September 26, 2024 • 52 mins
“Tempus is focused on the idea of bringing technology, and in particular artificial intelligence, to diagnostics by essentially combining a molecular result and then connecting that to clinical data for that patient.” Eric Lefkofsky, Founder and CEO of Tempus AI, tells Bloomberg Intelligence in this episode of the Vanguards of Health Care podcast. Lefkofsky joins BI analyst Jonathan Palmer for a discussion on how Tempus built a multipurpose platform that generates unique insights not only for treating patients but also in drug discovery and speeding recruitment for clinical trials. The two cover how oncology is just the first beachhead, why it’s still early days for the company’s data business and the future of using algorithms in treatment.
See omnystudio.com/listener for privacy information.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Speaker 1 (00:21):
Welcome to another episode of the Vanguards of Healthcare podcast,
where we chat with the leaders at the forefront of
change in the healthcare industry. My name is Jonathan Palmer,
and I'm a healthcare analyst at Bloomberg Intelligence, the in
house research arm of Bloomberg. We're happy to welcome Eric
Lefkowski for today's episode. He's the founder and CEO of
Tempa's AI. Prior to founding Tempess, he has an impressive
(00:43):
career as an entrepreneur, founding several successful companies, most notably
group On. His work at Tempes is transforming healthcare through
the application of artificial intelligence into precision medicine. Thanks for
joining us, Eric, Why don't we get started with an
overview of Tempest and kind of how you're changing the
health care landscape.
Speaker 2 (01:01):
Thanks for having me. So.
Speaker 3 (01:02):
Tempest is focused on this idea of bringing technology and
in particular artificial intelligence to diagnostic by essentially combining a
molecular result, so sequencing a patient producing some result, and
then connecting that to clinical data for that patient so
you can really personalize the result around that particular patient.
(01:25):
The benefits you get are pretty extreme. So in the
case of sequencing cancer patients. If I find a mutation
and I might recommend a drug, It doesn't help to
recommend a drug if the patient took that drug in
a prior line of therapy and failed, just as it
doesn't do any good to recommend a clinical trial that
the patient can't enroll in. So, for example, if I
knew my patient was a smoker, why I recommend a
(01:47):
clinical trial. We're the first exclusion criterias. You can't be
a smoker. So we could see, or I could see
when I started Tempest about nine years ago, that we
were likely going to enter this period where you'd be
able to generate enormous amounts of low cost, high quality data,
in particular because we were going to start sequencing most
(02:08):
cancer patients, and that this data would pile up in
huge ways and it would become increasingly complicated for physicians to, like,
in real time, interpret that data and then make decisions.
And so we thought, what if you could collect clinical
data for those patients, sequence those patients, and then put
the two together so you could kind of produce the
(02:30):
insights as part of the report itself. So if you
had something intelligent to say about that patient, you'd say
it in the report, and so that's how Tempest was born.
And you know, we went out started convincing some hospitals
to try to work with us, opened up a lab,
and pretty early on you could see the benefits of
(02:51):
our solution, Like it became apparent that these kind of
comprehensive tests that were also intelligent and more personal were
going to be you know, benefit official to physicians and patients,
and so they began ordering them at scale and we
grew and during that journey we also realized that not
only could you generate really personalized reports for patients, but
(03:14):
that when you connected clinical and molecular data, if you
de identified that data, you could produce really interesting data
sets for researchers and provide the same benefit to researchers,
whether they are researchers at economic medical centers or researchers
at biotech companies or researchers at pharmaceutical companies. And by
helping them do research, they would be more efficient, they'd
(03:35):
make better drugs and fail less often.
Speaker 1 (03:38):
That's a great overview to start. I mean, if we
rewind back to that initial inception of the company, did
you have that idea of marrying the two pieces the
genomic data and the and the you know, the longitudinal
data post post the test.
Speaker 3 (03:51):
Yeah, almost from from day one. We if you go
back and look there, there's this really cool thing hold
the wayback machine where you can kind of like look
at Internet sites at various moments time, you know, and
you can see us talking about big data and data
science and AI and combining clinical molecular data. Really back
seven or eight years ago, so very early after we
started a company, we began thinking about how this model
(04:14):
would evolve. What we didn't know back then is that
in order to bring in all this clinical data for patients,
we were going to inevitably become connected to thousands of hospitals.
There's about fifty five hundred hospitals in the United States,
and we're connected to something like twenty three hundred some
number like that. So it's approaching half of all hospitals
(04:36):
the United States. And so when you're connected to these hospitals,
it's like a cable company, right, It's like laying cable
to a house. All of a sudden you have a
direct connection to that house. And so by virtue of
these connections, we're able to pull data out of these
electronic healthcare record.
Speaker 2 (04:52):
Systems or EHR systems.
Speaker 3 (04:53):
We're able to generate diagnostic insights and put them into
the EHR system all through these electronic conections that we built,
and so that's allowed us to have a third business
line because today TEMPS is really made up with three products.
Genomics where we sequence patients and bill insurance, our data
business where we license the identified data, and our third
(05:14):
business units called applications, where we basically leverage these connections.
We have the thousands of hospitals and deploy purely algorithmic
diagnostics in real time to be able to close care gaps,
or route patients to the right clinical trial, or be predictive.
That business line we couldn't see if you go back
in time, like we could see that marrying clinical molecular
(05:36):
data would be powerful, but we couldn't see that you
could actually wire up the US healthcare ecosystem and then
leverage AI to really improve outcomes.
Speaker 1 (05:45):
There's a lot to unpack there, So maybe I'll start
with the kind of the core genomics business when you
first kicked off, You know, what was the moat that
you were trying to build VISA via the competition that
was out there. I mean other people were doing you know,
multi gene panels. You know, how did you really convince
the first practices to use Tempest? Was it the data piece?
Was that the most compelling part of the sale?
Speaker 2 (06:08):
Yeah?
Speaker 3 (06:08):
You know, when we first started, we knew we wanted
to marry clinical molecular data, but we didn't know we were
going to sequence patients. We actually thought, let's not open
up a lab, let's not sequence patients. Let's go to
the people that sequence patients and tell them that we're
pretty convinced that there's this emerging problem that's only going
to get worse, and let us be their tech partner.
(06:30):
And we went to a few of the big labs
at the time. They were entirely unwilling to give us
any data. They were basically like, you know.
Speaker 2 (06:40):
There's no way I'm going to give you any data.
Speaker 3 (06:42):
And so we said, well, without the data, we can't
actually solve this problem. So then we had to go
to hospitals and say, you know, let us sequence your patients.
I mean, we were fortunate, I think that we were
able to raise enough capital early on to build a lab,
to hire really credible people to join the company that
knew how to sequence patients that knew how to develop essays,
(07:05):
and so that credibility allowed people to feel comfortable that
we were going to be able to deliver. And then
obviously there are certain hurdles you have to get over, like,
for example, being accredited by CAP.
Speaker 2 (07:15):
And having a CAP clear certified lab.
Speaker 3 (07:17):
So we achieved that, but early on, the bigger hurdle
wasn't convincing people to let us sequence. I don't think
they weren't that happy with their existing sequencing partners. The
biggest hurdle was saying you should give us clinical data
for the patient to make the sequencing tests smarter. Back then,
you know, eight or nine years ago, people were very
(07:37):
adverse to turning over any data. Today that has almost
entirely changed. Now people recognize that first of all, it's
not their data, it's the patient's data. And second of all,
if this data can be leveraged to produce a better outcome,
that's all that matters.
Speaker 1 (07:54):
So it's funny because I would think in some ways
that today, where the value of data is so much
more imperant, actually be more challenging to get that data
from from providers. Do you see any hurdles or challenges
in that respect.
Speaker 2 (08:07):
No, I mean we have so much data.
Speaker 3 (08:08):
We have, like you know, we're approaching half of all
cancer pation data United States in some way, shape or form.
So we have a huge amount of data. So I
think people are very comfortable giving us data that We're
going to do everything we can to make sure it's secure,
to maintain privacy, to adhere to you know, regulations and
compliance and that, and that only used a de identified
(08:29):
version of that data for the benefit of patients. So
I think people are seem to be very willing to
contribute data to our platform. But certainly, you know, as
more and more people realize the value of this data,
you get more competition. And so the competition has grown,
and I suspect it will continue to grow.
Speaker 1 (08:47):
Yeah, I mean I was thinking about that in the
through the lens of you know, we've seen US Oncology
continue to consolidate practices and Kora has been buying one oncology.
Is that is that a hurdle for your business?
Speaker 2 (08:58):
It's not.
Speaker 3 (08:59):
It's not a hurdle really at all. I think it
actually could be an accelerant in that you know, this
consolidation will likely lead to more strategic moves that that
could benefit us. But as it relates to the testing side,
those decisions are almost entirely made at an individual oncologist level.
So even though you might have practices that are that
(09:22):
are part of one academic medical center system or owned
by a conglomerate, the decision as to which sequencing lab
I want to use is still almost entirely made one
on collegist at a time. So all the big labs
are effectively kind of five big next generation sequencing labs
and Cancer, Tempest, Foundation, Medicine, Caris, Garden, Ntera, and almost
(09:47):
all of us have you know, large sales forces that
are calling on doctors one at a time explaining the
benefits of our platform.
Speaker 1 (09:55):
And so when your doctors call, you know, how do
they differentiate between some of those other players.
Speaker 3 (10:00):
Our big advantage has historically been it is today technology.
So when we're competing with other people, if we're winning,
it's because we have a technology platform that delivers more
personalized results. The insights are more integrated and easier to
use by physicians, They're more comprehensive. Our platform is more comprehensive,
(10:23):
and so physicians feel like they are getting the totality
of information they need and they're able to treat their patients.
I mean, that's typically why doctors would choose US versus
somebody else. We also happen to have at this point
as one of the largest sequencers of cancerbations in the world.
We have a very comprehensive portfolio from inherited cancer risk
(10:44):
to solid tumor profiling, to liquid biopsy to minimal visual
disease detection and monitoring.
Speaker 2 (10:49):
So it's a pretty comprehensive.
Speaker 1 (10:50):
Port I'm smiling because you kind of led right into
my next question, which was just to really talk about
the milestones and from the portfolio of offerings that you have.
If you think about where you were to start to
where you were today, you know, you listed all those
different offerings, whether it's the hereditary or MRD, you know,
which are you most excited about today?
Speaker 3 (11:11):
We are. The biggest part of our platform today is
in a conference of genomic profiling or therapy selection. So
that's you know, a patient comes in, they're diagnosed with cancer,
they need to be sequenced quite often to figure out
what drug to give them, so therapy selection, and that's
where we have the biggest part of our portfolio. Both
in the case of solid tumor profiling, so like I
(11:32):
have a biopsy and I'm sequencing the tissue of that
tumor versus like liquid biopsy, which is I'm kind of
pulling those molecular insights from blood. And about a quarter
of our volume is liquid biopsy versus solid tumbor profiling.
So at our scale which you can seek US Republic,
we're very large. We're very large in both in both areas,
(11:52):
we do quite a bit of inherited risk profiling, so
that piece we also do quite at some The newest
offering for US, and the one that obviously has significant
potential going forward, is MRD we launched. So for those
that don't know, you know, there's been this emerging space
(12:13):
within cancer of monitoring and minimal situlecies detection, and this
is the idea that from again either blood or tissue,
I can effectively, you know, track patients over time and
try to figure out if their cancer has come back
before I might see it in a scan. And what
you see is you'll see certain patterns, whether they're methylomic
(12:35):
based or whether they're from actual like you know, fragments
of DNA circulating in the blood. You'll see signatures that
something a mutation looks like it's circulating in the blood
before I can see it in a scan. And this
is because if you think about it, you know, fragments
of DNA circulating in your blood. If you're looking at
(12:56):
a particular scan, looking at an image of a scan,
you might need to get a large volume of cells
together to be able to see something. You know, whether
it's a million or a billion, right, you need you
need a very large polk. So a lot of these
tests that are coming to the market right now are
picking up signs of cancer, you know, five months, six months,
(13:19):
seven months, eight months before you would see it on
a scan. So they're quite powerful in terms of their
ability to detect cancer earlier than a patient might normally
see it. And there is significant benefit people have published
on this, and there's more that will come about about
being able to intervene when you when you can see
that a patient has actually recurred before you might see
(13:41):
it in a scan. We're excited about our portfolio because
we launched both what's called tumer naive and tumor informed.
Tumor naive is essentially saying, right from the blood, we're
going to detect those signatures of early disease. Tumor informed
is saying we're going to basically, you know, take up
take a piece of that tumor, build a bespoke assay
(14:02):
of some kind, and then track it over time. They
both have benefits, right, just like there are benefits to
solid tumor profiling versus liquid. The benefit of the tumor
naive or the liquid profiling is that it's very logistically easy.
Speaker 2 (14:18):
You draw blood, you can run a test.
Speaker 3 (14:21):
The benefit of tumor informed is that it tends to
produce higher sensitivity and specificity and improved detection. And so
we have our own tumor naive assay that we launched
in colorectal cancer, and then we have a partnership with
a company called Personalis where they've launched an assay tumor
informed assay in lung cancer, breast cancer and IO response.
Speaker 1 (14:43):
Is the corectal assay just the beachhead to other indications.
Speaker 3 (14:48):
Over time, we expect to take our turmer naive MRD assay,
which right now is in colorectal, and move into other
disease areas. Some of those are easier to move into
than others. It's kind of tracking those signatures. The blood
is a function of you know, different tumor types of.
Speaker 1 (15:03):
Behavior, how much they're shedding that the end of the
blood like prostate doesn't do very well in these assays.
Speaker 3 (15:08):
Some some subtype shed a lot and some shed very little,
and that tends to be the distinction. So we we
spent a long time building our assay in in in
c r C, which we're quite proud of. It's got
great stats, and but we're on a constant effort to
you know, fine tune our bioinformatics and our ability to
(15:29):
boost sensitivity and specificity and the lower reality or limits
of detection.
Speaker 1 (15:33):
So what's what's the drivers behind that? Is it on
the informatic side like you just mentioned, or is it
on the chemistry side.
Speaker 3 (15:38):
It's it's both, It's both. The informatic side can only
take you so far. So at some point you boost.
Speaker 2 (15:45):
In chemistry, you can, but you can get far. You
can get far with informatics.
Speaker 3 (15:49):
I mean, you know, and you and and there's a
real benefit informatics with scale, right if you think about it,
informatics is you know, it's math and and and building
models and the more data you feed into a model,
the better the model gets. That's why you know chat
GPT four hosts better than three, and three is better
than you know, on and on. So I think we
(16:10):
will You'll see significant improvements over time by our ability
to fine tune the biotraumatics. But net net, if you
want to really have quantum loops forward, you need improvements
in chemistry, need to be able to basically pull signals
out at higher fidelity.
Speaker 1 (16:23):
And how long do those those milestones take to reach?
You know, if you think about your product roadmap on
the chemistry side, you know the next iteration does it
take twelve months? Does it take years?
Speaker 3 (16:34):
Well, we benefit, I think Tempest benefits. You know, given
our scale, we're fortunate there's a lot of other labs,
both private and you know, you know companies as well
as parts of public institutions like academic medical centers that
want to partner with us, So we benefit. We don't
have to develop all of these advancements on our own.
(16:55):
We have people coming to us constantly that you know,
have have novel breakthroughs that they want to see us
bring to market, So it'll be a combination of both.
Speaker 1 (17:04):
Well, maybe is a good segue into your R and
D organization, can you can you maybe talk about the
size and scope of what you're focused on.
Speaker 3 (17:11):
Yeah, I mean you can look so first of on
our in our in our public filings, we break down
R and D by you both tech and by science.
But I mean it's directionally. We have if you look
at headcount, somewhere between nine hundred and one thousand people broken
down between product engineering and our science teams. And the
science teams are maybe about three hundred PhDs and mds combined,
(17:33):
so it's a fairly large scientific team. And then it's
you know, five six hundred software engineers and a engineers,
so it's a fairly large technical team as well. The
if you look at our investments over time, and we've invested,
you know a lot of money. It's taking a few
billion dollars to get here, a lot of it's been
in those in those areas.
Speaker 1 (17:52):
So maybe that that leads to, you know, where do
you have to continue to make investments next as you
continue to scale? Is it is it just purely on
the R and D function? And do you have to
build more labs? You know, where do are those investment
dollars going to go?
Speaker 3 (18:04):
We have today three labs, one in Chicago, one in Raleigh,
one in Atlanta, and the capacity of those labs is
quite significant, so we wouldn't have to build more more
like lab space in order to grow quite a bit.
So I think from that standpoint, we've made significant investments
already for us. If we were to build another lab
(18:25):
at some point, it might be geographic, like do we
want to have, for example, a lab out west because
we like the logistical benefits, but we're fairly built out.
Most of our investments are really trying to bring the
benefits of new technologies, in particular generative AI in large
language models to our entire technology stack, both on the
(18:48):
product side and on the science side. And you know,
we're kind of at this really fortuitous moment in time
where you know, if you think about when when Tempest
was started, we could see this day coming, and the
background technologies we had were largely natural language processing and
optical character recognition. Now all of a sudden, you've hypercharged
(19:11):
that with large language models and general of AI, and
so all of a sudden, what you can do with
large data sets is like unimaginable.
Speaker 2 (19:20):
So if you want to.
Speaker 3 (19:21):
Take advantage of that as a company, you kind of
need two things. You need a proprietary data set and
the ability to distribute any insights you get from these models,
and Tempest happens to have both at scale. You know,
we have two hundred and fifty petabytes of proprietary data
that we can use to train models, and we're connected
to know almost half of all hospitals the United States,
(19:41):
So we have I think we're in a good position
to benefit from genera of AI, and so we spend
a lot of our time and money and energy on
how do we really bring you know, LLLMS to our
entire and so.
Speaker 1 (19:55):
Going forward, if I think about you know, you delineated
between the product and maybe the providers who are we're
using some of these AI tools, and then on the
data side, where you know, the biopharma companies are utilizing
your stack. You know, where do you think there's going
to be the more profound inflection.
Speaker 2 (20:11):
I think there will be in the near term.
Speaker 3 (20:15):
I would suspect both will be kind of similar in
terms of the utility of these technologies. I think in
the semi near term. And again I'm not able to
really predict with specificity when this is going.
Speaker 1 (20:29):
To happen, but we won't hold you to it.
Speaker 3 (20:31):
Maybe yeah, maybe in the next three to five years
or something, I think there will be truly profound impacts
on physicians and how they behave. I think companies like
ours and others will arm the average patient with enormous
(20:53):
insights about their care, and it will literally, you know,
it's it's going to be like I've never said this before.
Speaker 2 (21:02):
I just thought of it right now. If it's gould
be a horrible analogy, but I'll use it.
Speaker 3 (21:04):
It's a little bit like, you know, you show up
in a city like New York, okay twenty five years ago,
you get in a cab and they take you where
they take you. But now if you don't, you pull
up your phone, you go to Google or ways, and
you can tell them what route and he should go
and you're right, or at least as right as you
know you can be. So all of a sudden, the
(21:24):
cab driver can't say, oh, you know, we're.
Speaker 1 (21:27):
Going to take a left here.
Speaker 3 (21:28):
Yeah, don't tell me what to do. You don't know
what you're doing. You're like, no, I actually do know
what I'm doing. You know, Google's telling me you're going
the wrong way, and that I think is going to
be very similar to how the average patient, not just
an oncology but in neurology and cardiology and immunology is
interacting with their doctors. And so doctors have to be
armed with similar tools or they're going to.
Speaker 2 (21:51):
Really be at a disadvantage.
Speaker 1 (21:52):
That's fascinating. I Mean one of the things that I
thought of while you were speaking there was, you know,
just how you marry what you're doing with guyidelines and
you know NCCN guidelines kind of have a treatment pathway.
You know, where do those intersect with the data that
that you're providing and do they ever conflict?
Speaker 3 (22:11):
Yeah, I mean they don't conflict because we're not really
in the business of saying, you know, our data says this,
so you should do that. You know, there's no like
do this, do that, red, green, yellow, It's more function
of this is the data we have, You as a
physician are free to look at it and to the
extent it helps you make a decision that you should
(22:33):
And I think physicians, at least in oncology, you know
something like two thirds of the time, especially in later
stage disease, so people with stage three, stage four, you know,
third line therapy, fourth line therapy. There's really no guideline
to guide you. And in fact, I tell this to
people all the time. You know, if you if you
look at, for example, a pancreatic cancer patient at the
(22:53):
one year mark, who's in let's say, you know, third
line therapy, the I don't know the overall survive, the
two year over survival or three or like five percent.
So what that fundamentally tells you is that the standard
of care is going to produce death ninety five percent
of the time. So to a patient, that's not a
good that's not something one should fill that out. Yeah,
(23:16):
you'd prefer not to follow that standard of care, right,
You're like, I want to so quite often, unfortunately, if
you look at in the case of cancer, there's roughly
one point nine million new diagnoses a year and basically
about six hundred thousand people die a year, so roughly
a third of cancer patients die. And you know, you
don't want to follow the standard of care when you're
in that group, when you have a very high likelihood
(23:38):
of doing well, like you know, stage one ear positive
breast cancer, where you have whatever ninety nine percent five
years overall survival, then you want to follow the standard
of care. So I think we're arming physicians with the
kind of data they would need that when there's really
no good guideline, they can at least make an informed decision.
When the guideline should produce a good outcome, they're typically
(24:00):
following that guideline. So we're not in contradiction. What is
going to be in contradiction at some point is that
the guidelines will be unable to keep up with the
emerging state of evidence and information unless they too are
leveraging AI. So, for example, we do pharmacogenomic profiling of
(24:22):
patients that have major depressive disorder, bipolar disorder, severe anxiety.
So there's like seventy million people in this country that
have some form of anxiety. And we know that there
are certain antidepressants that aren't metabolized by the body in
certain ways. So for example, if I'm a SIP two
D six for metabolizer, I shouldn't be taking this class
of drugs. Fairly soon, we're going to also know that
(24:45):
if I'm a cancer patient taking a targeted therapy or
an immunotherapy or a chemotherapy, it is having this impact
on how I metabolize these drugs. So now you're going
to have to ask oncologists to think about the drug,
the prescribing, what it means for your depression, how they
should think about changing your dose so that you're not
(25:06):
experiencing adversements. It's just too it's overwhelming, and you know,
you can imagine a world where the number of data
points that a doctor has to consider go from pick
a number ten to a thousand and they just.
Speaker 2 (25:21):
Can't do it.
Speaker 1 (25:21):
It's impossible.
Speaker 3 (25:22):
It's impossible, and the guidelines would go from the guideline
And if you look at a non small cell lung
cancer guideline, it's three hundred pages, four inner pages massive,
So what's it going to go to ten thousand pages?
It's no longer a guideline unless you know, there's too many,
too many branches of the tree.
Speaker 1 (25:39):
So that's a good segue into you know, your your
other opportunities. You mentioned the neuropsychiatry. You know, where else
are you focusing this model? You know you've talked about
cardiology as well. Can we maybe use some other examples
of where you think the treatment paradigm is going to
change using your platform?
Speaker 3 (25:56):
Yeah, I mean cardiology is a great example. What we
also have products radiology and pathology, but cardiology is a
great example. In each disease area, if you're in the
business of trying to bring the benefits of artificial intelligence
and technology to diagnostics, you have to go disease by
disease and pick the diagnostic you think is most central
to how patients are treated. So in cancer, that's genomic profiling, because.
Speaker 2 (26:19):
That's that's kind of what it is.
Speaker 3 (26:21):
But in cardiology it's actually a kind of good old fashioned,
boring ECG. Right, that's the thing that most often is
frontline you know, twelve lead ECG or which also called
an EKG. So we began thinking about how would you
deploy technology to that test. And it turns out that
(26:41):
about three percent of patients who have a normal ECG,
and we run about three hundred million, I think, give
or take in this country, year, three percent of patients
are told they're fine, go home, everything's good, and yet
within a year they have a heart attacker stroke. So
clearly things were not fine. The question is what didn't
we see when we did that ECG. And it turns
(27:04):
out that the people that have a heart attacker stroke
typically have some undiagnosed condition, namely things like atrial fibrillation,
hypertrophic card in myopathy, low injection fraction, aortic stenosis, whatever,
pick one of your So we've been working with a
team that has been building models for years that we
(27:24):
brought into Tempest where we can basically predict these things.
And the first of which was our model to predict
atrial fibrillation from a normal ECG, which got FDA approval
like about a month ago. So that's a good example
of you know, you can imagine a world where you know,
hundreds of thousands of patients a year that are part
(27:44):
of a healthcare system, their ECGs will be uploaded every night,
will run that model, and then we'll call that three
percent and say, hey, wait a minute, even though the
doc told you yesterday or this morning that everything looks good,
you actually are at high risk for something, and we
want to run an echo. We want you to take
a stress test, we want you.
Speaker 1 (28:04):
To wear a monitoring at a minimum, go to a cardiologist.
Speaker 2 (28:07):
Go to a cardiologist.
Speaker 3 (28:09):
And you know, when somebody has a heart attack or
a stroke, it costs the system like one hundred thousand
dollars a year.
Speaker 1 (28:15):
So and running the algorithm costs.
Speaker 3 (28:18):
Right now, it costs zero because no one pays for it.
But at some point when it's reimbursed, and it should
be reimbursed, I mean, what will they pay a hundred bucks?
I mean, it'll be minimal compared to.
Speaker 1 (28:29):
The outcome that you're foregoing, right, Yeah, So maybe that's
a good good spot to talk about reimbursement around algorithms.
You know, what's this current state of play? You know,
these are very new, I guess from the perspective of
people who pay, you know, whether that's CMS or commercial
or medicaid. You know, can you talk about that landscape
(28:51):
and how you see that evolving over time?
Speaker 2 (28:52):
Yeah? I mean, and I'm fairly you know, outspoken about this.
Speaker 3 (28:56):
I mean, it certainly benefits tempest, So I acknowledge that
that conflict, but I'm still one thousand percent convinced that
it's right. The US healthcare system, from a financial perspective,
couldn't be more broken. Okay, it's five trillion dollars a year,
and the outcomes we produce and you can like go
to Google and look them up, are not good. I mean,
(29:18):
I don't even know where we are currently in terms
of like mortality.
Speaker 1 (29:21):
Rate behind some Scandinavian country seeing a lot of other people. Yeah, well,
I mean.
Speaker 3 (29:26):
And my guess is, like, you know, we probably spend
as much as the next five countries combined, easily. Yeah, easily.
So this is not where you want to be. And
so how do you solve this problem? Okay, we're not
going to solve it by you know, negotiating drug prices,
and we're not going to solve it by you know,
making small tweaks to Obamacare. These are just not going
to solve our problem. The only solution to this problem
(29:47):
is technology, is data, is AI, and what it really
comes down to is using technology to remove waste.
Speaker 2 (29:57):
So what's waste?
Speaker 3 (29:59):
Okay, so we used to, you know, go to movie
stores to get movies.
Speaker 2 (30:03):
Now that's not a bad thing. I love the Blockbuster.
Speaker 3 (30:05):
But someone had to build a Blockbuster. Someone had to
make a video, ship the video, put the video on
a shelf, I had to go drive my car, pick
the video up, put it in the VHS system whatever.
It was like a lot of stuff.
Speaker 2 (30:17):
Went into that.
Speaker 3 (30:17):
Today I can technologies allows me to watch a movie
on demand and none of that stuff has to exist.
No buildings, no driving, no whatever that's waste. So we
need to do the same thing in healthcare. We need
to basically, and payers aren't good at this, and the
government isn't good at this.
Speaker 2 (30:34):
We need to.
Speaker 3 (30:35):
We need to basically find a way to pay for
AI insights, pay for diagnostic insights, pay for the things
that spot the problem early on, so we can avoid
all the downstream cost, which is typically you know, multiples
of the other.
Speaker 1 (30:54):
So how do you actually catalyze that, you know, how
do we get that to happen on the hill?
Speaker 2 (30:59):
I think it takes time. First, I think it's inevitable.
Speaker 3 (31:01):
Like, you know, there was a time when no one
thought Blockbuster would be ending, but Blockbuster and then it's gone,
and I think it's the same thing here, Like there's
no doubt in my mind. Eventually, you know, the CMS
and payers will have to pay for algramic diagnostics, they
will pay for next generation sequencing. Broadly, they'll realize there's
(31:23):
still payers in this country today, massive large payers that
don't pay for sequencing, so they'd rather a patient not
be sequenced, get the wrong drug, including drugs that cost
like one hundred thousand dollars a year, not do well
be in the ICU all that kind of bad stuff
instead of just paying for a two thousand dollars test
that some percentage of the time not small, mind you, like,
(31:45):
if we're not like in the twenties, twenty percent of
the time we route patients to an entirely different therapeutic
by virtue of that sequencing result.
Speaker 2 (31:52):
So I think it's just.
Speaker 3 (31:54):
Takes It takes time. But I suspect sequencing will be
broadly reimbursed, and I suspect that we will find a
way to pay for EI insights and will pervasively deploy them.
And that's how you'll start to see. You know, it
may be, for example, that the diagnostic category generates an
extra one hundred billion of revenue, right which would be seismic,
(32:18):
but you'll see a trillion dollar reduction in health care expense.
And it's it's like it's that Proflee.
Speaker 1 (32:25):
I agree with you. You know, I've seen in my career.
If I think back to non invasive prenatal testing, you know,
the adoption there took probably almost a decade. You know,
if I think about sequenomeen, maybe twenty twelve. Hopefully this
time it happens a little bit quicker. But I think
the jury's out. Well, maybe that leads us to the
data business because we've spent a lot of time on that.
(32:46):
We spend a little bit of time on the algorithms,
we spend a little bit of time on genomics. You know,
as I think about your data business, can we just
talk a little bit about how how that's been built
and scaled and you know, why are the big large
pharma companies coming to you and kind of what the
value proposition that that data offers them.
Speaker 3 (33:03):
Yeah, the single most unique part of our data set
is that it's the first time somebody's actually aggregated from
the clinical process, like in other words, from patients being
treated yesterday, last.
Speaker 2 (33:18):
Week, a month ago.
Speaker 3 (33:20):
A really large number of files that are that have
matched rich molecular data typically you know DNA and RNA
and tumor profiling and germline profiling, and it's really rich
molecular data connected to.
Speaker 2 (33:39):
Outcome and response.
Speaker 3 (33:40):
Data for those patients. So I know, like what drug
did these patients take and how do they respond? But
I also know their molecular profile, not at a superficial level,
but at a deep level, like maybe I want to
understand the transcript Domak or RNA insights at a deep
level for that population. Once Tempest imassed enough of these files, right,
(34:01):
we were able to do something that historically people couldn't do,
which is we could go to biotechs and pharmaceutical companies
and says, what research questions are you trying to answer?
What exact patient population are you about to run a
study in and can we see if we have that
population in our data set. So somebody might say, well,
I think patients with triple negative breast cancer and a
(34:23):
pick three C A and RB one mutation who took
a checkpoint inhibitor are going to respond this way, and
so that's what they're Maybe they have a drug that
inhibits something, and so they couldn't find a thousand patients
that have triple negative breast cancer with a pick three
C A and RB one mutation that took a checkpoint
inhibitor for which they have response data, and we have
that and so all of a sudden they can license
(34:45):
that data on a d identified basis and answer really
very specific, profound questions.
Speaker 1 (34:51):
Well, you guys have had some huge contract wins. Can
you maybe talk a little bit about, you know, how
you see that landscape unfolding in the future. I mean,
there's there's a point where I think they're there's a
marginal number of contracts left, isn't there.
Speaker 3 (35:03):
Yeah, I mean, well we've only signed really kind of
you know, two or three, you know, one hundred million
dollar plus contracts, and I would suspect there'll be one
hundred biopharmer companies that sign that level of agreement at
some point. So I think in terms of the overall space,
it's still pretty undeveloped. And that's because most people still
(35:25):
haven't realized the value of this data. Most people, and
you know, there are thousands of biotechs and farms of them,
there's you know, one hundred at real scale, meaning they
have a market cap of a billion dollars or more.
Speaker 2 (35:37):
They have real money.
Speaker 3 (35:38):
Most people don't. Are still literally going to have the
same failure rate of Phase two's the same failure rate
of phase threes. And that's not okay. Like in other words,
if there's data out there that some people are using
to reduce those failure rates to increase the probability of
(35:59):
technical and regulatory success, and you're not doing that, it's
just a matter of time, right, There's a great there's
kind of a great stat I I can't remember. I
don't have it like, but I remember reading it. They
did a survey of companies that in the year two
thousand CEOs Fortune five hundred CEOs. They asked them how
many people thought the Internet was a fad. They looked
(36:21):
at the market cap of the companies twenty years later
that you know where the CEO said it was a
fad versus not and it was almost perfectly correlated, you know,
where those that were like, this is not a fad
had did well and those that thought it was did
not well. And it's the same thing here. I think
it's just a matter of time before I don't know
if they're all going to become Tempest clients, but it's
(36:42):
just a matter of time, I think before people have
to incorporate this kind of data into their discovery all.
Speaker 1 (36:48):
It's almost fair to say it's going to be table
stakes going for.
Speaker 2 (36:50):
I think it's table stakes.
Speaker 1 (36:51):
And then maybe talk a little bit about the competition
that's out there, and you know, maybe what differentiates you
from others. I mean, you mentioned maybe they won't all
go to Tempest, but you know, what does the state
of play look like? From a competitive landscape.
Speaker 3 (37:03):
Well, you know, right now, there's not a lot of
competition in the data space where we operate, so discovery
develop or at least discovery development. I mean, I've said
this before, Like you know, we if we call on
one hundred people to sell them our data, to license
them ourday identified data, you know, ninety five percent of
the time or something, it's never like there's somebody else
(37:25):
that they're looking at.
Speaker 2 (37:25):
It's you know, it's almost.
Speaker 3 (37:27):
When we get to know, it's because they don't have budget,
they're not interested in the project, they're not yet convinced
this would add value. It's not like, oh I was
going to go with you, but I went with this
other person. There's just and that's because it's the president
moment conversation at the present moment. You know, nobody has
our data set that has this much data that can
(37:50):
be interrogated to this manner. And we've also invested really
significant in terms of capital and time. The tools that
make this data useful, that help you analyze the data
are you know, built up at this point. So I
think we can do things today other people can't do.
And that's why folks are licensing our data.
Speaker 1 (38:10):
You know the other piece of that business, I guess
as your clinical trials business, can you can you talk
about the rationale for going into that space as well?
Speaker 3 (38:16):
Yeah, so this was I mentioned the first two businesses
that we could see from inception, you know, the genomics
business of the data business that made sense to us.
The one we couldn't see where these applications that could
only exist by virtue of having this connected marketplace, and
our clinical trials business is one of those. What happened
(38:37):
is is we as we started to get to scale,
as more and more hospital systems, we're saying, I want
to order tempest tests, I want you to be my
sequencing provider, especially we work with a very large percentage
of the NCI cancer centers. As more and more of
those hospitals were connecting, we began to realize like, oh
wait a minute, Like we have all this real time
(38:58):
data for these patients flowing through our systems, and what.
Speaker 2 (39:02):
Could we do with that?
Speaker 3 (39:03):
And one of the first things we realized we could
do was spot these patients that were a perfect fit
for a clinical trial that weren't going to go on
that clinical trial either because the provider didn't know the
doctor didn't know or because the clinical trial wasn't even
offered at that site, and we thought, can we close
that gap? And unfortunately, in oncology, one of the reasons
(39:26):
that the death rate is still so high. It's six
hundred thousand people died last year, give or take, it's
almost the same number twenty five years ago. We don't
do a good job in this country of getting patients
on clinical trials. Our enrollment rate is something like three
to five percent, and we know that patients do better
on clinical trials, the outcomes are materially better. So we thought, well,
(39:50):
we can help close this gap. We can spot these
patients that are a perfect fit, and then let's go
to the people that are running these trials and let's say, look,
if I find you a perfect patient in real time,
well you open up that clinical trial within ten days
at that site.
Speaker 2 (40:04):
And so that's.
Speaker 3 (40:05):
The model we've put forward. We've been doing this for
about three or four years. We have we've been onboarded
hundreds of clinical trials into the network. We've already enrolled
over a thousand patients onto trials that were at community
sites that never would have even been able to avail
themselves to that trial. So it's you know, it's starting
to get to some scale.
Speaker 1 (40:25):
Now we're people in the marketplace already. Do our competitors
in the marketplace already doing some of that? I mean,
I remember when you know, lab Corp brought Covance. There
was this idea that, well, we were running the tests,
we're going to be able to push people over to
Covance and put them in the trial, you know, pretty quickly.
Or was the reality that they just weren't that sophisticated
at doing it.
Speaker 3 (40:45):
Yeah, I mean, you know, in the world in which
we operate, which is basically looking at rich molecular data
because most of the trials that we are focused on
are connected to some kind of biomarker or some eleclar
insight and the ability to look at clinical data and
know that this patient is a fit. You have to
have all the tools we built a structure clinical data
(41:07):
and all the tools we built to sequence patients. Those
tools don't exist in a lab Corp. They don't structure
clinical data like, that's not what they do, So you
wouldn't be able to do everything we do. They might
be able to say, hey, if you're looking for patients
that you know, have this particular blood marker. I can
see those patients, so that's probably true. But the kind
(41:28):
of end to end, you know, I know that this
patient is a perfect fit for this trial requires the
ability to have access to that clinical data for that patient,
structure that clinical data in minutes, hours, and then have
structured the inclusion or of the trial and married the two.
(41:48):
And I'm not aware of anyone else that has our
scale or does that.
Speaker 1 (41:51):
That's great.
Speaker 3 (41:52):
Yeah, there's some farmer companies that have tried it in
very limited ways that I mean, no artists had a
just in time program. But it's more about rapid enrollment
of a rapid site, you know, selection and enrollment, scaling
the site versus Yeah, it's not it's not about doing
all the stuff we do define the patients.
Speaker 1 (42:11):
So if we think about the company and where you
want to take it, you know, what's a long term
view of where you want to be in five to
ten years.
Speaker 3 (42:19):
I mean, our our goal is the same now as
it was you know then, which is we want to
actually you know, fulfill the mission of bringing you know,
added diagnostics and healthcare, and we want to help patients
live longer and healthier. Lives and we want to basically
build a business that you know is important, uh and
and and significant. So you know, we're nine years into
that journey, and the problem is big. Our solution is
(42:46):
still early in terms of rolling out, and so I think,
you know, we were like heads down trying to actually
get the scale. We've achieved some scale, but you know,
if you get to a billion dollars of revenue in
a in a five trillion dollar space, you know you're
you're you know, quite still quite small. So I think
we're very focused on, you know, really working on the
(43:11):
problem set and deploying our solution. And one of the
best parts of our culture and we have, you know,
an incredibly talented team that that has is so you know,
multidisciplinary and has so much you know, unique expertise. But
one of the best parts of our team is that
we really are problem oriented. Our culture is problem oriented.
(43:31):
It's like, what's the problem, what's the solution? How do
we how do we you know, put our heads down and.
Speaker 2 (43:36):
Figure it out?
Speaker 1 (43:37):
So maybe that leads into where you're spending your time.
You know, you had the IPO earlier this year. I'm
sure that took a lot of brain power. You know,
where are you focused today? And you know where do
you want to spend your time over the next couple
of years.
Speaker 3 (43:50):
Yeah, I my time is pretty much the same. I mean, yeah,
Certainly there was a month or so where it was
a disproportionate amount of time talking to investors just because
there's the sheer volume of people you had to talk
to when you go public. But you know, my time
basically today is is identical to what it was a
year ago. We always talked to some number of investors
(44:13):
because we were you know, it was very expensive to
raise the capital we needed to raise to build Tempest
was not a it wasn't a small effort, so we
were always meeting with investors. We were always having those
kind of conversations. You know, we've had an independent board
for a long time. We we've we've always behaved and
had a certain amount of financial rigor and discipline and
(44:33):
structure as if we were public, just because that's kind
of how we think, that's our orientation, and so, you know,
the things I do today are pretty much the same
as the things I did I did a year ago. Certainly,
the the the notoriety is higher, and you know, your
(44:53):
results are public, and so you you you're you're kind
of inevitably more focused on h a quarter to quarter mentality.
But I try my hardest to not let that take Well.
Speaker 1 (45:06):
I noticed in your shareholder letter you said you guys
are going to be focused on the long term, and
that's that's diametrically opposed to Wall Street in many cases.
Speaker 3 (45:14):
It really is. Oh my god, it really is. I mean,
and when we tell people this, and I think as
a we we we we really. I think the best
way to behave as a as a as a public
company in many ways is to treat your shareholders the
same way you would treat them in as as they
(45:34):
were private company. Like in other words, you have a board,
you have shareholders, there's a there's a high degree of
transparency and a flow of information because you have respect
for those shareholders. And then when your public things kind
of change, right, you behave differently. One of the best
examples is, you know, when companies go public, they give
a range, like my hair, like my results are range.
(45:57):
Private companies don't have ranges, right, I mean, if you're
if you have sent a private board or whatever, we're
here's my number. Ye, if you if you're a private company,
you said, here's my range, they'd be like, what, pick a.
Speaker 2 (46:08):
Number, Like, just pick a number.
Speaker 3 (46:10):
So we do the same thing we say to We
said to our shareholders, this is what we expect to achieve,
this notion of like you know, we're gonna you know,
beat and raise and sandbag and this and that. Like
I think it's just a lot of gamesmanship that doesn't really.
Speaker 2 (46:27):
Change the outcome.
Speaker 3 (46:28):
And in the end, in the end, shareholders are are
quite sophisticated and smart. If you have a business that's growing,
that's stable, that that generates leverage, that looks like it
has high long term growth prospects, you're gonna get great
shareholders and you're gonna have a great stock price. And
(46:49):
you get these short term vacillations where it goes up
and it goes down and it's overpriced on our price.
But really, if you're not selling in the short term
or buying in the short term, it's just noise. And
you know, I try not to think.
Speaker 1 (47:03):
About it well as an analyst. I try not to
make these conversations about, you know, the financial journey. But
one thing you just mentioned turned on a light in
my head, which is just the leverage in the business model.
And I know you guys have talked about getting too profitability,
but what are the key drivers to that?
Speaker 3 (47:18):
Yeah, I mean the biggest driver is when you have
significant growth like we do, and you're we're fortunate that
our growth rate is pretty high, right it's twenty five
to thirty percent kind of general speaking massive, and you
have a relatively high margin. Our margins are fifty five
sixty percent right now.
Speaker 2 (47:35):
So if you have.
Speaker 3 (47:36):
Pretty good growth and pretty good margin, then generating leverage
is purely a function of what you reinvest in the business.
So if you look at Q one to Q two
of this year, and this is directionly right around the
numbers in front of me, but I think we had
like kind of negative forty three million of adjustity with
don Q one and maybe negative thirty one and Q
two or something, and so we generate about twelve million
(47:59):
of improvement, and that was almost entirely a function of
we had growth in gross profit dollars and we didn't
put them all back in the business. And we told
people that's a trend we expect to continue. Obviously, the
numbers will fluctuate from quarter to quarter, but we expect
there to be growth. We expect there to be gross
dollar growth, and we don't intend to reinvest all those
(48:19):
back in the business. And so we're on this kind
of relatively short path to adjusted ebitdaut break even and
a cash flow break even. Those two, those two should
trend relatively close together, you know, with stock based compets
some other things. You know, companies like ours tend to
the EPs number can lag a bit, but in terms
of free cash flow and adjustedy, but.
Speaker 2 (48:42):
We expect to be on a very short path to
being break even.
Speaker 1 (48:45):
Well that leads me you mentioned reinvestment. You know, if
we think about the technology that's coming to market around AI,
and I'm not an expert in technology, but what makes
you or what gets you excited about what's coming? And
you know, where have you identified air areas of new
technologies that are emerging, and whether it's the large language
models or the next iteration after that, that kind of
(49:07):
get you really excited.
Speaker 3 (49:09):
I think the biggest, look, the biggest lever that gets
me excited is large language models.
Speaker 2 (49:16):
Generally, I like it.
Speaker 3 (49:17):
It's just such a transformational technology. There's really no there's
no other there's like kind of no other technology that
I think for the next four or five six years,
that will that will that will you know, get get
close to it. I mean, this is the great moment
in time when these models. If you look at you know,
(49:39):
the kind of sophistication between a chat GIPT three and
a four to zero, it's mind boggling. And same thing
with Lama, same thing with Gemini, and I would suspect
the same thing will happen again when they launched the
next version of the model. So you have these models
that are getting incredibly sophisticated. You have these the context
(49:59):
wind sort of the kind of the ability to query
data is growing significantly, and the and the and what
ends up happening when you can throw in a bunch
of data into these models and the models are unbluably sophisticated,
you can do really profound things. And so I think
that's what I'm most excited about.
Speaker 1 (50:19):
That's great. The way I like to wrap these conversations
up is really to maybe talk about something that is
personal to you, maybe a life lesson from the guests.
You know, what drives you day to day. You know
you've been successful as an entrepreneur. You know many times
when you wake up in the morning. What's driving you
to fulfill this mission?
Speaker 2 (50:39):
Well, this is a this is a big you know,
this is a big one.
Speaker 3 (50:42):
So I mean, I think unlike other businesses that I've
been a part of, all of which you know, I think,
we're amazing experiences. You know, you don't have to find
motivation when you're in cancer or other disease areas, Like
the motivation is there. So one of the best parts
of the work we do. And I say this also
(51:04):
quite often, is that one of the coolest parts about
Tempess is if you were being entirely focused on economics
versus entirely focused on being altruistic, you do the exact
same thing, meaning this very if I said to people
to Tempa's just singularly focus on patients, don't care it
(51:25):
all about money, that would produce the exact same outcomes
as if I said, be singularly focused on money, don't
care at all about patients.
Speaker 1 (51:32):
Like, in other words, pretty rare that they are aligned.
It's pretty rare they're aligned.
Speaker 3 (51:35):
Right, I mean, so we live and we happen to
have a business where they're entirely aligned, and so it
allows us to I think, you know, have that kind
of perpetual motivation because everything we do is is benefiting
the business and patients.
Speaker 1 (51:50):
That's great, So I think we'll call it there. That's
Eric Lefkowski, CEO and founder of Tempa's AI. Thank you
so much for joining us for our latest episode. Please
make sure to click the thunt A button on your
favorite podcast app or site so you never missed a
discussion with the leaders in healthcare innovation. I'm Jonathan Palmer,
and you've been listening to the vanguards of healthcare podcasts
by Bloomberg Intelligence. Until next time, take care,
Welcome to another episode of the Vanguards of Healthcare podcast,
where we chat with the leaders at the forefront of
change in the healthcare industry. My name is Jonathan Palmer,
and I'm a healthcare analyst at Bloomberg Intelligence, the in
house research arm of Bloomberg. We're happy to welcome Eric
Lefkowski for today's episode. He's the founder and CEO of
Tempa's AI. Prior to founding Tempess, he has an impressive
(00:43):
career as an entrepreneur, founding several successful companies, most notably
group On. His work at Tempes is transforming healthcare through
the application of artificial intelligence into precision medicine. Thanks for
joining us, Eric, Why don't we get started with an
overview of Tempest and kind of how you're changing the
health care landscape.
Speaker 2 (01:01):
Thanks for having me. So.
Speaker 3 (01:02):
Tempest is focused on this idea of bringing technology and
in particular artificial intelligence to diagnostic by essentially combining a
molecular result, so sequencing a patient producing some result, and
then connecting that to clinical data for that patient so
you can really personalize the result around that particular patient.
(01:25):
The benefits you get are pretty extreme. So in the
case of sequencing cancer patients. If I find a mutation
and I might recommend a drug, It doesn't help to
recommend a drug if the patient took that drug in
a prior line of therapy and failed, just as it
doesn't do any good to recommend a clinical trial that
the patient can't enroll in. So, for example, if I
knew my patient was a smoker, why I recommend a
(01:47):
clinical trial. We're the first exclusion criterias. You can't be
a smoker. So we could see, or I could see
when I started Tempest about nine years ago, that we
were likely going to enter this period where you'd be
able to generate enormous amounts of low cost, high quality data,
in particular because we were going to start sequencing most
(02:08):
cancer patients, and that this data would pile up in
huge ways and it would become increasingly complicated for physicians to, like,
in real time, interpret that data and then make decisions.
And so we thought, what if you could collect clinical
data for those patients, sequence those patients, and then put
the two together so you could kind of produce the
(02:30):
insights as part of the report itself. So if you
had something intelligent to say about that patient, you'd say
it in the report, and so that's how Tempest was born.
And you know, we went out started convincing some hospitals
to try to work with us, opened up a lab,
and pretty early on you could see the benefits of
(02:51):
our solution, Like it became apparent that these kind of
comprehensive tests that were also intelligent and more personal were
going to be you know, benefit official to physicians and patients,
and so they began ordering them at scale and we
grew and during that journey we also realized that not
only could you generate really personalized reports for patients, but
(03:14):
that when you connected clinical and molecular data, if you
de identified that data, you could produce really interesting data
sets for researchers and provide the same benefit to researchers,
whether they are researchers at economic medical centers or researchers
at biotech companies or researchers at pharmaceutical companies. And by
helping them do research, they would be more efficient, they'd
(03:35):
make better drugs and fail less often.
Speaker 1 (03:38):
That's a great overview to start. I mean, if we
rewind back to that initial inception of the company, did
you have that idea of marrying the two pieces the
genomic data and the and the you know, the longitudinal
data post post the test.
Speaker 3 (03:51):
Yeah, almost from from day one. We if you go
back and look there, there's this really cool thing hold
the wayback machine where you can kind of like look
at Internet sites at various moments time, you know, and
you can see us talking about big data and data
science and AI and combining clinical molecular data. Really back
seven or eight years ago, so very early after we
started a company, we began thinking about how this model
(04:14):
would evolve. What we didn't know back then is that
in order to bring in all this clinical data for patients,
we were going to inevitably become connected to thousands of hospitals.
There's about fifty five hundred hospitals in the United States,
and we're connected to something like twenty three hundred some
number like that. So it's approaching half of all hospitals
(04:36):
the United States. And so when you're connected to these hospitals,
it's like a cable company, right, It's like laying cable
to a house. All of a sudden you have a
direct connection to that house. And so by virtue of
these connections, we're able to pull data out of these
electronic healthcare record.
Speaker 2 (04:52):
Systems or EHR systems.
Speaker 3 (04:53):
We're able to generate diagnostic insights and put them into
the EHR system all through these electronic conections that we built,
and so that's allowed us to have a third business
line because today TEMPS is really made up with three products.
Genomics where we sequence patients and bill insurance, our data
business where we license the identified data, and our third
(05:14):
business units called applications, where we basically leverage these connections.
We have the thousands of hospitals and deploy purely algorithmic
diagnostics in real time to be able to close care gaps,
or route patients to the right clinical trial, or be predictive.
That business line we couldn't see if you go back
in time, like we could see that marrying clinical molecular
(05:36):
data would be powerful, but we couldn't see that you
could actually wire up the US healthcare ecosystem and then
leverage AI to really improve outcomes.
Speaker 1 (05:45):
There's a lot to unpack there, So maybe I'll start
with the kind of the core genomics business when you
first kicked off, You know, what was the moat that
you were trying to build VISA via the competition that
was out there. I mean other people were doing you know,
multi gene panels. You know, how did you really convince
the first practices to use Tempest? Was it the data piece?
Was that the most compelling part of the sale?
Speaker 2 (06:08):
Yeah?
Speaker 3 (06:08):
You know, when we first started, we knew we wanted
to marry clinical molecular data, but we didn't know we were
going to sequence patients. We actually thought, let's not open
up a lab, let's not sequence patients. Let's go to
the people that sequence patients and tell them that we're
pretty convinced that there's this emerging problem that's only going
to get worse, and let us be their tech partner.
(06:30):
And we went to a few of the big labs
at the time. They were entirely unwilling to give us
any data. They were basically like, you know.
Speaker 2 (06:40):
There's no way I'm going to give you any data.
Speaker 3 (06:42):
And so we said, well, without the data, we can't
actually solve this problem. So then we had to go
to hospitals and say, you know, let us sequence your patients.
I mean, we were fortunate, I think that we were
able to raise enough capital early on to build a lab,
to hire really credible people to join the company that
knew how to sequence patients that knew how to develop essays,
(07:05):
and so that credibility allowed people to feel comfortable that
we were going to be able to deliver. And then
obviously there are certain hurdles you have to get over, like,
for example, being accredited by CAP.
Speaker 2 (07:15):
And having a CAP clear certified lab.
Speaker 3 (07:17):
So we achieved that, but early on, the bigger hurdle
wasn't convincing people to let us sequence. I don't think
they weren't that happy with their existing sequencing partners. The
biggest hurdle was saying you should give us clinical data
for the patient to make the sequencing tests smarter. Back then,
you know, eight or nine years ago, people were very
(07:37):
adverse to turning over any data. Today that has almost
entirely changed. Now people recognize that first of all, it's
not their data, it's the patient's data. And second of all,
if this data can be leveraged to produce a better outcome,
that's all that matters.
Speaker 1 (07:54):
So it's funny because I would think in some ways
that today, where the value of data is so much
more imperant, actually be more challenging to get that data
from from providers. Do you see any hurdles or challenges
in that respect.
Speaker 2 (08:07):
No, I mean we have so much data.
Speaker 3 (08:08):
We have, like you know, we're approaching half of all
cancer pation data United States in some way, shape or form.
So we have a huge amount of data. So I
think people are very comfortable giving us data that We're
going to do everything we can to make sure it's secure,
to maintain privacy, to adhere to you know, regulations and
compliance and that, and that only used a de identified
(08:29):
version of that data for the benefit of patients. So
I think people are seem to be very willing to
contribute data to our platform. But certainly, you know, as
more and more people realize the value of this data,
you get more competition. And so the competition has grown,
and I suspect it will continue to grow.
Speaker 1 (08:47):
Yeah, I mean I was thinking about that in the
through the lens of you know, we've seen US Oncology
continue to consolidate practices and Kora has been buying one oncology.
Is that is that a hurdle for your business?
Speaker 2 (08:58):
It's not.
Speaker 3 (08:59):
It's not a hurdle really at all. I think it
actually could be an accelerant in that you know, this
consolidation will likely lead to more strategic moves that that
could benefit us. But as it relates to the testing side,
those decisions are almost entirely made at an individual oncologist level.
So even though you might have practices that are that
(09:22):
are part of one academic medical center system or owned
by a conglomerate, the decision as to which sequencing lab
I want to use is still almost entirely made one
on collegist at a time. So all the big labs
are effectively kind of five big next generation sequencing labs
and Cancer, Tempest, Foundation, Medicine, Caris, Garden, Ntera, and almost
(09:47):
all of us have you know, large sales forces that
are calling on doctors one at a time explaining the
benefits of our platform.
Speaker 1 (09:55):
And so when your doctors call, you know, how do
they differentiate between some of those other players.
Speaker 3 (10:00):
Our big advantage has historically been it is today technology.
So when we're competing with other people, if we're winning,
it's because we have a technology platform that delivers more
personalized results. The insights are more integrated and easier to
use by physicians, They're more comprehensive. Our platform is more comprehensive,
(10:23):
and so physicians feel like they are getting the totality
of information they need and they're able to treat their patients.
I mean, that's typically why doctors would choose US versus
somebody else. We also happen to have at this point
as one of the largest sequencers of cancerbations in the world.
We have a very comprehensive portfolio from inherited cancer risk
(10:44):
to solid tumor profiling, to liquid biopsy to minimal visual
disease detection and monitoring.
Speaker 2 (10:49):
So it's a pretty comprehensive.
Speaker 1 (10:50):
Port I'm smiling because you kind of led right into
my next question, which was just to really talk about
the milestones and from the portfolio of offerings that you have.
If you think about where you were to start to
where you were today, you know, you listed all those
different offerings, whether it's the hereditary or MRD, you know,
which are you most excited about today?
Speaker 3 (11:11):
We are. The biggest part of our platform today is
in a conference of genomic profiling or therapy selection. So
that's you know, a patient comes in, they're diagnosed with cancer,
they need to be sequenced quite often to figure out
what drug to give them, so therapy selection, and that's
where we have the biggest part of our portfolio. Both
in the case of solid tumor profiling, so like I
(11:32):
have a biopsy and I'm sequencing the tissue of that
tumor versus like liquid biopsy, which is I'm kind of
pulling those molecular insights from blood. And about a quarter
of our volume is liquid biopsy versus solid tumbor profiling.
So at our scale which you can seek US Republic,
we're very large. We're very large in both in both areas,
(11:52):
we do quite a bit of inherited risk profiling, so
that piece we also do quite at some The newest
offering for US, and the one that obviously has significant
potential going forward, is MRD we launched. So for those
that don't know, you know, there's been this emerging space
(12:13):
within cancer of monitoring and minimal situlecies detection, and this
is the idea that from again either blood or tissue,
I can effectively, you know, track patients over time and
try to figure out if their cancer has come back
before I might see it in a scan. And what
you see is you'll see certain patterns, whether they're methylomic
(12:35):
based or whether they're from actual like you know, fragments
of DNA circulating in the blood. You'll see signatures that
something a mutation looks like it's circulating in the blood
before I can see it in a scan. And this
is because if you think about it, you know, fragments
of DNA circulating in your blood. If you're looking at
(12:56):
a particular scan, looking at an image of a scan,
you might need to get a large volume of cells
together to be able to see something. You know, whether
it's a million or a billion, right, you need you
need a very large polk. So a lot of these
tests that are coming to the market right now are
picking up signs of cancer, you know, five months, six months,
(13:19):
seven months, eight months before you would see it on
a scan. So they're quite powerful in terms of their
ability to detect cancer earlier than a patient might normally
see it. And there is significant benefit people have published
on this, and there's more that will come about about
being able to intervene when you when you can see
that a patient has actually recurred before you might see
(13:41):
it in a scan. We're excited about our portfolio because
we launched both what's called tumer naive and tumor informed.
Tumor naive is essentially saying, right from the blood, we're
going to detect those signatures of early disease. Tumor informed
is saying we're going to basically, you know, take up
take a piece of that tumor, build a bespoke assay
(14:02):
of some kind, and then track it over time. They
both have benefits, right, just like there are benefits to
solid tumor profiling versus liquid. The benefit of the tumor
naive or the liquid profiling is that it's very logistically easy.
Speaker 2 (14:18):
You draw blood, you can run a test.
Speaker 3 (14:21):
The benefit of tumor informed is that it tends to
produce higher sensitivity and specificity and improved detection. And so
we have our own tumor naive assay that we launched
in colorectal cancer, and then we have a partnership with
a company called Personalis where they've launched an assay tumor
informed assay in lung cancer, breast cancer and IO response.
Speaker 1 (14:43):
Is the corectal assay just the beachhead to other indications.
Speaker 3 (14:48):
Over time, we expect to take our turmer naive MRD assay,
which right now is in colorectal, and move into other
disease areas. Some of those are easier to move into
than others. It's kind of tracking those signatures. The blood
is a function of you know, different tumor types of.
Speaker 1 (15:03):
Behavior, how much they're shedding that the end of the
blood like prostate doesn't do very well in these assays.
Speaker 3 (15:08):
Some some subtype shed a lot and some shed very little,
and that tends to be the distinction. So we we
spent a long time building our assay in in in
c r C, which we're quite proud of. It's got
great stats, and but we're on a constant effort to
you know, fine tune our bioinformatics and our ability to
(15:29):
boost sensitivity and specificity and the lower reality or limits
of detection.
Speaker 1 (15:33):
So what's what's the drivers behind that? Is it on
the informatic side like you just mentioned, or is it
on the chemistry side.
Speaker 3 (15:38):
It's it's both, It's both. The informatic side can only
take you so far. So at some point you boost.
Speaker 2 (15:45):
In chemistry, you can, but you can get far. You
can get far with informatics.
Speaker 3 (15:49):
I mean, you know, and you and and there's a
real benefit informatics with scale, right if you think about it,
informatics is you know, it's math and and and building
models and the more data you feed into a model,
the better the model gets. That's why you know chat
GPT four hosts better than three, and three is better
than you know, on and on. So I think we
(16:10):
will You'll see significant improvements over time by our ability
to fine tune the biotraumatics. But net net, if you
want to really have quantum loops forward, you need improvements
in chemistry, need to be able to basically pull signals
out at higher fidelity.
Speaker 1 (16:23):
And how long do those those milestones take to reach?
You know, if you think about your product roadmap on
the chemistry side, you know the next iteration does it
take twelve months? Does it take years?
Speaker 3 (16:34):
Well, we benefit, I think Tempest benefits. You know, given
our scale, we're fortunate there's a lot of other labs,
both private and you know, you know companies as well
as parts of public institutions like academic medical centers that
want to partner with us, So we benefit. We don't
have to develop all of these advancements on our own.
(16:55):
We have people coming to us constantly that you know,
have have novel breakthroughs that they want to see us
bring to market, So it'll be a combination of both.
Speaker 1 (17:04):
Well, maybe is a good segue into your R and
D organization, can you can you maybe talk about the
size and scope of what you're focused on.
Speaker 3 (17:11):
Yeah, I mean you can look so first of on
our in our in our public filings, we break down
R and D by you both tech and by science.
But I mean it's directionally. We have if you look
at headcount, somewhere between nine hundred and one thousand people broken
down between product engineering and our science teams. And the
science teams are maybe about three hundred PhDs and mds combined,
(17:33):
so it's a fairly large scientific team. And then it's
you know, five six hundred software engineers and a engineers,
so it's a fairly large technical team as well. The
if you look at our investments over time, and we've invested,
you know a lot of money. It's taking a few
billion dollars to get here, a lot of it's been
in those in those areas.
Speaker 1 (17:52):
So maybe that that leads to, you know, where do
you have to continue to make investments next as you
continue to scale? Is it is it just purely on
the R and D function? And do you have to
build more labs? You know, where do are those investment
dollars going to go?
Speaker 3 (18:04):
We have today three labs, one in Chicago, one in Raleigh,
one in Atlanta, and the capacity of those labs is
quite significant, so we wouldn't have to build more more
like lab space in order to grow quite a bit.
So I think from that standpoint, we've made significant investments
already for us. If we were to build another lab
(18:25):
at some point, it might be geographic, like do we
want to have, for example, a lab out west because
we like the logistical benefits, but we're fairly built out.
Most of our investments are really trying to bring the
benefits of new technologies, in particular generative AI in large
language models to our entire technology stack, both on the
(18:48):
product side and on the science side. And you know,
we're kind of at this really fortuitous moment in time
where you know, if you think about when when Tempest
was started, we could see this day coming, and the
background technologies we had were largely natural language processing and
optical character recognition. Now all of a sudden, you've hypercharged
(19:11):
that with large language models and general of AI, and
so all of a sudden, what you can do with
large data sets is like unimaginable.
Speaker 2 (19:20):
So if you want to.
Speaker 3 (19:21):
Take advantage of that as a company, you kind of
need two things. You need a proprietary data set and
the ability to distribute any insights you get from these models,
and Tempest happens to have both at scale. You know,
we have two hundred and fifty petabytes of proprietary data
that we can use to train models, and we're connected
to know almost half of all hospitals the United States,
(19:41):
So we have I think we're in a good position
to benefit from genera of AI, and so we spend
a lot of our time and money and energy on
how do we really bring you know, LLLMS to our
entire and so.
Speaker 1 (19:55):
Going forward, if I think about you know, you delineated
between the product and maybe the providers who are we're
using some of these AI tools, and then on the
data side, where you know, the biopharma companies are utilizing
your stack. You know, where do you think there's going
to be the more profound inflection.
Speaker 2 (20:11):
I think there will be in the near term.
Speaker 3 (20:15):
I would suspect both will be kind of similar in
terms of the utility of these technologies. I think in
the semi near term. And again I'm not able to
really predict with specificity when this is going.
Speaker 1 (20:29):
To happen, but we won't hold you to it.
Speaker 3 (20:31):
Maybe yeah, maybe in the next three to five years
or something, I think there will be truly profound impacts
on physicians and how they behave. I think companies like
ours and others will arm the average patient with enormous
(20:53):
insights about their care, and it will literally, you know,
it's it's going to be like I've never said this before.
Speaker 2 (21:02):
I just thought of it right now. If it's gould
be a horrible analogy, but I'll use it.
Speaker 3 (21:04):
It's a little bit like, you know, you show up
in a city like New York, okay twenty five years ago,
you get in a cab and they take you where
they take you. But now if you don't, you pull
up your phone, you go to Google or ways, and
you can tell them what route and he should go
and you're right, or at least as right as you
know you can be. So all of a sudden, the
(21:24):
cab driver can't say, oh, you know, we're.
Speaker 1 (21:27):
Going to take a left here.
Speaker 3 (21:28):
Yeah, don't tell me what to do. You don't know
what you're doing. You're like, no, I actually do know
what I'm doing. You know, Google's telling me you're going
the wrong way, and that I think is going to
be very similar to how the average patient, not just
an oncology but in neurology and cardiology and immunology is
interacting with their doctors. And so doctors have to be
armed with similar tools or they're going to.
Speaker 2 (21:51):
Really be at a disadvantage.
Speaker 1 (21:52):
That's fascinating. I Mean one of the things that I
thought of while you were speaking there was, you know,
just how you marry what you're doing with guyidelines and
you know NCCN guidelines kind of have a treatment pathway.
You know, where do those intersect with the data that
that you're providing and do they ever conflict?
Speaker 3 (22:11):
Yeah, I mean they don't conflict because we're not really
in the business of saying, you know, our data says this,
so you should do that. You know, there's no like
do this, do that, red, green, yellow, It's more function
of this is the data we have, You as a
physician are free to look at it and to the
extent it helps you make a decision that you should
(22:33):
And I think physicians, at least in oncology, you know
something like two thirds of the time, especially in later
stage disease, so people with stage three, stage four, you know,
third line therapy, fourth line therapy. There's really no guideline
to guide you. And in fact, I tell this to
people all the time. You know, if you if you
look at, for example, a pancreatic cancer patient at the
(22:53):
one year mark, who's in let's say, you know, third
line therapy, the I don't know the overall survive, the
two year over survival or three or like five percent.
So what that fundamentally tells you is that the standard
of care is going to produce death ninety five percent
of the time. So to a patient, that's not a
good that's not something one should fill that out. Yeah,
(23:16):
you'd prefer not to follow that standard of care, right,
You're like, I want to so quite often, unfortunately, if
you look at in the case of cancer, there's roughly
one point nine million new diagnoses a year and basically
about six hundred thousand people die a year, so roughly
a third of cancer patients die. And you know, you
don't want to follow the standard of care when you're
in that group, when you have a very high likelihood
(23:38):
of doing well, like you know, stage one ear positive
breast cancer, where you have whatever ninety nine percent five
years overall survival, then you want to follow the standard
of care. So I think we're arming physicians with the
kind of data they would need that when there's really
no good guideline, they can at least make an informed decision.
When the guideline should produce a good outcome, they're typically
(24:00):
following that guideline. So we're not in contradiction. What is
going to be in contradiction at some point is that
the guidelines will be unable to keep up with the
emerging state of evidence and information unless they too are
leveraging AI. So, for example, we do pharmacogenomic profiling of
(24:22):
patients that have major depressive disorder, bipolar disorder, severe anxiety.
So there's like seventy million people in this country that
have some form of anxiety. And we know that there
are certain antidepressants that aren't metabolized by the body in
certain ways. So for example, if I'm a SIP two
D six for metabolizer, I shouldn't be taking this class
of drugs. Fairly soon, we're going to also know that
(24:45):
if I'm a cancer patient taking a targeted therapy or
an immunotherapy or a chemotherapy, it is having this impact
on how I metabolize these drugs. So now you're going
to have to ask oncologists to think about the drug,
the prescribing, what it means for your depression, how they
should think about changing your dose so that you're not
(25:06):
experiencing adversements. It's just too it's overwhelming, and you know,
you can imagine a world where the number of data
points that a doctor has to consider go from pick
a number ten to a thousand and they just.
Speaker 2 (25:21):
Can't do it.
Speaker 1 (25:21):
It's impossible.
Speaker 3 (25:22):
It's impossible, and the guidelines would go from the guideline
And if you look at a non small cell lung
cancer guideline, it's three hundred pages, four inner pages massive,
So what's it going to go to ten thousand pages?
It's no longer a guideline unless you know, there's too many,
too many branches of the tree.
Speaker 1 (25:39):
So that's a good segue into you know, your your
other opportunities. You mentioned the neuropsychiatry. You know, where else
are you focusing this model? You know you've talked about
cardiology as well. Can we maybe use some other examples
of where you think the treatment paradigm is going to
change using your platform?
Speaker 3 (25:56):
Yeah, I mean cardiology is a great example. What we
also have products radiology and pathology, but cardiology is a
great example. In each disease area, if you're in the
business of trying to bring the benefits of artificial intelligence
and technology to diagnostics, you have to go disease by
disease and pick the diagnostic you think is most central
to how patients are treated. So in cancer, that's genomic profiling, because.
Speaker 2 (26:19):
That's that's kind of what it is.
Speaker 3 (26:21):
But in cardiology it's actually a kind of good old fashioned,
boring ECG. Right, that's the thing that most often is
frontline you know, twelve lead ECG or which also called
an EKG. So we began thinking about how would you
deploy technology to that test. And it turns out that
(26:41):
about three percent of patients who have a normal ECG,
and we run about three hundred million, I think, give
or take in this country, year, three percent of patients
are told they're fine, go home, everything's good, and yet
within a year they have a heart attacker stroke. So
clearly things were not fine. The question is what didn't
we see when we did that ECG. And it turns
(27:04):
out that the people that have a heart attacker stroke
typically have some undiagnosed condition, namely things like atrial fibrillation,
hypertrophic card in myopathy, low injection fraction, aortic stenosis, whatever,
pick one of your So we've been working with a
team that has been building models for years that we
(27:24):
brought into Tempest where we can basically predict these things.
And the first of which was our model to predict
atrial fibrillation from a normal ECG, which got FDA approval
like about a month ago. So that's a good example
of you know, you can imagine a world where you know,
hundreds of thousands of patients a year that are part
(27:44):
of a healthcare system, their ECGs will be uploaded every night,
will run that model, and then we'll call that three
percent and say, hey, wait a minute, even though the
doc told you yesterday or this morning that everything looks good,
you actually are at high risk for something, and we
want to run an echo. We want you to take
a stress test, we want you.
Speaker 1 (28:04):
To wear a monitoring at a minimum, go to a cardiologist.
Speaker 2 (28:07):
Go to a cardiologist.
Speaker 3 (28:09):
And you know, when somebody has a heart attack or
a stroke, it costs the system like one hundred thousand
dollars a year.
Speaker 1 (28:15):
So and running the algorithm costs.
Speaker 3 (28:18):
Right now, it costs zero because no one pays for it.
But at some point when it's reimbursed, and it should
be reimbursed, I mean, what will they pay a hundred bucks?
I mean, it'll be minimal compared to.
Speaker 1 (28:29):
The outcome that you're foregoing, right, Yeah, So maybe that's
a good good spot to talk about reimbursement around algorithms.
You know, what's this current state of play? You know,
these are very new, I guess from the perspective of
people who pay, you know, whether that's CMS or commercial
or medicaid. You know, can you talk about that landscape
(28:51):
and how you see that evolving over time?
Speaker 2 (28:52):
Yeah? I mean, and I'm fairly you know, outspoken about this.
Speaker 3 (28:56):
I mean, it certainly benefits tempest, So I acknowledge that
that conflict, but I'm still one thousand percent convinced that
it's right. The US healthcare system, from a financial perspective,
couldn't be more broken. Okay, it's five trillion dollars a year,
and the outcomes we produce and you can like go
to Google and look them up, are not good. I mean,
(29:18):
I don't even know where we are currently in terms
of like mortality.
Speaker 1 (29:21):
Rate behind some Scandinavian country seeing a lot of other people. Yeah, well,
I mean.
Speaker 3 (29:26):
And my guess is, like, you know, we probably spend
as much as the next five countries combined, easily. Yeah, easily.
So this is not where you want to be. And
so how do you solve this problem? Okay, we're not
going to solve it by you know, negotiating drug prices,
and we're not going to solve it by you know,
making small tweaks to Obamacare. These are just not going
to solve our problem. The only solution to this problem
(29:47):
is technology, is data, is AI, and what it really
comes down to is using technology to remove waste.
Speaker 2 (29:57):
So what's waste?
Speaker 3 (29:59):
Okay, so we used to, you know, go to movie
stores to get movies.
Speaker 2 (30:03):
Now that's not a bad thing. I love the Blockbuster.
Speaker 3 (30:05):
But someone had to build a Blockbuster. Someone had to
make a video, ship the video, put the video on
a shelf, I had to go drive my car, pick
the video up, put it in the VHS system whatever.
It was like a lot of stuff.
Speaker 2 (30:17):
Went into that.
Speaker 3 (30:17):
Today I can technologies allows me to watch a movie
on demand and none of that stuff has to exist.
No buildings, no driving, no whatever that's waste. So we
need to do the same thing in healthcare. We need
to basically, and payers aren't good at this, and the
government isn't good at this.
Speaker 2 (30:34):
We need to.
Speaker 3 (30:35):
We need to basically find a way to pay for
AI insights, pay for diagnostic insights, pay for the things
that spot the problem early on, so we can avoid
all the downstream cost, which is typically you know, multiples
of the other.
Speaker 1 (30:54):
So how do you actually catalyze that, you know, how
do we get that to happen on the hill?
Speaker 2 (30:59):
I think it takes time. First, I think it's inevitable.
Speaker 3 (31:01):
Like, you know, there was a time when no one
thought Blockbuster would be ending, but Blockbuster and then it's gone,
and I think it's the same thing here, Like there's
no doubt in my mind. Eventually, you know, the CMS
and payers will have to pay for algramic diagnostics, they
will pay for next generation sequencing. Broadly, they'll realize there's
(31:23):
still payers in this country today, massive large payers that
don't pay for sequencing, so they'd rather a patient not
be sequenced, get the wrong drug, including drugs that cost
like one hundred thousand dollars a year, not do well
be in the ICU all that kind of bad stuff
instead of just paying for a two thousand dollars test
that some percentage of the time not small, mind you, like,
(31:45):
if we're not like in the twenties, twenty percent of
the time we route patients to an entirely different therapeutic
by virtue of that sequencing result.
Speaker 2 (31:52):
So I think it's just.
Speaker 3 (31:54):
Takes It takes time. But I suspect sequencing will be
broadly reimbursed, and I suspect that we will find a
way to pay for EI insights and will pervasively deploy them.
And that's how you'll start to see. You know, it
may be, for example, that the diagnostic category generates an
extra one hundred billion of revenue, right which would be seismic,
(32:18):
but you'll see a trillion dollar reduction in health care expense.
And it's it's like it's that Proflee.
Speaker 1 (32:25):
I agree with you. You know, I've seen in my career.
If I think back to non invasive prenatal testing, you know,
the adoption there took probably almost a decade. You know,
if I think about sequenomeen, maybe twenty twelve. Hopefully this
time it happens a little bit quicker. But I think
the jury's out. Well, maybe that leads us to the
data business because we've spent a lot of time on that.
(32:46):
We spend a little bit of time on the algorithms,
we spend a little bit of time on genomics. You know,
as I think about your data business, can we just
talk a little bit about how how that's been built
and scaled and you know, why are the big large
pharma companies coming to you and kind of what the
value proposition that that data offers them.
Speaker 3 (33:03):
Yeah, the single most unique part of our data set
is that it's the first time somebody's actually aggregated from
the clinical process, like in other words, from patients being
treated yesterday, last.
Speaker 2 (33:18):
Week, a month ago.
Speaker 3 (33:20):
A really large number of files that are that have
matched rich molecular data typically you know DNA and RNA
and tumor profiling and germline profiling, and it's really rich
molecular data connected to.
Speaker 2 (33:39):
Outcome and response.
Speaker 3 (33:40):
Data for those patients. So I know, like what drug
did these patients take and how do they respond? But
I also know their molecular profile, not at a superficial level,
but at a deep level, like maybe I want to
understand the transcript Domak or RNA insights at a deep
level for that population. Once Tempest imassed enough of these files, right,
(34:01):
we were able to do something that historically people couldn't do,
which is we could go to biotechs and pharmaceutical companies
and says, what research questions are you trying to answer?
What exact patient population are you about to run a
study in and can we see if we have that
population in our data set. So somebody might say, well,
I think patients with triple negative breast cancer and a
(34:23):
pick three C A and RB one mutation who took
a checkpoint inhibitor are going to respond this way, and
so that's what they're Maybe they have a drug that
inhibits something, and so they couldn't find a thousand patients
that have triple negative breast cancer with a pick three
C A and RB one mutation that took a checkpoint
inhibitor for which they have response data, and we have
that and so all of a sudden they can license
(34:45):
that data on a d identified basis and answer really
very specific, profound questions.
Speaker 1 (34:51):
Well, you guys have had some huge contract wins. Can
you maybe talk a little bit about, you know, how
you see that landscape unfolding in the future. I mean,
there's there's a point where I think they're there's a
marginal number of contracts left, isn't there.
Speaker 3 (35:03):
Yeah, I mean, well we've only signed really kind of
you know, two or three, you know, one hundred million
dollar plus contracts, and I would suspect there'll be one
hundred biopharmer companies that sign that level of agreement at
some point. So I think in terms of the overall space,
it's still pretty undeveloped. And that's because most people still
(35:25):
haven't realized the value of this data. Most people, and
you know, there are thousands of biotechs and farms of them,
there's you know, one hundred at real scale, meaning they
have a market cap of a billion dollars or more.
Speaker 2 (35:37):
They have real money.
Speaker 3 (35:38):
Most people don't. Are still literally going to have the
same failure rate of Phase two's the same failure rate
of phase threes. And that's not okay. Like in other words,
if there's data out there that some people are using
to reduce those failure rates to increase the probability of
(35:59):
technical and regulatory success, and you're not doing that, it's
just a matter of time, right, There's a great there's
kind of a great stat I I can't remember. I
don't have it like, but I remember reading it. They
did a survey of companies that in the year two
thousand CEOs Fortune five hundred CEOs. They asked them how
many people thought the Internet was a fad. They looked
(36:21):
at the market cap of the companies twenty years later
that you know where the CEO said it was a
fad versus not and it was almost perfectly correlated, you know,
where those that were like, this is not a fad
had did well and those that thought it was did
not well. And it's the same thing here. I think
it's just a matter of time before I don't know
if they're all going to become Tempest clients, but it's
(36:42):
just a matter of time, I think before people have
to incorporate this kind of data into their discovery all.
Speaker 1 (36:48):
It's almost fair to say it's going to be table
stakes going for.
Speaker 2 (36:50):
I think it's table stakes.
Speaker 1 (36:51):
And then maybe talk a little bit about the competition
that's out there, and you know, maybe what differentiates you
from others. I mean, you mentioned maybe they won't all
go to Tempest, but you know, what does the state
of play look like? From a competitive landscape.
Speaker 3 (37:03):
Well, you know, right now, there's not a lot of
competition in the data space where we operate, so discovery
develop or at least discovery development. I mean, I've said
this before, Like you know, we if we call on
one hundred people to sell them our data, to license
them ourday identified data, you know, ninety five percent of
the time or something, it's never like there's somebody else
(37:25):
that they're looking at.
Speaker 2 (37:25):
It's you know, it's almost.
Speaker 3 (37:27):
When we get to know, it's because they don't have budget,
they're not interested in the project, they're not yet convinced
this would add value. It's not like, oh I was
going to go with you, but I went with this
other person. There's just and that's because it's the president
moment conversation at the present moment. You know, nobody has
our data set that has this much data that can
(37:50):
be interrogated to this manner. And we've also invested really
significant in terms of capital and time. The tools that
make this data useful, that help you analyze the data
are you know, built up at this point. So I
think we can do things today other people can't do.
And that's why folks are licensing our data.
Speaker 1 (38:10):
You know the other piece of that business, I guess
as your clinical trials business, can you can you talk
about the rationale for going into that space as well?
Speaker 3 (38:16):
Yeah, so this was I mentioned the first two businesses
that we could see from inception, you know, the genomics
business of the data business that made sense to us.
The one we couldn't see where these applications that could
only exist by virtue of having this connected marketplace, and
our clinical trials business is one of those. What happened
(38:37):
is is we as we started to get to scale,
as more and more hospital systems, we're saying, I want
to order tempest tests, I want you to be my
sequencing provider, especially we work with a very large percentage
of the NCI cancer centers. As more and more of
those hospitals were connecting, we began to realize like, oh
wait a minute, Like we have all this real time
(38:58):
data for these patients flowing through our systems, and what.
Speaker 2 (39:02):
Could we do with that?
Speaker 3 (39:03):
And one of the first things we realized we could
do was spot these patients that were a perfect fit
for a clinical trial that weren't going to go on
that clinical trial either because the provider didn't know the
doctor didn't know or because the clinical trial wasn't even
offered at that site, and we thought, can we close
that gap? And unfortunately, in oncology, one of the reasons
(39:26):
that the death rate is still so high. It's six
hundred thousand people died last year, give or take, it's
almost the same number twenty five years ago. We don't
do a good job in this country of getting patients
on clinical trials. Our enrollment rate is something like three
to five percent, and we know that patients do better
on clinical trials, the outcomes are materially better. So we thought, well,
(39:50):
we can help close this gap. We can spot these
patients that are a perfect fit, and then let's go
to the people that are running these trials and let's say, look,
if I find you a perfect patient in real time,
well you open up that clinical trial within ten days
at that site.
Speaker 2 (40:04):
And so that's.
Speaker 3 (40:05):
The model we've put forward. We've been doing this for
about three or four years. We have we've been onboarded
hundreds of clinical trials into the network. We've already enrolled
over a thousand patients onto trials that were at community
sites that never would have even been able to avail
themselves to that trial. So it's you know, it's starting
to get to some scale.
Speaker 1 (40:25):
Now we're people in the marketplace already. Do our competitors
in the marketplace already doing some of that? I mean,
I remember when you know, lab Corp brought Covance. There
was this idea that, well, we were running the tests,
we're going to be able to push people over to
Covance and put them in the trial, you know, pretty quickly.
Or was the reality that they just weren't that sophisticated
at doing it.
Speaker 3 (40:45):
Yeah, I mean, you know, in the world in which
we operate, which is basically looking at rich molecular data
because most of the trials that we are focused on
are connected to some kind of biomarker or some eleclar
insight and the ability to look at clinical data and
know that this patient is a fit. You have to
have all the tools we built a structure clinical data
(41:07):
and all the tools we built to sequence patients. Those
tools don't exist in a lab Corp. They don't structure
clinical data like, that's not what they do, So you
wouldn't be able to do everything we do. They might
be able to say, hey, if you're looking for patients
that you know, have this particular blood marker. I can
see those patients, so that's probably true. But the kind
(41:28):
of end to end, you know, I know that this
patient is a perfect fit for this trial requires the
ability to have access to that clinical data for that patient,
structure that clinical data in minutes, hours, and then have
structured the inclusion or of the trial and married the two.
(41:48):
And I'm not aware of anyone else that has our
scale or does that.
Speaker 1 (41:51):
That's great.
Speaker 3 (41:52):
Yeah, there's some farmer companies that have tried it in
very limited ways that I mean, no artists had a
just in time program. But it's more about rapid enrollment
of a rapid site, you know, selection and enrollment, scaling
the site versus Yeah, it's not it's not about doing
all the stuff we do define the patients.
Speaker 1 (42:11):
So if we think about the company and where you
want to take it, you know, what's a long term
view of where you want to be in five to
ten years.
Speaker 3 (42:19):
I mean, our our goal is the same now as
it was you know then, which is we want to
actually you know, fulfill the mission of bringing you know,
added diagnostics and healthcare, and we want to help patients
live longer and healthier. Lives and we want to basically
build a business that you know is important, uh and
and and significant. So you know, we're nine years into
that journey, and the problem is big. Our solution is
(42:46):
still early in terms of rolling out, and so I think,
you know, we were like heads down trying to actually
get the scale. We've achieved some scale, but you know,
if you get to a billion dollars of revenue in
a in a five trillion dollar space, you know you're
you're you know, quite still quite small. So I think
we're very focused on, you know, really working on the
(43:11):
problem set and deploying our solution. And one of the
best parts of our culture and we have, you know,
an incredibly talented team that that has is so you know,
multidisciplinary and has so much you know, unique expertise. But
one of the best parts of our team is that
we really are problem oriented. Our culture is problem oriented.
(43:31):
It's like, what's the problem, what's the solution? How do
we how do we you know, put our heads down and.
Speaker 2 (43:36):
Figure it out?
Speaker 1 (43:37):
So maybe that leads into where you're spending your time.
You know, you had the IPO earlier this year. I'm
sure that took a lot of brain power. You know,
where are you focused today? And you know where do
you want to spend your time over the next couple
of years.
Speaker 3 (43:50):
Yeah, I my time is pretty much the same. I mean, yeah,
Certainly there was a month or so where it was
a disproportionate amount of time talking to investors just because
there's the sheer volume of people you had to talk
to when you go public. But you know, my time
basically today is is identical to what it was a
year ago. We always talked to some number of investors
(44:13):
because we were you know, it was very expensive to
raise the capital we needed to raise to build Tempest
was not a it wasn't a small effort, so we
were always meeting with investors. We were always having those
kind of conversations. You know, we've had an independent board
for a long time. We we've we've always behaved and
had a certain amount of financial rigor and discipline and
(44:33):
structure as if we were public, just because that's kind
of how we think, that's our orientation, and so, you know,
the things I do today are pretty much the same
as the things I did I did a year ago. Certainly,
the the the notoriety is higher, and you know, your
(44:53):
results are public, and so you you you're you're kind
of inevitably more focused on h a quarter to quarter mentality.
But I try my hardest to not let that take Well.
Speaker 1 (45:06):
I noticed in your shareholder letter you said you guys
are going to be focused on the long term, and
that's that's diametrically opposed to Wall Street in many cases.
Speaker 3 (45:14):
It really is. Oh my god, it really is. I mean,
and when we tell people this, and I think as
a we we we we really. I think the best
way to behave as a as a as a public
company in many ways is to treat your shareholders the
same way you would treat them in as as they
(45:34):
were private company. Like in other words, you have a board,
you have shareholders, there's a there's a high degree of
transparency and a flow of information because you have respect
for those shareholders. And then when your public things kind
of change, right, you behave differently. One of the best
examples is, you know, when companies go public, they give
a range, like my hair, like my results are range.
(45:57):
Private companies don't have ranges, right, I mean, if you're
if you have sent a private board or whatever, we're
here's my number. Ye, if you if you're a private company,
you said, here's my range, they'd be like, what, pick a.
Speaker 2 (46:08):
Number, Like, just pick a number.
Speaker 3 (46:10):
So we do the same thing we say to We
said to our shareholders, this is what we expect to achieve,
this notion of like you know, we're gonna you know,
beat and raise and sandbag and this and that. Like
I think it's just a lot of gamesmanship that doesn't really.
Speaker 2 (46:27):
Change the outcome.
Speaker 3 (46:28):
And in the end, in the end, shareholders are are
quite sophisticated and smart. If you have a business that's growing,
that's stable, that that generates leverage, that looks like it
has high long term growth prospects, you're gonna get great
shareholders and you're gonna have a great stock price. And
(46:49):
you get these short term vacillations where it goes up
and it goes down and it's overpriced on our price.
But really, if you're not selling in the short term
or buying in the short term, it's just noise. And
you know, I try not to think.
Speaker 1 (47:03):
About it well as an analyst. I try not to
make these conversations about, you know, the financial journey. But
one thing you just mentioned turned on a light in
my head, which is just the leverage in the business model.
And I know you guys have talked about getting too profitability,
but what are the key drivers to that?
Speaker 3 (47:18):
Yeah, I mean the biggest driver is when you have
significant growth like we do, and you're we're fortunate that
our growth rate is pretty high, right it's twenty five
to thirty percent kind of general speaking massive, and you
have a relatively high margin. Our margins are fifty five
sixty percent right now.
Speaker 2 (47:35):
So if you have.
Speaker 3 (47:36):
Pretty good growth and pretty good margin, then generating leverage
is purely a function of what you reinvest in the business.
So if you look at Q one to Q two
of this year, and this is directionly right around the
numbers in front of me, but I think we had
like kind of negative forty three million of adjustity with
don Q one and maybe negative thirty one and Q
two or something, and so we generate about twelve million
(47:59):
of improvement, and that was almost entirely a function of
we had growth in gross profit dollars and we didn't
put them all back in the business. And we told
people that's a trend we expect to continue. Obviously, the
numbers will fluctuate from quarter to quarter, but we expect
there to be growth. We expect there to be gross
dollar growth, and we don't intend to reinvest all those
(48:19):
back in the business. And so we're on this kind
of relatively short path to adjusted ebitdaut break even and
a cash flow break even. Those two, those two should
trend relatively close together, you know, with stock based compets
some other things. You know, companies like ours tend to
the EPs number can lag a bit, but in terms
of free cash flow and adjustedy, but.
Speaker 2 (48:42):
We expect to be on a very short path to
being break even.
Speaker 1 (48:45):
Well that leads me you mentioned reinvestment. You know, if
we think about the technology that's coming to market around AI,
and I'm not an expert in technology, but what makes
you or what gets you excited about what's coming? And
you know, where have you identified air areas of new
technologies that are emerging, and whether it's the large language
models or the next iteration after that, that kind of
(49:07):
get you really excited.
Speaker 3 (49:09):
I think the biggest, look, the biggest lever that gets
me excited is large language models.
Speaker 2 (49:16):
Generally, I like it.
Speaker 3 (49:17):
It's just such a transformational technology. There's really no there's
no other there's like kind of no other technology that
I think for the next four or five six years,
that will that will that will you know, get get
close to it. I mean, this is the great moment
in time when these models. If you look at you know,
(49:39):
the kind of sophistication between a chat GIPT three and
a four to zero, it's mind boggling. And same thing
with Lama, same thing with Gemini, and I would suspect
the same thing will happen again when they launched the
next version of the model. So you have these models
that are getting incredibly sophisticated. You have these the context
(49:59):
wind sort of the kind of the ability to query
data is growing significantly, and the and the and what
ends up happening when you can throw in a bunch
of data into these models and the models are unbluably sophisticated,
you can do really profound things. And so I think
that's what I'm most excited about.
Speaker 1 (50:19):
That's great. The way I like to wrap these conversations
up is really to maybe talk about something that is
personal to you, maybe a life lesson from the guests.
You know, what drives you day to day. You know
you've been successful as an entrepreneur. You know many times
when you wake up in the morning. What's driving you
to fulfill this mission?
Speaker 2 (50:39):
Well, this is a this is a big you know,
this is a big one.
Speaker 3 (50:42):
So I mean, I think unlike other businesses that I've
been a part of, all of which you know, I think,
we're amazing experiences. You know, you don't have to find
motivation when you're in cancer or other disease areas, Like
the motivation is there. So one of the best parts
of the work we do. And I say this also
(51:04):
quite often, is that one of the coolest parts about
Tempess is if you were being entirely focused on economics
versus entirely focused on being altruistic, you do the exact
same thing, meaning this very if I said to people
to Tempa's just singularly focus on patients, don't care it
(51:25):
all about money, that would produce the exact same outcomes
as if I said, be singularly focused on money, don't
care at all about patients.
Speaker 1 (51:32):
Like, in other words, pretty rare that they are aligned.
It's pretty rare they're aligned.
Speaker 3 (51:35):
Right, I mean, so we live and we happen to
have a business where they're entirely aligned, and so it
allows us to I think, you know, have that kind
of perpetual motivation because everything we do is is benefiting
the business and patients.
Speaker 1 (51:50):
That's great, So I think we'll call it there. That's
Eric Lefkowski, CEO and founder of Tempa's AI. Thank you
so much for joining us for our latest episode. Please
make sure to click the thunt A button on your
favorite podcast app or site so you never missed a
discussion with the leaders in healthcare innovation. I'm Jonathan Palmer,
and you've been listening to the vanguards of healthcare podcasts
by Bloomberg Intelligence. Until next time, take care,
Host
Jonathan Palmer
Popular Podcasts
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
The Joe Rogan Experience
The official podcast of comedian Joe Rogan.
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Special Summer Offer: Exclusively on Apple Podcasts, try our Dateline Premium subscription completely free for one month! With Dateline Premium, you get every episode ad-free plus exclusive bonus content.