively. We also make sure that humans are in the loop, so our models are trained to bring in clinicians and others for health questions.

Dr. McDonough:
It's interesting, Kate. I train young doctors. Obviously, they're all using various forms of artificial intelligence in residency training, and we're always looking at the different options. But it is clear that part of their future is using these tools, using them effectively, and making them part of day-to-day care. And even for the board exams, many of the physicians watching or listening who take those exams are now allowed to use various forms of research—not necessarily AI yet, they don't want you looking things up—but they want to model what physicians and other healthcare providers are doing.

We know that AI has limitations, and as we all know, sometimes, it provides biased or inaccurate information, and that can cause patients unnecessary stress or even mislead them into thinking they're fine when they're not. What guardrails do you have in place to mitigate these risks and help with patient safety?

Kate Rouch:
So, as I mentioned, we have been doing a huge amount of work to fine tune the models to make sure they're responding effectively in a huge range of situations and appropriately pointing people to the right external references in different health situations.

So, the best things we have are things like health benchmarks, like I mentioned before. HealthBench is a specific model eval that we turn to to understand at scale how these models are performing, how accurate they are, et cetera. And so, based on what we're seeing at scale, we obviously do see the models responding in a way that is quite sophisticated and effective.

Dr. McDonough:
I want to tap into your extensive experience. You've worked in a lot of different areas, so you've dealt with public perception, We know there's these risks. We know there's these benefits, and we know it's going to get better over time. We're all counting on that, and we also know patients are still using AI. So for our clinicians out there, how would you suggest that they ensure their patients are using AI as safely and responsibly as possible, and more as a person who communicates as a career, what would you suggest? How should they handle that?

Kate Rouch:
I think the biggest thing here is open dialogue, information, and education. So, we know that a lot of people are turning to these tools, and so bringing that in the front door of conversations with their clinicians is really powerful so that it's clear that you can have a single dialogue. We encourage people to always talk to their clinicians about various questions that they may be using ChatGPT to get more information on.

I think on the clinician side, the same holds to really help people understand the best ways to understand how they may be using ChatGPT and to just have an open dialogue with them about that.

Dr. McDonough:
This gets us into maybe the most exciting part of the interview, at least for me, and that is the future. You're looking at the past where you used the tool, and clearly, it impacted you and your own personal health story. We're looking at what we have in the present, and we know things are going to improve. We're going to look back in 20 years and say, "Wow, this was the early stages." What do you see for the future? What's your vision? And again, we don't always predict the future, but in a sense, where do you see this going?

Kate Rouch:
I think we'll see, just based on what we've seen over the last couple years. ChatGPT was only introduced in 2023, which is pretty mind blowing in terms of the pace of innovation, research, and progress, both in how people are using these tools and then how quickly the technology itself is improving. My personal hope is that we can really, on the frontier science side, start to see improvements in things like clinical trials.

Obviously, I'm personally motivated and have a stake in this, but things like clinical trial enrollment—so many clinical trials struggle to even identify patients and get the right folks in the door. We're seeing some AI companies work with hospital systems on improving that process. I think we're starting to see some scientists begin to use AI in their workflows in terms of drug discovery and in the wet lab. And I really do think that we are going to be in a time where things that simply weren't possible, start to become possible in terms of managing disease and finding new therapies in a way that I think could be quite transformative and I find extremely inspiring.

Dr. McDonough:
Kate, I did want to ask you—looking at your personal story, you mentioned in San Francisco—certain facilities are heavily involved in this. Do you see this going across the board, where hospitals, small community hospitals, and big healthcare systems will all be, more or less, partnering? And what's the future of that look like?

Kate Rouch:
Yes. I think it's a particularly important development for systems that are more resource strapped, like hospitals in rural communities or where there aren't as many clinicians, nurses, et cetera to do things like reduce the administrative overload on clinicians so that they can spend more time with patients and make sure that the patients themselves in those places who may be waiting longer for appointments are coming into those appointments with a higher level of preparedness and able to make better use of those appointments.

We are actually seeing in our data that the use of ChatGPT Health actually indexes more towards so-called “hospital deserts” where there isn't as easy access to care as in a large city like San Francisco or New York City. So, we do think it can play a particularly helpful role for both clinicians and patients in those settings.

Dr. McDonough:
That's exciting to see. One of my concerns, being a family doctor, is the social determinants. It became very clear during COVID that some of the patients who I thought I was doing the best care with weren't necessarily taking their medicine, they couldn't get their medicine, or they didn't have the education. They just came in a couple times a year. They were the people who had preexisting conditions. They were the people who got the sickest because we couldn't get to them. It appears to me that the technology and a lot of the things we're talking about on The Convergence are tools that we can use to reach more people, get them the care they need, and make it easy for them.

Kate Rouch:
That's exactly right. And that truly is what we're seeing with 40 million people using this for health and wellness every day. And a lot of it is in things like lifestyle management—"Hey, I have a heart condition. How can I be eating differently every day?" but in a way that's much more specific and grounded to the individual, so therefore much more accessible for people.

Dr. McDonough:
Kate, thank you for taking the time to join us. It was an exciting and interesting show, and I'm so happy that your health is good.

Kate Rouch:
Thank you so much. I really appreciate it.

Dr. McDonough:
A very interesting conversation with Kate Rouch. One of the things we focused on is something really important to us as clinicians, and it’s that people are going to come into the office armed with information. They're going to be asking questions based on ChatGPT. And, as we can see, ChatGPT is getting better and better, as are many of the sources of artificial intelligence, but they're not necessarily always ready for primetime.

And as a result of that, we have to be prepared with the information to talk with our patients about this. I know from years of working in the media that when people see something either on television, on the radio, or now, through ChatGPT, they tend to believe that expert and really want to go with that expert, whether they’re right or wrong. So, as clinicians, we have that one-on-one relationship, and we're going to have to be prepared to have that conversation and maybe say, "Yeah, this is new and exciting, but it may not be where you want to be." Or, "Yeah, that diagnosis you were given, we see in people, but it's only in two or three percent. The vast majority of people don't have that."

So, we're going to have to, as the future unfolds, calm people down, support them, and give them better information. And the doctor's visit that we have traditionally used is going to dramatically change because conversation will be thrust upon us. I don't think that's a bad thing, because the more we communicate, the better we care for our patients. So, we're all going through this together, and we'll see what happens.

You've been listening to The Convergence on ReachMD, where innovators and physicians explore the technology transforming medicine. To hear about other technological advances shaping the future of medicine, visit The Convergence on reachmd.com where you can Be Part of the Knowledge. Thanks for listening, and we'll see you next time.