Episode Transcript
Transcripts are displayed as originally observed. Some content, including advertisements may have changed.
Use Ctrl + F to search
0:00
This episode is brought to you by TIGUS,
0:02
the modern research platform for leading investors. Stretch
0:05
your research budget with flexible expert calls
0:07
you can trust. At a
0:09
fraction of the cost of traditional expert
0:11
networks, TIGUS customers pay only one of
0:13
the expert charges with zero markups and
0:15
no confusing call credits, netting an average
0:17
of 70% savings. Don't want to
0:20
conduct a full hour call? TIGUS offers the
0:22
ability to schedule 30-minute meetings, an offer
0:24
you won't find anywhere else. And
0:26
they don't stop there. With white
0:28
glove custom sourcing for every project
0:30
and robust compliance measures, including a
0:33
dedicated 50-plus person analyst team that
0:35
vets every call transcript, TIGUS ensures
0:37
your privacy and protection. As
0:39
the industry innovator for qualitative insights, TIGUS helps
0:41
you find the right experts you need at
0:43
a quality and speed that can't be matched.
0:46
For a limited time, as a listener, you
0:48
can trial TIGUS for free by visiting tigus.co.patrick.
0:52
I'm endlessly fascinated by people who are doing
0:54
their life's work and David Sendera gets about
0:56
as close as you can in his crafting
0:58
of the Founders Podcast. Each
1:00
week, David gives you ideas you can use
1:02
in your work revealed by the life stories
1:04
of history's greatest entrepreneurs. He connects
1:06
the legacies of the likes of Kobe Bryant to
1:09
Estee Lauder to Winston Churchill and the list goes
1:11
on. He recently put out
1:13
a new favorite episode on Anna Wintour. Founders
1:16
Podcast will undoubtedly light a spark of ambition
1:18
and aid all of us in our quest
1:20
towards our life's work. David
1:22
isn't stopping anytime soon. In fact, if you
1:24
ask him, he's just begotten. I'm
1:26
proud that Founders is part of the Colossus network and
1:28
you can find a link to this treasure trove in
1:30
the show notes. Hello
1:35
and welcome everyone. I'm Patrick O'Shaughnessy and this
1:37
is Invest Like the Best. This
1:40
show is an open-ended exploration of markets,
1:42
ideas, stories, and strategies that will help
1:44
you better invest both your time and
1:46
your money. Invest Like the
1:48
Best is part of the Colossus family
1:50
of podcasts and you can access all
1:53
our podcasts, including edited transcripts, show notes,
1:55
and other resources to keep learning at
1:57
joincolossus.com. Patrick
2:00
O'Shaughnessy is the CEO of Positive
2:02
Time. All opinions expressed by
2:04
Patrick and podcast guests are solely their
2:07
own opinions and do not receive the
2:09
opinion of Positive Time. This
2:11
podcast is for informational purposes only and
2:13
should not be relied upon as a
2:15
basis for investment decisions. Clients
2:17
of Positive Time may maintain positions
2:19
in the securities discussed in this podcast.
2:22
To learn more, visit
2:24
psuheap.com. My
2:29
guest today is Alex Stillford. Alex
2:31
is the founder of Convoke, a software
2:34
platform to help streamline drug development and
2:36
commercialization. He has also been writing
2:38
frequent blog posts on the biotech industry since 2019,
2:40
keeping a pulse
2:42
on the direction of innovation. He
2:44
joined me today to talk about the history
2:46
of the pharmaceutical industry and what's becoming possible
2:48
in medicine in the coming years. Alex
2:51
helped break down the complexities of investing
2:53
in new drug development, breakthroughs in gene
2:56
therapy on the horizon, and the dance
2:58
between timely progress and restrictive regulation. This
3:01
industry has a ton to unpack and Alex
3:03
thoughtfully lays out the landscape. Please
3:05
enjoy my conversation with Alex Stillford. Alex,
3:10
we're going to spend a ton of time
3:12
today to give out pharmaceuticals. Not just the
3:14
industry, some of the companies, but the process
3:16
by which the world has created incredible
3:19
technologies for our health and also the
3:21
impediments to creating more of them in
3:24
the future. Could you give
3:26
your version of just a
3:28
sweeping overview of the
3:30
process and evolution of how we find drugs?
3:33
Alex Stillford, MD Yeah. So, it
3:35
starts with academic research. This
3:37
is mostly funded by government research and
3:39
it occurs in universities. So, the
3:42
NIH is the biggest vendor of biomedical
3:44
research in the world and they support
3:47
scientists trying to understand the basic science
3:49
of how diseases work. What are
3:51
the mechanics of diabetes? What are the mechanics of
3:53
heart disease? What are the
3:55
different genes, proteins involved in those
3:58
pathological processes? to
4:00
some extent, how could we modulate
4:03
those processes to affect disease? But
4:05
that's only really a small part of the
4:07
story. So once you have an idea for
4:09
how you might intervene a disease, you then
4:11
have to do a lot of work to
4:14
convert it into a drug. So these initial
4:16
hypothesis is to use these target specific proteins,
4:18
RNA, whatever, so you can modulate to affect
4:20
the disease. I've then taken
4:22
up by mostly small profit
4:26
biotech companies who then do
4:28
all the hard work of turning this into
4:30
a drug, using a molecule that can modulate
4:32
these targets, either inhibit it or
4:34
increase the activity in a certain way.
4:37
And then running everything through
4:39
first these in vitro systems,
4:42
just basically cells in a Petri dish, then
4:44
up to animal models, so testing them in
4:46
mice, monkey models of disease
4:48
potentially. Once it's safe enough, then you
4:50
can go into human trials. You
4:53
have a steadily escalating sequence of trials
4:55
starting from phase one, which is just
4:57
basic safety trials in human volunteers and
4:59
dosing. Phase two, we try to
5:01
get a sense of initial efficacy of the drug. And
5:04
then phase three, which confirms the efficacy and
5:06
safety to the extent we can. And then
5:08
you have to go to the regulators, FDA,
5:11
EMMA, the MDA, and they'll
5:13
review the package of information and make a
5:15
determination for whether or not the drug can
5:17
be marketed for a certain specific use case,
5:19
which is called an indication. That
5:22
long chain produces the things we use. What
5:25
was so interesting to me reading all your
5:27
work was the rate of discovery and
5:29
the rate of development seems to have changed
5:31
a lot. You read about this guy, Paul
5:34
Jansen, maybe you can describe who he was,
5:36
but it seemed like in the early days
5:38
of this research, a single person or lab
5:40
could produce dozens and dozens of things that
5:43
get used for a long time. And now,
5:45
I think he said somewhere in there, like
5:47
the average researcher will not work on anything
5:49
that ever gets to production and use in
5:52
trials or practice. So give us a history
5:54
of how efficient we are at finding new
5:56
stuff and what the important timeline points are
5:58
in that development. So, if we
6:01
go back to, let's say, the end of
6:03
the 19th century, where you have
6:05
a really nascent pharmaceutical industry. So,
6:07
you have pharmaceutical companies that came
6:10
out of apocasaries, which were selling
6:12
extracts of plants like you might
6:14
still get today, naturalistic medicine, and
6:16
extracts of organs. So, there
6:18
was this idea that nature produces some bounty of
6:20
molecules that are somewhat useful to treat health. But
6:23
we don't really know what parts of
6:25
the natural extracts actually treat health. So,
6:27
you have some companies that are extracting
6:29
just those natural products and selling them
6:31
in extracts of cyberscranes, whatever. And then
6:33
you have another set of companies that
6:35
came out of the chemicals industry, so
6:37
the dye industry. And around when
6:39
our ability to understand chemistry
6:42
and use chemistry to develop our
6:44
own molecules was developing to
6:46
this back in the 1900s, some
6:48
of these dye manufacturers were figuring
6:50
out ways to just adapt dye
6:52
chemicals and use them for therapeutic purposes.
6:55
Those companies eventually became more sophisticated
6:57
at the types of chemical
7:00
manipulations we could do, the type of extraction we could
7:02
do, and purification of these natural materials. And
7:04
that started to evolve into what
7:06
became the pharmaceutical industry around the
7:08
early 20th century. So,
7:11
then you had some
7:13
point, the golden age of antibiotics
7:16
between the two world wars became
7:18
the first great success story of the
7:21
pharmaceutical industry. So, moving away from products
7:23
like these natural extracts that didn't really
7:25
work in most cases towards antibiotics, which
7:27
had an obvious and noticeable effect on
7:30
the ability to cure these infectious diseases.
7:32
And a lot of these antibiotics were found
7:35
through just experiments. Go into the wild, find
7:37
soil samples, find drugs in these soil samples
7:39
from bacteria, and would purify them. And
7:43
that worked for a while, limited
7:45
to infectious diseases for the most part. And
7:47
then in the sort of 40s and 50s, you
7:50
start getting people like Jenssen
7:52
coming about. So, Jenssen
7:55
was a Belgian doctor. He trained in
7:57
medicine. And his dad was actually
7:59
one of these... importers of some
8:01
of these natural products, organ extracts. Janssen
8:04
recalled when he went to school that
8:06
in medical school, his contemporaries were making
8:08
fun of him or being part of
8:10
a family who sold these sort of
8:12
ineffective organ extracts. And
8:14
so Janssen had a bit of a chip on his shoulder, I think, from that. And
8:17
he wanted to figure out, adapt, how
8:19
can we actually improve on some
8:21
of these compounds that nature has given us? So he
8:23
went to the US for a while. He realized, I
8:25
think, when in 1948, there were some
8:28
labs that were springing up that were trying
8:30
to do more rational drug design, trade
8:32
their own new drugs rather than just take what existed
8:34
already in nature. And he brought that idea back to
8:36
Belgium and started his own lab in 1952 and started
8:38
trying to tinker
8:42
with natural compounds and use some of the
8:45
emerging tools of chemistry to just adapt some
8:47
of these existing compounds that had been discovered
8:49
already and find new uses for them. So
8:52
the 50s, there's a lot of opportunity in
8:54
the sense that you had these newly evolving
8:56
molecular tools, and you had a lot of
8:58
space to apply those tools to
9:00
discover new useful compounds. So you started with these
9:02
functional compounds that we found from nature. We had
9:05
a lot of starting points, morphine,
9:07
pettitine, and atropine, that could
9:09
be tinkered around with to find useful
9:11
things that are close to those natural
9:13
compounds, but not exactly the same. There's
9:15
some slightly different effects. And you had
9:17
an emerging suite of molecular tools. And
9:19
then you fast forward over time, the
9:21
kind of low-hanging fruit have, to some
9:24
extent, been exhausted with a lot of
9:26
those molecular tools were developed in that
9:28
period of time. So the pharmaceutical industry,
9:30
as it previously existed, has been declining
9:32
in its efficacy of finding new products
9:35
because you've exhausted the
9:37
opportunities to develop new drugs because we've
9:39
tinkered around a lot of what
9:42
is easy to find. And then you
9:44
have in the 80s, the biotech industry
9:46
arose. So just as you have this
9:48
declining pharmaceutical industry, which is small molecules,
9:51
medicinal chemistry, you have this rising
9:53
biotech industry that's built
9:56
on this idea that you can get bacteria to
9:58
produce any kind of product. of
10:00
arbitrary protein by inserting it into its DNA and
10:02
that's how you get a lot of those things like the
10:05
humanized insulin, the monofilamptibodies, there's all
10:07
these like biotech products that are now
10:09
somewhat still in the sentence. So
10:11
you have sort of two curves overlapping.
10:13
The fall off of the kind of
10:15
exhaustion of low-hanging fruits in traditional pharmaceuticals
10:18
and then the growth in what we
10:20
can do with biotechnology tools like recombinant
10:22
DNA and more recently CRISPR things like that. Yeah,
10:24
it'd be interesting to zoom all the way to
10:27
now and talk a little bit
10:29
about the best contenders for the
10:31
next explosion like what we saw in
10:34
small molecule pharma. What is the low-hanging
10:36
fruit of today? I love hearing you
10:38
talk about biologics or anything else that
10:40
you think about the major categories of
10:42
discovery or new enabling technologies that will
10:44
allow us to do more faster and
10:46
maybe have another explosion like what we
10:48
saw early in the industry's history. If
10:51
you look at the history of technologies in
10:54
biotech, it takes a long time to
10:57
commercialize something. Going back to the idea of people
10:59
discovering targets in academia, as soon as you have
11:01
an idea for a target, it takes something like
11:03
20 years for that to turn into a drug.
11:05
The best of time period 20 years plus
11:08
minus 10 is fairly consistent of how long
11:10
it takes to translate this sort of idea
11:12
of how you might treat a disease and
11:14
so much of what goes into commercializing and
11:16
developing a drug is turning it from this
11:18
idea into a product that is actually manufacturable,
11:20
scalable, doesn't trigger all these unwanted side effects
11:22
and many of these things you don't know
11:24
about until you've tried it out in humans.
11:26
Monoclonals now are the top
11:28
selling drug in the world, Keytruda,
11:31
previously with Humira. They're both monoclonal
11:33
antibodies and now the processes to
11:35
develop, produce, and sell those
11:37
monoclonal antibodies at scale are pretty established. It
11:39
took a long time to get there. I
11:41
think if you want to answer the question,
11:43
what is the next set of technologies that
11:45
are going to be really impactful in the
11:47
next 10, 20 years you need to look at
11:50
what is just nascent today
11:53
and getting approved. So things
11:55
like gene therapies, we have a few gene therapies
11:57
that have been approved. So-so, gentlemen, the treatment for
11:59
spiciness. It's just really devastating infant
12:01
neuromuscular disease. The patients who have that
12:03
disease would die before they are one
12:05
year old and the most serotype. But
12:08
now they can be seemingly almost cured
12:10
with these gene therapies and other modern
12:12
therapeutics. And that's one example of
12:14
an early success story, but we haven't had many
12:16
other examples of gene therapies being commercially successful. But
12:19
once you have one launch on success, you
12:21
can then iterate and refine the processes for
12:23
developing these drugs and you will have down
12:25
the lines future successes. So I think another
12:27
10 years, another 20 years, we'll see just
12:29
the process for developing manufacturing gene therapies.
12:32
All those things get worked out and
12:34
they'll become applied to many more conditions
12:36
at a larger scale. Also things like
12:38
cell therapies. So there's some
12:41
type of drug called CAR T-cell, which
12:43
is way more complicated than just a
12:45
typical pill. Immusals extract it from your
12:47
body. They're flown to a manufacturing site.
12:49
They're gene edited to have this sequence
12:52
protein inserted into their membranes that binds
12:54
a specific type of protein found on
12:56
cancer cells and it's reinfused into the
12:58
body and then it goes and eliminates
13:00
these pathogenic blood cancer cells. And
13:03
that's an extremely complicated manufacturing process.
13:06
You're taking these cells out as a human,
13:08
you're flying them to another country probably, you're
13:11
gene editing them and then you're flying them back and you have
13:13
to do that in a very short amount of time. One
13:16
because the viability of the product and
13:18
two because these patients have very severe types of
13:20
cancer and if you don't get it fast enough,
13:22
they're going to die. There's a lot of kinks
13:24
in those processes that need to be ironed out
13:27
before these technologies become scalable
13:29
beyond the niche use cases. One interesting
13:31
example is this company Bristol Myers Square,
13:33
they bought a company called CellGene who
13:35
are pioneers in this type of CAR
13:38
T-cell therapy. And they have
13:40
the same amount of employees working on delivering
13:42
these compounds as they have crazy patients. So
13:44
something like 4,000 each. The
13:46
high touch process expertise and scale you
13:49
need to deliver these super complex next
13:51
generation therapies like CAR T's, like gene
13:53
therapies that are somewhat personalized is just
13:56
way beyond traditional pharma where
13:58
you make a pill in a factory and then you put the the pill
14:00
on the shelf, you just have a pharmacist give it out.
14:03
So I think we're going to see a
14:05
lot more of these type of process-like products
14:07
that are way beyond just a standard pill
14:09
and it's much more
14:11
a whole complex, large,
14:14
the actual drug itself is a complex
14:16
product. There's also a whole process around
14:18
the drug of delivering it to patients in
14:20
a timely fashion, manufacturing,
14:22
which is very complicated. Another good
14:24
example is radiopharmaceuticals. So a lot
14:26
of companies have been investing in
14:28
this lately, buying up biotechs, developing
14:30
these drugs. And these are compounds
14:32
where you have a targeting element,
14:35
let's say cancer, and then you have a
14:37
radionucleotide, which emits radiation. You infuse these drugs,
14:39
it binds the cancer, it emits radiation, the
14:41
radiation kills the cancer cell in a targeted
14:44
way. And these are really complicated to deliver
14:47
because half-life assignment of these radioactive compounds is
14:49
something like seven days. You can manufacture and
14:51
deliver it to the patient within a few
14:53
days before they lose efficacy. So
14:55
these complex therapies are
14:57
one way of farmers trying
14:59
to build up moats to
15:02
distribution of products. Can you
15:04
say a few words? There's four categories there that
15:06
we'll focus on, three that are terms that I
15:08
think people have heard, but they may not know
15:10
exactly what it actually means. So
15:13
the first is monocle antibodies or
15:15
biologics. The second is gene
15:17
therapy. The third is cell therapy. Can
15:19
you just describe what those mean and
15:21
what they're doing as categories of drugs?
15:24
So I'll start with this antibody, right? So when
15:26
you get infected with some sort
15:28
of virus or bacteria, your
15:30
immune system will generate antibodies against
15:33
the invading threat. And these
15:36
antibodies are specific to foreign
15:38
proteins. Your body's always generating all the
15:40
time all these antibodies. And
15:43
when a virus infects you, then
15:45
if you have an existing antibody that
15:47
binds that virus, it'll multiply within the body, and
15:49
then it will bind to viral copies
15:52
and it will stop the infection by
15:54
just binding and neutralizing these viruses or
15:56
bacteria. And that was recognized
15:59
as Antibodies
16:01
can bind proteins on these
16:04
foreign microbes. It can
16:06
also probably be useful for binding
16:08
other molecules or other proteins
16:11
in the body that aren't necessarily foreign
16:13
but may impact disease in certain ways. So we can
16:15
use it to knock out proteins that we would want
16:17
to knock out to treat diseases. So
16:20
one example is
16:22
Humira, which is until recently the top selling
16:24
drug in the world, something like 20 billion in revenue.
16:26
It's an antibody that binds to a protein
16:29
called TNS alpha, which is
16:31
an inflammatory protein. People
16:33
who have arthritis in these inflammatory conditions
16:36
will have an excess of this protein
16:38
that just causes an immune response that
16:40
leads to inflammation and joint pain. If
16:43
you can put in this antibody that binds this
16:45
protein, takes it out, then you can treat some
16:48
of these diseases that are pathologically too many of
16:50
a certain protein. And if you want to treat cancer,
16:53
cancers have a different composition of proteins
16:55
on the surface than normal cells, you
16:58
can put in antibodies that
17:00
bind those cancer cells and kill them
17:02
specifically. So where the monoclonal comes in
17:04
is that a while back
17:06
there was some scientists figured out that
17:08
you could take these immune cells that
17:10
were producing antibodies and combine them
17:14
with a cancer cell essentially and
17:16
immortalize the cells that produces antibodies.
17:18
That's how they make them just
17:20
scalable, manufacturing process. You have these
17:24
cells that are artificially immortalized that produce the
17:26
antibody anyway, and they just produce tons and
17:28
tons of this antibody product that you can
17:31
just grow it up into a bioreactor vat
17:33
and then up to a certain level grow it and
17:35
then you skim off the antibodies. Gene therapy
17:38
is this idea that you can insert
17:40
genes into the body to compensate
17:42
for effective genes, which you have something
17:44
like this spinal muscular atrophy example where
17:47
patients who have that disease, they lack
17:49
a functional copy of a gene called
17:51
SMN1, which produces a
17:53
protein that your motor neurons need to
17:56
stay alive essentially. And if
17:58
you can package up that gene
18:00
and deliver it into a cell, the cells
18:02
that are missing this functional copy will then
18:04
produce the protein. You can restore that function.
18:07
That seems like quite a promising approach. But
18:09
then, of course, the difficulty with fixing these
18:11
mutations is actually getting the gene into
18:13
the cells. So a lot of the
18:15
innovation around gene therapy was figuring out how to deliver
18:18
genes to cells. And the approach that
18:20
we've landed on now in the industry, the
18:22
dominant approach is to use a virus,
18:25
essentially. So what Christmas is doing
18:27
is making specific let's
18:30
say cuts of the DNA to
18:32
either eliminate genes that we
18:34
want to eliminate or potentially down the
18:36
line to insert new functional copies of
18:39
genes. And then cell
18:41
therapy is this idea that you
18:43
can edit cells, immune cells in
18:45
the body, to program them in a way to
18:47
do something that you want them to do. They
18:49
might otherwise be inclined to do so. You
18:52
take immune cells that are very good at killing other cells, they're very
18:54
good at killing foreign cells and
18:56
microbes. But the body has
18:58
a lot of systems to prevent immune cells
19:00
from killing your own cells. So
19:02
what cell therapy is doing is taking
19:04
out some of these immune cells and
19:07
treating them so they'll bind and recognize
19:10
pathogenic cells like cancer cells that
19:12
may look to the unedited immune system
19:15
similar to a normal cell and then
19:17
redirecting the immune system towards removing
19:20
those pathogenic cells. You tell
19:22
me a little bit about the
19:24
potential speed of all
19:26
of this and the problems that slow
19:28
things down. I was really struck by
19:30
the AIDS example, your post, and
19:32
also just by this, I think it's Moore's Law backwards,
19:35
or E-R talk
19:40
about that episode with the AIDS epidemic and the frustration with those that
19:42
could have benefited from some drugs and just the slow
19:44
process of being able to get them and E-R and
19:47
the things that are slowing down the
19:49
iteration speed cycling of, can we make
19:51
a discovery? We've got patients in need.
19:53
How do we shorten the time between
19:55
a discovery and an implementation? The
19:58
big tension in how you
20:00
regulate drugs is how
20:02
do you balance the need to develop
20:04
drugs quickly while also keeping patients who
20:06
are participating in clinical trauma safe and
20:09
keeping people who take the drugs when it
20:11
comes to market safe. So it's really
20:13
difficult I think for regulators to find
20:15
an appropriate balance and it's almost an
20:17
impossible problem because if you come down
20:19
too hard on drugs and on
20:21
manufacturers and make them jump to too many hoops,
20:24
then it's just going to be uneconomical to develop
20:26
drugs and there's going to be a massive invisible
20:28
graveyard of people who could
20:30
have been saved had the drug come to
20:32
market faster but it was blocked by various
20:34
regulatory processes. But then if you're too lenient
20:36
on drug makers, some of them will take
20:39
advantage and will push drugs onto market that
20:41
shouldn't have been there. They may be acting
20:43
completely good intentions and they'll test according to
20:45
the process that regulatory agency demands and bring
20:47
the drug to market and it turns out
20:49
later that it causes some harm that we
20:51
could have foreseen if you had more rigorous
20:53
testing. So there's
20:56
no one size, that's all solution
20:58
to fix this tension. So
21:00
you have to always make some compromises
21:02
and coming back to the issue about
21:04
we're running out of low hanging fruit,
21:07
one of the biggest problems is that
21:09
we already have so many good drugs
21:11
for many conditions. So if
21:14
you have diabetes or something, then
21:16
we have insulin, we have many
21:18
varieties of insulin, we have the
21:21
GLP1 drugs now, we have a number
21:23
of other drugs that treat diabetes effectively
21:25
and they're very effective drugs. So it's
21:27
hard to make the case that you
21:29
need to bring another drug to
21:31
market with great haste to treat large
21:33
number of patients who are being well
21:35
treated by current drugs. So
21:38
under that situation, you
21:40
probably as a regulator are inclined to
21:42
put really high burdens on drug makers.
21:45
So you have to absolutely demonstrate that anything you bring to market
21:47
is really safe, you have to jump through all these tubes, you
21:49
can do mortality studies, all these really onerous
21:51
requirements. On the other hand, if
21:53
you have a disease where there's no effective standard
21:55
of care and patients are
21:58
dying or suffering greatly. from
22:00
their condition, then it makes
22:02
sense to potentially relax the regulatory burdens.
22:04
Compared to the AIDS issue, when you
22:07
had the AIDS epidemic, there
22:09
were a number of drugs that seemed to
22:11
be promising. The pharmaceutical
22:13
companies were developing at
22:15
speed, like they did with COVID, right? A
22:17
lot of companies were developing all their anti-COVID
22:19
drugs and vaccines. And the AIDS patients were
22:22
quite rightly saying that we're going to die
22:25
anyway. So we have a death sentence. You
22:27
should let us try these drugs
22:29
and see if they're effective because
22:32
the alternative is death. And
22:34
the FDA, until that time, didn't
22:37
really have a process to deal
22:39
with these differences in need between
22:41
conditions. So they would treat
22:43
diabetes similar to how they would treat
22:45
AIDS. So after a lot
22:47
of protesting by the AIDS community and
22:49
a lot of patient advocacy, the FDA
22:51
ended up speeding a number of these
22:54
compounds through development, so AZT and
22:56
DDI to the drugs. These turned out to
22:58
not be particularly effective drugs,
23:00
but you could argue that potentially by
23:03
starting off that process of defining a pathway
23:05
for how you treat AIDS, getting some drugs
23:07
on the market, getting some initial revenue, you'd
23:09
then spurred the process to iterate and invent
23:11
drugs that are now very effective. So now
23:13
there are very effective treatments for AIDS. This
23:16
initial speeding of the regulatory process during
23:18
the AIDS epidemic turned into
23:20
what became the accelerated approval pathway.
23:23
And that is for drugs that have
23:25
seen promising in disease a very high
23:27
unmet need like AIDS was, like
23:29
a lot of cancers are. Drugs can get
23:32
through regulators with much
23:34
less evidence than they can in other conditions
23:36
like diabetes where the standards might be higher.
23:38
So now what you've seen in
23:40
the industry is regulatory arbitrage, where a lot
23:43
of companies are investing in things that are
23:45
easier to get past the regulators. You don't
23:47
need to do all these really huge phase
23:49
three trials because there isn't a good standard
23:51
care, where there's not so many patients. So
23:54
you see a lot of this investment in
23:56
cancers and greater genetic diseases.
23:58
The requirement for evidence is... lower and
24:00
the need is greater. One of the things that
24:02
you write a lot about is just
24:04
the discovery process itself. And there's a
24:06
very clever video game analogy to Super
24:08
Mario and what we can learn about
24:10
speed running video games and relating that
24:12
to the process of discovering things that
24:14
might be useful in the pharma world
24:16
and the biotech world. Maybe you
24:18
flush that idea out of just your interest in
24:21
the process of discovery itself and what we can
24:23
learn from other domains like games. Yeah,
24:25
so I think if you look at
24:27
how things get discovered in the pharma
24:29
industry, it's often very
24:31
serendipitous. So you'll have
24:34
these stories of people tinkering away on
24:36
some idea they think is promising for
24:38
years and years and then eventually
24:40
that becomes a drug. So like the GLP ones
24:42
is a good example where you
24:45
had decades of the scientists at
24:47
Novo Nordisk pushing for this idea,
24:49
tinkering around the boundaries of GLPs
24:52
and trying to figure out how to
24:54
actually take this idea from just
24:57
a concept to an actual drug. So
25:00
I think one way to help
25:02
make the industry more effective is to
25:05
try and find ways of
25:08
promoting that artful tinkering
25:10
because we just haven't proven to be very
25:12
good at developing drugs
25:15
to a pre-described positioning if you
25:17
like. The really important and powerful drugs
25:19
get discovered from this bottom-up process where
25:21
you have scientists who are really passionate
25:23
about an idea, they push through their
25:25
management saying you should stop this, it's
25:28
clearly not working and they try to
25:30
find ways to work on this without
25:32
management finding out. They try to
25:34
get resources wherever they can and champion
25:37
despair, loss of opposition and then eventually
25:39
it turns into this drug that becomes
25:41
a mega blockbuster and helps a huge
25:43
amount of patients and then the management
25:45
thinks they'll say that we do it
25:47
all along or something in retrospect. But
25:49
you can't prospectively figure out what drugs
25:51
are going to be really impactful a
25:53
lot of the time. So
25:55
I think it's quite dangerous in an industry
25:57
like pharma which is so innovation. driven
26:00
and science driven to impose these
26:02
top-down, pre-described notions
26:05
of what people should invest in, you're very
26:07
likely to be wrong. Until a few years
26:09
ago, I think Novo Nordisk was regarded as
26:11
a pretty boring European
26:13
pharma company, not very innovative,
26:16
just doing insulin. But now
26:18
they're like the 16th
26:20
biggest company in the world or something and
26:22
everyone's excited about the GLPs. You would have
26:24
necessarily predicted that would come out of Novo
26:27
Nordisk, but Novo Nordisk was able to produce
26:29
GLPs because they had this really long-standing interest
26:31
in a specific area, edible diseases, they support
26:33
the scientists there and they've just been tinkering
26:36
in this area for a long time. So
26:39
to come back to the food moray
26:41
idea, that post is about this
26:43
idea that one of the
26:45
ways you might productively use
26:47
technologies like artificial intelligence and
26:49
simulation tooling is to have
26:52
an intuition about what areas of
26:55
science might be promising and then
26:57
use automation to do scalable
27:00
tinkering around that nucleus of
27:02
an idea. So the example
27:04
in Super Mario was that there is this idea
27:07
that you could do a specific type of
27:09
jump to get through a level in Super
27:11
Mario much faster than people have been able
27:13
to do so far, but no one knew
27:15
it was possible. So until a software developer
27:17
developed this tool called Scattershot, which simulates millions
27:20
and millions of Marios jumping up this level in
27:22
every possible configuration of jumps and starting positions, he
27:24
managed to figure out there actually was a way
27:26
to get up to the top of the castle.
27:28
They were trying to jump up and finish a
27:30
level much faster than previously possible. It's changed this
27:33
whole way that speedrunning is approached for that game.
27:35
Speedrunning itself is probably out of interest to that
27:37
many people, but the general principle I think is
27:39
interesting that you have this intuition
27:41
that something is possible. For Jansen, when he
27:44
was developing his drugs, he had an intuition
27:46
that you could improve this compound
27:49
pethidine to make
27:51
a better drug, but he didn't exactly know
27:53
how to modulate it to make it the
27:55
best drug it could be. He was just
27:57
tinkering around. So if you could use technologies
28:00
to automated scientists to feel the
28:02
process of tinkering, trying out solutions,
28:04
testing them, then you can tighten
28:07
the feedback loop around the
28:09
industry and maybe develop drugs a lot faster
28:11
than you might otherwise be. I think
28:13
one of the big problems with pharmaceuticals and industry is
28:16
that the iteration speed is super low
28:18
compared to something like tech. Like in
28:20
tech, if you have
28:22
an idea for a product, you can
28:25
spin up a demo in
28:27
a few days and you
28:29
can show it to some customers, you
28:31
can get some feedback on that, you
28:33
can then iterate it, layer instantly in
28:35
code effectively, and then show the next demo to
28:37
customers, get more feedback, you can do these A, B
28:39
tests at massive scale. You can move super fast and
28:41
learn very quickly. But pharma is
28:44
a very far domain to learn quickly
28:46
because everything takes so long. Clinical
28:48
trials will take around 10 years to get
28:51
everything through clinical trials. Even
28:53
do animal experiments, it takes weeks
28:55
or months to do animal experiments in certain ways.
28:58
Making all these chemicals takes a long time. You're
29:00
testing even in the very earliest stages. So
29:03
you learn at a very slow rate about
29:05
what works. And if you could
29:07
just find ways to speed that up across
29:09
the whole process, then you'll do a
29:12
lot to, I think, address some of these
29:14
problems with the pharma industry like drug
29:16
prices, just a huge healthcare
29:18
spend, things like that. I was
29:20
AI-figured to all this, obviously alpha-holed within
29:22
the mainstream consciousness. I think Jensen Wong,
29:24
the CEO of N videos, recently talking
29:27
about how much explosion of
29:29
activity there is using AI in the
29:31
world of medicine and biology. So talk
29:34
about the role that this new set
29:36
of tools might play in both discovery
29:39
and speed and tinkering
29:41
and all these different ideas. I
29:43
think I'm short-term, pessimistic, long-term optimistic
29:46
when it comes to AI for
29:48
drug discovery. And I'm much
29:50
more optimistic in the short-term about using AI
29:52
to automate some of the processes along with
29:54
drug development, not necessarily the designing of drugs
29:57
itself, but all the steps to bring a
29:59
drug to mind. market. So if
30:01
we start with drug discovery, like finding a drug, I
30:04
think it's just very difficult to see
30:06
how AI is anything more than just
30:08
another tool in the expanding toolbox that
30:10
drug developers use to produce and
30:13
test molecules. AlphaFold's an
30:15
example where now you can predict the
30:17
protein structure from a protein sequence, which
30:19
is a problem that we didn't know
30:21
how to do to a high degree
30:23
of accuracy until before AlphaFold. And
30:26
a lot of people are saying, oh, AlphaFold is
30:28
going to revolutionize drug discovery. But
30:30
in reality, it's just one little piece
30:32
of the hugely complex puzzle where sometimes
30:34
you'll have situations where you don't know
30:36
the structure of a protein you're trying
30:38
to make a drug against. And
30:40
AlphaFold can be a helpful starting point for that
30:42
specific circumstance. But
30:45
actually, AlphaFold has problems
30:47
with the very granular
30:49
predictions of the configuration
30:51
of the sites where drugs might bind. So
30:53
it's not completely accurate. So we need more
30:55
accurate models for it to be really useful
30:57
for drug discovery. And
30:59
to improve AlphaFold, we need more
31:01
data. It's just taken a long time to collect
31:03
all this data. So AlphaFold has trained on
31:06
decades of protein structures that have
31:08
been collected painstakingly by researchers growing
31:10
crystals in dark rooms. It's
31:12
definitely speeded up parts of
31:15
that process, but a very small part.
31:17
And I think the way AI is going to impact
31:19
the pharma industry looks similar to that. A lot of
31:21
little tools that are going to be very
31:24
useful and accurate, but they're going to take a long time
31:26
to deploy and figure out how best to use them and
31:28
work them into the process. One
31:30
thing I should also mention is that there's this
31:32
idea that drug discovery is rate limited
31:34
by our ability to design
31:36
molecules and design new drugs. And
31:39
that's not really true. Our ability
31:41
is much more rate limited by the downstream
31:44
clinical development and
31:46
testing these drugs and gathering information. We
31:49
need to actually justify approving these drugs
31:51
for sale and working them. While
31:53
I'm optimistic about over the
31:56
long term using things like AI to
31:58
predict what drugs might be
32:01
effective to do drug discovery and
32:03
drug candidate selection. I'm
32:05
much more optimistic about using AI in the
32:07
short term to help drugs to market faster
32:09
than they otherwise might have been. So things
32:12
like can we use AI
32:14
to help prepare
32:16
regulatory documents quicker? Can
32:18
we use AI to do things like
32:21
help companies figure out what
32:23
the markets for their drugs are and prioritize what
32:25
opportunities they should invest in? How can we use
32:27
AI to help identify patients who might be a
32:30
good fit to enroll in a certain trial? So
32:32
things like trial enrollment just takes away
32:35
too much time. So just
32:37
better tooling to identify patients who might
32:39
be a good fit for a trial
32:41
and enrolling them seems pretty impactful. If
32:44
you were just a czar of this
32:46
entire universe and godlike powers, what two
32:49
or three changes would you
32:51
make regulatory, technology-wise, industry structure?
32:53
Anything that you could change.
32:56
What would you change that you think would
32:58
most benefit patient outcomes or the overall
33:00
system? I would make regulatory
33:03
changes. I think I would want to
33:05
formalize the idea that different diseases require
33:07
different standards of evidence and then pre-publishing
33:09
the degree of evidence that we require
33:12
for each different disease. So something like
33:14
diabetes should have quite a high standard
33:16
of evidence. So you should have to
33:19
enroll very large numbers of patients and
33:21
test mortality because we have
33:23
extremely good drugs. And in other cases,
33:25
we should have a lot of diseases where our standard of
33:28
care is not as effective. We should
33:30
formalize this idea of smaller trials, maybe approved drugs
33:33
with a weaker standard of
33:36
evidence. And we are seeing some moves
33:38
towards that with rare diseases. So
33:40
you have a lot of these conditions that
33:42
are too small to really be able
33:44
to tell. They maybe have
33:46
too few patients, but they may have too few patients plus the disease
33:49
evolves over too long of a time for you to reasonably
33:52
be able to figure out the standard
33:54
drugs are really working to any high
33:56
level of confidence. So I think in those conditions,
33:58
you'll have to be able to say... Look,
34:00
we're just not going to be able to determine
34:02
this in a trial. We have to, based on
34:04
some principle of the reasons, evidence we've
34:07
collected about how the drug works in animals
34:09
and maybe some initial human testing and
34:11
some preliminary biomarker evidence, we can then
34:13
approve this drug and roll it out
34:15
and then collect information more rigorously once
34:18
on the market. Because you're just never
34:20
going to get drugs for certain diseases,
34:22
just uneconomical to achieve this vision of
34:24
hyper-personalized medicine where you're
34:26
developing a drug for everyone's individual
34:28
condition and biology without some ability
34:30
to make inferences about what is
34:32
likely to work based on early
34:34
evidence and genetics and reasoning
34:36
about the mechanism of action. So
34:39
I think the second thing I would do
34:41
is just be much more rigorous about data
34:43
collection post-marketing. So you have a lot of
34:45
drugs that go on to the market and
34:47
then they're approved and then we're not really
34:49
following up sufficiently to be able to
34:51
tell us they really work. So there's been some instances
34:53
of drugs that have got this extolibrated approval on
34:56
some fairly flimsy evidence but they seem promising.
34:58
But then the company is meant to do
35:00
a confirmatory trial to determine, okay, actually, we
35:02
have a good reason to believe this drug
35:05
is going to work but we
35:07
don't actually know that until we've collected this evidence
35:09
over five, ten years or whatever. So
35:12
companies drag their heels on doing the confirmatory
35:14
study. I think the regulators should be much
35:16
more stringent in forcing companies to collect that
35:18
data and report it and take it out
35:21
of the market. That is a combination of
35:23
being more open to approving
35:25
drugs based on a lower standard evidence when
35:27
there's no good treatments and when it's just
35:29
unfeasible to do with standard clinical trial coupled
35:31
with more rigorous data collection and then taking
35:34
things off the market when they're not working.
35:36
I think that will help increase the sort of range of learning
35:38
in the regulatory process. Surrogates as
35:40
well, I think, are really important. So
35:42
surrogates are measures like biomarkers that you
35:45
can use as a proxy to tell whether a
35:48
disease is being treated effectively or not. And so
35:50
for cancer, what you really care about when someone
35:52
has cancer is stopping them from dying but you
35:54
can get a proxy for that by looking at
35:56
the size of the tumor and it's not always
35:59
a perfect correlation. you shrink the tumor.
36:01
Most of the time, that's a good
36:03
thing. Sometimes you shrink the tumor, but it
36:05
doesn't actually help the patient survive longer in the
36:07
end. So you're exposing them to toxicity and
36:09
you're not helping them survive. So it's actually you
36:11
doing that harm with that drug. The idea
36:13
of a surrogate is that you have a
36:16
measure like tumor shrinkage that
36:19
predicts benefit down the line and that
36:21
helps you develop a drug much faster. A lot
36:23
of drugs have been, not to market, much faster
36:25
in cancer than they otherwise would have
36:27
been many very effective drugs because they've
36:29
used these surrogate markers of efficacy like
36:31
tumor shrinkage. But it's harder to
36:33
apply this principle in other conditions like
36:36
more complex chronic conditions where you don't
36:38
have an objective measure that's as simple
36:40
as does a tumor get smaller or
36:42
not. If you're treating a disease that's
36:45
eventually fatal, if you see that this
36:47
biological response happens early on, then that's
36:49
very highly predictive of improved survival down
36:51
the line. And just finding more of
36:53
these surrogates is a really, I think,
36:55
valuable activity that something like the NIH
36:58
and government agencies should invest a lot
37:00
more money in developing because as soon
37:02
as you have an established surrogate that
37:04
the regulators could use, you really speed
37:06
up how quickly you can iterate on
37:09
developing drugs. You can take it to the clinic, you can
37:11
see if it works in the surrogate, and then you can decide whether
37:13
or not to continue or discontinue based on that. You don't have to
37:15
wait five, 10 years to just
37:17
survive or not. So something like aging, we're
37:20
never going to get drugs to prolong
37:22
people's lives until we figure
37:25
out a surrogate for aging. Can
37:27
you talk about RCTs, randomized controlled
37:29
trials, and something like vitamin D?
37:31
I love the idea that this
37:33
tool, RCTs, is an incredibly important
37:35
process that's helped us learn a
37:37
lot. What, if any, limitations are
37:39
there in your mind to RCTs
37:41
and learning? So the big
37:44
limitation to RCTs is they
37:46
take time, they're complex to run, and then people
37:48
don't want to be in the control group, essentially.
37:50
So you have to expose patients to something that
37:52
they don't want to take really because you're entering
37:55
an RCT for a drug, you want the active
37:57
drug. But maybe take a step back. So the
37:59
idea behind... an RCT is that
38:02
you have patients
38:04
or people who are going to take
38:06
some intervention and you want to figure
38:08
out does this intervention really work
38:11
or not. And you have to find some
38:13
way of assigning the intervention randomly because if
38:15
you rely on anything other than randomness, there's
38:17
going to be some bias that creeps into
38:19
the study. Let's say you have this
38:21
new drug and you take it to doctors and you say,
38:23
I want to test this new drug, then the doctors
38:27
may be more inclined to give
38:29
the drug to patients who are sicker, who
38:31
really need it. And they're less inclined to give
38:33
the experimental drug to patients who seem like they're
38:35
going to recover anyway. So if you just do
38:38
that non-random allocation, you're going to see that
38:40
people who take the drug die at a much higher
38:42
rate or have a much worse illness than people who
38:44
don't take the drug because you have
38:46
this selection bias of doctors giving the drug to
38:48
people who are less likely to have a good
38:51
outcome. You need to find a
38:53
way to eliminate those biases and you can do that
38:55
with just total randomization. And so coming back to the
38:57
vitamin D example, one thing in
39:00
data and retrospective data that
39:02
looks at vitamin D levels
39:04
and outcomes is that
39:06
you see people with low vitamin D levels
39:08
have generally worse health outcomes.
39:10
So they have high rates of mortality,
39:13
they have cardiovascular disease. I think it's
39:15
probably been correlated with every bad thing
39:17
you can probably get. And there's a
39:20
natural inference there that we should just
39:22
supplement vitamin D to the people
39:24
who are low levels and then they'll have less
39:26
rates of all these bad diseases. So then people
39:28
do the trials and they do
39:30
a randomized control trial when they assign
39:33
people randomly to groups and they
39:35
give supplementation of vitamin D and then they
39:37
measure outcomes in all these things like mortality
39:39
and health. And then what they find is that
39:42
actually vitamin D has very
39:44
minimal or no effect on both of these
39:46
outcomes, even though some of the
39:49
data from retrospective analysis looks pretty strong. So
39:51
how did that happen? What
39:53
you find out is something like vitamin D
39:55
is actually a marker for poor health in
39:57
general. So vitamin D is produced by your
39:59
boss. body in response to sunlight.
40:02
And if you are ill
40:04
or poor health generally, you're
40:06
less likely to go outside or you
40:09
may be less mobile in general, so you may
40:11
get less exposure to sun.
40:13
And then you'll see the correlation between just
40:15
general poor health and vitamin D. It's like
40:17
with pushups. So they say, if you can
40:19
do less than 10 pushups, then you're much
40:21
more likely to die. Not literally the ability to
40:23
do pushups determines how likely you are to die
40:25
is just being able to do pushups
40:28
is a decent proxy marker for general health.
40:30
So RCTs are just the most
40:33
effective way we know of eliminating all
40:35
these biases that you don't know exist
40:38
from the start. Because otherwise, it's just everything
40:40
you try gets really confounded and
40:42
you don't even know how it's confounded. Can
40:45
you imagine an alternative or do you
40:47
think this will be and needs to
40:49
be the method by which we learn
40:51
truth and efficacy across medical interventions? So
40:54
I think RCTs are always going
40:56
to be the gold standard. And
40:59
I think when we can do RCTs
41:01
recently, we should do them because
41:03
they just eliminate so many of these
41:05
problems that you're just unaware of. That's the biggest
41:07
issue if you control for the sources of bias
41:09
that you think are likely to exist. There'll be
41:11
hidden sources of bias that you just can't control
41:13
for. So I think we want
41:15
to keep RCTs as a control inventory method.
41:17
But there are other tools in
41:19
these instances where RCTs aren't feasible that have a lot
41:21
of promise. So things like if
41:23
you do a single arm study, which is
41:26
just one treatment group, and you compare the
41:28
progression of a disease with a
41:30
matched digital twin of these patients, you
41:32
can train with information from natural history
41:35
data. We can make a good assumption
41:37
about the rate at which these patients would
41:39
have progressed had they not had
41:41
the drug. So there's been a few approvals
41:43
that have used some of these matched natural
41:46
history data as an alternative cohort. And those
41:48
have been in cases where there's been too
41:50
few patients to really do largely
41:52
randomized trials. And then you have
41:54
these instances of super rare diseases where you
41:57
may have just a handful of patients
41:59
in the US, for instance, is the
42:02
same possible thrown RCT and have any
42:04
reasonable statistical outcome. You need to use
42:06
things like digital twins or
42:08
surrogates, single arm trials, you
42:10
can't use RCTs. There's a lot of innovation
42:13
in this digital twin idea and natural history
42:15
cohorts. Obviously, a big driver
42:17
of all this is profits. We
42:19
talked about how people don't like pharma,
42:22
in part because they charge so much
42:24
for really valuable drugs and they make
42:26
big profits. And that bothers people as
42:28
it relates to their health. But profits
42:30
and revenue drive. The motive
42:32
for discovery and the US has always been a
42:34
leader here. Can you talk about
42:36
what you've learned about blockbuster drugs specifically? You
42:39
can define what you mean by a blockbuster
42:41
drug, but it seems like the biggest drugs
42:43
that represent, there's sort of a mini power
42:45
law here that the top handful
42:47
of drugs represent a huge percent of the
42:49
entire industry's revenue. So talk about the business
42:51
side here, the distribution of revenue, the role
42:54
of blockbuster drugs, the good, the bad, the
42:56
ugly. Love to hear what you've learned there.
42:58
Yeah. So a blockbuster drug is a drug
43:00
that makes a billion dollars in annual revenue.
43:03
So blockbusters are really have an
43:05
outside influence in the pharmaceutical industry
43:07
because they account for a
43:10
huge proportion of the overall revenue
43:12
of the industry. Well, the way
43:14
the economics operate is pretty analogous
43:16
to venture capital and these long
43:19
tail models where you have many
43:21
losers who don't make much money or don't recoup the
43:23
money that has invested in them and
43:25
a few really huge winners that will go
43:28
off and generate billions of billions
43:30
and billions of dollars and pay for all
43:32
the failures many times over. So the industry
43:34
is in many ways worth
43:36
investing in for investors and
43:38
for companies because there's always this potential
43:41
of hitting it really big and having a mega
43:43
blockbuster like Humira or Kritrudar, the COVID vaccines that
43:46
generate tens of billions of dollars in revenue a
43:48
year. Because if you look at all the drugs
43:50
that are getting launched, it's sad that you have
43:52
all this work that goes into developing a drug
43:54
and it takes me 10 years to
43:56
get it to market or 12 years to get it to market. And
43:59
it... launches and then it stops. 55%
44:03
of drugs that launch, that get to the left process,
44:05
they don't even recoup enough money to
44:07
pay the average development cost of
44:09
a drug. And then you
44:11
have a very small number of drugs, blockbusters, that
44:14
30% to 40% of revenue of the
44:16
whole industry is made by these blockbuster drugs, which
44:19
are quite a small fraction of all
44:21
drugs. But when I look at the numbers, something
44:23
like 170 blockbuster
44:25
drugs that were actively
44:28
generating revenue, and then the
44:30
whole long tail of other drugs is
44:32
making very small sums relative to blockbusters.
44:35
So because the industry operates like
44:37
a lottery type model where you just want
44:39
to really hit a big, you have a
44:41
venture capital, it distorts all the incentives. So
44:44
you see a similar thing in
44:46
biotech investments that you do with VCs
44:48
where the farmer is only really interested
44:50
in developing a drug if it can
44:52
potentially become one of these blockbusters. Because
44:54
anything less than that is just unlikely
44:56
to recoup the investment. So you get
44:58
a lot of these small markets that
45:00
patients really would like to have drugs
45:02
for. They could be really beneficial and we
45:04
know that we could potentially develop a drug against this
45:06
disease. We have a good understanding of
45:09
the mechanism, like many small genetic diseases, but it's
45:11
just not worth big farmer's time or even biotech's
45:13
time to invest in developing these drugs because the
45:15
economics don't work out. Even if
45:18
you get to the whole process of developing this drug and
45:20
it works and you get to clinical trials, you
45:23
maybe make a few tens of
45:25
millions a year and that's just absolutely not worth it because it
45:27
will cost you 100, 200, 300
45:29
million just to get the drug through the whole development process.
45:32
So there's a lot of, I
45:34
think, alpha that could be unlocked
45:36
in making the process much cheaper
45:39
in how drugs are tested and validated because
45:41
it would mean that it's actually
45:43
worth developing all these drugs that are not
45:45
worth developing. What do you
45:48
think about drug pricing as a key variable
45:50
in all of this big equation that blockbuster
45:52
drugs are driven by a price times an
45:54
amount? What drug pricing sometimes comes up
45:56
all the time as predatory
45:59
or strange or... or the US system
46:01
subsidizing a lot of the rest of
46:03
the world. We make an outsize percent
46:05
of the discoveries and you want to be
46:07
compensated for that. It seems like a very complicated equation.
46:10
So what, if anything, have you learned
46:12
about drug pricing that you feel is different than
46:14
the norm or how would you change things? So
46:17
I think drug pricing is truly difficult. One thing I
46:19
guess to say about the US is the US has
46:21
by far the highest prices in the world. I think
46:23
people recognize that. It's about twice as high as Europe
46:26
on a net price basis. And the
46:28
US market for drugs is something like
46:30
40% of the global market and it's something like 60%
46:33
by revenue of newer drugs. So drugs
46:35
launched in the past 10 years or so. So
46:37
the US, it's kind of where outside share of
46:40
the revenue that drug companies make. And part of
46:42
that is because the US system is, for the
46:44
most part, some of this is changing, but it
46:46
is mostly a free pricing system where you can charge
46:48
whatever you want. So when you're pricing
46:50
a drug for the US market, you'll often, as
46:52
a company, just try to charge whatever you think
46:54
you could get away with, what if you think the market
46:56
will bear. And in a reasonable world where
46:58
all the incentives are aligned, what the
47:00
market will bear is close to the actual
47:03
value of that drug. In healthcare,
47:05
you have all these strange incentives in markets that don't
47:07
function properly. But in the rest of the world, you
47:09
have systems where you
47:12
have governments doing the
47:14
negotiation on behalf of the population.
47:16
So they'll make assessments of the value
47:18
of a drug based on metrics like how
47:20
many quality adjusted life years does this drug
47:22
give us, what's the benefit of this drug
47:24
over the existing standard of care, and how
47:26
much better is it, and how much for
47:28
price premium can we give it on the
47:30
existing drugs. So both systems have flaws. I
47:32
don't think it's possible to find a perfect
47:34
system to satisfy everyone. If you look
47:37
at the profit margins of drug makers, they make
47:39
something like 10% to 20% profit
47:41
after you take out the cost. And the growth margins are something like
47:43
80%. Really,
47:46
that's a distorted picture of the
47:48
actual profitability of
47:51
drug companies because the distortions between
47:53
when R&D expenses are paid and when the drug revenue comes
47:56
in. So it's not necessarily a good picture of how much
47:58
of a profit you take out of the drug. drug
48:00
companies getting for their investment. And if you look at
48:02
how much drug companies are getting for their investment, actually,
48:05
most of them are pretty near zero or very
48:08
low return on invested capital. The
48:10
industry as a whole is not that great of a business. It's
48:12
very much like a lottery model that
48:14
is just type of lottery that's
48:16
attractive to people with biomedical PhDs,
48:18
where most people are not
48:21
making that much money, but
48:23
a small number of drugs and companies are
48:25
making supernormal profits. And you can
48:27
point to the examples of people
48:29
who have very high prices and making supernormal
48:31
profits like Humira, 20 billion seems
48:34
like an incredible amount of money for a drug
48:36
to make. And I think if you
48:38
look at that isolated example, you can say,
48:40
just seems unreasonable that anyone is making 20
48:42
billion off this drug per year. It's way
48:45
more than what someone should reasonably
48:47
get for producing this drug. But you
48:49
have to think more about the system
48:51
of incentives. You want to be worth
48:53
investing in developing drugs. And so the
48:55
incentive needs to be very strong. And
48:57
because such a small number of drugs
48:59
produce so much of the revenue, like
49:02
a real Pareto distribution in revenue,
49:04
you need these lottery winners
49:07
to make it worthwhile investing in the
49:09
system, which is an
49:11
unfortunate reality that because drug development
49:13
is so inefficient, you need high
49:15
prices. And you need these
49:17
superblockbusters to make the economics work out.
49:19
When you're doing R&D, you're paying money
49:22
that you have now for revenues in
49:24
10 years. So from a temporarily discounted
49:26
point of view, you need to think
49:28
you're going to get a really huge
49:30
amount of money in the future
49:32
to be worthwhile you paying hundreds of millions of dollars
49:34
in the next few years. I think
49:37
it's a really difficult problem. If you put
49:39
pressure too much on drug prices, you can
49:41
very easily remove the incentives people to invest
49:43
in the industry. Quite a fragile ecosystem that's
49:45
built up once you start cutting down the
49:47
tall poppies, you actually
49:50
are removing a lot of
49:52
the incentives to even develop any
49:54
drugs. And there's just a few things that
49:56
you can do outside of damage by cutting
49:58
down even just a small
50:00
number of these huge lottery winners,
50:02
if you like. You have to accept a bit of
50:05
a trade-off where you are trading off
50:07
innovation for how much you're spending on
50:09
drugs. What about prevention?
50:12
We've talked entirely about
50:14
interventions where something bad happens
50:16
in the body and we've
50:18
developed ways of treating that thing or making
50:20
it better. What about similar
50:23
research that could go into, whether
50:25
it's lifestyle or other things that we
50:27
do before these bad things happen to
50:30
us that would prevent them from happening
50:32
in the first place? Any thoughts
50:34
on that side of the ledger? Yeah,
50:36
I think prevention is difficult. People
50:38
in the US often say that the
50:40
reason why the US doesn't invest in
50:42
prevention is because a lot of people
50:44
are on these employee insurance plans and
50:47
there's a lot of insurance plan turnover.
50:50
So any insurance company that invests in
50:52
prevention for its members is unlikely to
50:54
reap the benefits of those
50:56
investments when they switch to another employer and
50:58
get a different plan. But the
51:01
problem is that if you look at other countries
51:03
that do have these nationalized healthcare systems like the
51:05
NHS, other European countries, they actually don't
51:07
invest in that much in prevention either
51:09
even though they should in theory because it's a
51:12
whole nationalized health system. They should be
51:14
wanting to invest in those things. So is the
51:16
problem really that there's a lack of supply of
51:19
preventative treatments or is it really there's a lack
51:21
of demand for preventative treatments? And
51:24
I feel like the issue is probably more
51:26
on the demand side where people unfortunately don't
51:29
have a strong demand for preventative
51:31
medicine and they don't take up
51:33
preventative medicine in many cases when
51:35
it's offered to them. So it's
51:37
the whole thing with the argument
51:39
about othempic and why don't
51:41
people just practice diet and exercise. That's
51:44
a more sustainable way of dealing with obesity. The
51:46
reality is that people just don't do that. So
51:48
they would rather just take othempic when
51:50
they are obese and have that treat
51:52
their obesity. So I think this whole
51:55
consumer attitude shift that needs to happen
51:57
before preventative medicine ever becomes really established
51:59
needs to be demanded. So you think
52:01
about like the healthcare system, who is really the
52:03
end consumer of pharmaceuticals and healthcare system? I think
52:05
you can make the argument that it's actually healthy
52:08
people who are enrolling in the plans and are
52:10
paying the majority of the money that goes into
52:12
the system. And they want that
52:14
system to represent value for money for when they
52:16
do get sick. So you have to have
52:19
the people who are healthy have
52:22
a greater demand for products of
52:24
prevention that then incentivize companies
52:26
to invest in them. And that's a really important
52:28
thing to do is to provide the services. I'd love
52:30
you to talk about two extremes. The
52:32
things that have you in this entire
52:34
world, the most excited for the future,
52:36
and the things that have you the
52:38
most worried. I think what I'm
52:40
most excited about is not a specific technology,
52:43
but more this Cambrian explosion
52:46
if you like those modalities
52:48
and biotech. So
52:50
for a while, I think
52:52
back maybe 10, 20 years ago,
52:54
there were that many different types of different treatment
52:57
classes. And since the rise
52:59
of biotech, you have all these different ways
53:01
of treating disease that are becoming established and
53:04
have a lot of potential. So you
53:06
have CRISPR, RNA interference, these CAR T
53:08
cells we talked about, base editing, different
53:10
variations of CRISPR. Our ability to control
53:13
our biology is getting much
53:15
better than it was even just a
53:18
few decades ago. So
53:20
this increase in our ability to control
53:22
our own biology and intervene in biological
53:24
systems and very precisely
53:26
use these molecular scaffolds to push
53:28
a system into a desirable state
53:31
is really interesting. And just
53:33
this flywheel of improving on the
53:35
modalities that we do have. So
53:38
you look at something like CRISPR, we just got our
53:40
first CRISPR-approved therapy in the UK and the US. But
53:43
CRISPR is already in some ways becoming a bit of an
53:45
outdated technology. You have a lot of investment in what's next.
53:47
So instead of making cuts, you're going to have these
53:50
base editors that are changing specific
53:52
letters in the DNA sequence. And
53:55
that's a more effective way of precisely treating
53:57
certain genetic diseases than making cuts at CRISPR.
54:00
cuts. And then maybe after these base editors you
54:02
have prime editing, it's even more versatile in the
54:04
type of edits it can make. And you have
54:06
different types of editing. So just a lot of improvements
54:09
in tooling for how we intervene in
54:11
biological systems to treat diseases and
54:14
seemingly like an increase in the rate
54:16
at which we're developing these tools and
54:18
applying them to the clinic to solve
54:20
biological problems and to treat diseases. Like
54:23
general explosion in technology, our ability to
54:25
control systems is really interesting. I'm interested
54:27
in just general new business models of
54:30
the industry. So one
54:32
trend you've seen over the past, I
54:35
think it started probably in the 2000s,
54:37
is this greater externalization research.
54:40
So pharma companies, big pharma companies used to do
54:42
a lot of research in-house. They've
54:44
mostly externalized that to smaller biotechs
54:47
that are venture capital funded. So
54:49
the big pharma companies are really
54:51
just like commercialization machines all the time. And
54:53
they do maintain some research labs, but the
54:55
purpose of the research labs is in many
54:57
cases just to validate external opportunities and test
54:59
them in-house and have the expertise to actually
55:01
meaningfully evaluate these external opportunities they're going to
55:04
bring in. Maybe you'll see greater and greater
55:06
externalization of more and more functions. Now
55:08
we have innovation R&D externalized to
55:10
little biotechs. We have running
55:13
clinical trials being externalized to these
55:15
clinical research organizations that
55:17
specialize in that. You can argue whether that's been
55:19
a good thing or not, but it's just a
55:21
hard externalization trend and you have more and more
55:23
just pieces being externalized. Every commercial analysis
55:25
will be externalized in the future as
55:28
well. So you may end up with
55:30
pharma companies just being very specific IP
55:32
holding companies, plus just
55:34
managing the finances of actually selling and distributing
55:36
these drugs. Every little piece of
55:38
the ecosystem is focused on some specific component
55:40
of the process and maybe that
55:42
will help to be more efficient. So I'm interested in
55:45
that. I think you can argue that it
55:47
may not be helpful. Externalization hasn't been completely helpful.
55:49
There's been some trade-offs there. But there's
55:51
a lot of redundancy in the industry. And if
55:54
you're in a biotech who's getting ready to
55:56
launch, people having to build up redundant capacity
55:58
to run and execute trials in an efficient
56:00
way. And then what I'm worried about is
56:03
there's a genuine worry that it's
56:05
not going to be worthwhile
56:07
for a lot of companies to continue
56:10
to invest in developing new drugs. So
56:12
I'm worried that if we don't get
56:14
the balance, rewards innovation and incentives to
56:17
innovation, we'll end up with a system
56:19
where companies will decide, okay, I'm not
56:21
actually incentivized to really invest in doing
56:24
meaningful fundamental research and spending 30 years
56:27
tinkering on some opportunity that may
56:29
eventually bear fruit. I'm actually just
56:31
going to market the drugs I already have
56:33
that are old. I'm going to invest in
56:35
drugs that are very hard to copy. So
56:38
even though once they do go generic, no one's going
56:40
to be able to copy me. So things like these
56:42
radio, radio fair treats, talk about Cartes, they
56:45
have a whole process of sojourn with them, so they're
56:47
quite hard to copy. And then you're just going to
56:49
sit on existing treatments and try to milk
56:51
them as much as possible and use
56:53
tricks and techniques to extend the patent life
56:56
and make it hard as a copy and
56:58
try and extract as much value for as long as possible. And
57:00
maybe not invest in this fundamental
57:03
research that really drives meaningful products
57:05
and has this element of unpredictability to it.
57:08
One thing I think is quite dangerous in
57:10
the pharma industry is just this over-reliance on
57:12
an accounting type mindset. Rather than a recognition
57:15
that there's a science-driven industry, there's
57:17
an innovation-driven industry, the commercial
57:20
parts of big pharma should be in
57:22
service to the R&D portion of the
57:24
organization or the ecosystem. And
57:27
you should really be hiring good
57:29
people with good intuitions about what's worth
57:31
developing and letting those scientists tinker
57:33
for as long as they need to on some
57:36
of these really challenging ideas at the forefront of
57:38
what is possible and give them enough time, 10 or 20
57:40
years, for these ideas to mature into
57:43
actual products and then you can commercialize them.
57:45
And if the returns innovation are down too
57:47
much and maybe you get too many fully
57:49
managerial positions who are focused on the accounting
57:51
aspect of it, there's one of milk-existing products
57:53
and copy the big ones that we don't
57:55
have an existing market, then you're not going
57:57
to get a kind of meaningful process like...
58:00
like Keytruda, these immuno-oncology drugs, the GLP-1
58:02
agonists that we're seeing now. So
58:04
yeah, that's one thing I worry about. So just a
58:06
totally fascinating overview of one of the most
58:08
interesting parts of the world and of the
58:10
business world, of just the technology and innovation
58:13
world. I've so appreciated everything you've written and
58:15
sharing so much here with us today. In
58:18
these interviews, I always ask the same traditional closing question.
58:20
What is the kindest thing that anyone's ever done for
58:22
you? I think this has a
58:24
pretty obvious answer in my case. Many
58:26
people have done many kind things for me, right?
58:28
I think I just have to go with probably
58:30
what is the most common answer, which is my
58:33
parents have always been an unending well of support.
58:35
And they bailed me out of many low points
58:37
in my life. So I'm very thankful for that.
58:39
And now while other people have done me many
58:41
kindnesses, nothing really compares to what
58:43
my parents have done. Alex, thank you
58:45
so much for your time. Thanks. If
58:49
you enjoyed this episode, check out
58:51
joincolossus.com. There you'll find every episode
58:53
of this podcast complete with transcripts, show notes,
58:56
and resources to keep learning. You can also
58:58
sign up for our newsletter, Colossus Weekly, where
59:00
we condense episodes to the big ideas, quotations,
59:02
and more, as well as share the best
59:04
content we find on the internet every week.
59:28
Thank you.
Podchaser is the ultimate destination for podcast data, search, and discovery. Learn More