Up next in 10
AI in Healthcare: Balancing Technology and Human Touch
2K views
May 16, 2025
Experts from Northwell Health, Philips, and Mount Sinai discuss the potential of AI to reduce administrative burdens, improve patient care, and build trust among healthcare professionals and patients, while emphasizing the importance of human-centered technological innovation. Speakers: Dr. Jill Kalman, Chief Medical Officer, Northwell Shez Partovi, Chief Innovation Officer and Chief Business Leader of Healthcare Informatics, Philips Dr. David L. Reich, MD, Chief Clinical Officer, Mount Sinai Health System; President, The Mount Sinai Hospital, NYC
View Video Transcript
0:00
I'm Diane Brady. Thank you so much for joining us, first of all. I know many people want a piece of your time, and I think it's wonderful that you've taken it with us. We're going to have a great discussion tonight
0:11
I'm Executive Editorial Director here at Fortune, and for a few years now we've been having gatherings like this
0:19
I have to say it's the greatest joy of my job, especially now to have this opportunity to talk about the issues you're facing with my colleague, Jason Del Rey here, who's going to do the heavy lifting
0:31
We gather together. No, you don't have to get up. You can stand up. You can join me
0:35
It's a tag team exercise. Hi, Jason. How are you doing? So our conversation tonight is how the AI revolution is saving time, strengthening care
0:45
I know that life is about hard choices and some of you we've talked already tonight about what's
0:50
happening in the policy environment what's going to go on up here is on the record the conversations
0:56
we have the table or shatom house that means that you know you can't unhear anything but if it's
1:02
interesting we might come back and say would you like to talk about this but the goal really is
1:06
to share to feel like you're in a safe place we really want a chance to get smarter
1:12
So this is your host, Jason, and I'm just going to hand it straight over to you because you're going to introduce the speakers
1:21
I am. And then we're going to have the table conversation. Thank you for joining us here in very good hands and look forward to a great evening
1:28
Thanks again. Thank you, Diane. Oh, you're going to laugh, Mike. We can clap for Diane
1:37
Well, I'm going to get right into it. It's my pleasure to introduce our panel of experts who will be discussing as many topics as we can in the next 20 minutes
1:47
I'm going to start with Dr. Jill Kamen, Executive Vice President, Chief Medical Officer, and Deputy Physician-in-Chief at Northwell Health and lifelong Knicks fan like myself
1:59
We promise only to check the scores once we sit down back at the table
2:04
You guys are all allowed to. Shez Partobi, Chief Innovation Officer and Chief Business Leader of Healthcare Informatics at Phillips
2:13
Our co-host, hey Shez. Hey. And Dr. David Rich, Chief Clinical Officer of Mount Sinai Health System and President of the Mount Sinai Hospital
2:25
David? Good. You said we're going to get extra controversial tonight. Do I have that right
2:29
Well, but we're following your lead. Okay. All right. Well, that's part of the goal
2:35
Welcome Jill, Shez, and David. Let's dive right in. Shez, I actually wanted to start with you. I know
2:40
Phillips recently completed an annual survey that touches on, among other things, trust among
2:49
stakeholders when it comes to AI in this space. Can you sort of start us off and give us sort of
2:55
the landscape? Very briefly, I'll do that. Future Health Index is the 10th year, but it's 1,900
3:02
healthcare professionals that we surveyed and 16,000 patients in 16 countries. So a couple of
3:08
interesting things. First, voice of patient, voice of clinician. What did we hear? Voice of patient
3:14
What we heard from the patients is that they are concerned that technology is going to actually
3:19
disintermediate and reduce the amount of time that they have with clinicians. So they are
3:24
interested in technology, but only if it helps increase face time between a physician and a
3:30
patient or a nurse in a patient. Voice of the clinicians, what we heard from the clinicians
3:35
is that they are overwhelmed because of the fact that they have to do so much administrative work
3:39
In fact, one of the data points was that for every hour of clinical work
3:44
80% of the respondents said for every hour of clinical work, they're spending an hour or more of administrative work
3:49
And they are really eager to see AI reduce that administrative burden
3:54
And then on the trust side, it's the last data point, which is really interesting
3:57
physicians, healthcare professionals, about 80% of them were optimistic about AI, like you are
4:05
80% of you were in a conversation earlier. But patients had only 60% optimism
4:12
And there's this 20% gap in how healthcare professionals are looking at the value of AI versus patient
4:20
Now, what was the last and most interesting part to me, and the surprise to me was
4:23
contrary to what you hear actually in mainstream media is the fact that patients said that their
4:31
trust goes up if the AI is actually combined with input from physicians and nurses. So you're all
4:38
saying not a surprise we have room full of clinicians here but that may not be the general
4:42
sentiment that you know. So empowering. Exactly. So this idea that AI is great, I want technology
4:48
I don't want to dissent you maybe from their clinician and I will trust it more if it's
4:52
informed by the professional. So I'll stop there. There's so much more we can talk about, but those are so the highlights
4:59
Can we jump in? I'm going to add a couple of elements of that on the patient side
5:03
I think when patients say, I don't want me distant from my physician, I think that's
5:09
also generational. So you're going to have some patients who are going to be like, I don't really care
5:13
to look at anybody as long as someone is caring for me in the right way, and I think there
5:17
is generational. And you can go through those different stages. I think the trust of the clinician
5:22
I think clinicians are innovative I would say mostly excited so 80 is probably spot on And then there is the doubt of is it right can I trust it in terms of the diagnostic validity accuracy
5:36
and the accuracy of that and doctors are always going to doubt that. I think there's also one other element of trust is the health system
5:43
You have a health system that has all of these huge reams of data. Who owns it
5:50
Who monetizes it? Who wants it? and that's part of obviously AI
5:56
And the last part of the trust is also on the patient side of data
5:59
Who owns a patient's data? And I think there's fear on that
6:04
Well, I'll add that one of the only smart things I ever did related to AI
6:08
was to recruit someone who started as a nurse who is now our chief of digital innovation
6:14
And I show a slide in my AI stump speech where there are nurses protesting in front of headquarters for Kaiser in California
6:25
And it was actually funny because I was at the Consortium for Healthcare AI while the Kaiser guy was explaining how wonderful they were
6:30
while the nurses were outside protesting. We're not algorithms, don't replace us
6:36
But the reason we haven't had that yet, I'm sure there will be some controversy at some point
6:43
is that when you obsessively build AI into workflows in ways that make people's jobs better, then you develop that trust
6:52
You know, I'll just give an example. When we developed an algorithm that predicted severe malnutrition in the hospital
6:59
it's a very underdiagnosed condition, and there's a use case for economically supporting it
7:06
because we both treat the malnutrition, and we also get reimbursed more for the higher complexity of the care
7:12
And the dieticians at first were a little skeptical, but they were involved in the process
7:18
and they are now three times more likely to diagnose and treat severe malnutrition than they were
7:25
before we implemented the algorithm. So it actually made their job better and more satisfying
7:30
And this is predictive ytics. This is a predictive world. And the other thing is, when we get to ambient AI and documentation
7:39
the most recent study I saw was that nurses spend 51% of their time documenting
7:43
And I don't think anybody went to nursing school to study documentation. So the ability to interact directly with people as Ambient AI is listening to the conversation, doing the initial documentation, which of course is overseen by the licensed professional
8:01
I think that if you find use cases that make people's lives better, and we're talking about professionals, that creates that trust
8:10
And I think the trust of the healthcare professionals will then extend to the patients
8:16
I think that's exactly, if I want to come back to the time point, but Jill, you said something that reminded me that actually in the actual fish eye survey, the impression you had was exactly echoed, which is 45 was the over-under of the age
8:30
Under 45, there was 66% trust and optimism, and over 45 was only 33%
8:36
Yeah, I think it's remarkable. So you have this bimodal distribution of the age population with respect to optimism
8:42
You know, on the side of time, I'm curious. Actually, we were talking about this earlier on automation versus sort of augmentation
8:50
Because what you're talking about is this, look, in AI, you have the potential to automate things completely
8:55
And I think we're seeing a lot of demand from health systems, clinicians to say there are things I should not be doing
9:02
Just automate those away. Are you talking on the operational side versus clinical
9:07
Well, this is the discussion we were having before, which is from a perspective of what I'm talking about here is, yes
9:13
operational things that are menial and below the certification level of the individual should be done automatically by AI
9:21
And then in the case of dispensing care, I think augmentation is currently where, and also you hear from patients, that's where they want it to be
9:29
And I think there's a lot of potential in augmenting the ability of physicians and nurses to actually deliver better care for more people
9:34
But I think this idea of what do you automate completely and what do you augment is we're still trying to find that needle
9:41
But operational stuff is pretty straightforward. If you can automate it, make it go away
9:44
I would agree. I think that the operational use cases for AI
9:49
is very powerful to reduce that burden. What right now is going on inside your organization
9:55
that is able to maybe build trust on the operational side because it's so useful
10:02
Stuff for jail. Yeah, yeah. So one that's really interesting is prior authorization, which everybody hates
10:11
But it's funny, you know, you could be speaking about something like this across the world
10:17
and they're talking about innovation, and they're still talking about how to get patients into a skilled nursing facility
10:21
It's really quite remarkable. So we have actually, we have a partnership with a venture studio called Aegis
10:26
that we spin off companies specifically about operational issues and utilizing AI
10:32
It's actually a unique relationship. The first one that was around transitions of care
10:36
So what we do have automated is all the calls that have to go to the insurance companies
10:42
to the skilled nurses, back again. They're usually done by nurses or other skilled people on the floors
10:48
and now is reducing that burden. And I love cases like this because it takes incredibly mundane tasks that no one has to do and that where I think the automation is And there no risk to the patient I mean that also So I think the operational value is incredible
11:05
Well, I'll give to you. It's all based upon the use cases, finding the right use cases
11:10
The first one is just an obvious clinical one. We have a vendor that automatically reads the CT scans when someone comes in with crushing
11:20
chest pain, has a CT angio. and when it diagnoses an aortic dissection
11:27
which for those of you who may not be familiar with the condition, it's where the aorta starts to fall apart
11:31
has a mortality rate of 1% per hour. So you've got to get to the aorta very quickly
11:36
Our cardiothoracic surgeons are getting a notification before the radiologist has even seen the report
11:42
So in that case, you build trust because the surgeon feels that
11:48
I think another use case that we're just starting, that we just implemented is using an agentic AM model
11:54
that replicates the nurse calling a patient before a cardiac catheterization. That is one of the most boring things you could do
12:02
Are you on Coumadin? Not on Coumadin. No, but the point is it's boring for the nurses
12:08
and it's very, very robust. So how does it work exactly? Well, I won't talk about the persona
12:15
that was chosen by the cardiologist, the gymnast very well. But the implementation..
12:20
Now I want to know. That'll be over dinner when the mic is off and the chaplain rules apply
12:27
But the point is that taking repetitive tasks are very important. That being said, I asked the question when they were presenting it, so where's the return on investment here
12:38
I had those particular nurses, and they were like, oh, we'll just use those nurses for something else
12:42
I have to say, as a hospital leader who suffers with budgetary discussions on a very frequent basis
12:49
that until there's an ROI where you actually can either deploy people differently
12:56
increase productivity in terms of maybe doing more cases with the same personnel
13:00
unless there's an economic or very strong clinical ROI like this reduction in mortality
13:08
because you do more rapid diagnosis, it's very challenging for us to implement something
13:12
unless I can both demonstrate strong ROI, which isn't always financial, but mostly
13:18
and scalability. I think that my question is... May I? Yeah, please
13:23
I want to... And in fact, in the ideal scenario when you implement an AI tool
13:28
it actually has multidimensional benefits. So, for example, if you're doing cardiac echo
13:33
cardiac echo might be 20 minutes of scanning the heart with an ultrasound. And then afterwards
13:38
typically you have to lay out the sonographer, lays out the heart movie
13:42
picks the ideal images, clicks in all the wall outlines, and then does something
13:47
to do the wall. Does the calculation. Does the calculation. Okay, so 20 minutes of scanning
13:51
about 20 minutes of post-processing, give or take. So 40 minutes on this
13:55
And so in one of the Philips scanners, the AI that we've embedded in it
14:00
once the cardiac scanning is done, it finds the right best images
14:04
does the automatic outlining, does the wall stress motion calculation, does everything and puts it there
14:09
So that's 20 minutes less to just one click. So now ROI. Patient 40 minutes to 20 minutes
14:15
Sonographer can do either two cases or your day is half as long savings
14:19
or you do twice as many cases as revenue. Of course. So this, for sure, tying an AI tool to these impacts are critical
14:30
It's beyond, and I'm glad you mentioned this in the financial streams, because beyond building trust
14:35
that you believe the measurement, you'd actually have to have a real benefit
14:40
that productivity is improved or patient experience is improved because there's value in that as well
14:44
And the ideal AI solution hits on many vectors like this particular one
14:49
Improved revenue, improved experience, less time, and measurement. And a person that's a sonographer is working at the highest level of certification
14:56
because they're not doing Photoshop, right? You mentioned Venture Studio and that work
15:02
I'm curious about the question. And, I mean, big tech companies have these questions
15:08
Healthcare networks have these questions of buy versus build, partner versus, you know, owning all, keeping all our data in house. Obviously, there's the
15:18
regulatory aspect, which maybe we'll talk to you at the table. I'm curious how, sort of
15:24
yeah, how you think about that? I think we, buy versus build and partnerships, I think is a great
15:29
way of categorizing it. I think we probably have examples in all of those categories. I think that
15:35
when you can build it, the question is, what's the pathway? And how are you going to actually
15:40
speed to development and proof of concept and then utilization. So we have, I think David has one also
15:46
an early decompensation model using AI machine learning. We have a group that
15:52
is building it within and we're working with a small tech company that's also built it but they're
15:56
further ahead. So the question is, and then we see how can we work together
16:00
And they're aware of that dynamic. They're completely aware. But I also
16:04
think the partnerships with companies are evolving differently than now that we've ever been before. Typically
16:11
in the past, I believe, and chime in on the vendor side, that vendors came, this is our product
16:17
and this is what we're going to do. It's much more partnerships now. What do you need
16:21
How do we develop it together? Let's develop it together, pile it together
16:25
scale it together so that it really fits into your healthcare delivery as opposed to saying we see it this way we see it this way And I think that partnership is everywhere that we starting at each of these
16:37
So we have categories in all, and I think it's which pushes ahead
16:41
and can be effective in those categories. My comment about buy versus build is I think it's very dependent upon, again, use case
16:48
So when it comes to imaging, we've certainly seen with the companies
16:53
with whom we've collaborated, that an image in the Middle East looks the same as an image in South America
16:59
as it does in New York for the most part. But when we worked with a very large electronic health record
17:05
company based in the Midwest, and they did a sepsis prediction algorithm, and it didn't work at all in New York
17:12
I made a joke to their head of AI. Well, I said, if you develop a model on corn farmers
17:16
in Nebraska, it doesn't work in East Harlem. And he said, how did you know I used the Midwestern data set
17:21
And so the answer to the joke is that I would turn into a very own political joke
17:27
Instead of all politics is local, I would say in certain cases, all predictive ytics is local
17:33
Yeah. I think there's actually something to the population. Obviously, the populations that your data is coming from can look very different
17:38
Yeah. But once again, with imaging, I think it's much easier to have a model which is more universally deployable
17:46
but we have the problem of making it work for pretty pretty ytics in our local environment
17:52
so that's i think why we're going to be you know stuck to a certain extent uh maybe vendors that
17:58
will be able to process your data locally uh but unless there's local validation you know for
18:04
example even with that non-nutrition example i gave we obsessively looked at each of the six
18:08
hospitals where we implemented it to make sure it actually worked the same in all those hospitals
18:13
Well, I think your comment on partnerships and actually dovetailing to your comment on imaging
18:19
in fact, given where the industry is today in terms of innovation and with large language models and vision models
18:24
it actually is more catalytic to being partners because the data that's needed to build is exactly
18:30
So I have thousands of data scientists and engineers, but the data sits in your data center
18:37
And this is where the partnership really comes to place. And also, working back from the bedside rather than from the technology forward is the way, at least Philips, our model is invest in the physician's problems, not in your own ideas
18:51
Invest in the nurse's problems, not in your own ideas. And so by working backwards from the bedside, then you can innovate on their behalf and go forward
18:58
So actually, partnerships for sure is the way to go. And I don't think it's always been that way. No
19:03
Why the change? Well, for Philips, it's been that way, but I'll let you go
19:09
first of all I think she's right it is our data and it is our
19:13
patients and I think there's an appreciation of that and I think also it's just an appreciation, they're a large health system
19:19
there's healthcare delivery, how do we fit into yours and there are a lot of
19:23
companies around so you're not just the one game and now that there are many, people want to work together
19:27
I think that's what I've seen and that's been the most effective. Yeah, if I could
19:31
go a little futuristic for just half a second Yeah, that's where I was going to actually end us
19:35
before we sit down, go futuristic Yeah, I think Alex Charney is here in the room somewhere
19:40
He's head of the Mount Sinai Million Discoveries Program. So after a lot of discussion with our IRB, we're up to 300,000 into a million exome ysis
19:50
Regeneron has been a good partner in that. And so we'll have the genomic data
19:54
We have the electronic health record data. We have imaging data. We have waveform data
19:58
You don't. And so sometimes we do need the imaging companies to give us very important advances forward
20:08
but having that total picture of the patient is something that is more regional
20:14
and it lives within the health systems. So I think that future of that partnership has to recognize who has the data assets
20:22
and how we collaborate. Because if we just do imaging in isolation and we see there's coronary calcification
20:28
as an incidental finding, you've diagnosed that on a CT scan, We have to put that together with the lipid profile and with the genomic profile
20:38
if we're going to really make the right decisions for patients. So we can be too siloed in our approach
20:46
unless we start to think more broadly about how we make it patient-centric
20:51
and not, let's say, industry-centric. I mean, at this point in history, and I know you want to finish
20:56
I just want to come back to this idea. We've been doing partnerships for innovation for a long time
21:01
but now even more than ever to your point under the comment you made
21:05
it's actually not even just a health system and for example a fortune 500 company
21:09
but there are small young companies that are doing the main things
21:13
so we actually our approach has been an open ecosystem our AI technology, AI managers open
21:19
we have interoperability as our foundation so we actually believe this really is a
21:23
multidisciplinary between and in fact you really need regulators in there so it's
21:30
where we are No one has the answer. You have parts of it
21:34
We have parts of it. There are some great young companies that have amazing technology. And they need to also work with industry
21:40
You make a great point. We work with small, medium, and obviously very large
21:44
And I love each of those conventions. We're going to now continue the conversation at the table, get everyone involved
21:50
I know how many smart people we have in this room. Thank you so much, Jill, Shez, David
21:54
And we're going to enjoy some great food and wine, too. Thank you
21:58
Thank you so much. Thank you so much
#Health Education & Medical Training
#Public Health