Modernization Hub

Modernization and Improvement
AI: Saving Lives and Resources 24/7/365 (Cloud Next ’19)

AI: Saving Lives and Resources 24/7/365 (Cloud Next ’19)


[MUSIC PLAYING] DOUG BOURGEOIS:
I’m Doug Bourgeois. I’m a managing director
in Deloitte’s Government and Public Services
consulting practice. And I’m also the lead alliance
partner for our relationship with Google in the
government industry sector. A couple of my colleagues
will be joining me on stage in just a few minutes. Sean Conlin, who is a principle
in our Analytics and Cognitive Government Public
Services practice. He’s going to focus on the
persona today of a data scientist. Meera Kanhouwa is a
medical doctor and also a managing director in our
Life Sciences and Health Care practice. And she’s going to address the
situation from a practitioner’s perspective. So before we get into
the subject of today, the opioid epidemic
and a solution to help look at it a
little differently, I’d like to spend a few minutes
talking about our relationship with Google and why
it’s so important. Actually, I’m going to
focus on a few keywords here around the relationship,
three to be specific. First, is transformation. Deloitte is really all
about transformation. We’re about transformation
for our clients. We help our clients plan for
and execute broad transformation initiatives to drive their
strategic agendas forward and leverage innovation
in the process. Google technologies, I’m
sure you’re all aware, because you’re here,
Google technologies are transformative. They often form the catalyst
of broad transformation and innovation for our clients. And that’s one of the
reasons why joining together and joining forces, Deloitte
and Google is so powerful. So more specifically, some
of the things that Deloitte brings, we wrap our services
around Google’s technology and Google services,
and we address things like program
management, for example. We bring broad experience
and proven methods to help our clients
plan and execute successful
transformation programs. We also address project
management expertise, both the mechanics of
project management, but also bringing domain experts
in our clients’ mission areas, and also experts in the
technologies at hand around Google that we
bring to those projects, and execute from a
project standpoint successfully as well. In addition to that, as you
can imagine, with a focus on transformation,
change management is potentially a substantial
issue for our clients. So we bring again,
experienced practitioners and proven methods to help our
clients plan for and manage the organizational
change that comes to light invariably when
you’re driving transformation for them. And then finally, and this
is an area where we’ve really made substantial
investments from a firm standpoint over the
last several years, is the technology, engineering,
is that we essentially provide scale to Google
engineering and Google technology by bringing
several thousand trained Google engineers
to the table as well. One quick note on our
capability, so the second word. If the first word
was transformation, the second word is capable. We are very proud
about the announcement earlier this week that for
the second consecutive year Deloitte was named Google’s
Global Systems Integrator Alliance Partner of the Year. And that is, of course, the
result of a substantial amount of investment, a substantial
amount of commitment by a tremendous group of
leaders and individuals from a broad scale perspective. And I think it’s primarily
indicative of both the breadth– it’s global– and it
spans all industries as well, and the depth, as I
mentioned– addressing strategy through architecture,
engineering, and even into operation. So it’s indicative of our
breadth and our depth. And you know some indicators
of that, of course, we have a premier partner
status with Google Cloud. We go to market around key
cloud solutions like SAP and Salesforce, and then
not the least of which some of our specializations,
and relevant to today’s conversation, machine
learning and cloud migration
specializations as well. So the third word– if the first word
was transformation and the second word was capable,
the third word is solutions. From a go-to-market
standpoint, we’ve really focused on developing
solutions using Google Cloud and Google technologies that are
addressing some of the broader kind of impactful challenges
that our clients face everyday to drive the transformation
of their own mission and their own objectives. So the first– I’m not going to go
through all of these, but a couple that
are worth mentioning is the analytics solutions
with MissionGraph. It’s primarily a platform. And it’s a platform that ingest
data across disparate data sources, looks for
relationships among that data, and then has some really strong
and powerful visualization capabilities to help us
position our clients to gain substantial insights
into the information that they’re looking at. And that is actually
the platform that was used to develop the
Opioid360 solution that we’re going to go through here
in just a few minutes. Second, fraud detection. Of course, I’m in the government
practice and my colleagues as well. But if it involves money
changing hands or transactions, fraud is potentially
an issue broadly. So we’re addressing that again,
with a platform approach, bringing multiple
solutions to address fraud detection from a variety
of perspectives to the market. And then finally IoT. IoT is a hot topic these days. But we see it as very much
a convergence of cloud, of analytics, and
IoT at the edge. And we’ve got some really
interesting solutions out in the market
that we’ve been able to partner with Google to
develop and help our clients look at some of
their new challenges out at the edge
of their networks. Of course, security comes
into play there as well. So again, transformational,
capable, and a solution-focused. And then, of course
Opioid360 is the area of focus for today’s discussion. So in just a few
minutes my colleagues are going to come up on stage
and start to maybe enlighten us a little bit around
thinking around this particular challenge. But first, what
we’re going to do is we’re going to show a quick,
very brief video to kind of set the stage for the way we’re
looking at the problem. [VIDEO PLAYBACK] – The opioid epidemic is
like nothing we’ve seen. Each day it claims an average
of 130 lives in the US alone. Its impact can be felt across
families and communities, and it is being fought by first
responders, law enforcement, and health care providers. But its complexity
means no one group is able to offer a
complete solution. To combat the opioid epidemic,
those at the front line must be swift, deliberate,
and collaborative. They need to intervene early
on to help focus resources and protect lives. They need to get to
the future faster. That’s why in collaboration
with Google and DataStax, Deloitte has created
an analytic solution for the people and organizations
responding to this challenge. Opioid360 is sophisticated
and flexible, synthesizing de-identified
data from disparate and siloed sources and providing
insight into the crisis like never before. And it leverages
our deep experience with AI and predictive
modeling, helping your organization take the
right action at the right time. Imagine those on the
front being able to dial into the needs of
specific populations. Imagine the risk of addiction
mapped against barriers to treatment, such
as housing security, unreliable transportation,
or lack of insurance. Imagine all of this
powered by a dedication to privacy and security and
the empathetic knowledge that the people are
more than data sets. People require
tailored interventions, ones that match
their medical needs and address any barriers to
care, like transportation assistance for Chris so he
can get to his local treatment facility, or the
support Maria needs to keep her family of four
together while she receives treatment, and information
for Alex’s medical provider so he can make the
best choice about how to manage Alex’s back pain. We don’t have one opioid
epidemic, but many, striking in tandem. With Opioid360, we gain
much needed visibility. Its predictive models help
identify at-risk populations and treatment barriers
earlier in the lifecycle, leading to timely,
proven interventions, and its flexibility
makes it a viable tool for each of the multiple
organizations trying to solve this complex, costly,
and interconnected problem. A future with lowered rates of
opioid misuse can start today. It’s time for Opioid360. [MUSIC PLAYING] [END PLAYBACK] SEAN CONLIN: Good afternoon. Many of you in this room fall
into one of two communities. One of the
communities is made up of health practitioners,
policymakers, scientists, and others who operate on the
front of the opioid response. And for all of those in
that community, thank you. There’s a second
community that’s made up of the data scientists. And the data scientists
are operating. They are driving across
and making a difference in our economy and
in our society. And the time is right
for that community to have a big impact in
helping with the opioid crisis. My partner Meera and I are
going to share with you today a solution, an innovative
solution called Opioid360. Its holds the promise
of potentially being that solution from
the data science community that can make a big difference
for those on the front line. The way we’re going to
structure today’s conversation is just that, a dialogue
between those two communities. I am going to be the voice of
the data science community, and Meera is going to be the
voice of the first responders– or the front– sorry. MEERA KANHOUWA: Front line. SEAN CONLIN: Front line. Sorry. The front line community. Myself, I’m Sean Conlin. I am a principal in
Deloitte Consulting. I bring AI solutions to
government organizations. The organization
I am a leader in has about 1,000 data scientists. That’s about 10% of our
national data scientists. Meera, who I’m
going to introduce– will introduce herself
in just a minute. But the opioid solution
that we’re bringing forward is an innovative solution
that brings AI and large data sets together, and holds
tremendous promise. We’re excited to
share that with you. Meera, could you introduce
yourself and offer perspective on the opiates crisis? MEERA KANHOUWA: Sure. Thank you, Sean. Everybody, it’s a pleasure to
be here with you this afternoon. My name is Meera Kanhouwa. I’m an emergency
medicine physician. And I’ve practiced on the front
lines of inner city emergency departments from
almost 20 years. I have physically laid
hands on 80,000 patients, not including those I treated in
a war zone while I was in Iraq. I’ve seen every aspect of the
opiate situation from point of injury, acute injury,
to heroin overdoses, to deaths in my
emergency department. And let me tell you what. There’s nothing that’s
more painful than trying to tell somebody at 3:00 in
the morning who is begging, begging and crying for
350 milligrams of Demerol that they can’t have it
because you know that they’re on the road to addiction. And it’s a very, very difficult
and complicated disease to treat from a
provider perspective. So I’m going to speak
to today as a provider. I left clinical medicine
after 18 years of practice because of the
promise of technology having an ability to
truly transform the way we deliver care. And in an industry
where our data is growing at the rate
of petabytes per year– petabytes per year, right? We on the front lines taking
care of patients are data poor. Imagine that irony. How do you solve that problem? So what I would really
love for technology to do for me and
all of my colleagues that take care of patients
day in and day out, is make us smarter. Help me take care of
my patients better. Help me do it more efficiently
and more effectively so that they don’t
keep bouncing back to the emergency department. And in our nation’s
emergency departments, when I started practicing, we
had under 100 million visits a year. Currently in the nation’s
emergency departments, the growth to
emergency departments has increased unfettered
at the rate of 8% to 12% per year for the last 25 years. Why is that? People don’t have primary care. They have no way
to go to get care. They can’t be seen
on a routine basis. Virtual health isn’t deployed. They don’t have access. They don’t have transportation. They don’t have
money to get there. And we know that
when they do this, morbidity and
mortality increases in a manner that is documented
in peer-reviewed literature. And we’ll share some of
that with you later today. So with that, I think the best
way of putting this opioid use disorder into perspective
is in the time that we are all sitting
in this room today, four souls will die. Four people will die while
we’re sitting here in this room. And that’s really
tragic, and especially when these deaths
are 100% preventable. So we wanted to
build a tool that was able to move the needle to
the left that really helped us identify those patients who were
at risk so that you could offer them better treatments
and better interventions. And with that, I’m going to
turn it back over to you, Sean. SEAN CONLIN: Great. So thanks, Meera. So from a data science
community perspective, why don’t we start
unpacking what are the kinds of questions,
what kinds of predictions, what would be helpful to you
in the front response community that would be helpful? MEERA KANHOUWA: Yeah. Sure. So imagine you’re in an
emergency department– because easy for me to do– and you get two people who
both come in with back pain. Both of them have been
prescribed T3s, Percocet, Ativan, Flexeril, whatever the
cocktail du jour is that they got, and they’re both there. It’s impossible for
me to know who’s going to be at higher
risk of addiction, because I’m treating acute
pain at a point in time. But that’s also the best
time for me to intervene. It may be the only time
they see medical care. So if we can figure out
who is at higher risk, I could then pull
my case managers in. I could get them to follow up. I could do whatever I had to
do to get systems in place to get them in care. So that would be
the first question. SEAN CONLIN: All right, so we
guessed the first one right. Who is at risk? OK? MEERA KANHOUWA: Yeah. And I think the next question
is– can we turn down the volume on me a little bit? I’m sorry. I’m hearing a lot of reverb. Thank you– is really
understanding what the barriers to treatment are. So medical treatment for
opiate addiction does work. It is very effective. There is some recidivism. And that is because when
you are addicted to opiates, you really have changes going
on in your neuronal pathways in your brain. So there is a rewiring of
your brain that does occur. So the medication
treatment protocols have to actually bring you back. And only about two
people out of 10 that actually would benefit
from medical treatment are actually
receiving that today. So it would be great to know
what the barriers to people getting that treatment are. SEAN CONLIN: So we guessed
right on the second one too. But let’s just emphasize
that for a minute. So nationally, the
scientific community has done an amazing job. We have medical
treatment that can help those who are addicted. MEERA KANHOUWA:
Yeah, absolutely. SEAN CONLIN: Very effective. The evidence is in. MEERA KANHOUWA:
The evidence is in. SEAN CONLIN: That’s
probably the only ray of sunshine in this whole mess. But as a nation– think about this–
of all of those who need to be on medical
treatment, and it works, only two out of 10 individuals
are getting access to that. And as a nation, what
are those barriers keeping people from getting
access to that treatment? The country is spending
a fortune on treatments, but people are
not getting there. And unpacking that and
understanding what that is and how AI and
analytics and data can help the decision
makers better manage that is an important
issue for the nation. MEERA KANHOUWA: Absolutely. Absolutely. And I think to better
understand this, it would really be good to know
what kinds of interventions? Now typically, when you hear
a doctor say intervention, you’re thinking oh some kind
of procedure, a doctor thing. But not really, right? We do know at this point that
social determinants of health are 80% of the equation when
it comes to our own health and determining
how healthy we are. And when you think about
social determinants, think about getting good
sleep, exercise, diet, and how you live,
where you live, who you are surrounded by. These are all social
determinants of health. So understanding what
the interventions are, we’re actually talking about
human services interventions, not medical treatments. So I look at those
as different things. SEAN CONLIN: So let me just
say back when I heard there. So there’s medical treatment. And that’s not what we’re
talking about with Opioid360. But a very big influence
around whether somebody gets healthy or not,
or is successful against the addiction,
is whether or not they have the right social
services around them? MEERA KANHOUWA:
Social determinants of health, absolutely. So there is increasing
peer-reviewed literature in many of the leading
journals about how those social determinants
actually improve or decrease our health. SEAN CONLIN: OK. So from a data science
community perspective to those on the front
line, if we can help you better understand early
in the lifecycle who’s at elevated risk of
opiate use disorder, if we can help you identify
at the individual level what the barriers
to treatment are, and then if we can help
make suggestions of what are optimal interventions
for individuals, that would be helpful. MEERA KANHOUWA: It
would be really helpful. If you could move the
needle to the left and get those people who
are at highest risk– finding that needle
in a haystack, and get them into care
programs before they get addicted to the
point where they end up seeing me from their
heroin overdose, that would be really great. SEAN CONLIN: Excellent. OK, so let’s unpack Opiod360
and talk about first the data. Before we get into the
details of the data, let’s level set on this. First and foremost, personal
identified information is protected here. We de-identified
the data and then we bring it together
in Opioid360. There are three
major areas of data that are fed into this system. And this is very
large data sets. The primary users of it
currently and probably for the foreseeable
near term future are governments, state
governments, in particular. And if you think about
a state government– this is the first data set,
the public sector data– you’ve got three major areas
of the state government that are dealing with
this opiate crisis. You’ve got the health,
the human services, and the law enforcement parts
of the state government. And if you peel that back,
that’s 25 or 30 different data sets typically, if you
go from state to state. And being able to bring
that data together, de-identified–
because keep in mind, you have CJIS laws and
HIPAA laws, and all sorts of complex legal requirements. So you de-identify the
data where it sits. And then you bring
it together and you can link the de-identified
data together. And the purpose of Opioid360
is not at the individual level per se, but it’s more to be able
to look at millions of examples over 10, 20 years, and look
for patterns in those data sets that are strongly
indicative of those three questions– elevated
risk, and what’s worked, what interventions have
been successful for people who fit into that persona,
and which ones have not been– so you can more
intelligently make decisions around resource allocation. That’s the first of the
three kinds of data. The second one is what’s
called a lifestyle data set. Meera talked about social
determinants of health. So there is publicly
accessible data that gives indications of
particular things on lifestyle. And we’ll look at
this in just a minute. But that’s an
important data set. And then there’s a third
one through our partnership with Google, a very
important data set that’s from Google Analytics. And there’s some very
interesting research. One of the most recent ones
was in a “Scientific American” article from December that
shows a positive correlation between certain
Google data being an early indicator of what
becomes opioid use or even opioid-related death. And it’s through the analytics
engine we aggregated up to a geography footprint. PII is strictly protected, just
as any of the other data sets. So that’s the data
that we bring in. One thing I’d emphasize too–
and then I’ll pull you back in, Meera– is we crafted
this data very carefully. One of the biggest
problems that we’ve heard from those
on the front line is that the data tends to be
very old, relatively speaking. MEERA KANHOUWA: Yeah, so when
you’re looking at public sector data, typically claims data
is going to be aged anywhere from 60 to 90 days or more. ED visit data for
those registries, you can think about
aging of the data all the way up to
about 180 days. PDMPs, or Prescription
Drug Monitoring Programs, could have anywhere from a
30 to 90 day lag as well. So as you’re trying to identify,
and predictably identify who these patients are, what we
really need is real-time data, something that would– either leveraging the
local health data exchange for the medical piece. But again, keep in mind,
the medical data is only 20% of the story. The social determinants
and the lifestyle data you talked about, that’s
the 80% of the story that I hope you’re going to
tell us moves the needle. SEAN CONLIN: Exactly. So the Google data is
the most current data. The lifestyle data
is updated monthly. So that’s a 0 to 30 data set. And then you’ve got
the public sector data that is after 30 days. So now you have a complete
continuum of timely data that the community just
doesn’t have today. So that harnessed
and brought together in an integrated fashion,
is very important. Maybe an example
would be helpful. So imagine that we’d
looked at millions of these cases over
many, many years, and we had extracted
clusters or personas that were indicative of groups
that were relatively safe, and ones that are very much at
risk of developing addiction. And we’re going to
compare two groups. And we’re going to start
with just the medical data. What Meera was talking about– and this is the data that
our front line practitioners typically deal with today. This is what they
were left with. In each of these
populations, you’ve got individuals who
had a back pain. They had been in rehab
for three to six months. They were put onto an
opiate prescription. And the question is
what’s the likelihood that an individual in
either of those personas would develop an addiction? We’re not talking
the individual, but what is the tendency? A good analogy is when we buy on
Amazon, they make suggestions. They say, if you buy this,
you might want to buy this. If you’re in one
of those personas, it’s not deterministic,
but there’s an indication that you may be at much
higher risk than somebody else that you would
develop an addiction. MEERA KANHOUWA: Yeah. And Sean, this is actually what
I was talking about before. When you’ve got two
patients who show up, they both have low
back pain, they’ve both been on similar
meds, you have no way of knowing at the
point of care who’s at higher risk of addiction,
even if one of them took an ambulance to come see
you at 3:00 in the morning because their back is spasming. I mean, we just have
no way of knowing. So why don’t you show us with
this lifestyle data does. SEAN CONLIN: All right. So let’s take a closer
look at persona A. We’re going to pull in lifestyle. And we’re going to pull
in financial information. MEERA KANHOUWA: So
you know, financial is pretty interesting. I mean, it turns out that 1/3 of
all bankruptcies in the United States are due to medical costs. And there is an
absolute correlation between your overall health
and your financial status. And not only that,
there is now starting to become a socioeconomic
divide for those people in certain populations
whose lifespan has actually decreased. In the United
States our lifespans are decreasing based on
your socioeconomic status. Now that’s appalling
in this nation. We are the richest
nation on earth, and we have never, ever been
in the top 10 OECD countries in terms of delivering care. We’ve never even made
it to the top 20. Frankly, we never
made it to the top 30. When I started looking at the
data submitted by the World Health Organization, we were
about number– in the 80s. Now we’re somewhere in the 40s. So as you start
understanding how the financial aspects really
impact overall health, I think it’s critical. People don’t look at this
and they don’t know about it. SEAN CONLIN: Thank you. So people in persona
A tend to be married. They tend to have kids. They tend to own their home. They tend to eat healthy. They start exercise pretty
quickly after they’ve– for example, into this scenario,
they’d be back exercising. They have short commutes. They have strong
financial indicators, good financial stability. And they live in
economically growing areas. And if you look at
the neighborhood or the geographic
area where they are, they tend to be in areas where
the Google data indicates that there isn’t a lot of
activity that correlates to high risk opiate abuse. Now if we look at persona B,
it’s a very different picture. People in this persona tend to,
again, it’s not deterministic, but they tend to have
a high rate of divorce. They tend to have unhealthy
eating habits, long commutes. They don’t exercise. There’s a lot of bankruptcy
and financial distress in this area. And they tend to live
in areas that are not economically stable or growing. And if you look at the
geographic footprint from the Google data– you can’t go to PII– but in that general
area, it tends to have a high activity
around high risk search terms that correlate with
opioid use disorder. The predictive models,
when we started here, based just on the
health data, they predicted 10% to 15% chance. And they were both about
the same prediction. Now with this new
data, we revisit it. Persona A actually
dropped significantly. And persona B, individuals in
that group, went up fourfold. Massive increase. And this actually
shows up in the data. It very much
correlates in the data. And the machine learning
predictive models show that. Let me just give you a very
brief anecdote, if I can. MEERA KANHOUWA: And I’d
like to share one as well. SEAN CONLIN: Yeah,
why don’t you share. And then I’ll share a
war story with folks. MEERA KANHOUWA: We
love war stories. I think that providers
might ask you well, how is Google data going to
help me be a better doctor? And my answer to them, quite
frankly, is the following. As you look at these
lifestyle data– this is based on
millions of people– as you look at drugs that are
going through clinical trial, you’re lucky to get a sample
size of 10,000 patients. Like, that’s an awesome trial. All right? Most of these sample
sizes for FDA trials are 5,000, 8,000 patients. This is millions of patients. So as we think about is
this data statistically significant– I mean, we haven’t done a
chi-square regression analysis to actually answer
that question. But you could intuitively
gut understand that the sample size is probably
going to be very impactful. And I’m excited about
the kinds of studies that we could do
based on this data, and showing how it really can
help us determine outcomes. SEAN CONLIN: Excellent. I’d like to draw your
attention to the predictions. So there’s an
interesting set of work done for the state
in the Midwest, for their Medicaid population. And we were given well
over a million records that went back over 10 years. And the question was build
a predictive AI model that can identify within
that population who is at elevated risk of
opioid use disorder. That was the question. So we built the
machine learning model. And when we brought
the results forward, the client shared with us
that they had actually built their own predictive model. And they wanted to
compare that two. So we said, great,
let’s compare them. They used a fundamentally
different approach. They had used a
rules-based approach. It had hundreds of rules
that were from experts like Meera that
said things like, if somebody goes more
than 30 miles to get an opiate prescription
filled at a pharmacy, that’s a risk factor. If they’ve got more
than one doctor that gives them an
opiate prescription, that’s a risk factor. And they had hundreds of rules
they developed over years. And we used a machine
learning model that just looked at the data. But it had a massive data
set to go train off of. Their predictive
model identified 15% of the target
population, those who are at elevated
risk of opiate abuse. Our are identified 85%. They had a 50% error rate. We had a 10% error rate. And probably most
importantly, theirs took about nine
months of runtime for a given individual
from the time they first started their
opioid prescription until the system had enough
data to make a prediction. Ours could do it in four to
five months, half that time. And we’ve already got models
in the works that are going to pull that off much shorter. And early prediction,
early intervention, early identification is
the key to success here. And it’s showing up in the data. It’s showing up in the models. MEERA KANHOUWA: Yeah. Just for the audience,
this is really only showing those personas that are
impacted by medical prescription as their start to opiates. We know there’s another persona
which is through illegal means, getting heroin, IV fentanyl,
stuff like that on the street. So that’s a whole other
way to evaluate risk. SEAN CONLIN: And we’ve
actually made progress on that. That is a difficult problem. But instead of one
clean area where you can trace it through
the prescriptions through the doctors, there
are about eight to nine different variables
that can be analyzed to identify those
at-risk populations, and basically be a surrogate
for the data that’s in the clean
prescription data set. MEERA KANHOUWA: Great. SEAN CONLIN: So why
don’t we take a look and share some specific
examples, some actual use cases. And Meera, this one is an
example of a statewide policy executive. Think Medicaid policy
executive, who’s looking across an
entire state population. And to tee this up, because
we have this lifestyle data, for demo or for
example purposes, what you’re about to see
here didn’t have the benefit of the state data. And it didn’t have the benefit
of the Google Analytics data. There’s just the lifestyle data. So what we did is we pulled the
data down on 7.7 million people who live in a southeast state. All the PII is stripped out. They’re just random
numbers, you’ll see that here in a minute. But we’ve got lifestyle
data on 7.7 million people. So in this one screen,
you’re interacting with that complexity. Then on top of that, we
apply predictive AI models in a series of them. So if you go across
the top there, one of the things we said is– for every one of them,
we put a score on them– what’s the likelihood of being
at elevated risk of opioid use disorder? And you’ll see there
about 140,000 individuals. Kind of depends on where you
put the tear line of what’s high risk. But it identified that. There about 2.2 million
who are at high risk of being uninsured. This is not a Medicaid
expansion state. And the state confirmed that
that’s probably about right. When you integrate it
with the state data, that number gets dialed
in quite tightly, although they even
struggle with that number. And then on the left hand side– I’m not going to ask you
to actually read it– but there are a series of
filters that we can apply. So Meera, for example, one of
the high risk populations– and I know you had to deal
with it in the emergency room– is what if we could look across
7.7 million and identify based on generic individuals–
so the PII is protected– pregnant, likely at high
risk of opiate addiction, uninsured, housing insecurity,
and they have children at home. That’s a very high
risk population. Any thoughts? MEERA KANHOUWA: Well,
it’s even worse than that because a pregnant mom
is going to give birth to an addicted baby. And that baby will be premature. And that baby will
spend on average 45 days in the neonatal
intensive care unit. That baby will have
IVs and monitors, will likely be intubated. And that baby will go
through withdrawal. That tiny little
neonate will withdrawal in front of your eyes. And it’s terrible. And these babies cost about
half a million dollars to make it through
that NICU alive. And they don’t all
make it out alive. So this is a really
huge problem when we’re talking about collateral
damage from opiate addiction. I mean, you know, these are
youngest, youngest souls coming into this world. And if you could move
the needle to the left and not have that
pregnant mom get addicted, and not have her give
birth to an addicted baby, you’d actually be saving lives. SEAN CONLIN: I don’t
know what to say to that. It’s a good reminder
that this stuff matters. MEERA KANHOUWA: It does
matter, absolutely. SEAN CONLIN: We’re
not just talking data. We’re talking lives. So what we just did
there, we actually applied all those filters. Depends where you put high risk. Out of 100 scale, where
100 is extremely high risk of addiction, this
is 60 and above. And there 5,000
individuals in this state that were identified, fitting
what we just talked about. If you move it up to
90, so 90 out of 100, that number reduces
down to about 500. But that’s still 500. And by way, this is data
from the last 30 days. So that’s 500 people
who fit that criteria, that you can see by the map
which counties they live in, where the densities are. We also, in the
upper right corner, again, it’s hard to see
in a room like this, but for each of
those individuals– again, the PII is protected– we use data to take
a educated guess at what are the
barriers to treatment that individual is up against. And then based on
that, we aggregate it. So we can look across the
500 or the 5,000 individuals and aggregate that up for
the policy decision maker. And they can say, if
I pull these policy levers in these
particular geographies, how many people can I help? And I’ve got limited resources. And how do I get those out
there and get the most benefit for society and for
these individuals? And this is data and capability
and predictions they’ve never had access to before. One last thing, Meera. If we took away the filters
that we just put there, and we just said, let’s look
across the entire 7.7 million and we said, who’s
at 90 and above– very high risk of addiction
for whatever reason– it comes out to about
18,000 individuals. And now if you look
down at the bottom there, it’s a little hard
to see again on this screen, but when we aggregate what are
those barriers to treatment? What you find is
they’re about 10%. Again, this is with data
from the last 30 days. About 10% would benefit
from a HUD housing credit, because their main
barrier to treatment is that they have
housing insecurity. And about 40% would benefit
from Medicaid expansion. They just don’t have insurance. But otherwise, there seems
to be reasonable stability in their life. And those barriers
to treatment, that’s where a lot of the action is in
the marketplace to study that. But these are the kinds
of tools that the people on the front line can be given
to start studying that problem and see what works and
what doesn’t, given a particular barrier. MEERA KANHOUWA: So you
know there’s actually peer-reviewed data that shows
that transportation is actually a barrier to treatment. And in fact, Dr. Harold Thomas,
who’s a surgeon in the Bronx, did a study on breast cancer
individuals over a 10-year period, and started the
Patient Navigation Institute as a result of that, and
was able to successfully demonstrate that
there is a 60%– six zero percent– increase
in morbidity and mortality when something as simple
as I don’t have a ride to get to the doctor. So what’s really interesting
as a result of this is that Medicare
has now approved Lyft and Uber for patients
to get to their appointments. Anthem, Cigna, Humana
are all allowing for people who fit
a certain category to be able to actually get
transportation to get there. And there is an
interesting study done about you’re much
more likely to live if you take an Uber to
the emergency department than if you take an ambulance. So just keep that in mind. If you want to go to
the ER, call an Uber, unless you’re having
a heart attack, then call your local ambulance. But it’s pretty interesting. So transportation– you
wouldn’t think it’s a big deal, but it is a pretty big deal. SEAN CONLIN: So that
example we just went through is more at
the policy level. A different persona for a user
for this kind of a capability would be more at the
caseworker level. Now we said all the
PII is protected. And you’re actually going to
see to the extent you can see it on the screen. Can you guys still hear me? I just hit my mic. We’re good? It’s still genericized, but
at the caseworker level, if you think a Medicaid shop
or human services or child services using this
kind of capability too, that’s more of a
caseworker use case where they’ve got
control of the data. They’re using PII today. So they could use it just
on their limited data. And what you’re
looking at here is– the center nodes are patients. And it’s a patient– we’re looking at patients,
doctors, everybody in the ecosystem, the
entities, pharmacies, treatment facilities. These are the high
risk ones in it, somebody’s area
of responsibility. So at a glance, you immediately
see where the risk factors are. The scoring is based on the
outputs of predictive AI models. And now this is a 360 view,
hence the name, of a patient, patient 400. It’s real data. Well, it’s patient 400. It has a red halo
around her or him. And that is indicative that the
predictive score says they’re at high risk of addiction. They’ve got doctors
associated with them. They’ve got
lifestyle information associated with them. You can see them
appearing, and they’re receiving food stamps
and financial counseling. The power of this
is that we talked about health human services
and law enforcement data. When that starts
getting integrated, you can now do this network
analysis to see connections. I can see how much
is being spent on an individual
in the health area, but also their human services
area, and law enforcement, and not only them,
but their dependents. So we can start
connecting this data up. I said that we’re using
AI to connect the data up. You could never, never
connect this kind of data up if you had to do it manually. But using some machine
learning models was one of the big
breakthroughs we’ve had here in the last two years. We’re doing this at scale
both at the federal level and now down at the state level. And that connectedness is key
to being able to drive and feed the machine learning models. MEERA KANHOUWA: I would love to
have had this in the emergency department. We had a patient, we’ll
call her Patient X, who used to come to our ED. And it took us about 18
months to figure out she was actually four
different people. And she would come in dressed
as four different people, different wigs, different
clothing styles. And she would get prescriptions
for percs, Demerol, T3s, et cetera, and sell them. And she’d sell the scripts. And she was using the money
to take care of her kids. Her kids hadn’t been
signed up to CHIP and so they weren’t
getting any benefits. And so this was how
she was earning money. And it really took our ED
docs kind of saying you know, she looks really similar to that
girl I saw just a few days ago. But it took us about 18
months to connect the dots that she was the same person. And then we had to
track her effectively, to say look she might
have different hair, she might have
different clothes. But had we had
something like this, we wouldn’t have had
to spend 18 months. Ultimately, we were able to get
her associated with the care team that she needed and
with the case management that she needed. And I think that benefited
her and her family greatly. SEAN CONLIN: And part of
the tragedy of that story too is, if you look at her
barriers for her and her children, she was eligible for
many, many different programs and just wasn’t getting care. MEERA KANHOUWA:
Absolutely, that’s right. And just wasn’t getting it. SEAN CONLIN: Yeah, excellent. I’d like to just
do a shout out too for our Google Partners on
this and the alliance we have with them. This is the December
“Scientific American” article I referenced earlier. If you’re curious
about this, I’d encourage you to
take a look at that. And with that, we’re going
to bring Doug back up. And we are happy to share some
details with you on Opioid360. DOUG BOURGEOIS:
Thanks, Meera and Sean. And I know for those of you
on the front end of this, they were skeptical
of my explanation of the power of the partnership
between Deloitte and Google. I think now maybe you have
a different appreciation for what I mean. Talk about powerful. In order to bring our
capabilities together, transformative looking at
this widespread epidemic and this problem in a
very different manner, looking at it very
holistically, bringing in things that I had never
heard about before I was aware of this solution,
things like HUD housing credits and how rides and other sort of
interventions other than simply medical can make a difference
in people’s ability to overcome the addiction
and the challenges that they face
associated with that. Very, very
problemsome situation. So we pretty much burned
through our time here today. So I encourage you to
do a couple of things. One, scan in that
barcode right there. It will take you right to
our website on Opioid360. At that website,
not only is there more information
around this solution, you can also schedule
a demo with our team if you’d like to do that. Also please visit our
booth 1303, right out on the showcase floor there. Walk in, look up,
look for Deloitte. You can’t miss it. Our team’s there
doing demos as well, and can answer any questions
you have around the solution and get more
information around that. So I think with that,
I know you know, maybe we have one more minute. It’s one thing that you know,
when I’m talking in the booth, and I get a lot questions
from people who stopped by, one thing that I get
quite a bit is can we take this outside the US? I mean, does it work? You know, is there
similar availability of the information
in order to bring it to an international perspective? SEAN CONLIN: Yeah, so
the short answer is yes. We’re actually
pretty aggressively in conversations with
other countries on this. Deloitte partners in
those other states. And I’d say there’s actually
two dimensions where there’s a lot of
activity to expand this. One is internationally,
and the second is the predictive
models that sit on top of the data can
also apply to other health and disease topics. So think Obesity260,
Diabetes360. There’s actually about
50 different areas where we’ve developed
predictive models into. So we see Opioid360 as
being the start of something that really could be big. DOUG BOURGEOIS: And thank you. And one last question,
the final question is, is this smoke and mirrors? I mean, just showing
us really cool stuff, or is it actually
in use anywhere? I mean, has it been– SEAN CONLIN: No. Well, first of all, many
of these technologies, the confluence, there are many
pieces that have come together, have really matured here just
in the past couple of years. And we have two active states
that are implementing it now. And we’re talking
to many others. And I think it’ll
start in the states. But we’re already
starting to talk to some of the private
firms in the health space. DOUG BOURGEOIS: Terrific. Thank you. I’d like to thank Sean and
Meera for this very enlightening perspective around this
very challenging epidemic. And thank you all for attending. And again, come
by our booth at– MEERA KANHOUWA: Thank you. [MUSIC PLAYING]

Leave a Reply

Your email address will not be published. Required fields are marked *