Professional Documents
Culture Documents
STAT Ebook How Technology Is Shaping Future Health Care
STAT Ebook How Technology Is Shaping Future Health Care
UNDERWRITTEN BY
TA B L E O F C O N T E N T S
B O S TO N • WA S H I N GTO N • N E W YO R K
SA N F R A N C I S C O • LO S A N G E L E S • C L E V E L A N D
Those advances were front and center at this year’s STAT Health Tech Summit,
which brought together researchers, policy experts, and top executives working at
the intersection of health and technology, including leaders from Google Health,
Teladoc, Livongo, 23andMe, and Fitbit.
This e-book includes all of STAT’s coverage of the event, as well as select
coverage of other topics in health tech. Those stories include a look at how
Lyft’s growing health business is trying to close gaps in health care access and
an examination of how AI could reduce existing biases in health care — or
exacerbate them.
Taken together, the stories capture the enormous promise of technology to make
health care more accessible — and underscore the pitfalls that could get in the
way of health tech reaching its true potential.
HOW TECHNOLOGY IS SHAPING THE FUTURE OF HEALTH CARE LETTER FROM OUR EDITOR
How Lyft’s growing medical
H E A LT H T E C H
Just a few years earlier, Callahan, now the vice president of health care at
Lyft, had been diagnosed with breast cancer. She saw firsthand how much
of a barrier transportation could pose to care.
“That was a moment where I thought, [this] is a problem I can get after, be-
cause I can understand what those patients are trying to deal with,” Callahan
told STAT’s Erin Brodwin on Thursday at the STAT Health Tech Summit.
In the past two years, Lyft has significantly expanded its medical transportation
business, which is designed to help patients get to and from medical appoint-
ments. Callahan also spoke with STAT about how that strategy can help
tackle some of the social determinants that shape a person’s access to care.
Typically those experiences are often public transport often loading patients into
a multi-load vehicle so think of like, a van with six to eight people in it. That takes
much, much longer to get to their appointment than it would if they were taking
a direct route… That leads to a tremendous amount of patient dissatisfaction…
The other thing that it does, from a member perspective, is if instead of two
hours to go to a doctor appointment, it’s going to take you four to six hours
because you’re waiting on average one to three hours for your ride to show up
for you. That means that you probably are going to forgo those appointments
because you can’t afford the additional childcare. You can’t afford to take time
off of work. And then you skip those preventative appointments, and often what
happens is those people end up in some kind of an adverse event.
Is there something specific that Lyft adds to the patient transport solution
that other transportation providers are not currently providing?
I would say that the on demand solution is very timely. [With Centene], they
were able to decrease wait times from 28 minutes to seven minutes.
…The way that we all expect now to interact with our life, it is not to sit on a
phone with a call center agent and schedule a ride three days in advance. And
then if something happens… and you have to move your appointment, [you
have] to get back on that phone call with that call center agent again.
How can you overcome patient privacy concerns for substance use disorder
patients whose records are protected by HIPAA and [other federal policies]?
What do you see at Lyft that you couldn’t see while you were working
at McKesson?
You know, I think one of the things that really attracted me to Lyft was to get
closer to the patient… I was diagnosed with breast cancer towards the end of [my
time at McKesson]. And that also really impacted my decision to come to Lyft,
because at that time — it was 2014 — rideshare wasn’t prevalent. I’m a fairly well
off woman in the health care sector, and I was trying to figure out how I could
get from my home into UCSF to get care for 18 months on an ongoing basis. And
when Lyft called, that was really a moment where I thought, you know, that is a
problem I can get after because I can understand what those patients are trying
to deal with.
Tell us more about what Lyft does to address social determinants of health.
Social determinants are really focused on where you live, where you work, where
you play, everything around you. Your ZIP code is is much more predictive of
your health than your health care and even your genetic code. So that incorpo-
rates things like transportation, your job. do you have access to healthy food. All
of those things are really encompassed into social determinants of health.
One of the things that I think we have a challenge with is connecting social
determinants of health into our clinical systems and. One thing that is become
very clear to me is that we don’t have a consistent way to pay [clinicians to]
evaluate people for social determinants of health issues [or] barriers that they
might have that preclude them from getting to healthcare. And we don’t have
a good way to code it, and often in healthcare, if you don’t code it, you can’t
measure the effectiveness.
It took a lot of late-night and early-morning Zoom calls plus some socially
distanced in-person meetings to create the first health tech giant, all conducted
during a pandemic and in near-total secrecy. Now that telemedicine provider
Teladoc Health and diabetes coaching company Livongo are moving closer to
clinching their $18.5 billion deal later this year, leaders of both companies say
they’re ready to provide a single solution for care that will satisfy consumers,
providers, and payers.
“We know there’s a lot of waste in health care, but we will trim a lot of that
waste out while also improving clinical health outcomes,” Jennifer Schneider,
president of Livongo, said Thursday at STAT’s Health Tech Summit. “I think
in five-ish years … you’re going to see a different model where the experience
of being a person receiving health care — and that’s all of us — is going to be
convenient. You’re actually going to not hate it.”
Teladoc, which went public in 2015, charges subscription access fees to employers
and insurers so that their employees and members can gain access to doctors on
its platform; patients also pay visit fees when they use the service. The company
relies on doctors who are independent contractors, rather than employing
providers directly. More than 70 million people in the U.S. have access to
Teladoc’s platform, and the company has expanded internationally as well.
Livongo, which went public last year, makes most of its money charging em-
ployers and insurers to provide diabetes coaching and monitoring to their work-
ers and members. The company also has growing businesses in other chronic
diseases including hypertension, weight management, and behavioral health.
There’s only a 25% overlap between clients of both companies, Gorevic said,
so there is room to take advantage of the new company’s combined offerings.
Telehealth as a whole has seen a steep rise and later fall in demand for remote
services, tied to the coronavirus pandemic lockdowns earlier this year, followed
by gradual reopening that allowed more in-person medical visits. But Teladoc’s
and Livongo’s business remained steady, Gorevic and Schenider said, because
the chronic nature of many patients’ conditions means their needs stay constant.
Asked if their vision for telehealth to reach more people depended on wide-
spread broadband connectivity, Gorevic said the combined company’s model
can improve care by being an “equalizer,” no matter which neighborhood you
live in or how close you are to the best health care.
Looking back at how news of their merger surprised some investors, Schneider
turned to a running analogy.
“Jason sits on the board of New York Road Runners Association and I’m a
runner and racer. And you know that when you’re running fast, you don’t slow
down, you accelerate,” she said. “So from that standpoint, we had a massive
lead, both companies, and it was exactly the time to accelerate, not to sit back.”
Remote monitoring technologies let doctors keep tabs on how you’re doing,
even when you’re nowhere near the doctor’s office.
That means there’s already reimbursement and training worked out, he said.
But there are still a host of challenges, including inclusive adoption strategies,
better software development, and meeting patients where they are.
We’ve seen the pivot to virtual care. We haven’t seen the full pivot to remote
monitoring. But Stanford, for example, went overnight with the flip of a switch
to 90% virtual care. And initially, everybody loved it. Patients loved it. Doctors
loved it. Allied health professionals loved it. And as things went on, we actually
now realize there’s a whole new set of what I’m calling last-mile problems for
digital health.
And so that’s what keeps me up at night: What is the net effect of this? Are we
actually building the right solutions? And do we really know where we need to
go? And hopefully not learn it the hard way. So, for example, in some ways, if
you’re sick, is remote care just delaying the need to go in? Because we haven’t
figured out how to deliver complex care at a distance. And then so related to
that would be: How do we do that? What are the ways that we can do that
synchronously and asynchronously?
But the interesting issue now is that if you look at the valuations of a phar-
maceutical company — we just heard a great discussion on vaccines or other
drugs — it’s tied to the randomized trial data. It’s tied to the evidence. But if
you look at the valuations of the tech companies, it’s tied really to projections
of revenue, which, in my opinion, are often not grounded in reality with
respect to adoption of clinical evidence.
And so it is important not just to do the observational study or the small pilot
or to have really squishy outcomes of engagement — but to really do the hard
things, to randomize patients, because that’s when you remove the problems
that observational studies are fraught with. And to really look at meaningful
and sustained outcomes, not six-month outcomes, but to go a bit longer. And so
the good news is we’re starting to see that across the board with digital health
trials and interventions.
So a lot of digital health research has historically been focused on the sort
of proverbial cyclist on Sand Hill Road in Silicon Valley. That’s the kind of
very health conscious and often wealthier person who owns lots of pricey
wearables. But your group is trying to expand beyond that population
to develop digital health interventions for Latinx populations, for Black
populations, as well as for gig workers. Could you tell us about that work?
And when we looked at the trials now published, we saw that we enroll 25,000
people over the age of 65. And many, many people had real, hard comorbidities:
diabetes, hypertension, heart failure, prior heart attack, to name a few. And so
what we realized is, actually, there is adoption. And what we’ve seen in an annual
report that we do in partnership with Rock Health is that digital adoption is real
across the whole U.S. There are barriers, but people are doing this. The problem
is that the products really aren’t clinically facing a wider variety of people. And
so we’re trying to think through where those use cases are. And if you look at the
people who have cell phones, it’s everyone.
And so how do you engage, for example, gig economy workers? We’re going
through a major potential legislation change on how gig economy workers are
viewed, whether they’re employees or contract workers. But that has a lot of
issues around health care benefits, insurance access, health care access. We
know, for example, from a classic study called the London Transport Study and
from studies of New York taxi drivers that they bear high cardiovascular risk
for a variety of reasons. So how can we deliver care where they are? And so one
example we’re working on is hypertension and really trying to design products
that aren’t disintermediating their regular doctor or relationships they may
have with brick and mortar — but are letting that physician use this to guide
and optimize hypertension care. And so that that’s another theme here:
disintermediation and reintermediation of clinicians.
We’re in the long haul for digital health — where it really is health and it’s not an
adjacency to regular health. And right now, startups trying to get early revenue,
show viability, a product market fit, understandably react to a health care system
or doctors or practices being too slow to adopt by going around them. And so,
yes, we saw great risk management and great early, mostly observational, though
non-randomized studies of how employees are using these tools. The model that
goes around their doctor completely, to some extent, works because you might
be able to do that for diabetes or cardiovascular risk. But none of that is really
getting back to the doctor in a meaningful way. So it’s not well integrated.
The other area that we’ve seen telehealth take off is transactional health care.
And so that’s an urgent care visit. And now it’s a Covid visit. And there’s a
market for that: a company called Sesame just launched their own marketplace.
And what’s fascinating — I checked out their website this morning — is how
little, in terms of revenue, we’re seeing doctors willing to take for these short
visits. But there’s no continuity. They’re focused, and they’re transactional. And
maybe that’s OK, but it really doesn’t fit into the broader landscape of what you
need for health. And I think that’s the tough, harder problem to figure out, but
the important one.
What’s the path for novel digital measures of health to make it into electronic
medical records where a clinician can see and use them? Are there other
reasonable avenues of getting those measures into the hands of clinicians?
And I think there’s kind of a tale of two paradigms here: The first paradigm is
you can go to your local big box pharmacy and buy a blood pressure cuff. No
problem. And you can get it on your own. And there’s no manual tool kit on
EMR integration; those have been around before smartphones and before the
internet. There were no issues, and doctors were not saying: Oh my god, we
should not be selling blood pressure cuffs to patients because how do they know
what to do? The same is true of diabetes to the extent that patients manage
their own therapies. So the data — for a clinician to actually want it — have to
be actionable. We really don’t want data that we can’t really use. And so part of
the design spec is to figure that out.
The other use case I think is very mature is remote patient monitoring with
sensors. It’s is actually not a new paradigm. I’m an arrhythmia specialist, and
we’ve been doing this since the 1990s. So there are implantable devices that
transmit to nightstand devices and now cell phones that then go to the cloud.
And that data is filtered and sent back to the clinician in an actionable way.
There’s reimbursement, there’s workflows, there’s even training pathways for
allied professionals to be good at this. And so we don’t see that in the current
landscape of non-implantable devices. And I think that’s what needs to happen.
So EMR integration is certainly important. But you could argue then: How
come we don’t integrate someone’s social media whereabouts on the EMR, too?
The data needs to be actionable; first or foremost, it needs to be presented to the
patient in an actionable way. And then what we’re not seeing is building software
products that are based on disease use cases, heart failure, atrial fibrillation, etc.
It’s a really great question. I don’t think the need is that different. I think that it
really is how we get there and that’s being done in different ways.
So you have a couple of things going on. You have large healthcare systems like
ours at Stanford trying to figure out the whole “make versus buy” argument.
And we’re large enough and we have so many programmers, data scientists,
and a great medical center, but also a great group of faculty and engineering
and everything where we don’t have to buy — we can make and we can handle
it. But that’s not true of everyone. And so some people are going to build their
own stack.
The extreme would be a company like Forward that goes top to bottom and
has not vendorized pretty much anything in their stock of care, including their
sensors. And then the others are going to have to tag on. And then you have the
question: How do you give low-cost health care in really meaningful ways where
you can use the technology?
And so that’s the thesis of Todd Park and his company Devoted Health, for
example. They’re trying to understand: if they build out a stack in a managed
Medicare or Medicare HMO population, how can we reduce costs? So all of
these experiments are happening, and we’ll just see where they shake out. What
I think is not going to work, though, is with a subscription model where you’re
across different platforms and you’re paying for diseases. So, you know, I can’t
go to Netflix to manage my hypertension and then go to Hulu to manage my
diabetes.
So I want to ask you a little bit more about the Teladoc and Livongo
merger. I’m curious how you see that alignment and the combination of
the two companies fitting in with this paradigm you describe involving
two models: One involves bypassing the traditional health care system
and selling to employers and kind of having your own health care system.
The other model tries to integrate in with a continuous provider that the
patient has a relationship with in-person.
First of all, both companies have been wildly successful at what they do, and
they’ve shown great product market fit and they’ve developed products that work
in those use cases. But what we’re seeing now is this theme, I think, and I’d love
to hear what they say, but, you know, the opportunity for reintermediation. So
you have a great virtual cure platform. And then you have a great set of disease
management tools within a company. And so it makes obvious sense to put those
together. The real question is, you know, how do we envision digital health work-
ing and remote monitoring? Is it going to be a parallel universe that a patient has
kind of almost for their alter ego to health care, but they still anchor to a brick
and mortar physician because they have insurance? I hope not. I think it all
needs to consolidate. And that’s where we see this opportunity of what I think is
reintermediating physicians. So physicians also, you know, may not be working
for a single platform anymore. We have seen over the last 20 years now, basically
health care systems, buying practices and consolidating, which gives them pur-
chasing power. And now we’re seeing an increase of supply with a lot of these
virtual solutions. So now, you know, my hope is that doctors can become free
agents as part of this and really deliver care across many different platforms.
Many reimbursement policy experts have looked at the various payment pay-
per-performance incentives that we’ve seen, for example, in Medicare. And they
they sometimes work, and they have a small effect, but they often don’t work at
all. So if you want to break this thing and start over, that would be great. But it’s
much easier said than done. That goes all the way down to the fact that in
America, the government cannot negotiate with drug companies.
So, brand-new drugs that get released — there’s a whole suite coming out for
heart failure — are going to cost more in our country versus others. So the
way to get around that has been to show better demonstration of value, for
example, to managed Medicare populations, direct-to-consumer payment,
going to employers to show value. And obviously, everyone is working around
those areas.
But we also have the ability for Medicaid and insurance programs where you
have expansion pools where people can select where they really haven’t matured.
So I think this is a long haul. And I really, truly don’t know how this is going
to change. One major change could be in a post-pandemic world, depending
on where people are with respect to jobs. We’ve seen unemployment numbers
nosedive and start to go up as they may broker their own health care in really
interesting ways. So they may get high-deductible health plans and then hedge
against catastrophic illness with that. But then for the daily stuff — that runny
nose, the sniffles, general wellness checks — they may use low-cost telehealth.
So with remote monitoring, there is new reimbursement now, and it’s stable, it’s
working. And we’re seeing a lot of companies try to get in with that. We haven’t
seen major consolidation across different areas. The cardiac stuff is happening
mainly in companies whose core business is is ECG outpatient monitoring. And
so we’ve not seen multiple disease states under a single solution, nor have we seen
a lot of development in great software products that can work asynchronously
between doctors and patients. So I think that will definitely continue.
I’m really looking to see what the traction is of remote monitoring and virtual
visits. We saw virtual visits skyrocket, but in the last quarter the claims have come
down. The number of people has gone down. We’ve seen more face to face. What
does that steady state look like? And can we learn if we’re just going to delay
the inevitable, that you have to come in? Or can we think creatively on on how
we can have complex diagnostics at home? Is it possible to get workers to go
out and do a ECG or echocardiogram or other studies, meeting patients where
they are? And can we scale something like that? Is it possible that patients
can do AI-guided diagnostics themselves in a safe way where they can avoid
coming in and mail things back? So I’m really interested to see how we think
through those ideas and build products around them.
23andme has long been known as a consumer genetic testing unicorn. But CEO
Anne Wojcicki describes it differently, as “the people’s research company.”
Any such decisions are years away, a 23andMe spokesperson later noted, which
should give the company time to figure out a way to ensure its customers benefit.
The first medicine being developed by the partners, an anti-cancer antibody, is
only now starting clinical trials.
The experimental cancer drug that entered clinical trials in July “was actually
one of the early programs under Richard Scheller,” Wojcicki noted. Scheller,
23andMe’s former chief scientific officer who was considered an instrumental
figure in the company’s push into drug development, left in July 2019.
In 2018, the company signed an exclusive four-year deal with GSK, with the
option to extend their work an extra year. The companies are working on about
30 drug programs together.
At the time the deal was announced, 23andMe faced some backlash for
providing information distilled from its consumer clients to pharmaceutical
companies. Both companies have argued that they are transparent about how
consumers’ data is used; the press release announcing the deal noted that
23andMe customers would have to opt-in before their de-identified data was
shared with GSK.
“I think what you see is that for people who are really sick, they want to make
a difference. They want to make a difference in their life or they want to make
a difference in the lives of their children,” Wojcicki said on Wednesday. “People
participate all the time in studies at Stanford or with Pfizer or with other groups.
People want to see their information used for good.”
As part of the collaboration, GSK invested $300 million in the genetic testing
company; the companies agreed to evenly split the cost of the work done as part
of the collaboration, which GSK hoped would “improve the probability of R&D
success,” according to its 2018 annual report.
GSK’s $300 million investment was enough to buy 14.5% of 23andMe, which
is still privately held; Wojcicki noted Wednesday that after 14 years, it is “not a
profitable company.”
23andMe’s core business is selling genetic testing packages to people who are
interested in learning more about their ancestry, their predisposition to certain
genetic diseases, or their ability to metabolize certain drugs.
In 2017, the company received clearance from the Food and Drug Admin-
istration to market its tests as a way to detect a select group of hereditary
conditions: Parkinson’s disease, a type of Gaucher disease, several hereditary
blood disorders, and alpha-1 antitrypsin deficiency. Another FDA clearance
followed in 2018 for pharmacogenetic tests, which look for genetic markers of
drug metabolism.
Covid-19 may turn out to be digital health’s tipping point, two Silicon
Valley venture capitalists said Wednesday. The industry’s rapid shift could
be beneficial for patients, providers, and some industry players — but it also
might prove detrimental to companies that swerved toward Covid-19 and
away from the bedrock of their businesses.
“I’ve been worried about all of these companies that have suddenly pivoted
to Covid overnight,” Lisa Suennen, an investor who leads the firm Manatt’s
venture capital fund and its digital and technology businesses, said in a panel
at the STAT Health Tech Summit.
“I feel like every business plan, they crossed out ‘AI’ and wrote in ‘Covid-19,’”
she added.
Suennen said it’s encouraging to see so many health tech companies pitch
in on the Covid-19 response. Tech companies have launched symptom bots,
built digital contact tracing technology, and created new tools to analyze case
data and spot emerging hotspots. But, experts said, a full pivot to Covid-19
— at the cost of their core capabilities in other areas of health tech — might
not be a smart strategy for every company.
Stead said she and many other venture capitalists and investors saw a flood of
companies shift their focus to Covid-19, hoping to address some aspect of the
health crisis. For some companies, Stead saw the effort as “wishful thinking.”
“I think because of the scope, breadth, and speed of a pandemic, unless you
were already working in the space and had a core competency to do whatever
you were proposing to do … we didn’t believe it was a realistic business model,”
she said.
But it’s clear that the pandemic has dramatically accelerated digital health’s
adoption with such speed and volume that it’s difficult to see health care turn-
ing back from tech tools. In some ways, it’s taken 20 years for digital health to
become an overnight success, Suennen said.
That’s easiest to see in telehealth, which has allowed patients to see their doctors
when in-person visits were — and may still remain — inadvisable.
But digital health also extends to remote patient monitoring and synthetic clinical
trials, which combine health records from many patients to test therapies, Stead
said. The growth in those areas could prove crucial for pharmaceutical com-
panies whose patient recruitment efforts for clinical trials may be lagging due
to physical distancing practices.
That strength may outlast the pandemic. “This is not just a stopgap,” Suennen
said about telehealth. “The energy behind proof is growing more and more, as
it should.”
At best, that technology has the potential to make it easier to detect and diagnose
diseases, streamline care, and even eliminate some forms of bias in the health
care system. But if it’s not designed and deployed carefully, AI could also
perpetuate existing biases or even exacerbate their impact.
“Badly built algorithms can create biases, but well-built algorithms can actually
undo the human biases that are in the system,” Sendhil Mullainathan, a
computational and behavioral science researcher at the University of Chicago’s
Booth School of Business, told STAT’s Shraddha Chakradhar at the STAT
Health Tech Summit this month.
Mullainathan also spoke with STAT about about the importance of commu-
nication in developing AI tools, the data used to train algorithms, and how AI
could improve care. This conversation has been lightly edited and condensed.
And when they did that on a bunch of X-rays, they noticed it was looking in a
particular region. They were like, well, this is odd. So they zoomed in and as they
zoomed in and clarified what was going on in that region, what they saw were
these pen marks. And they’re like, “What is this doing here?” And it turned out
that in their dataset, when radiologists noticed something interesting, they would
put pen marks there. And the algorithm wasn’t identifying disease so much as it
was identifying pen marks.
And you can see how in the data they had, they had what appeared to be a good
performing algorithm because pen marks was associated with disease. But it’s
unlikely to do that anywhere else, especially if what you were thinking was, “Let’s
automate radiologists out of there.” So in a perverse way, the algorithm was
seeing something quite different than by all accounts, it appeared to be seen.
Every radiologist knew they put pen marks. Something was broken in the
human communication system that when the data was handed over, they didn’t
say, “By the way, every X-ray that has disease also has pen marks, every X-ray
that doesn’t, doesn’t have pen marks. So you might want to watch out for that.”
Perhaps the best example of your work emerged around this time last year
when you found that racial bias in a commonly used hospital algorithm
was working to perpetuate racial bias in patients who were in that hospital
system. Can you talk to us about that?
This is one of the strangest pieces of research I’ve ever worked on, just because
you rarely have a chance to do something that’s on this scale. So this is a category
of algorithm that, you know, depending on how you count either 60 million or
100 and something million of patients are exposed to.
And the way these algorithms work is they’re just predictive — they say,
“Hey, based on everything I know about you, how much care do you tend to
use? And if you look like you’re going to use a lot of care, let’s put you in these
expensive care coordination programs.” They make a lot of sense. And so
what we did is we said, “OK, let’s take these programs and let’s look at how
they do.” … They do a good job of finding the people who need a lot of care.
And that’s why people buy them.
But the surprise came, as you alluded to, in the racial element of it. When you
looked at how well they did for whites versus Blacks. What you found is at the
same level of illness, Blacks were much lower ranked than whites. To the extent
that if you were to equalize, you would more than double the number of Blacks
being put into these programs. So it’s as if for the same level sickness, Blacks
were given a much lower score.
It’s tempting when you think of algorithms being biased to imagine there’s
something nefarious going on. But when you dug into this, what happened
was another just fairly simple communication error — but very consequential
communication error. So I keep using the words “find the people who need lots
of care.” So what does that mean, care? If we unpack it, there are two ways we
could define care.
We could go to your data. Look at your claims and say, “Oh, here’s a person
who we’ve spent a lot of money on.” Now, that’s the easiest data to get, because
that’s claims data. You can also go and look at care and say, “Oh, this is a
person who ends up being very sick.” … These two are used interchangeably
in health care a lot.
Now, here’s a tragic fact in the United States. Health — physical health and
health as measured by dollars — don’t relate in the same way for Blacks and
whites at every level of sickness. We spend more on whites than we do on Blacks.
So what the algorithm was actually asked to find is expensive patients. Expensive
patients are disproportionately white patients because we spend more on them.
So this subtle miscommunication that crept in and it actually there were there
were six algorithms of this variety and it actually crept in apparently to all of
them. So it’s not as if these people were stupid. It’s not as if you just get better
data scientists. You just get a better team.
It is true that a poorly built algorithm will end up embodying the biases that we
have as humans. That’s what we see here, because costs are a biased function of
health. We train on costs and we got this problem. And I think the first line of
defense we have against all of this is just to check.
And the nice thing about algorithms is that they sit in a box and we can look
at their behavior, we can tweak them, we can keep working on them. I can’t go
in and tweak what’s inside a doctor’s head. … So algorithms [that are] poorly
designed really are quite a big problem. But they actually offer this amazing
opportunity for us that if we’re careful, we actually can do a lot more good
things with them.
I don’t know if you’ve ever read this book “The Diving Bell and the Butterfly.”
This is about this guy who basically — to make a long story short — he basically
had to blink. He was trapped in his body. And the only muscle he could you can
move where his eyelids. So you could blink. So you think of how horrific that is.
You just can’t move anything else but your eye. But he wrote this entire book to a
sequence of blinks. They had a whole code …
So why am I telling you this? There is now work that over the last 10 years has
said, wait — for people in that position, we can put EEGs on their head, and
actually read out the brain signal, build an algorithm that translates that brain
signal consistently into their attempt as to where are they looking on a particular
keyboard, for example? Which allows people to basically type with their minds,
which is like science fiction. And it is something that seems unbelievable …
The workings of the algorithm, which predicted that her risk was low, were
familiar: Barzilay helped build that very model, after being spurred by her
2014 cancer diagnosis to pivot her research to health care.
Barzilay’s work in AI, which ranges from tools for early cancer detection to
platforms to identify new antibiotics, is increasingly garnering recognition:
On Wednesday, the Association for the Advancement of Artificial Intelligence
named Barzilay as the inaugural recipient of a new annual award honoring
an individual developing or promoting AI for the good of society. The award
comes with a $1 million prize sponsored by the Chinese education technology
company Squirrel AI Learning.
With the new award, AAAI aims to counterbalance the widespread messages
of concern circulated in the news media and by other commentators about the
potential negative impacts of AI. “What we wanted to do with the award is to
put out to the public that if we treat AI with fear, then we may not pursue the
benefits that AI is having for people,” Gil said.
Barzilay’s treatment was successful, and she believes her clinical team at
MGH did the best they could in providing her with standard care. At the
same time, she said, “it was extremely not satisfying to see how the simplest
things that the technology can address were not addressed” — including a
delayed diagnosis, an inability to collect data, and statistical flaws in studies
used to make treatment decisions.
Barzilay said she thinks it’s incumbent not just on the AI community, but also
people outside of it, to turn the abundance of research on AI in health care
into tools that can improve care.
To try to change that, Barzilay has delved into drug development, building
a machine learning platform that was used to identify a novel antibiotic that
effectively treated a gastrointestinal bug in mice in a study published earlier
this year in the journal Cell.
She’s also co-leading the team developing the AI model for assessing breast
cancer risk that was used on her own mammography data in June. In a study
published last year in the journal Radiology, Barzilay and her team trained,
validated, and tested their model on historical data from about 40,000 women
who were screened for breast cancer. They found that their model could dis-
criminate risk better than an older, widely used risk evaluation tool, known as
the Tyrer-Cuzick model, that relies on breast density to assess risk.
For now, with the model still in development, only a subset of MGH radiologists
are actually viewing those predictions. They typically do not share the results
with individuals who are screened. (There are occasional exceptions, such as in
cases in which a woman has read about the model in the news media and makes
a specific request to her radiologist to pull up those numbers.)
Another exception was Barzilay, because of her scientific interest and her role
in the model’s development. When she went in to MGH for her most recent
mammogram, she and Constance Lehman — the hospital’s director of breast
imaging who is co-leading the project with Barzilay to develop the model —
pulled up the model’s output and discussed the predictions.
Barzilay doesn’t remember the precise probabilities that the model spit out about
her personal risk of recurrence, but the overall forecast was reassuring: “There
was nothing remarkable there,” she recalled.
“Millions of Americans suddenly asked themselves, ‘Can I solve this care need
without showing up in person?’” said Sean Duffy, chief executive officer and
co-founder of Omada Health, a virtual diabetes care provider. “That consumer
expectation change is going to be the thing that writes history the quickest.”
For established companies like Apple, Amazon, and Alphabet, the exploding
popularity of health tracking is a boon to their push to make sizable inroads
in health. Those companies are courting the new patient-consumer with a
device-first strategy, transforming their bestselling wearables into health tools
with medical capabilities.
Meanwhile, health tech companies like diabetes care providers Omada Health
and Livongo are taking a platform-driven approach, catering to patients with
remote monitoring programs that connect them with health professionals and
provide useful data.
T H E TAC T I C A M O N G T E C H G I A N T S
Tech companies are starting to chart their path to remote monitoring by trans-
forming consumer gadgets to medical devices, with an eye on clinical evidence.
Apple was the first to enter the space this way, publishing a large and entirely
virtual clinical study of its Apple Watch and embedded electrocardiogram,
or EKG, which records the heart’s electrical signal. The study, which Apple
brought to the Food and Drug Administration as part of its work to get the
watch cleared as a medical device, showed the device could spot the heart con-
dition atrial fibrillation, or A-fib. Fitbit, which was acquired by Google last year,
is following the same path. The company launched a similar virtual study of its
wearable in A-fib in May and plans to present the data to the FDA.
Even Facebook, which has yet to make its own wellness wearable, appears to
be edging toward the heart monitoring space.
The tech- and consumer-driven shake-ups are already creating ripple effects
throughout the health care system. Clinicians, for example, are increasingly
being asked to interpret the results of Apple Watch EKGs in patients who are
hesitant to come in for a visit during the pandemic.
“You’re really seeing a shift where it’s consumers and consumer electronics
deciding things more than a doctor deciding which device to use,” said Ritu
Thamman, cardiologist and assistant professor of medicine at the University
of Pittsburgh School of Medicine. “We’re being pushed by the consumers
themselves, and that’s creating the competition and the drive to create the
best user experience.”
“Just because you have a watch that tells you things doesn’t mean you have
remote monitoring. The platforms and the integration need to start,” said
Mintu Turakhia, a cardiologist and executive director of Stanford Medicine’s
Center for Digital Health.
T H E P L AT F O R M - F I R S T P U S H BY H E A LT H T E C H C O M PA N I E S
Virtual care businesses, in contrast to tech giants, are jumping into the health
tracking space with a platform-centric strategy. Companies including Omada, as
well as Alphabet subsidiaries Onduo and Verily, offer care delivery programs that
harness remote monitoring hardware made by other companies and use fleets of
faraway health coaches to help patients interpret and understand their data.
But virtual care companies’ business models often rely on reimbursement or buy-
in from health insurers or employers, meaning their success depends on being
able to consistently demonstrate their effectiveness with research. And while
many of these companies have published small and short term studies, academics
and researchers say larger and more comprehensive research is needed.
“There’s very little clinical trial data” for remote devices, Turakhia said.
“We are definitely thinking about ways to reach outside hospital walls,” said
Anders Wold, vice president and chief executive officer of clinical care solutions
at GE Healthcare.
If those changes are to have real sticking power, however, companies including
tech giants and health tech providers will need to figure out how to make their
devices an established, long-term component of the existing health care system,
rather than simply a temporary or one-off solution.
“There are a lot of consumer devices out there with [FDA] clearance,” said
Turakhia. “But when you’re talking about remote patient monitoring, you’re
really talking about the whole system.”
Emerging tech and digitalization have increasingly shaped innovation and the
pace of change in health care and life sciences. The coronavirus pandemic is
accelerating technology adoption as health care systems and governments work
to slow its spread; health care providers seek to care for patients safely; research-
ers work to keep clinical trials on track; and pharmaceutical manufacturers seek
to maintain the global supply of life-changing therapies and bring vaccines and
treatments for Covid-19 to patients.
In the coming months and years, we will see how innovation born from the
challenges of the pandemic will affect health care and life sciences long-term.
At the STAT Health Tech Summit in September, virtual attendees were
asked about the biggest challenges and opportunities of artificial intelligence
and other new health tech today and in the future.
8%
21%
3%
14%
Nanotech
Wearables
Gene Therapy
30%
Robotic assistance
54%
23%
39% Analytics for forecasting the
pandemic’s impact
30%
9%
Legal concerns
41%
Regulatory hurdles
20%
Relevant expertise
3%