IHPI Seminar: Development of the System for Opioid Overdose Surveillance (S.O.S.)

– Welcome to the IHPI Seminar Series. Just a couple of announcements
before we get started. This is our last seminar for the summer, but we do have one summer
talk that is gonna be at a similar time slot and that you guys might be interested in, which is some discussion
of current things going on in Washington in healthcare
policy by Jonathan Cohn. That will be here in
this auditorium on July, is it July 19th, Kara? – Yes.
– July 19th, so keep that on your calendars. And there will be some snacks
in the lobby after the talk. So, if you wanna stay around and have some informal conversations with your peers, there will
be some snacks out there, so don’t run off too quickly. So, for today, we have, our guest speaker is Dr. Mahshid Abir, and Dr. Abir is an
emergency room physician and a health services researcher who’s been with IHPI for many years. And she has a joint appointment with the University of Michigan and RAND. She’s the director of a center, called the ACUTE Care Research Unit, here at the University of Michigan. And she’s had research in
a lot of different areas. She’s done some studies on the interaction between emergency room services and inpatient work, ambulatory care, mostly focused on how to make
those connections work better and provide better outcomes for patients. Today, she’s gonna be talking to us about a really fascinating
project that involves connecting a whole bunch of
different stakeholders and data around the opioid crisis. So, with that, I’m gonna
turn it over to Dr. Abir. – Good afternoon, thank you
so much for the introduction. I appreciate everyone being here on a really beautiful, sunny day, where you could be outside
enjoying the good weather. I appreciate the opportunity to present the System for Opioid
Overdose Surveillance, S.O.S. And this is a project that is funded by the CDC Injury Center here
at the University of Michigan and by the High Intensity
Drug Trafficking Areas and by the Michigan Department
of Health and Human Services. So, as Caroline mentioned, I’m an ER doc, and we are on the front
line of everything. We’re the safety net. And I’ve been practicing for 10 years now, and, prior to that,
four years of residency. And I’ve practiced in various places across the United States, the Northeast, Midwest, the South, in academia, community
hospitals, academic affiliates. And what I can tell you for sure is that the opioid crisis does not spare anyone. Doesn’t matter what you look
like, where you come from, how old you are, what you
have in your bank account, so it does not discriminate in that way. And, you know, for me and my colleagues in the emergency department,
it’s very personal. Okay, so we deal with people who come in half dead or completely dead, and some of them we
save, some we don’t save. And I’ll share a couple of stories from my own experience recently. So, a couple of months
ago, on a Sunday morning, around nine in the morning, I had a shift at the Adult
ER here at Michigan Medicine. And suddenly, we get two critical patients from the parking lot, two young
adults, a male and a female. And the respiration rates for
both was in the single digits. And they had shot up some
kind of opioid in the car, with the hopes of being found in the ER parking lot
so they could get saved if they used too much of the opioid. So, in that situation, we
happened to save both of them, and many others we don’t. And also, it’s getting scarier. So, the kinds of things
that seeing right now is, for example, you know, NARCAN, which is the anecdote for opioid overdose, is also being abused. So, on one shift, we
had three kids come in, and I use kids because now I’m in my 40s, so I can do that, young adults, who, the chief complaint was NARCAN party. So, if you can imagine,
what is NARCAN party? Literally, that was the chief complaint. You can’t make this stuff up. So, apparently they were
all at a party in Ann Arbor, and they had NARCAN. And there was a designated NARCAN shooter, okay, like a designated driver, so that if the others
at the party overdosed, that the person could administer
NARCAN and save the people. That’s scary. And the other scary part is
that a lot of people don’t know, depending on how much opioid you use, that one dose of NARCAN is
not gonna last that long, so that it’s not gonna save you ultimately if you don’t have enough backup NARCAN. So, a pretty, pretty scary
kind of like outlook here. And, you know, in the ER, we usually deal with
the tip of the iceberg, no matter what the condition is. I mean, we see a lot of
kind of smaller things, like toe jam or my hair hurts, whatever. But we see a lot of things,
the extremes of disease, right? So, opioid overdose is the extreme of the opioid abuse disease. And that is, again, something that I
gravitated to understanding out of my own personal
experience as an ER doc. So, how big is the problem? So, as I was making this PowerPoint, I was sitting with my eight-year-old son, and he was really interested
in this particular one. And he was describing to me
that, “Mom, the one on the red “looks like a really, really bad problem.” Yes, it is a very bad
problem, this opioid crisis, but we really don’t know how big it is. We know it’s really big. We really don’t know how
big it is, and why is that? Because we don’t have
ways to do surveillance of fatal and non-fatal
overdoses in a timely manner. So, for example, in Michigan, you know, how bad is the problem? You know, where are the hot spots? Which communities are in most trouble? And this is information that’s
really, really important. Why? Because, as much as the Federal Government and many other entities are throwing money to understand and address
the opioid crisis, there’s not enough money
to go around, right? So, understanding which
communities are in most trouble so that you can target both public health and law enforcement interventions
is absolutely critical. So, what is the current
state of opioid surveillance in the country? It’s in really kind of bad shape. So, for the most part, a lot of individual counties
or health departments, and this is not just in Michigan, this is in many other places, they’re doing kind of
one-off surveillance. They’re doing their own thing. And when you do your own thing, your system doesn’t talk to
the other systems in your state or to other states, and it’s not scalable. And a lot of times, it’s not sustainable. So, doing one-off kind of initiatives is probably not the long-term solution. Also, using outdated or
manually collected information, okay, so when you look at a
lot of the talks that folks in Michigan do and talking
about the crisis in Michigan, you will notice that a lot
of the data is from 2016, and you’re lucky if you see
any data more recent than that. So, if we assume that
this is a static epidemic and that both geographically and its size, it’s not changing, then,
yes, relying on data from 2016 may not be a bad approach. But we need to really understand what this epidemic is doing, how it’s moving in, you
know, between communities, across the state, and
across the United States. And in order to do that, we need real-time or near
real-time surveillance. Another approach that people are taking is syndromic surveillance. And I’m sure there’s a
lot of public health folks in this audience today, and
you all know what this is. So, essentially, it’s a chief complaint-driven
surveillance system, so that you, based on,
for example, ER visits and chief complaints when
people go to the ER visits, you have, you track to understand you know, whether there’s
an epidemic coming up, for example, for influenza. So, what is the problem with that? As many of you may know,
a chief complaint is when someone goes to
either a doctor’s office or an ER or urgent care, and
they express to the person who’s registering them
in, why am I here today. And in the case of the ER, a lot of times, it’s the nurse’s opinion, or the patient is not
even responsive, and it’s, the team decide, you know,
well, patient is here for unresponsiveness, and
they assign a chief complaint. That is not the final
diagnosis of the patient. So, because there’s no medical
workup that’s been done, it’s purely an impression, an
initial impression, imprecise. And so, for example, the State
has a surveillance system that’s based on syndromic surveillance, which, again, many places
use, not to put it down. However, again, it’s
chief complaint-driven. And according to the State,
they have never been able to foresee an upcoming epidemic, ever, and it’s been around for years. So, that tells you that it’s
probably not the best approach. And when I talk about chief complaint, it’s compared to an ICD-10 code. So, the ICD-10 code, which used to be an ICD-9
code until recently, is the actual diagnosis, that after the clinical
workup has been done, what is the diagnosis of that patient? That is a lot more accurate. ‘Cause, at that point, so, for example, in the case of overdose,
you have a history. EMS or the ambulance tells you that there was a needle
in the person’s arm. You know, you have a urine drug screen. You may have other information
that helps you decide whether or not the person
had an opioid overdose. Another approach people
are taking is taking counts of opioid overdoses by
counting NARCAN administration either by police or by
ambulances or EMS agencies. And, you know, again, there, you know, for just, and a lot
of you may already know this, but NARCAN is a pretty harmless medication that is meant to reverse
an opioid overdose. Now, when EMS or police
find an unresponsive patient or person in the field, they, especially given the current epidemic, they go ahead and give
them NARCAN as they should. However, if someone had passed out from a cardiac
reason or a lung reason or they had a stroke or
a bleed in their brain or a completely different reason unrelated to opiate overdose,
they will be counted in. And that’s a problem, so
that is also inaccurate. And, you know, talking about NARCAN and using that for
surveillance, there is a system that some folks across
the country are using, particularly in the
Northeast, called ODMAP. And ODMAP is a very clever system. It maps geocodes where overdoses occur, and it relies on a couple
of pieces of information, so incoming data from police
and EMS, from the field, on individuals with location
of suspected overdose. Sometimes they suspect overdose, again, if there’s a
needle in the person’s arm or there’s drugs on the scene or there’s family or friends who say that, yeah, the person was shooting up heroin. Sometimes it’s based on
NARCAN administration. So, again, NARCAN administration, any unresponsive or poorly
responsive individual should get NARCAN. Some people do come out of, even if it’s an overdose,
sometimes it reverses it, sometimes it requires multiple doses. But if you’re counting all the ones that didn’t respond as overdoses, then you’re overcounting opioid overdose in cases where they actually weren’t. So, what kind of challenges
do we have here in Michigan for kind of getting an idea of how big the opioid problem
is and doing surveillance? Just like many other states, medical examiner data in
Michigan is not centralized. So, there’s not one place or one entity that actually all medical examiners across the state report their deaths to. Imagine if we had that system,
then we could keep track of all the overdoses that lead to death. And it would be in one place, and we could all understand
the scope of the problem, you know, as far as opioid
deaths are concerned. Emergency department data
also is not centralized. There are a couple of
health information exchanges across the state. There are two bigger ones that have data for most of the state. So, one is Great Lakes Health Connect, which I’ll talk about in a minute. And they have data for almost
all of the Lower Peninsula, except for a couple of
health systems and hospitals, and they don’t have any
of the Upper Peninsula. And there’s MiHIN, which is another health information
exchange that the State, Michigan Department of Health
and Human Services uses. So, but in that situation even,
it’s not all of the states, it’s not all of the emergency departments, and currently not being
used for the purposes of keeping track of overdoses. Then there is emergency medical services NARCAN administration. There is a centralized EMS database that Michigan Department of
Health and Human Services keeps, called MI-EMSIS. And in MI-EMSIS, every one to three days,
you get data refreshed on NARCAN administration
for the entire state, so that’s pretty comprehensive. But again, it’s just
NARCAN administration, and many people who are not overdoses, who are unresponsive or poorly
responsive do get NARCAN. So, I wanna briefly talk
about this recent data. We actually crunched
these numbers last week. This piece of the project is funded by the Michigan Department
of Health and Human Services. And the purpose of this piece
of the project was to look in Washtenaw County to see how, what is the positive predictive
value and sensitivity of NARCAN administration for
detecting opioid overdoses? So, we used 2017 data
for Michigan Medicine, so our own emergency department, and all transports to our ED
by Huron Valley Ambulance, which is the main ambulance
agency in Washtenaw County. And this is 2017 data, and we measured the positive predictive value
of naloxone administration as an indicator of opioid overdose. So, we had 145 cases of naloxone transports to
Michigan Medicine by HVA. Of those, 59 cases received an ICD-10 code of opioid overdose. So, the positive predictive value is .41. Number two, we measured the sensitivity
of naloxone administration as an indicator of opioid overdose. So, 108 Michigan cases with an opiate
overdose-related ICD-10 code were transported to the ED by HVA. Of those, 58 cases
received naloxone from HVA. So, the sensitivity of
naloxone administration for picking up cases of overdose is .54. So, you know, given that
this is a small sample, it’s not that big of a sample, we’re waiting on data for St.
Joe’s Emergency Department, which is the other main
ED in Washtenaw County, and we’ll have a larger sample. But at least based on this data, you know, to make a
case that you should use NARCAN administration alone as a way to detect overdoses is probably not a very good idea. So, we sought, with the funding sources that I described earlier,
to form a partnership between the University of
Michigan Injury Prevention Center and HIDTA, High Intensity
Drug Trafficking Area, in order to come up with a public health, law
enforcement partnership to understand the best approach to surveillance for opioid overdoses, both fatal and non-fatal,
in the state of Michigan. And the funds that are coming from the High Intensity
Drug Trafficking Areas are from the Office of
National Drug Control Policy. So, what are these High
Intensity Drug Trafficking Areas? So, as the name implies, they are counties in states
across the United States where drugs are dealt the most, they’re bought and sold the most. And HIDTA is a coalition of state, local, state, and federal law enforcement that have joined forces to
address the opioid crisis in designated geographic areas
across the United States. So, all states across the United
States have HIDTA counties, and the mission is to
assist law enforcement, disrupt and dismantle drug
trafficking organization and money laundering organizations. It is a program of the Federal
Government, so, you know, these are cops essentially
that we’re partnering with. And I will tell you that this
complement of public health and law enforcement is
incredibly powerful. Because, you know, as, Amanda Kogowski, who’s sitting right here, who is the manager of the
ACUTE Care Research Unit, which I run, and she’s
the public health analyst, so she’s the glue between the cops and the doctors and the
public health experts. As she will tell you, when we have sat across the
table with our cop partners, there has been so many aha moments. ‘Cause we don’t have their perspective, and they don’t have ours. And when the two sides come together, there are solutions that come apparent that each group alone cannot come up with. So, it’s incredibly powerful, and it’s been a really positive
experience for our team. So, HIDTA is all about
interagency collaboration. It was established by Congress in 1988, and it supports criminal
intelligence sharing and interagency collaboration. In 2015, five HIDTAs came together to create the Heroin Response Strategy in response to the opioid crisis. And the main principle
behind the HRS is that you need to have a partnership
between public health and law enforcement in order to come up with the most effective
strategies to address this crisis. So, essentially the goals of the HRS, so there are actually 22
HRS-participating states across the United States, and the goals are to improve collaboration between public health and public safety, increase the timeliness and
quality of overdose reporting, and develop strategies to reduce fatal and non-fatal overdoses. And, as part of this, effective surveillance is
one of the top missions. So, we got around the table to think about how to approach surveillance in Michigan and for anywhere else, for that matter. There were a couple of things that came to mind that
were incredibly important. So, before I get into these points, I want to talk about some examples of surveillance across the country that are currently recognized
as the gold standard. So, for example, in Massachusetts, Massachusetts links over a dozen datasets to understand the
epidemic, which is great. You know, so they do
inpatient data, ED data, data from outpatient providers, from rehab centers,
prescription practices, a lot of data linked across the way. But, you know, Massachusetts
is a very resource-rich state. Which other state can
replicate that model? And the other question, as
a researcher, for me, is what do you really seek to understand from linking that many datasets? What actually usable, concrete,
actionable information are you gaining from that? So, when we sat across the table with our law enforcement partners, one of our main goals was what are the fewer complement of datasets that will give us the most information? And also, how can we design
a system that could be scaled and also apply not just here
in Michigan but elsewhere and serve a model for the country? So, that was the initial intent when we started doing this work. Also, we wanted to understand how we can identify the hot spots. So, yeah, the problem
is big across the board, and I think everyone
understands that to some degree. But where are the problems the worst? Which communities need the help the most? And where do we need to get to first? So, that’s how you
maximize limited resources, and that was part of
the goal of the system that we set out to design. Also, it had to be timely and accurate. So, again, data from 2016 is not going to inform law enforcement. So, if there’s suddenly a cluster of opioid overdose deaths in a certain neighborhood in Michigan, if this happened two years ago, there’s nothing you can do about it now. It may not be the case anymore. But if they know it happened yesterday and they know today where
those hot spots are, maybe you can do something about it. Same thing with public health. So, if you know where
the trouble areas are, and this is at the population level, it’s not at the individual level, you can target your interventions to those communities that need the most, again, both law enforcement
and public health. So, we started off, in year one, which was the
first phase of the project, in Washtenaw County, to test our idea of how we can come up with
a scalable, timely system. So, we basically chose three datasets. We wanted to keep it
simple, three datasets. So, one was EMS data or
ambulance agency data. So, that is information on all individuals who were transported to Michigan Medicine, who were administered
NARCAN by ambulances. So, that’s the EMS piece. Then, there’s emergency department data. So, Michigan Medicine data,
and St. Joe’s data is coming, and then we’ll have the whole
Washtenaw County ED data. But for anyone, be able to pick up anyone who was seen in any of these EDs, who was either admitted or discharged with an ICD-10 code
related to opiate overdose, and also medical examiner
data from Washtenaw County. So, imagine, if someone overdoses, and an ambulance picks them up on the street or at their
home and administered NARCAN, it will get captured in the EMS data. When they get transported
to the emergency department and they get either
admitted or discharged, if they’re alive, it gets captured there. And that’s a confirmation
or a validation step that that NARCAN that was administered by EMS actually occurred in a person who actually had an overdose or didn’t. And lastly, if that person
did not make their ED visit, they died, they end up on
the medical examiner’s table, then you also see that there. So, we have visibility. You can track the same individual from the time they overdosed on the street to the time they were dead in
the medical examiner’s office, right through, without
overcounting or undercounting. In a lot of states where
they’re looking at similar data, they are not linking these datasets. They’re looking at them separately. So, the same individual that would show up in each of the datasets
would be counted three times, and you would have no
idea that that’s redundant ’cause you’re not linking the data. What we’re doing is
that we’re using these, all three datasets, the EMS,
ED, and medical examiner data, in a fully identified way. So, we have patient identifiers, and we can link the three datasets through probabilistic matching
to remove the redundancy and be able to track the same individual across the three settings. And so, once we identify
these folks, we update S.O.S. And the data sources that we get from each of them are
essentially very compact. We’re not wanting any other information but to identify fatal
and not-fatal overdoses and geocode them. So, we wanna know ideally two pieces of
information for all overdoses, where it occurred and where
does the patient live. And sometimes those two
locations are the same, and sometimes they’re not. But that’s the information we want ’cause we wanna hot spot the areas that have the biggest
problems with overdose. That’s the goal. So, to accomplish that, we’ve kept the list of
variables that we’re asking from each of these
datasets pretty compact. So, we have person’s name,
the date of the overdose, date of birth, incident
and/or home address, outcome of the overdose,
whether it’s fatal or non-fatal, race, gender. And from the EMS dataset, we’re going to know that they
were administered NARCAN. From the ED dataset, we’re also gonna know their ICD-10 code, so their final diagnosis, and we’re going to have
a urine drug screen. And from the ME data, for
the deaths that occur, within four to eight weeks, depending on the medical examiner, we’re gonna actually get
the toxicology report, so we’re gonna know the
offending agent or agents that resulted in the fatal overdose. So, this is the data that we have. So, this is for 2017, so, again, this is the
pilot from phase one. And, you know, just to give you some sense of the total numbers, the total non-fatal overdose is 147, and the total fatal is 71. And across the board, you have more males than females and more white individuals. And the peak of overdose for the non-fatal is in the 25- to 34-year-old range. But clearly, we’re going to
have access in this system to some of this demographic information, but, again, the biggest
goal here is geocoding and understanding where
these are occurring. So, this is data from January 1st, 2017 to December 31st, 2017. And what this is representing is the EMS data, the ED data, and medical examiner
data were, again, linked, and all the redundancies were removed. And this is actually
showing you two things for all the non-fatal locations. Orange is the residence location, and the green is the incident location. And based on this data for that full year, we identified the hot spot zip codes, and they’re the ones that were listed. We have a similar map, so for the fatal overdose locations. So, the red dots are
the incident location. The blue ones are the resident location. And the hot spot zip codes were identified for the overdose deaths here. And in case you’re wondering, about 40-some odd percent of the time, the location of the overdose
and the home location overlap. And our law enforcement partners tell us that they’re not surprised by that because the majority of the time, the place where the drug is bought, the person usually uses it within a mile of where they buy the drug. So, they weren’t surprised
by seeing the numbers that we came up with. And we have some very recent data to share with you from last week, again, importance of having timely data. So, this is for ED data
and EMS data, linked again, and you see the non-fatal overdoses in this map
that show the hot spots and some demographic information. So, I can stop for some
questions here before I go on. Go ahead. – [Audience Member] That
hot spot map doesn’t take into account the population
in each of the count, of the zip code areas. How did you deal with that? – So, we haven’t dealt with that yet, but the system itself is going
to be projecting the rates. And it’s gonna, once it’s
actually up and running, will be based on the
rate of the population, so that you’re not, for
example, in the Upper Peninsula, where there’s that, it’s not populated, you’re comparing apples with
apples at the end of the day. So, the data that I showed
does not account for that, but ultimately the system will. – [Audience Member] ‘Cause
it seems like it was just picking out the most populated zip codes. – It probably is, at this point, the data that I reflected,
because it’s not quite, the population rate is
not really accounted for. – [Audience Member] Thank you. – [Audience Member] Are there
any, among the ED patients who got naloxone but did not have a ICD-10 diagnosis of opioid overdose, were there any patterns
among that group in terms of, is there a systematic way
we’re undercounting these? ‘Cause it seems like your non-fatals were kind of undercounted, and I can imagine there
are varied reasons why. Were there, and then,
also, in a similar way, were there (speaking faintly)
would be to run a drug test? – We haven’t looked at those things yet. So, this is, the project, the
validation step just started about a month and a half
ago, and this was the first analysis that came out of it. So, what the State has asked
us to do is actually look at some of what you just alluded to, which is looking at
gender and race and age and kind of trying to figure
out if there are patterns where we can actually understand
why we’re undercounting, I mean, to explain
exactly what we’re seeing. We haven’t done that yet, and open to talking to you about what else we should
potentially be looking at. – [Audience Member] ‘Cause I
imagine there are probably some (speaking faintly) that
occur (speaking faintly), heroin overdoses or overdoses in home. But I imagine there’s
probably a significant, but almost all of them probably were overdoses on (speaking faintly) overdose. And my guess is that, if
you take most of those ones that have discharged after being found cyanotic and reversed with naloxone, those are all (speaking
faintly) same thing. Similarly, in EMS, I don’t know
how (speaking faintly) was, but there are several studies
on (speaking faintly). And my guess is, again, the ones who are found apneic, who get naloxone (speaking faintly) the majority of overdoses. – No, you’re absolutely right. I mean, so I think, at the end of the day, you’re pointing at the ED setting, I think that the EMS and ED datasets are gonna have a lot of problems. We’re also gonna be
missing a bunch of people. So, if someone goes to an urgent care, it’s not captured here. If they go from the scene,
they refuse transport by EMS, that’s not gonna get, ’cause you have to get transported for that to even count, right? So, I think that, at the end of the day, once this is up and running, there’s gonna be many kind of disclaimers of what this is exactly, what it isn’t, and it is not perfect. But, I think, really early on, we made a decision that
we weren’t gonna let perfect be the enemy of the good. And I think that this is something, is like a LEGO block mentality of building pieces one by one and making it better over time. And, Caroline, going back
to your question earlier, I just want to add something,
that the purpose of the data that I’m presenting and the talk overall is more so how we built it. The data that we’re presenting
is very much a pilot, and so it’s not so much
with an intention of pinpointing the location so much. But the pilot test kind of goal is to test whether this
probabilistic matching and the entire process is doable or not. And there’s a lot of kinks and glitches that we need to think
through and kind of fix before it’s kind of ready for prime time and kind of sharing the actual data. – [Audience Member] I am
intrigued by the sensitivity of the administered data of the naloxone. I noticed there’s, how it was so low. And I’m curious if you
could speak to that, like what the implications
of that are for surveillance and just for treatment in general. If it’s a safe med to give, is it just that people aren’t that sick and they don’t really need it? Or how is overdose defined here? Like, I would think that
they would (speaking faintly) for an overdose (speaking faintly). – That’s an excellent question. So, you know, we, Amanda and I, before we set up to do this analysis, we didn’t find that anyone else has done this particular analysis. Now, it may be lurking
somewhere out there. I want to put that out there for a second. Number two is, based on my
own clinical experience, just my own anecdote, my guess was that it would
actually be closer to 33%. And I thought it would be
even more dismal than that. Because, I think, in the ER, at least when I give NARCAN to my patients who are poorly responsive or unresponsive, I’m lucky if a third of
the time they come about. So, you know, so we’ll see
what the St. Joe’s data says. But going back to your question of what is the implication
for all of this, when we shared this
data actually yesterday on a phone call with our
law enforcement partners, who are a big fan of
the ODMAP that I shared, they love that, their first reaction was, oh no, so we’re giving all
these people NARCAN and they, should all these people be getting NARCAN? And that’s when I said no, no, no, no, that’s not the takeaway
message here at all. So, NARCAN is a harmless drug. This is a really, really bad epidemic. People are dying, more so than
motor vehicle collisions now. So, if someone is poorly
responsive or unresponsive, they should be given NARCAN. However, the take-home message is NARCAN administration
alone should not be used for surveillance and
for counting overdoses, so that you shouldn’t say
that, oh well, you know, in such and such county,
you know, last month, 30 patients got NARCAN, so that means that
there were 30 overdoses. That you can’t do because of the numbers, at least based on what
I just shared with you, which is, again, preliminary. – [Audience Member] (speaking
faintly) looking at people who have a (speaking faintly)
opioid versus having naloxone. Naloxone, isn’t, you know, if any of us (speaking faintly)
naloxone (speaking faintly). For the people who are either acutely or chronically on opioids, we will often start (speaking faintly)
with nausea, vomiting. They will often get (speaking faintly), sometimes causing injury
to themselves or the staff. And then, of course, if they’re
treating (speaking faintly), there are (speaking faintly) pain. So, unless people are kind
of (speaking faintly), generally we otherwise will
avoid giving them naloxone, even if they are (speaking
faintly) overdose because it actually makes matters worse. And they’re better
(speaking faintly) overdose. – [Audience Member] I’m
curious because I saw Dr. Joyce deJong was the medical examiner, you’re probably familiar with her, on the west side of the state. She did really interesting
work recently where she started doing tox screens
on non-opioid-related deaths, things like car accidents, and
found the presence of opioids in a lot of those patients, so, which kind of (speaking
faintly) the belief, the opioid epidemic,
especially (faintly speaking), has been a lot more impactful
when we have numbers for it. Does your research have a way
to kind of capture that yet? Or is that part of the
future iteration of this? – Sure, absolutely, so
I actually was on a call with her this week, and we were talking about a synergy, potentially,
between the two projects. And the way I understand her project is, let’s say, even there’s someone
in a motor vehicle collision who’s even a passenger,
who’s not even the driver, and, you know, doing tox screens on people who are fatal crashes. And, you know, I think that it is more an indication of abuse. It doesn’t necessarily tell you that there wasn’t an overdose, but there’s no way you
can prove an overdose, just solely based on a positive
drug screen for someone, detecting opioid in their
system when they are found dead in a car crash or in a different setting. So, I think that it’s a more accurate count of the abuse problem, and I don’t know, in that situation, how you would go about determining if it was truly an overdose or not. And I think that, for me, it is so important
and has been that, to look at this crisis as a spectrum and that overdose truly is at the extreme, and I think that it
requires kind of a focus and an attention all by itself. Now, it can be a sign of someone who is, you know, especially if
it’s a non-fatal overdose, which is probably more
interesting to know about because you can potentially
do something about and intervene on than fatal ones, you really wanna, there’s a difference
between someone who maybe was at a party, a young person, who usually doesn’t use
opioids, used an opioid once, and then ended up dead, not breathing. That’s one situation, but then it’s someone who
has a history of abuse, who escalates and ends
up dead with an overdose. But I think the situation
of escalation and death from an opiate overdose
is what we’re focusing on, and, you know, I really, and I’ll talk about that
in a few slides down, of how important it is to have a kind of like razor-sharp
focus on that piece ’cause it really merits
attention by itself. So, this has been a fascinating ride. And, you know, I’m clearly a
health services researcher. I’ve had other projects that
have had community engagement, stakeholder engagement,
you know, obstacles, but I think that I have to say that I have never been
involved in a project that has had this many turns in the road and this many obstacles. And I say that partially kind of with excitement
and partially with pain, but overall it’s been a
fascinating learning experience for me and my team and collaborators. So, there’s a number of
issues in this phase two. So, phase two is year two,
and year two ends in October. And there is, going from one
county, which was the pilot, to envisioning a real-time
surveillance system for fatal and non-fatal overdoses for the whole state is quite a task. And since, you know,
none of us on our team had designed a surveillance system before, there was a lot of learning as you go. And for starters, I want to say that apparently you can’t just go off and do surveillance on
the population (chuckles). You can’t just one day pick
up and decide to do that. You have to have authority to do that by the State or Federal
Government, so, entity, so that’s one part of it. There is legal issues. There is compliance, IRB issues, stakeholder, funding issues. So, this overall has just
been a fascinating exercise in what has turned out to be
not really a research project but a public health initiative. So, one of the steps in
the expansion process was identifying the key stakeholders. So, I talked about EMS data. So, in the pilot, we
got the data from HVA. This is local. They send us the data. Every one to three days,
we get data from them on NARCAN administration for transports to our emergency
department, so we had that. But we need to talk to
someone who actually has data for the whole state, right? ‘Cause it has to be
the whole state system. Also, for the ED piece, you know, we have Michigan Medicine data. We have access to it every day. We have IRB approval to do research on it and look at it in every which way. We’re getting St. Joe’s data. We have that. But again, how do you go from that to getting ED data for all of the state and also in near real-time? So, you need to get that in a timely way. And same thing with the
medical examiner data, right? So, again, we established a
relationship for the pilot with a medical examiner
in Washtenaw County, but that is a long way from getting medical
examiner data in a timely way in a state where there’s
no centralized system, so lots of challenges there. So, we found out what our needs were. We found out who in the state may have some of the solutions, and we met with the
leadership of those groups. And we established some common ground to see how we can move
forward in phase two with some mutually beneficial goal so that all parties are getting
something good out of this. So, that was kind of the goal
for stakeholder engagement. So, and these are the partners that we actually came up with. So, the goal for phase two is to expand S.O.S. from Washtenaw County, which is one of the 12
HIDTA counties in Michigan, to an additional three to five, so that’s the goal, by October 2018. And we partnered with
a group called MDILog, which is an online medical examiner data collecting system
and reporting system. And you may wonder, why
didn’t I mention that earlier? So, if you can believe this,
after multiple conversations with the Washtenaw
County medical examiner, which is our partner and
we work closely with them, after a few meetings, we found out that they
actually run this system. So, that wasn’t apparent in the beginning, and this online reporting system reports on about 50% of the
counties across the state. It’s not for the whole state,
but 50% of the counties. And from the time a dead body is delivered to a medical examiner’s office,
within three to five hours, data on a suspected overdose
is reported in the system. How awesome is that, right? It’s not the whole state,
but it’s a lot of the state. And essentially, when
the dead body gets in, they can tell usually if
it’s a suspected overdose. There’s drug paraphernalia, or there’s drug on the scene,
or there’s family or someone. And when the report is completed, that’s submitted to the medical examiner, they usually have some data pointing to the fact
that this is an overdose. And then, depending on the
medical examiner’s office, within four to eight weeks, they have the confirmation
and the toxicology report. So, this ended up being
our death data partner, and we basically set up a subcontract with them to get the data. And then partnership with
Great Lakes Health Connect, which, as I mentioned earlier, is one of the health information
exchanges in the state, and they have real-time data every time someone is
admitted to the hospital or discharged from an emergency department with an ICD-10 code that’s
related to an opioid overdose. This data will come to us, for the hospitals that are
willing to participate in S.O.S., within three to five hours. Okay, again, it doesn’t get
more real-time than that. So, that’s our ED partner. The problem with this system is that they have none of
the Upper Peninsula, and they have almost all
of the Lower Peninsula, with a few exceptions. And, as you know, the rural
areas have a big opioid problem, so we need to somehow
get that data over time through agreements and
come up with a solution of how to get the Upper Peninsula. But, as of yet, that’s not something that we can get through Great Lakes. And also, through prior relationships with the Michigan Department
of Health and Human Services, they gave us the initial green light for getting data from
them for the whole state on NARCAN administration for
Upper and Lower Peninsula, for all ambulance transports
that administer NARCAN. And our IRB, once we got the IRB, which I’ll get into in a second here, from Michigan Medicine,
which is not regulated, once we had that, we were able to proceed with the IRB application for the State and application to get their data. And that is going through,
it’s in progress right now. They preemptively told
us we will be getting the data eventually, in
the next couple of months. So, that, you know, that’s a huge piece of
this three-dataset puzzle. That, again, will be linked
through probabilistic matching and ultimately on the S.O.S. website, which I’ll show you a
preliminary prototype of, will be mapped and geocoded, so that we know where the hot spots are, and refreshed every one to three days. So, I’ve mentioned the
capabilities so far, but I’ll kind of go through
them again a little bit here. So, S.O.S. will be updated for most of the data every 24 hours. And for the death data, you know, again, initially it’s gonna
be suspected overdoses, an event which is gonna
reflected on the map in a certain color. And once the confirmation comes, the color changes to a confirmed overdose. And we’re going to have
the toxicology report of what the offending agent or agents are, which is huge for law enforcement, so that they know exactly
what drugs are on the street and what drugs are killing people. And then so, as I mentioned, one of the most powerful
aspects of this approach is eliminating over- and undercounting. So, that, again, a lot of places are using similar databases,
but they’re not linking them so that individually,
when you look at them, you don’t know how much,
where the redundancy is, and where you’re
overcounting the overdoses. The system will present both rates and raw numbers of events. So, again, the data that I
showed you is just raw numbers and not the rates, but the system itself will show the data in both ways. We’ll have a view that’s
rate and a view that’s, or raw numbers. We’ll provide both location
of home and location of death for fatal overdoses and non-fatal EMS and allows for tracking the movement of the epidemic across
communities in the state. The county-level data will
be open to the public. So, imagine, you log on
to this S.O.S. website, anyone can see the map
at the county level. But for everyone else, which
are the key stakeholders who provided the data for the system and others who need to know this data, it’s going to be password-protected. And you will go in, and you’ll
be able to see the data, the overdoses at the census tract level. Now, from a legal perspective, we would’ve been able
to actually give access to the public to the census tract data. And what we would’ve done is that, within the same census
tract, the dot of where the location of the overdose
was or the home location was wouldn’t have to be the exact place, but a randomly located dot
within that same census tract. But we really didn’t feel that, although legally and
from an IRB perspective, we didn’t need to be this careful, we didn’t feel like anyone
needs to know that information, outside of people who have
to do something about it. So, we want to protect
this data and be kind of very careful with it, so we are not gonna, the average person is
not gonna have access to the census tract-level data. So, regulatory review, so
this was quite a process. So, again, we had IRB approval for the county of Washtenaw County. And for that, it was a research project, so we were able to do number crunching and ask all sorts of
questions from the data. But we were now moving
to the larger kind of, the statewide system vision. And to do that, there was an obstacle, and the obstacle was that, to use Great Lakes Health Connect data, so that’s the ED data, we would not be allowed
to use it for research. Big problem, right? So, that we could use it for surveillance, but we cannot use it for research. So, we had to somehow make a case to the IRB here at Michigan Medicine that this is surveillance, that, yes, the Washtenaw County
one was a research thing, we were pilot testing it, now we’re moving on to bigger things, and this is now surveillance
and not research. And then we had to learn a
lot about the differences. And this is fascinating to me, and hopefully you’ll find it interesting. So, public health
surveillance can be classified into, again, research or non-research. And when it’s research, it’s meant to produce
generalizable knowledge. When it’s non-research,
it monitors the population for frequency of occurrence
of a health condition. If it’s researched, it
may be used to invoke public health prevention
or disease, injury control, but the primary goal is not
informing public health. For non-research, invokes
public health mechanisms to prevent or control disease or injury. And if it’s research, the primary intent is usually producing
generalizable knowledge. And for non-research, the primary intent is produce information
about the population. And in the research
aspect of surveillance, again, the scope of the inquiry is broader than assessing occurrence
of disease or injury, and non-research is
not hypothesis testing. So, again, it’s tracking a certain disease or condition among a population
and not generalizable. So, we had to do a lot of homework and meetings with the IRB
and understanding this so that we can make a case, you know, having the obstacle,
again, of not being able to use one of the main data
sources out of the three, the ED data for research,
so we couldn’t do that. So, we had to make this a
purely surveillant system, and we had to make a case that it’s not just public
health surveillance, it’s non-research public
health surveillance. So, this is kind of a summary
of what we went through. So, first and foremost, we met with the Office
of the General Counsel to understand the legal hazards ahead. So, for me personally,
there’s nothing in life that I’m more afraid of, of
the category of knowledge that I don’t know I don’t know. You know, when you don’t know
you don’t know something, that’s kind of dangerous. So, we kind of went in
with the General Counsel and kind of told them this
is what we want to do, tell us everything legally
that we want to be aware us, and pretend like we don’t know anything. And we really didn’t know anything about the laws surrounding,
particularly studying substance abuse in patient populations. I mean, so that’s very, there’s laws around that
that you have to be aware of. So, it was really good that
that was our starting point. That informed our subsequent meetings with the IRB Michigan Medicine, of which there was many. And in order to finally
get this approval of, this designation of not regulated, the CDC Injury Center and
Dr. Rebecca Cunningham got a letter from the CDC
saying that yes, indeed, the project that they’re trying to do is public health
surveillance, non-research. And that was like, that
morning when I got up and saw that letter was like
my birthday, it was incredible. So, and that’s when we got the designation of not regulated. So, you know, was quite a
process, a lot of obstacles, but it was really necessary. And, you know, so, I want
to add one thing here, is that I mentioned
the three data sources, so obviously the EMS data
that we’re going to get from the State, the ED data
that we’re gonna get from Great Lakes Health Connect,
and the medical examiner data. And this project has been
amazingly a community and stakeholder-based
participatory research project every step of the way, and
it will be toward the end, I’m convinced of it. The reason why I say that is,
although we have a subcontract with Great Lakes Health
Connect for them to give us this data from the EDs and
hospitals across the state, every hospital that is gonna participate is gonna have to give
an approval for the data that they’re giving to
Great Lakes Health Connect that will come to us to
be used for this purpose. So, that’s an additional step. And in order to get
buy-in from this hospital, we’ve got buy-in from a number already, but in order to get buy-in
kind of like en masse, we actually have a co-sponsored meeting in Lansing next week with
Great Lakes Health Connect, where all the hospitals
that are participating in Great Lakes are hopefully coming. They’re invited, they’re hopefully coming, so we can present this and get buy-in in a larger scale from people. And same thing with the
medical examiner data. So, MDILog gets data from 50% of the counties across the state, but every medical examiner
that gives their data to MDILog will have to
approve them sharing that data with S.O.S. for this purpose. So, we have our work cut out for us. So, this is a picture of the
prototype of the website. It’s very preliminary. Again, there are many glitches that we have to sort out
and challenges ahead of us, but preliminarily we have a website. So, you would log in, and you would see this kind
of hello, welcome page. And ultimately, there
would be a map of Michigan. And as we get data from
the other counties, this map will be filled in. And if you’re a general public person, you go on the website, you
can see how many to date, you know, there’s been overdoses, fatal and non-fatal, are in your county. And when you log in, if you have a login and your privilege, you know, you’re able to get in, you see it at the census tract level, which is, again, very helpful. But even in that situation, the exact location of overdose
in home for both fatal and non-fatal overdoses is
never gonna be revealed. The dot is gonna be randomly located somewhere within the census tract, but we believe that this is enough detail for public health and law
enforcement interventions at a population level to
inform, again, interventions. And there would be a heatmap view. There’s gonna be just different
views of the data presented. And again, refreshed
every one to three days. And there’s gonna be some demographics, so you have some idea of
the age, gender, and race of individuals who are
experiencing these overdoses. So, what are the next steps? So, our goal is to, as I mentioned, there’s 12
HIDTA counties in Michigan. Of the 12, Washtenaw County is one. The three to five that
we’re going to be adding are gonna be HIDTA counties, so we’re starting to get
a big bang for our buck. So, we know already
these HIDTA counties are where the drugs are sold most commonly, so we believe that this is
the place to start with S.O.S. We’re adding those counties. But ultimately, the goal is to get a statewide system in three years, and we’ve already had many conversations with the Michigan Department
of Health and Human Services. I presented the project at the Governor’s
Commission a few weeks ago, with the explicit intent of asking for them to support this
as a statewide system. And what kind of
implications does it have? So, I think I mentioned
that, as a researcher, it pains me not to use this
system for research, right? I mean, it’s kind of like
oh my God, it’s all here, we should be doing so much more with it. And in time, I believe, you know, I think if we started off that way, we would never be able to pull this off. And as it is, there’s
many challenges ahead. But down the line, I
believe once people trust us and they give the data
and they see it’s working and with the right permissions, that there’s a ton you can do. So, you can take that from
a population-level system to individually tracking people. So, imagine if you do
some kind of modeling around how many non-fatal
overdoses would it take for your next one to be fatal, right? Some predictive modeling around that. And imagine the guy or
a woman comes to the ER, and you give ’em a slip that, you know, if you continue at this
rate, you may die next time. I don’t know if IRB would
approve that or not, but I’m just saying that there would be kind of
individual-level implications. And also, a lot of the
questions I get as well, you know, linking them
to prescribing datasets and all sorts of other datasets, down the line, absolutely, but I think that keeping
this system simple right now or, again, with explicit purpose of identifying fatal
and non-fatal overdoses is a heavy, heavy lift in of itself. And if we complicate it, we
may not accomplish the goal. And again, implementing interventions for multiple repeat offenders. So, there’s one person
in the dataset from 2017 that had, I believe, seven
or eight non-fatal overdoses. So, you can imagine that the risk of that person dying is
really, really high, right? And that’s just in our ER, who knows where else they’re going. So, you can imagine there
would be many other uses for this system down the line. Thank you so much. (audience applauding) (audience member speaking faintly) – [Audience Member] You
talked about standardizing and matching the data from
the different sources. Can you talk a little bit more about that (speaking faintly)? – Sure, absolutely. So, as I showed in that slide, there’s multiple patient identifiers. So, the way that we’re gonna
do the probabilistic matching is that we’re gonna basically say that if it’s the same name and
it’s the same date of birth and it’s the same home address, then it’s probably the same person. And that’s gonna be one of the
disclaimers on the website. There’s gonna be many disclaimers, that’s gonna be one of it. It’s not perfect, but it’s pretty close. Thank you so much. (audience applauding)

Leave a Reply

(*) Required, Your email will not be published