Subscribe
Share
Share
Embed
Where we share our weekly news debriefs and discussions with industry experts. These are lo-fi recordings aimed at giving our readers more opportunities to engage with our analysis and a view into some of the conversations that shape it.
Martin: Shiv, welcome to the show.
Shiv: Hey, thank you so much.
Sorry I'm late.
Add some technical.
Martin: No, no worries at all.
Uh, how are you doing today?
Shiv: I'm doing great.
Yeah.
Happy Monday to you guys.
Yeah,
Martin: happy Monday to you.
I.
Actually, I'm gonna kick it over to
Kevin, 'cause he was actually at JPM
and, and has some, some thoughts on,
Kevin: I love listening to
your JPM presentation, as
always hearing you talk shiv.
One of the, um, one of the, the, the
framework that I left that presentation
with and then saw again in your guys, um,
news about the partnerships with jama New
England Journal of Medicine was this idea
of the pre-visit conversation during visit
conversation, post-visit conversation.
It seems like the, the partnerships
y'all announced were, are, are focused on
the during visit conversation, clinical
decision support as part of that.
But I'd be curious if you could give
us the lay of the land of how you
think about the product offering today.
Expand on that pre, during post, and how
this content partnership fits into that.
Shiv: Absolutely appreciate that.
Um, so high level, the mantra inside
of the company, inside of a bridge
is that we wanna aspirationally,
we wanna save time, we wanna save
money, and we wanna save lives.
And as a cardiologist saving
lives, we take that very seriously.
It's aspirational.
To be able to demonstrate
that takes, takes time.
But you gotta start somewhere.
You gotta start with product.
And when we think about product and the
product, user experience, chassis, if you
will, we think about what happens before
I walk in the room to see my patient.
What happens during that moment
when I'm with that patient?
Then what happens after all the work
that happens afterwards in terms of
looking things up or creating artifacts
that I'm entering into any number of
different systems of record, for example.
But before the conversation, I
might be prepping for that patient.
I'm about, I'm about to see, I used to
have a, a Monday morning clinic, uh,
a cardiology clinic and every Sunday
night I'd be in front of Sunday night
football just prepping my charts.
I'd be looking up all
the complicated patients.
It would bring my blood pressure down.
I'd feel a lot more at ease the next
day that I could actually do a good job.
Not only could I be timely, but I
could actually be a prepared mind
walking in the room, feel like a
little bit more of a superhero.
But with this technology, we can,
you know, put that on steroids.
Now, before I walk in the room, I
can really know who this person is.
I can even be queued on what
questions I need to ask or
what articles I should read.
And those articles
could be outta UpToDate.
We've announced that
partnership in the past.
It could be from New England
Journal of Medicine or from jama.
But I can kind of walk in that
room with a lot of the data,
a lot of the evidence sort of
contextualized to who this person is.
And then I walk in the room and our
ethos continues to be no distractions,
eye contact being present, no
beeping sounds, no flashing lights.
But when I look down or look, look to
the computer screen, I should be able
to get that cue and maybe, you know,
you should be asking these questions
or considering this differential.
And then afterwards, obviously I can
continue to query against all the context
now, not just from the record, but
also from the conversation that I had.
Hopefully that kind of
makes, makes more sense.
Kevin: Uh, I'd be curious, um, if
you wouldn't mind walking through
a specific use case of this, like
in my head I've got it as kind of.
Google search, I type in a question and
it, it, it references the literature, but
how, how does it fit into your product and
what's the, are you typing in questions?
Is it ambient as well?
What's the, how does it fit in?
Shiv: Yeah, so maybe a couple ways
I can describe how it fits in.
One is through the, the technology
at actually very politely, very
hopefully, tastefully giving you cues.
So, um.
Saw a patient recently and and Chase,
one of our clinician scientists,
will bring up this example as well.
The patient happened to have a certain
type of medical history and also had
a certain BMI and I was seeing this
patient for chest pain and the usual
course of action for chest pain in
a patient who's got a lot of risk
factors is to do some sort of risk.
Factor stratification or like a
stress test of some sort to sort
of figure out what's going on.
And I ordered a certain type of
stress test, a nuclear stress
test, but then afterwards the
technology queued me very quickly.
It didn't call me an idiot, but it
said, Hey, do you want to reconsider
the nuclear modality and think
about doing a pet like a mibi?
Um, and, uh, maybe it would be a
better, potentially a better choice.
And you wanna read more on why.
And if I hit that button, I could very
quickly, um, look at the evidence, you
know, where that's coming from, whether
it was an UpToDate article or a New
England Journal of Medicine article.
In that case, like I can actually go
down all the way to the data and build a
lot of conviction and changing my plan.
If I wanna change my plan, I can simply
free text that I can say, okay, adjust
my plan to reflect this and also adjust
my orders to reflect this, this, you
know, evidence or this data that I just.
Um, so everything's like
very deeply integrated with
our enterprise positioning.
I'd say like that's the most important
aspect of a, like all, all the product
features that we build is, is trying
to build as much trust as possible and
feed as as much context as possible.
Martin: One of the things that I find
fascinating about this space is like, it
feels like up until maybe like five years
ago or four years ago, clinical decision
support was like highly algorithmic.
It was like, okay, there's a drug drug
interaction and we can like surface that.
And what that led to, I think was
some, some alert fatigue for doctors.
They were getting kind of bombarded
with like, you got these alerts.
Um, how are you thinking about, you
know, you have, you've announced
these partners with the New England
Journal of Medicine and with jama.
How are you thinking about.
The sort of the, the right context
and, and sort of fitting those in
so that it's not detracting from
the sort of patient interaction.
Shiv: Yeah, it's, it's a great question.
I remember those clinical
decision support as a term too.
It doesn't feel as cool anymore.
It feels like clinical intelligence
is the new way to describe this space.
Um, I used to be a corporate
VC of sorts at a large health
system at UPMC years ago, and we
invested in multiple CDS companies.
And at the time, you're right, it
was basically best practice alerts.
It was alert fatigue, it was popups.
And like, the best you could do is give
a surface, a popup at the right time, you
know, at the right moment in a workflow.
And, um, try to introduce
friction that a clinician might
hopefully perceive as good.
Um, but this is a new moment where,
especially when you think about.
Products from like an AI first, from
an agent first sort of standpoint.
There's so much work that you can do
in the background and there's context
that you can kind of engineer the new
term as obviously not just context
engineering, but harness engineering and,
um, and, and how can you, how can you
kind of safely, um, lead the clinician
to the, to the right data to help?
Augment their decision making, improve it.
So to, to do this in a way that
really works, um, for us evals,
um, have looked multifold.
Like one is just quality.
We score all of our outputs against
a physician built rubric system.
Another is just testing for adversarial
use because I think you're speaking
to, or we're speaking to the user
experience where it's, it's, it can feel
pretty open-ended and so jailbreaking
can be, um, a risk and that you need
to build the right guardrails into
that harness to protect against that.
And then another is just like straight
safety from the standpoint of surfacing
the wrong research papers or the,
like, the wrong drugs or, or, or
missing key parts of the context that
you're feeding in, in the background.
But we've now used, you know, uh, well
over a thousand really, really deep.
Test cases to build rubrics developed
by clinicians that kind of got
us to this point of conviction
that this is ready to go ga.
Um, and that's what we, what, what we
just announced, but it's a treadmill,
like we'll always be running on it,
just the way we run on all things
related to like mid revenue cycle
and clinical documentation for that.
Kevin: As an aside, as a MBA intern at
Medtronic in 2010, I remember putting
together a slide deck on clinical
decision support and alert fatigue and
how hot of a market it was at the time.
But just on this intellectual level,
the idea of Medtronic getting involved
in clinical decision support, I think
played into some of that alert fatigue.
And when you think about the
underlying incentives there,
I'll leave that aside for now.
I, I'd be curious if to, to go back
to, um, JP Morgan presentation.
You talked about, I think a
bridge is live in something like
200 plus health systems today.
Uh, a lot of providers are using it.
It's used in a lot of
patient conversations.
I'd be curious to hear you reflect
on state of play from a go to market
perspective as an AI solution.
Obviously we have seen, uh, an explosion
of health system interests and these sorts
of tools over the past several years.
It strikes me that.
If I'm a health system leader,
I've probably thought about this
question, have put in place the rails
of what my AI deployment strategy
is within the system at this point.
Shiv: Yeah.
Kevin: How do you think about
conversations with existing systems, with
potential customers about feature set?
How you're building that out,
how something like this fits in?
Shiv: Yeah, it's, it's go to market is
everything, especially in healthcare.
Um, and you know, like people think
that healthcare is one homogenous
market and obviously it's not.
And it's, we think it's really difficult
to swim upstream, but it's way more
possible feasible to go sort of, sort
of go downstream from like those big
systems from, from the IDNs and the IDFs
and the payer providers and the AMCs.
So we, we yoed it.
You know, in 20 22, 20 23, and went
to the big systems, and now over time
we'll start to expand and extend.
But within those big systems, the, the,
the barrier for good enough, um, against
even the first use case, like still
high, like, uh, by that I mean like the
threshold for good enough because they
need a solution to work in an enterprise
grade way per the CIO that can help
all their clinicians and all their
specialties and all their spoken languages
and all their different care settings.
But then, you know what, what we
quickly discovered is that there's so
much more that you can extend into,
whether it's move upstream into what
happens before that encounter or
downstream into what happens after.
And like as an example, like,
uh, you, we might sit with a
sit down with the CIO who's.
Thinking about, you know, an ambient
solution, but they're also separately,
maybe a couple years ago they were
thinking about a CDI solution, and
then you need, you sit down with them
and you help them understand, actually
behind the scenes we can shift left on
a lot of that CDI work and get it done,
um, even before the node is generated.
And now you don't need to worry as
much about CDI solutions for XY, Z sort
of, you know, um, you know, VA value.
And it's an epiphany for them.
And, and so it's the same
story for prior authorization.
There are a lot of health systems who
are, you know, were experimenting and
especially a couple years ago with prior
auth, uh, solutions that might live
like downstream and they're aggregating
data across a lot of different systems
and trying to help make the process.
For my backend administrators, a little
bit more simple, helping them kind
of get into those payer portals and
upload all the PDFs or, or fax data,
you know, in a more efficient way.
And then we would try to help them
understand, well, actually what happens
if you just sort of shift left again
and have these agents and subagent
understand not just the context of
who this patient is retrospectively,
but here are the conversation.
Cue the clinician on gaps, they
should close, and then create all
the right documentation and send
it automatically to those portals.
And even in, in the case of a couple
different payers, actually round
trip the authorization in real time.
And it, it, it starts to, it helps
them sort of understand that actually,
like you, you can kind of collapse a
lot of care delivery work workflows.
Um, if you start to, um, you know.
Just challenge your priors on, on, you
know, what it could look like if you
redesigned the entire system from scratch.
Um, and I think that's really the
opportunity with AI is to just
like, think, think about, you know,
uh, collapsing and compressing
workflows as much as possible.
Martin: We, I know we're, we're
getting close to time here, so last
question for you is, one of the things
we think about a lot here is like,
okay, where are we gonna start to
see the impact of AI in healthcare?
I think, you know, there's some people
who say, we're starting to see it
in coding, maybe in sort of ambient
scribing and, and productivity.
There are other people who are
skeptical and they say, this is a
silos paradox thing where you can see
it everywhere except for, you know,
the consolidated financial statements.
From your perspective, where do
you think we, we start to see the,
the impacts of AI in healthcare?
Like, what are the things that you're
most excited about and where should
we be looking for those screen shoots?
Shiv: Yeah, I, I have been
thinking about this quote a lot.
Like there's a Bezos quote from years
ago that when the anecdotes and the
data disagree, believe the anecdotes,
and that was very much like the story
for us in like early days of scaling.
We're now live in well over 270 large
health systems on a run rate to be a
part of well over a hundred million
encounters in the next 12 months.
And hopeful a lot more
than that, um, this year.
And.
Early days though, before we got to that
scale, we'd have doctors tell us that
for the first time they're experiencing
Los Angeles rush hour traffic.
Um, and that felt like real,
that, that felt like data that
felt like, you know, a big win.
But then we would go to that same system
and that same users sort of like data
from existing telemetry systems and there
was no indication that they had saved any
time and we're like, what is going on?
And so over time, and it's still like
a moving target, we've realized that
a lot of different systems in in are
instrumented in different ways and
it's, it can be pretty difficult.
So you try to find all the
proxies you can for the value
that you're trying to create.
Um, save time, save money, save lives.
I think that money piece is so important
to us because like, it's one thing to
deploy an AI system and deploy it into
the existing system without thinking
about compressing and collapsing
workflows and creating those efficiencies
impacting not just the bottom line,
but also figuring out how you can, you
can find positive some sort of models.
Between all the rate that the stakeholders
in healthcare and thinking about
payers and providers specifically.
But that has to be the goal, um, in, in
our estimation, because otherwise we're
just gonna be deploying, obviously, like
agents on top of existing workflows.
And that bal cost disease curve is gonna
look even steeper than it already does.
So for, for us, it's, it's a finite
number of use cases that we're
putting a lot of energy into that
can demonstrate that win-win, where
we're threading the needle through
the most important people, patients,
and their care teams, those end users,
but also those existing stakeholders.
Prior auth is a really
good example of that.
But, um, certainly a lot of the work that
we're doing, um, like the announcement
that we had with Availity, for example,
at JP Morgan too is, is all about that.
Martin: Shiv, this has
been hugely helpful.
Thanks so much.
If folks wanna learn more about a bridge
though, I, I think everyone listening
has probably already heard of a bridge.
Where is the best place on the
internet for them to do that?
Shiv: Yeah, well we have to, we'll do a
better and better, like, we'll do a better
job of putting a lot more information
on the website, but that's a good spot.
Um, LinkedIn is, I feel like, where
you have to live as an enterprise
healthcare company, and so probably
our LinkedIn account is dropping as
much information as anything else.
Martin: Great.
Thanks so much for your time today.
Shiv: Awesome.
Thanks for joining
Kevin: us.
Shiv: Yeah.
I really appreciate it.
Thank you.