← Becker’s Healthcare Podcast
Loading summary
Transcript31 lines
- [00:00]
A
Hello everyone, this is Jacob Emerson with the Becker's Health it podcast. Thrilled today to be joined by Dr. Nigam Shah, who is the chief data scientist at Stanford Healthcare. Dr. Shah, thank you so much for taking the time to be with me on the podcast today.
- [00:15]
B
Oh, great to be here.
- [00:17]
A
Yeah. And before we dive into everything, we want to talk with you about nagam. Can you tell us a little bit more about yourself, your background in healthcare and what it is that you do today at Stanford?
- [00:27]
B
Sure, sure. So I often joke I've never had a real job. I got my PhD from Penn State, came to Stanford as a postdoc in 2005 and never left. Been on the faculty since 2011, tenured in 2015, and then I was doing the usual, you know, writing papers, real world evidence, machine learning on EHRs and so on, and got sucked into working on things during the pandemic in the health system and found it very rewarding to do things in an operational environment. So in 2022 we set up this new role of chief data scientist for the healthcare system where the job basically is to translate things into practice such that patient care directly benefits.
- [01:13]
A
Fantastic. So, Nigam, we're here to talk with you today about an announcement that Stanford Healthcare made earlier this year saying that the system was going to be launching an ambient AI pilot with Atropos Health. Integrate real world evidence within the electronic health record via ambient AI listening from, from Microsoft Dax. So talk to us a little bit about this. How is this going to be working, this pilot, and how is this going to help your physicians access real world evidence to make better decisions for your patients?
- [01:47]
B
So real world evidence has been part of my work almost since 2011, 2012. And as you probably know, I mean, most of the times real world evidence is an observational study that takes anywhere from six to nine months or maybe one year. It gets published in a journal like JAMA or New England Journal or Lancet or what have you. And maybe a physician reads it or doesn't read it and the chance that it affects patient care is very low. There's some very shocking numbers out there saying where physicians track the number of decisions they make. They go back and ask how often was some evidence available particularly relevant to that case at hand. And often the answer is under 5%, under 3%, things like that. That's what I'd been researching. We did a project here called the Green Button Project that took that nine month time frame down and got it to under two days. So at the bedside a physician can make a Request and and a team produces a written report in two days. Once that was done, we spun out Atropos Health, which took those 48 hours, turned around and got it down to under three to four hours. So at the bedside, you have a question, you ask something, what happened to similar patients? And you get a written report before the end of the day. So then ambient AI comes around and a number of physicians are super excited about how it's now saving them time and freeing up their sort of cognitive bandwidth to think more about the patient. On the other hand, Atropos adoption is going fine. About 1,000 or 1,200 consultations are requested and reports being produced within a day. Now, separately, not at Stanford, Atropos had been developing this solution, which they call Chat rwd, which essentially automates this procedure of producing observational studies on demand. Takes the human mostly out of the loop and can get an answer for a lot of cases in under five minutes. Now, you look at these two things together and you ask, all right, we have a patient visit that's 15 minutes now getting transcribed, and in about a minute or two you get the full transcript. We have this separate ability that Stanford has already purchased where a physician, instead of a day, can get an answer in five minutes. What would happen if we put the two together that led to the current project you're talking about?
- [04:25]
A
Wow, that's some incredible numbers you just cited there in terms of five minutes and getting an answer back. That's incredible. I wonder, have you heard from clinicians across the Stanford enterprise in terms of how this is being received, what their are, is this being received? Enthusiastically.
- [04:46]
B
So lots of enthusiasm to pilot it and be, you know, the first to get access to it 100%. The Atropos service, the one that produces this report, has been live on our campus since almost 2022 actually, or 2021.
- [05:03]
A
Okay.
- [05:05]
B
And that typically is used in the inpatient setting because there you can tolerate a one day or a four hour turnaround. Right. Ambient scribes are used in the outpatient setting. And so this will be the first time that real world evidence will be offered in the outpatient setting on demand, at the point of care. So everybody's super excited about it.
- [05:26]
A
Wow. No, it's incredible. And is it my understanding that this is going to be continued to be rolled out this year? Is that the general timeline for this pilot?
- [05:36]
B
Absolutely, definitely. Before the 2025 ends, you know, most likely the first actual user will get it in the next 60 days or 90 days.
- [05:45]
A
Wow. And this is all your physicians Across Stanford?
- [05:49]
B
No, we'll actually be piloting this with what we call our community practice providers or ump, University medical partners or sometimes called Stanford Medical partners. So these are physicians who are like not at the university, they're in the community, practices all over the Bay Area. And they're the ones who are super excited that finally technology is at a, at a stage, at ease of use that they can incorporate into their care.
- [06:17]
A
Wow, that's amazing. And, and I know we did briefly just touch on this, but if from the patient's perspective, how would, how would this on the ground? What does this look like to me? I come into to the office and I receive those answers within five minutes. Or can you, can you detail that for us a little bit?
- [06:35]
B
So imagine before all of this technology, you go to a physician, you're speaking, they're looking at a computer screen, typing away furiously so as to not to miss anything. They maybe glance at you once or twice and then, you know, once their note is done, they look at you, tell you three things you should be doing and off you go. Right. Ambient scribes come around and now the doctor's actually looking at you and there's this thing in the ether that's transcribing. And by the time the conversation ends in about a minute, a note shows up which they can turn the screen around and show it to the patient directly. In fact, our CMIO, Dr. Topher Sharp has done even video interviews with reporters showing them how powerful this is in terms of the physician being able to pay attention to you. In normal encounter, everything is fine. This automation describing does the job. But then if there's a question, what does a physician do today? They will go to up to date or some other literature summary, or if they had a question that can't be answered by literature, they will trigger this real world evidence report from Atropos. But that's a manual step, both of them. Now you can imagine a situation where we already have the transcript and it's electronic. Can I distill a question out of it? Send it to literature evidence, send it to Atropos for data driven evidence, produce both of them back if something exists and is relevant and put that as two links in the note that is being shown to the physician and then they can choose to say, all right, yeah, this looks good enough, I should click on it and read it and talk to the patient about it.
- [08:21]
A
Understood. No, this is incredible and it really reflects what we're hearing from a lot of other health systems around the country piloting and Introducing this type of technology as well. So in that vein, Nigam, what, what would you say have been some of the key challenges that systems, maybe Stanford alone, but from other systems as well that you're hearing challenges they're facing when trying to integrate ambient AI tech into their existing workflows.
- [08:47]
B
Yeah. So there's sort of two broad buckets of challenges. The first one are sort of the, the seemingly daunting ones are the easy ones. I would call them technical challenges. What exact API are you going to call? What's your latency? How fast can you get the answer? How will you maintain security, privacy, compliance, all of that, excuse the word, junk that goes in with health IT in general in doing things with duct tape and bubble gum and string.
- [09:18]
A
Sure.
- [09:20]
B
Then the second bucket are the challenges we haven't yet thought of, which is how are we going to evaluate these things? Because given a Google query, for example, you always get an answer. Given an up to date search, you get an answer. Given an real world data search, you'll get an answer. And there's some safeguards built in that if the cohort's too small or the phenotypes are not clearly defined, the system will refrain from answering. But when you put all this together, how do you evaluate? How do you know it's adding value? Those are two different kinds of evaluations. Those are the things I think in the next year everybody is going to be thinking about.
- [09:59]
A
Yeah, no, absolutely. It's a great point. And that leads me to my next question for you is how do you think that ambient AI tech within the hospital, how's that going to continue to evolve over the next few years and what's the broader impact that you foresee this having on, on the industry?
- [10:16]
B
So I mean, again, I'm speculating just like I'm guessing, like everybody else.
- [10:20]
A
Sure.
- [10:21]
B
Ambient AI entered the scene as an automation, as in something that a human was doing, typing in a suboptimal manner, or you might have had a human Skype present. And that thing is now automated, which immediate return an ROI is the physician now has the attention to spend on the patient. So that's great, check. But now we're collecting a new kind of data in electronic form that we never had before. The real time conversation of what is happening, many things can come from that. This evidence generation on demand is near and dear to my heart. But you can imagine other things like documentation for billing could be completely automated. This whole idea that a biller and a coder has to now review everything and then send a Query back to the physician saying, hey, doc, do you remember? Was this the case with the patient or not? That whole thing can go away. We might be able to produce summaries that are directed to the patient. So the same note can be pushed through another kind of summarizer, AI, that translates it to whatever language you need and outlines the three things the patient needs to watch out for. So I think it's going to enable a whole new class of care activities that we did not think were even possible.
- [11:57]
A
Absolutely. So it sounds like, I mean, you're predicting not even just an improvement in patient and quality outcomes, but really massive repercussions for the compliance and reimbursement perspective within the health systems as well. So it's absolutely fascinating to think about.
- [12:11]
B
And even patient care. So, like, imagine if, you know, you went to. You went to a physician and they said, you know, watch out for these symptoms once every week, if we've recorded that conversation, we can trigger an agent that checks in with you by text, phone, or email, or calling you, saying, hey, Jacob, are you having this in the past seven days? We're never going to do that manually. Sure.
- [12:40]
A
No. It's really exciting. And certainly, I know here in the Chicago area, we've been seeing health systems personally as well, rolling out this technology. And it's very clearly from the physician perspective already and in the data as well, reducing that administrative burnout that so many of them face. So really exciting things on the horizon at Stanford as well, in this atmosphere. But before we go, Nigam, what else are we missing? You know, you've got the years of a lot of other CIOs, iOS and health IT executives in health systems from around the country. What else do you want to share with them about this new pilot with Atropos or about this technology that we've discussed in general?
- [13:24]
B
So I would like to sort of plant this idea that one of our faculty members here, Eric Brynjolfsson, calls the Turing trap. I'm sure all the listeners that have. Everyone's heard of the Turing Test, where a computer passes off as being a human. The Turing trap is the opposite of that, where we as humans only imagine doing with the computer what we already know how to do. That's why we think about compliance and billing. But we don't follow up with our patients routinely as much as we do. We just don't have the time. But what if technology enables that? The way I would encourage people to think of it is that, you know, AI writ large, or I like to think of it more as a data plus algorithm configuration. It started out doing simple things, you know, giving alerts and nudges and classifications and predictions. And then it matured into giving recommendations, giving observational studies and bedside evidence. Now it's getting further advanced into actively listening and summarizing and translating. But what's next? What other actions can it take? Can it do follow up? Can it check in on your meds? Can it check in on your glucose levels? Can the conversation trigger an agent that checks in on your CGM monitor and its values automatically? So I think what we need to be thinking about is what are the actions that we should be taking but are not yet taking that can be done for cheap and fast with AI.
- [15:05]
A
Fantastic. Great parting words for our listeners. So, Dr. Shah, I want to thank you for taking the time to sit down with us and for sharing about all the impactful work going on under your leadership at Stanford Healthcare. We really appreciate it.
- [15:19]
B
Thank you for having me.
- [15:21]
A
And to our listeners, if you'd like to listen to more podcasts from Becker's Healthcare, you can visit Becker's Hospital review dot com.