The Role of Health IT to Improve Medication Management: A National Web Conference

The Role of Health IT to Improve Medication Management: A National Web Conference


I’d like to welcome everyone to today’s
AHRQ Web conference on the role of health IT to improve medication
management. Although a few people are still logging in, we’re going to go ahead
and get started on time. My name is Derrick Wyatt and I will be moderating
this webinar. I currently serve as the division of Health IT research grants manager in the Center of Evidence and Practice Improvement at AHRQ. This slide shows the agenda for today’s webinar. Please also note that we are recording
this webinar and the recording will be available via the AHRQ Health IT YouTube channel in about two weeks.
Copies of the PowerPoint slides were emailed to each of you earlier this morning and were also available for download as you logged in today. We will also be sending them to participants via email following the webinar. We are pleased to have with us today an esteemed group of presenters. They include Dr. Karen Farris from the University of Michigan College of Pharmacy, Dr. Jeffrey Schnipper from Brigham and Women’s Hospital and Harvard Medical School and Dr. Margie Snyder from Purdue University College of Pharmacy. This webinar event is accredited by the Professional Education Service Group. For those of you who are interested in receiving continuing
education credit for participating in this activity, information about how to
claim your credit will be presented at the end of the presentation. It will
also be emailed to you after this webinar. For the purposes of
accreditation, let me note that neither AHRQ, Tista, RTI, PESG as well as planners,
reviewers or myself as the moderator, have any financial interest to disclose.
Presenter disclosures: Dr. Snyder has no financial interest to disclose,
Dr. Farris is a consultant with QuiO, and Dr. Schnipper is the principal
investigator for a study sponsored by the Mallinckrodt Pharmaceuticals.
Lastly, please note that no commercial support was received for the
development of this learning activity. Just a brief note about questions. We have
reserved time at the end of the presentation to address participant
questions. However during the presentations, feel free to submit questions that you
have for the presenters using the Q&A panel located on the right of the
PowerPoint slide. As a reminder, participants are in listen-only mode so
to ask questions, you will need to use the Q&A panel. This slide shows the
learning objectives for today’s webinar. By the end of this presentation, you
should be able to explain the benefits and challenges for using reinforcement
learning-guided text messaging to impact medication adherence, discuss the
evaluation of the smart pillbox used by patients during care transitions,
describe the extent to which clinical decision support for community
pharmacist-delivered medication therapy management aligns with establish human
factors principles, and discuss the usability and usefulness of MTM CDS for community pharmacist. And now it is my pleasure to introduce our first presenter: Dr. Karen Ferris. Dr. Ferris is a
Charles R Walgreens the third professor of Pharmacy Administration at
the University of Michigan College of Pharmacy. She is the chair of the
department of Clinical Pharmacy and also faculty lead of digital education at
the Institute for Healthcare Policy and Innovation. Dr. Farris studies
medication adherence and reason for non-adherence. She is currently working
with the Michigan Oncology Quality Consortium to improve how medication
adherence to oral oncolytics is assessed in oncology practices and how medication
self-management can be improved. And now, allow me to turn this over to Dr. Farris. Good afternoon everyone.
So I’m going to be talking with you today about our reinforcement learning
agent that we developed to guide text messages. The name of that project,
the acronym for it was AIM at BP. So using artificial intelligence medication
adherence texting for BP, for blood pressure. I’d like to acknowledge the
funding that we received from the University of Michigan through an M-Cubed
initiative, from our CTSA MICHR we received a pilot grant, and then also our
AHRQ funding as you can see there. So as almost all of you on this webinar know,
medication non-adherence is a problem and it costs our healthcare system a
significant amount of money. Importantly when we think about hypertension, about
30 percent of patients have uncontrolled hypertension despite treatment.
We also know that SMS interventions or texting interventions
can improve patients’ adherence, but what we don’t really understand is the
frequency of texting, how long we need to be texting, does it need
to be for two weeks or six months, and importantly we know that things happen
in people’s lives and our premise is that we need our text messaging to adapt.
And so that’s the point there in health interventions as currently designed have
limited ability to engage patients over time because they may not be adapting. So
our objective was to apply AI method specifically reinforcement
learning, which is a type of artificial intelligence, to develop a medication
adherence system that would adapt text messages. Today I’m going to be talking
about two different studies that we did. Study one was a prospective single group
trial involving 19 individuals. Of course individuals had to use an
anti-hypertensive medication, they had to be texters, they had to have internet, and we collected data with self-report
as well as bottle openings. And I’ll spend just a bit of time on that study. In
study two, we did a randomized trial involving 50 subjects and we partnered
with a Health Plan where individuals at the beginning of our study were
reported to have an anti-hypertensive PDC, or Proportion Days Covered, with
medication of less than 0.5 in the past year. And for this study, we used
self-report bottle openings as well as prescription claims for our evaluation. This is the model that we used to think
about the system that we developed, and what’s important here are the reasons
for medication non-adherence. And so the blue boxes really highlight the types of
text messages that we developed for our project. So the first one you think about
intention, so we think about what does a person think about, treating their
hypertension, and then if we think about medication beliefs, that would be
necessity beliefs or concern beliefs, and then an individual has to remember to
take those medications provided that they have that intention to do so. And
then in our studies we also provided what we’re called positive reinforcement
messages and I’ll show you some examples of those messages in just a bit. So
importantly this was the basis for how we thought about our project and how we
built our messages. So this slide shows you what, really what reinforcement
learning is. So let me just spend a bit of time walking through this diagram. So
the agent is what we call our RL agent, and that’s really the set of algorithms
that makes the decision to take a particular action. So in our particular
study, the action was to send a text message, but importantly, we also included
no message as an option so the agent could send one of five text messages or
no message. That could be its action choice. So then the environment is actually the individual who is receiving the text
message. In both of our studies we set these up so that the text message would
go about 15 minutes before their daily dose of their medication.Then the next
thing that’s important to think about is what’s the reward for the agent. And so
what that means is that the agent makes a decision to send a necessity belief
message, then it waits on the individual to do something to send some sort of
reward, and that reward in this study was opening the pill bottle. So the message
would be sent and then the agent is waiting to know if a pill bottle was
opened. At the same time we recognize that there is a context in which this
process occurred. And so the state variables are those characteristics
about the patient that we entered at baseline, as well as we develop some
additional variables. For example whether a person opened their pill bottles the
last five times they received a necessity belief question for example. So
those are the components of the RL system and importantly the agent is able
to learn and then what’s called exploit that learning, meaning that it knows what
kind of message may work best for each individual and it makes that calculation
each time it’s taking an action. But it continues to explore whether a different
message could also be effective. So the notion of RL is that it exploits what it
knows but continues to explore different action choices so that it can come to
the best application over time with many subjects. So I’m going to move now and
talk about the two studies that we did and in study one, I’m going to focus on
really what our questions were: can we get this thing put together
and can we make it work? So number one, is it working? And then
importantly are the messages adapting? Because recall that’s really the premise
of what we’re saying. Our thinking is that we don’t want to
just send the same old message every day to an individual: take your medicine, take
your medicine, take your medicine. So we want to determine if our messages are
adapting. So I wanted to include this relatively complex picture to show you
all the parts of our adherent system and if you look on the right side of the
screen you can see the AARDEX backend. And so what that’s telling you is that’s
the pill bottle, and you can see there a picture of the pill bottle and I have a
better picture of that on the next slide. And so the AARDEX backend would tell us,
and we connected to it, and it would tell us if the bottle was opened. So then you
have the database in the center that’s bringing together the study entrance,
which is the web interface, the RL engine, which is the algorithms and is making a
choice about what kind or whether to send a message, and then we have the
messenger component, which is directing Twilio, which is what we use to actually
send messages to individuals. So that’s the entire system that we put together
and so our real question was: did it work? So this slide shows you the picture of
the pill bottle that we use for the reward. And so this was in 2015 so I
would assert that we have even better technology now that we could use. So in
this particular instance, the individual had a MEMS cap that sat on top of that
bottle that you see. But to direct the information to the database at AARDEX,
they had to put that pill bottle on that little, it looks about the size
of a coffee cup holder, and they had to put that on that device upside down. So
you should already be thinking that’s a bit obtrusive, that’s an extra behavior
that a person had to do to receive their messages. And so we agree with that
and we have some thoughts about that for future work. So this is a description of the five
types of messages. If you think back to that model I showed you, these will all
link to those particular items that I talked about. And so you can see examples
of disease beliefs messages, an example of a remembering strategy message would
be: to help remember your blood pressure medications try putting your bottles
near something you see every day like your toothbrush. So again our agent would
select one of these messages or no message every day to interact with the
individual. Now if it was no message and of course there was not an interaction,
okay just to be clear about that. So one important thing that we need to
be thinking about is, when we develop messages and we’re experts and we sit
around and we develop our messages, it’s important to know if they are perceived
in the manner in which we intended. So we did two different card sorting
studies. We did one with University staff and we did another one with student
pharmacists. We gave them random samples of our five different kinds of messages
and we asked them to put them in stacks or buckets as I call them, different
categories if you will. And so this is a heat map that shows you for categories
three, four and five which are on the right, those are very well defined. The
two in the left upper quadrant which would be necessity beliefs and disease
beliefs, there’s less discrimination. But we know that in our second heat map
which I don’t have here, we had a little better differentiation. But we determined that these were sufficiently different messages from
this type of analysis. So this slide shows you the results from our first
study, which is just looking at: did the messages change over time?
So what you see here is the distribution of messages that were sent in the second
month compared to the distribution of messages that were sent in month 6 or
the last month. And you can see that the distribution did change. So for example,
32 percent of the messages were necessity in the second month and they
turned into 24 percent by the sixth month. You can see remembering strategies was
13 percent and that increased to 23 percent. So based on that work which was our before/after, we had the messages being sent, people were receiving them, and we could
see that the distribution of those messages was changing over time. So
therefore our view was that our RL agent was adapting. So in study two, we were
interested in whether the device continued to work, the agent continued to
work, but importantly was adherence changing? And what do participants think
about it? So these are the baseline comparisons of the individuals who
participated in our study. And let me remind you that this was a randomized
trial, everyone in the study had the reward, had the pill bottle, treatment and
control both had the pill bottle. And the intervention was really around the
messaging. So people in the intervention group were receiving messages but
everyone was monitored for pill bottle openings. What this information shows you
here is the baseline comparisons of individuals who were in the study.
Importantly you can see that that PDC was very low according to the records
that we received from our health partner, and there were no other statistically
significant differences based on those demographic information or in their
literacy or actually in self-reported adherence even though it looks different,
the numbers were not there to show that that was different.
This is a similar representation you saw earlier. It’s the distribution of
messages over time: month 1, month 3 and month 6. What we did a bit differently in
this trial was we, for lack of a better term, we biased the agent to make a decision
toward not sending a message. And so what that means is if the agent determined
that the messages had a similar probability of being selected, and if no
message was an option, it would go with that option. And we did that based
upon some feedback from our first trial, which I didn’t share with you, that said
we might have been messaging a bit too much. These are the actual pill bottle
openings and so you can see the control and then you can see the group that
received the messaging. And what you can see is that the green is perfect
adherence, the red is very poor adherence and the yellow is less than 100 but
certainly still really good above 80. So what you see is more green in the
messaging. Importantly you see everybody had pretty good
adherence when they started this trial. Even though our PDC was quite low, when
we brought them into this trial and everyone was being monitored, they were
quite good at taking their medication. So you see more perfect adherence in the
messaging group particularly early on in months 1, 2 and 3.
You also see more variability over time in the messaging group, but remember we
also were sending fewer messages at that time as well. These data show you the
self-report and so these are differences between baseline and three months for
the people who got messaging and the people who did not get messages. And what
you can see is that at month three, there was an improvement in medication
adherence, a difference between the two groups, and at 6 months that difference
was no longer there. And again this was a self-report of medication adherence. This
has a typo, that’s too bad, it’s supposed to say PDC
for the antihypertensive medication by group over time. So the red line
represents the individuals who receive text messages, the solid line is the
group that only had the pill bottle, and these timeframes represent, on the most
left one year out before this trial, the midpoint is six months before the trial,
and then the rightmost data points represent during the six-month trial. And
you can see those PDC’s based on the claims we received at the end of the
trial at the bottom. And so you might want to note the PDC’s just prior to the
trial are quite a bit higher than what we had been reported or told that they
were, so but this is still our calculation of the change over time, it
is not a statistically significant difference, the trend is in
the correct way but it is not different. And then quickly this slide represents
some early analysis that we’d done that said to us that there are a couple of
different groups responding differently to different messages. And so if you look
at the reddish color, that individual is actually quite different and was not
responsive really to any of the messages, whereas you see the blue bar, those folks
were responsive to every message and you couldn’t even differentiate what type of
message was best for them. If you look at the bottom group of individuals, you can
see that medication necessity would help them more so than the other types
of messages. So again this is very preliminary and we really need honestly
way more people to get a good sense of what type of individuals respond best to
what types of messages. And then my last data point is just to share with you
some feedback and of course we have way more feedback than this, but this is just
to give you a sense of it. So at three and six months,
individuals were telling us that the frequency of messaging was about right.
So remember we biased the agent to no messaging over time so individuals were
getting fewer messages and were still very happy with that. But again what we
also saw was that their adherence was slightly poorer if you remember those
pill bottle opening data. You can see here in the middle that they were saying
one message a week, a few times a week. Again I think we don’t have good data, we
don’t have good results in this trial or in other studies about this
messaging frequency. And then asked if they would enroll in something like this
again, forty to fifty percent said yes that they would, and about the same
amount said no they wouldn’t. So certainly there are a group of people
out there who would enjoy having this type of support with their medications.
So what did we learn? Or what are some of the issues that I think we face moving
forward? What we know is that this agent adapted over time, its impact on non-
adherence is mixed, the self-report was improved at three months. I think early
on, the adherence we saw was really positive in the messaging group but over
time that did deteriorate. Importantly an intervention to improve medication
adherence needs to be delivered to individuals who are not adherent or it’s
too difficult to show a difference over time and it’s really not needed. So while
our early data showed that these were individuals or were reported to be
non-adherent, when we looked at the claims afterwards, it certainly didn’t
seem that that was quite correct. And so there’s some disconnect between that. My
point would be that if we had people who were really adherent already trying to
improve that is difficult. Our recommendation would be to think about
how we could recruit into a study like this the uncontrolled disease. And then
importantly in the next point, how we could use an unobtrusive measure or
unobtrusive reward such as number of steps. So we’re
thinking about and we’re doing some work toward moving this agent into treating
heart failure or helping to manage heart failure, where number of steps would be
the reward and we would have adherence messaging, but we would also have other
types of messaging such as an individual managing their diet. I think importantly
even with an RL system, the system can learn to send no message so that you’re
not just constantly messaging individuals, but I think we need a lot
more work around if they’re a loading dose for messages, do we really need
daily messages or can we load and then booster, for lack of a better term, over
time? Okay so in closing, text messages improve self-reported adherence at three
months not at six months, pill bottle openings however showed little
variability. The adaptation of text messaging did work. Individuals indicated
one message per day or one every two to three days was
generally preferred. And our next steps as I said are really to put this in a
place where we can work with heart failure patients but I would be really
interested in how we might partner with someone around using this agent with
high cost specialty medications. And then finally, I must thank this great group of
investigators; we had engineers, we had health services researchers, I studied
medication adherence, so I don’t think that this was a project that could have
been done with me alone or even with the engineers alone, so we really
needed this team to get this done and I owe a debt of gratitude to these
individuals shown here. Thank you. Thank you Dr. Farris. As a reminder we will be
taking questions after the presentations so please submit any questions you have
for Dr. Farris using the Q&A panel. Let’s move on to our second presentation of
this webinar which is by Dr. Jeffrey Schnipper. Dr. Schnipper is an associate professor of
medicine at Harvard Medical School, associate physician at Brigham and
Women’s Hospital and the director of clinical research for the Brigham Health
Hospital Medicine Unit. He is also the director of a new Harvard
Brigham fellowship in hospital medicine research. He received his MD from Harvard
Medical School in 1996 and completed a residency in internal medicine and
primary care at Massachusetts General Hospital. His research interests focus on
improving the quality of healthcare delivery for general medicine patients.
And now I would like to turn the controls over to Dr. Schnipper. a smart pillbox being used during
transitions of care. So here’s the outline of what I’m going to cover in
the next 20 minutes or so. First a little bit of background on why we conducted
this study. I’ll spend some time giving you a description of the smart pillbox
intervention. I’ll show you the flow diagram of how everyone did in the trial, spend a lot of the time talking about barriers to implementation of this
intervention, and then move on to a discussion of what it would take to make
this intervention part of usual care and I’ll then I’ll do next steps in
conclusion and as that commander Wyatt just said we’ll do Q&A after all three
presentations are over. So first a little bit of background. So as many of you
probably know, the transition period between the inpatient and the outpatient
settings is a really potentially dangerous time for patients especially
when it comes to medication safety. Adverse drug events or injuries due to
medications often occur after discharge. So they’re very common and they’re
potentially serious. Two of the reasons why these adverse drug events occur are patients’ misunderstanding of their medication regimen so they’re not even
sure of what medications they’re supposed to be taking after they leave the hospital
and then even when they are aware they’re often not adherent with those
regimens. And we’ve done a number of studies using patient education and
feedback and phone calls and they’ve all been less effective than expected when
they’ve been studied rigorously. So while technology isn’t always the answer to
all of our problems, we thought this was a good place to try some and there’s
been a lot of new technology that’s been created over the last few years. So what
you see in front of you is a smart pill box. So picture a blister pack of
medications for the week, so up to four times a day all seven days of the week,
that slips inside this pillbox. When it’s time to take your next dose of
medication, you perforate the top of that blister pack and you remove the
medications from that well as you can see in the picture on the bottom left. And at the bottom right you can see an adherence report. So let’s unpack this a
little bit and talk about the different features of this smart pillbox. So first
of all and most obviously it sends visual and audible med reminders. So in
other words, when it’s time for a patient to take their next dose of medications
that well of the blister pack lights up and a
little alarm goes off and then the pillbox then senses whether those
medications are in fact removed from that well on time. If you do not remove
the medications on time then it can send a phone message, a text message, or an
email alert either to the patient or to their caregiver, so for example a family
member, and let them know that the medications have not been taken or at
least not removed from the pillbox. And then it takes all that information that is
learned over the course of the week and over the months that the
patient is using the intervention, and sends that adherence data in a report to
the patient’s primary care physician. So there’s a link that we put embedded in
our electronic health record, which takes the PCP to that report and that report
is also available to a pharmacist case manager. So let’s talk about the details
of the intervention a little bit more. So the way that we did it is that the first
month of meds was supplied by our hospitals outpatient pharmacy right
before the patients left the hospital. Some hospitals call this a meds to beds
program. This guarantees that the patient left the hospital with the correct
medications and now with their blister packs and their smart pillbox.
Subsequent medication trays were then mailed to the patient every two to four
weeks either by our hospitals pharmacy or by a collaborating pharmacy depending
on the patient’s insurance. Any medications that are at risk for unexpected changes
so for example warfarin or a diuretic in a patient with heart failure where that
dose of the diuretic changes a lot, these meds are withheld from the pillbox
because it would be too hard for them to change constantly if you’re only getting
a new shipment of trays every two to four weeks. In addition any as needed
medication so PRNs, opioids or other controlled substances
are also not supplied in these pillboxes, in these medication trays. We had a
pharmacist care manager that would call the patient if there was evidence of non-
adherence, so three days in a row of less than 80 percent adherence would trigger a phone
call to the patient and see how things were going. And then the intervention
lasted for six months, and at that point we made sure the patient had a safe
landing back either to their usual pharmacy, they can continue to use just
the blister packs without the pillbox if that’s what they liked, and if they
really really wanted to use the pillbox, the company that supplied it was willing
and generous enough to let them continue to use it
for an indefinite period of time. So the goals of our study
were to implement the use of this smart pillbox in the transitional care
setting. This kind of a pillbox has been used in the stable ambulatory setting in
prior studies but we wanted to see it in this particularly high-risk period of
time when patients leave the hospital and go back home.
We wanted to evaluate the effects of this intervention on medication
discrepancies, on adherence and on markers of chronic disease control, and
also determine barriers and facilitators to implementation. And this study was
approved by our primary care practice-based research network, our PBRN.
So here’s the design of our study. So the inclusion criteria included adult
English or Spanish-speaking patients who were admitted to our general
medicine cardiology or oncology teams, were on five or more chronic
medications, were going to be discharged home and had a primary care physician in
one of our participating hospital-based practices, we have a number of those. We’ve
been randomized by practice to one of three arms: so the smart pillbox, two a
normal pillbox so we can see the incremental benefits of the smarts of
the pillbox and then the third arm is usual care. And the goal was to enroll 133
patients per arm. Outcomes were followed for six months and included the
following: so discrepancies between the regimen that is documented in our
electronic medical record so what are they supposed to be on and what was actually
being dispensed either by their usual pharmacy or by our pharmacies filling
these medication trays, medication adherence and we’ll talk about what that
means in a few minutes, and measures of chronic disease control. And the
intervention lasted from January 2017 to June of 2018. So there were rules that a
lot of providers in our system had to do in order for this intervention to go
smoothly. So on the inpatient side the inpatient team had to perform discharge
medication reconciliation and send those prescriptions to our outpatient pharmacy
as early in the discharge process as possible so that the pharmacy had time to
do what they needed to do. They would then send an electronic message to the PCP or
the specialist with the planned outpatient regimen. And this was based on
feedback from our PCPs who said, you know, if you do discharge med rec incorrectly
or I don’t like what you’ve discharged the patient on and I’ve got to change
it three days later and you’ve given me a pillbox or I’m stuck with
that regimen for two or four weeks with that patient, that’s going to be a
problem and so this was our compromise to say, okay we’ll make sure that you
agree with the regimen before we send the patient home so you’re unlikely to
change it when you see them back in the office the following week. And then
finally the inpatient team had to complete a bedside medication delivery
form; this is what triggers the meds to beds program: providing an estimated date
and time of discharge, know if there are gonna be any anticipated last-minute
changes to that discharge medication regimen, right sometimes you’re waiting
on like what’s that last dose of furosemide gonna be before the patient
leaves, so to give the pharmacy a heads-up that that might change, and then
note if any of those medications needed to be withheld from the pillbox due to
risk for unexpected changes. In terms of getting the adherence report, so it was
available from a native screen within our electronic health record so to the
PCP it just looks like it’s built into an EHR, it would then provide a link to
the adherence report login screen and the default login and password was the
same as the providers’ credentials in our regular electronic health record where
possible, it wasn’t always possible, but for the most part hopefully it was
something that the PCP could remember so they can get into that adherence report.
So let’s look at the adherence report. So by clicking on the dashboard, the
providers would see at the top first a list of all the patients who needed
outreach, so those who had three days with less than 80 percent compliance. And
then if you scroll down further the providers could see their full
list of patients that were involved in the study. And then if you clicked on one
of those patients what you would see is an adherence report. So as you can see
here, just an illustration, this graph displays the percent of doses taken
based on the pillbox information over the previous weeks and you can see every
dot is a different week going back in time. And then you can adjust the
timeframe by instead of clicking on the weeks button you click on the months
button and then you can see it month over month. Remember all this information
is being sent over the airwaves through a cell phone signal and captured so we
can tell people how they’re doing over time. And then this is what I call the
heat map. So this tells you exactly which wells of the medication tray the
patients took and which ones they didn’t take. And so this, in this first view here,
this is in a given week you can see each of the four times a day, each of the
seven days of the week. Did they take their pills on time? Did they not take
their pills on time? And then you get to see how late or how early they took the
medications compared to when they were supposed to, when the alarm goes off. And
then in aggregated format, you can see, you know, over the course of weeks, you
know, which particular days of the week, which particular times of the day are
patients having trouble with. So in this particular example, you see the patients
never take their medications on Saturdays and they’re all dark blue
while there may be some other times a day, other days of the week when patients
are more likely to be adherent and to take their medications out of the tray.
So what was the role of the outpatient providers? So we encouraged PCPs or the
practice managers in their practices to review these adherence reports
periodically and if they saw evidence of non-adherence, we suggested that
they engage their patients as the practices saw fit. If the collaborating
pharmacy did reach out to patients, because remember we were tracking this in the
background and contacting patients if they were not adherent, they would write
a note about the outreach that they did and then we would add that to the
patients’ chart. And finally if there were discrepancies between the EHRs and
the pillboxes medication list, so when it was time to send a new supply of
medication trays, then the practice would be contacted to try to resolve the
discrepancies. So in the ideal world, the pharmacy was always sending out the
regimen if the patient was supposed to be on at least according to our
electronic health record. Outcome assessment as I said was during the
180 days after discharge from the hospital. The first outcome were
medication discrepancies. So again other differences between what was being
filled by the pharmacies and what the active medication list said in our
electronic health record at any moment in time when those medications were
being, when those prescriptions were being filled. The second outcome was
medication adherence so using proportion of days covered (PDC), just like
in the previous talk you heard, and then we did a different measure of adherence
called the Daily Polypharmacy Possession Ratio which I can get into details
during the Q&A if you’re interested, but basically the idea here is that on
every single given day you say, okay how many medications is the patient supposed
to be on that day? And how many of those medications did the patient have a
supply of based on the prescription refill patterns? So it’s a little bit
more nuanced than a PDC. And then finally we looked at measures of disease
control. So if the patients had hypertension or diabetes or
high cholesterol, we looked at their blood pressures, their hemoglobin A1Cs
and their LDLs respectively over the six-month period. Specifically in the
intervention arm patients, we looked at a few other things. So for example, what
portion of their medication regimens were we able to put in the pill trays? How good
was our on-time delivery of trays? How often did providers use these adherence
reports? And then how often did they document that they actually took action
in response to those adherence reports? And then what we’re doing right now is
qualitative analyses, so we are interviewing patients, inpatient
providers and outpatient providers and asking them about the perceived effects
of the intervention on patient care, on workflow, about barriers and facilitators
and implementation, and then finally suggestions for improvement. So here is
the flow diagram. So we assessed 961 patients for eligibility of which 719
were eligible. The ones that were not eligible were mainly because they were
unlikely to be discharged home or they weren’t on five or more medications, some
that had cognitive impairment and no health care proxy. Of those 719, 512 were
not enrolled: 401 patients declined, 85 patients we
were never able to approach, they were discharged before that time. So that led us to 207 patients who were randomized. Because we were having
trouble randomizing enough patients, we actually ended up getting rid of the
basic pillbox group, so arm 2, so those 24 patients that you can see in the middle
group that we ended up excluding from the study, so that gave us enough
patients so we could really do the main comparison which was usual care compared
to the smart pillbox. As you can see, 73 patients were allocated to the
intervention, 93 were allocated to usual care. Of the ones allocated to
intervention, as you can see, we had 15 post enrollment exclusions because those
patients in the end were not actually discharged home or run fewer than five
medications. Of the 73 who are allocated to the intervention, 40 actually got
it, 33 unfortunately were unable to, because they were discharged over the
weekend when our hospital pharmacy was closed or there wasn’t enough time to
receive it or we had staffing issues and I’ll talk about some of these logistics
in a few minutes. And then as you can see, of the 93 that were
allocated usual care, only 1 withdrew. And then again in terms of that’s who
received the intervention or usual care, and then in terms of again those who are
allocated the intervention, 12 patients stopped the intervention early
either because their PCP wanted them to, the study team felt like it was unsafe
for them to continue to use the pillbox, the patients sometimes withdrew, but we
still analyzed 72 out of the 73 patients, everyone who didn’t withdraw from the
study. So this is an intention-to-treat analysis and compared to the 93 who were
allocated to usual care. Okay so let’s talk about some of the barriers that we
encountered and as you can probably get a sense from this flow diagram, there
were a lot of barriers. So the first set of barriers were about enrolling patients
in the study. So as you can see in this table we have the barrier
and then how we try to solve that problem or should try to solve that
problem in the future. So the biggest one that we found about enrollment was that
patients denied they had any previous problems with adherence. And even though
we know again from some of the things that were said during the last talk, that
you know 50 percent of patients are not adherent with their meds, but there is a
lot of pride, there’s also a lot of shame, and a lot of guilt around patients non-
adherence and so it was tough sometimes to break through that, especially if you
know, now we say, well we have a system that’s gonna be monitoring your
adherence really really closely, some patients just didn’t want to be a part
of that. And so we did a lot of scripting to try to reduce the stigma of accepting
this intervention, saying we’re gonna give it to everybody just because you’re
on five meds and going home, we didn’t you know, pick you out of a crowd because
we knew you’re having problems with adherence in the past. Sometimes we tried
to leverage patients caregivers or providers. The patients were kind of in
denial but maybe the other two were in our court and saw the need for this
intervention. Perceived portability issues with the pillbox; it is kind of
bulky and for patients who leave the home a lot, some of them didn’t want to
enroll in the study for that reason. And so we talked to patients, that they could
always remove the pills you know, out of the tray early in the day and just keep
them with them, you know wherever they travel, but that was definitely a
barrier. Too many medications dispensed outside the pillbox, right you hit some
kind of threshold in a patient’s mind where they say well so many of my meds
are not in the pillbox because they’re controlled substances or they are
insulin injections or they are that dose of furosemide that changes a lot, then
they don’t see the value of the pillbox anymore. And so we were able to provide
text reminders for non-pillbox medications to remind them to take them
so they still had that benefit and then there was a lot of patient education
that we tried to provide about using the pillbox under different situations. It
was possible for some patients co-payments to increase. So if they were
doing 90-day fills and now they had to go down to a 30-day fill for these
medication trays, sometimes that meant the co-payments went up and so we did
emphasize that the benefits of the intervention may be worth the copay
increase. Some patients believed that, some did not. And then of course there are always
gonna be some patients who are resistant to participating in any kind of research
study and so you know we tried to highlight the benefits
to patients and to the general public, appealing to this altruistic side.
What about barriers at discharge? So the biggest barriers here were around
logistics. So you know as I said, one of them was turnaround time. So the pharmacy
often received the discharge prescriptions for patients less than two
hours before the anticipated discharge and this just did not give them a lot of
time. And so we encouraged clinicians to provide the prescriptions as early as
possible, to do discharge med rec as early as possible, facilitate early
communication between the pharmacist and the clinician. The time required to
dispense initial medications and enter the information into the pillbox
application is actually pretty long, and so what would really need to happen is
develop software interfaces to just speed that amount of time to get the
data into the system to both interface with our EHR and with our
medication dispense system. As I mentioned during the flow diagram slide,
our outpatient pharmacy, a lot of outpatient pharmacies are affiliated
with hospitals, are closed on weekends and they close at you know five or six
o’clock every night. And so this made it hard for patients who were gonna go home
during those times. We did develop protocols for patients who were
discharged over a weekend to return on Monday to get their pillbox but not
every patient wanted to come back. And then lack of insurance coverage for
early prescription refills. Right so if a med was filled right before the
hospitalization and now you wanted to fill it again so you could put it in the
pillbox, you know that costs money, because the insurance would deny the
early refill. We had a fund to cover those costs but that would be hard to
reproduce in the future. And you’re not allowed to take the pills out of a
previously dispensed bottle and stick them in a medication tray; most states have
regulations against that. And finally barriers after
discharge, difficulty reaching patients to confirm those refills so we tried to
reach patients in a variety of ways. Difficulty obtaining prescription
refills from providers especially if there are multiple prescribers per
patient and so we had to develop a bunch of standard operating procedures to make
sure we got those prescription refills. Sometimes there were issues with pillbox
connectivity, so poor signal in some locations and sometimes we had to work
through with patients where to put the pillbox in their home so if the cell
phone signal was the strongest to send those adherence reports. And then and
that will have to be fixed with future pillbox enhancements. And finally
sometimes if it was just one small pill in a tray in a well, the pillbox would
not detect that pill had been removed and so again some planned enhancements. We also tried to group medications in those different wells to prevent this from
happening. So what would it take to make this intervention part of usual care?
Well if it’s no longer a study, several issues go away, right? There’s no more the
issue of resistance to participating in research. The issues of stigma and denial
may or may not go away but it may help if the intervention is recommended by
one’s own providers as opposed to you know a researcher. The logistical issues
as I said were prominent. Some would be resolved by productization, right? So if
we worked harder to ensure compatibility of the pillbox software with
our electronic health record, with our medication dispense system, that would
definitely help. Having multiple pharmacists who can program the software
dispense blister packs to take advantage of efficiencies of scale. We would
definitely need more concerted efforts to facilitate communication between
clinicians and pharmacists and provide those prescriptions early. Some of the
logistical issues are harder to correct so this tension between the time
constraints to set up an electronic pillbox and the rush and the
unpredictability of hospital discharge definitely paramount. And then the
restricted hours of most hospital-based pharmacies make evening and weekend discharges challenging. Some issues may require more
systemic changes, right? So should insurance companies and can insurance
companies agree to a waiver of an early refill and a reduction in co-payments to
90-day levels in exchange for using this intervention?
Because it really does improve adherence. Is there a sustainable business model
for pharmacies to do the extra work and if not who pays for it? What about the
former pharmacies that used to fill up patients’ meds but now have been
taken out of the loop? They may object to the loss of business. And you know this
is really a paradigm shift; we’re talking about regimen-based prescribing as
opposed to medication-based prescribing. Is it okay for an antihypertensive to
not change for two weeks until the new shipment of trays arrives in the
patient’s mailbox? And then some missions require iterative technological
improvements, so things about signal strength, pillbox connectivity, thresholds
for intercepting pill removal, portability. And then finally some
patients may not be ideal candidates for the intervention. And the question is: is
this the best time to initiate this intervention? There are a lot of
advantages to doing it at the time of hospital discharge, when we know so many
things go wrong with patients medication regimens, on the other hand, is it
logistically really just too hard to do this at the time of hospital discharge.
So our next steps are to complete our interviews with patients, PCPs and
inpatient providers, complete our outcome assessment, work on dissemination
activities. And in conclusion a smart pillbox has the potential to decrease
med discrepancies and improve adherence. Otherwise ideal candidates may resist
this electronic intervention for a variety of reasons. And some patients may
not be ideal candidates because of frequent travel or the nature of their
regimen. Many patients have been really pleased with this service, we really did
make a difference in a lot of patients and we’re looking forward to looking at
the outcomes. And although logistically difficult the potential benefits of
using this intervention during this high-risk period of hospital discharge
may warrant further efforts to modify and refine discharge workflows
to really make it possible. So I want to thank all of my collaborators, and this
is just a partial list of them, without whom the study would have been
impossible. And here’s my email address if anyone has questions afterwards and
I’m looking forward to the question-and-answer session at the
conclusion of this presentation. So thank you so much for listening. Thank you Dr. Schnipper. As a reminder we will be taking
questions after the presentation so please submit any questions you have for
Dr. Schnipper using the Q&A panel. Let’s move on to our final presentation of
this webinar which is being led by Dr. Margie Snyder. Dr. Snyder is an
associate professor of pharmacy practice at the Purdue University College of
Pharmacy. She received her doctor of pharmacy and masters of Public Health
degrees from the University of Pittsburgh. Dr. Snyder’s expertise is in community pharmacy and health services research. And her research is focused on strategies for optimizing the delivery
of medication therapy management through application of qualitative and mixed
methods. And now I would like to turn the controls over to Dr. Snyder. Thank you
Dr. Wyatt. As Dr. Wyatt said, my name is Margie Snyder and I’m an associate
professor at the Purdue College of Pharmacy. And I am the principal investigator for an
ongoing AHRQ-funded health IT study called Enhancing Clinical Decision Support
Applications for Community Pharmacist-Delivered Medication Therapy
Management. The aims of this study are first to evaluate
the extent to which computerized clinical decision support or CDS for community
pharmacist-delivered medication therapy management aligns with established
human factors principles. And two, to assess the usability of
MTM CDS for community pharmacists, as well as pharmacists’ perspectives on
the usefulness and usability of these technologies for patient care. During the webinar on September 13th,
I did present more information about the study. However due to
concerns from potential journals where manuscripts for this project will
be submitted, they’ve asked that we wait to make the complete slide
deck available. So therefore, if you have any questions regarding this
research, please don’t hesitate to reach out to me and my contact
information is there on the slide. Thank you. This concludes the content portion of
the webinar. Now we have a few minutes left for questions. You can type them in
the Q&A section of the WebEx portal. We may not be able to get to all
of them, however we will answer them all in writing
and forward them out to all the attendees. So the first question I have
is for Dr. Farris. A participant asked: how does the RL agent
make decisions? So the RL agent makes decisions based
on a set of computations that our engineer built and so it’s taking information
about baseline characteristics of the patient. So we actually asked them about their beliefs
in treating their hypertension the extent to which they
thought their antihypertensive was necessary, their concern beliefs.
So we asked those baseline values, our baseline items as well as other
demographic information and then we set the agent originally based upon their
responses to those. So if the person said they didn’t really think they needed
their antihypertensive medication or they were lower on that item, we would
send them a lot of necessity messages because the agent would say oh here’s an
area where we think we can…, we set it up that way. Then over time the agent is
learning what kind of message is working for an individual and that’s based upon
a set of computations that fit within you know an algorithm that sit within
the agent. I’m not sure if that answered the question because that’s all I can get
in right here, how about that? Thank you. The next question for you also Dr.
Farris. Can you talk a little bit about the cyber security precautions that
would apply to the technology you researched? Yeah that’s a really good
question. So if you remember that diagram that I showed you about the entire
system. So the entire system was sitting within our health system,
which has servers that are protected. So that kind of data was managed in a
HIPAA-compliant environment. In terms of texting we did not include information
that would have been a PHI problem. I know that now we can do
HIPAA-compliant texting and I think in future applications we definitely need
to do that but we were very cognizant at the time that we did not need to
put in a message that said: “hey Karen, you need to be taking your
whatever medication for your high blood pressure”, so that we would
name the patient, we would name the drug and we would name their
condition. So we did not tailor for lack of a better word, messages to that
degree because we didn’t have a HIPAA- compliant text program at the time. Those
do exist now. Thank you so much.
Dr. Schnipper, the next question is for you: Do you have any preliminary results
on the impact of a pillbox on patient care? Yeah, it’s a great question.
We’re just looking at the data right now. We certainly have some
anecdotal stories of patients who were very effective users
of the pillbox and really found it incredibly helpful, the convenience of
having their regimens mailed to their door, the reminders being there. Fewer
stories about the caregivers, you know, being part of the solution, which is, you
know, not what I wanted to hear. I really wanted to hear that caregivers were now
much more involved in care because, you know, they would get that text message,
for example, that their loved one didn’t open their pillbox on time. But the
final results, we’re analyzing as we speak. But you know I think some
patients benefited incredibly from this, but not all did. And I think what we
really need to look at is, you know, who are the kinds of patients who benefit the
most from this kind of an intervention. It’s certainly not for everybody but if
we had that better idea then we could certainly target this, you know, much
more efficiently. So stay tuned. We’ll know more soon.
Thank you, we look forward to the results that you have. The next
question is also for you, Dr. Schnipper. If you had to do the study over, what
would you have done differently? Yeah, that’s a great question. You know,
most of the studies that I work on are considered low-risk studies from
a human subjects standpoint, which means we can have a
non clinician research assistant enroll the patients in the study. However, in
hindsight, you know, the logistics of pulling this study off are so embedded in
how our healthcare system works, in how our hospital work that I actually
wish we had had a clinician be the one enroll the patients in the study
and then trigger all the subsequent actions that needed to happen both the
inpatient team and the hospital pharmacy team, because it does really require a level of sophistication in order to anticipate
what problems might occur and to prevent them from occurring. And so I think
that’s probably the main one that I took away from this. There are
others as well but I think that was probably the main one,
would be to have an actual clinician enroll the patients in the study
from day one. Okay, thank you for sharing that with us.
Dr. Snyder, the next question is for you: How do findings and recommendations
vary by alert type? For example, safety-related alert versus
adherence-related alert? Yeah so that’s actually something that
we’re still looking into but as I did note on the one slide, although all
of the alert types scored similarly as far as a numerical score, on the modified
I-MeDeSA, we did see a slightly better score among the adherence alerts. And on
the slides where I indicated the tables of our findings thus far, in
situations where it would say most alert categories versus all, we did find the
exception was for adherence alerts. So it’s interesting that
there do appear to be some heuristics that were better met among
our adherence alerts compared to other alerts and that’s something we’re
continuing to consider and think about. Thank you. We also had a question for you
in regards to… the study was focused on community pharmacists as the end users, but pharmacists have varying levels of experience with MTM and are often
supported by other types of users, such as pharmacy residents or students.
Can you kind of speak to this? Yeah, that is something that we noted in the prior AHRQ-funded
study that I mentioned, as it was seeing this kind of influence of the
user type and how they interface with CDS. You’re correct, for this study we did
focus on community pharmacists. I think we had maybe one community pharmacy
resident that participated. However actually since we started
this study, we have worked to collect some additional data for a follow-up study
where we’ve interviewed both community pharmacists as well as community
pharmacy residents. And we’re currently analyzing those data to better
understand their information needs as they’re completing a medication therapy
review, so we can think about how we might be able to best design CDS that could
better meet the needs of different user types. Okay, thank you.
Dr. Farris, the next question is for you. Is the pill bottle opening a good
reward for the agent? That is an excellent
question and in my opinion it is not. It may be a good reward when there’s true
variability in adherence but when adherence is as good as ours was, all the
messages end up with the same usefulness because individuals are opening their
pill bottle. So a reward that is really more clinical, so I mentioned the number
of steps, if there’s a way that you could have a blood glucose if you were doing
something in diabetes. I think we’ve got to think about what those better rewards
are for an agent in future work. Well we have thought about it and that’s my
point, right there. Okay, thank you.
Dr. Schnipper, the next two questions are for you. The first question is: can you talk some
more about the pros and cons of initiating this intervention at hospital discharge
as opposed to stable ambulatory patients? You know, absolutely. You know,
from a logistical standpoint there’s no question that is easier to do
in a stable ambulatory patient. You can wait until most of their prescriptions
are going to be up for renewal and therefore will not be denied by their
insurance when it’s time to you know refill them, you can pick a stable moment
in time when it’s a good time to, you know, set all this in motion, you probably
have a couple of days to set up the pillbox program, the software, fill
the blister packs, and there’s not this crazy rush and then you know, just have it
mailed to the patient’s home. On the other hand, there’s probably also less
immediate benefit, you know, if we know in reality right now at the time of
discharge, that you know, three days after discharge half of our patients are
taking the wrong regimen and even when they know what regimen they’re supposed
to take another half of them aren’t actually taking it, you know, there’s such
potential for benefit. So I think you know, the stakes are higher, it’s much
more complicated but I think the benefits are also higher as well at the
time of discharge. And so the question is: can we get through these logistical
problems to you know, to benefit patients more? We’re going to know more
when we finish you know, crunching the results of the
study you know, if those logistical issues were so paramount that we
couldn’t actually see a lot of benefit in the transition setting,
then I would say we should just go back to doing this in stable ambulatory
patients. But my guess is that in the real world, there’ll be patients
in both settings where we’re gonna want to initiate this
kind of an intervention. It’s not for everybody but I think for the right
patient you could do it in either setting. But you just have to be aware of
what the logistical issues are if you’re gonna do it during a transition in care.
Thank you. The second question is: do you know if
the smart box data was able to be used in the ED area to
obtain trauma or unconscious type patients med. Yeah that’s a great question.
You know I saw that in the Q&A panel. You know so picture a patient who is unconscious, you don’t know
what medications they’re on, if you could get access to this adherence report and
say: okay this is what they’ve been you know, taking and really taking because
it’s all in their pillbox and you know when they’ve been opening their pillbox. That would be really valuable information for someone to know. And you
know again, could they be taking some meds outside
the pillbox? Absolutely so it may not be 100% complete but it still would tell
you a good deal of information. I think that’s a great use case for this
and as the Internet of Things kind of explodes over time I
think we’re gonna know a lot more about patients. We have to be careful about
issues around cybersecurity of course, but you know, all kinds of use cases
where you can see the benefits of this. Someone did ask me a question about
cybersecurity and yet this is completely secure information; you can’t
hack into a smart pillbox. However obviously there are always these
concerns anytime you’re sending stuff over the airwaves.
Thank you so much and I think that’s going to conclude the question and answer period
for us. We’re going to close out the webinar from here. We have reached the
end of our time for this webinar and thank you for all attending. For those interested
in obtaining continuing education credits for participating in this
webinar, please visit the URL shown on the slide. You will select today’s webinar which will be indicated by the date and title
and then complete a brief evaluation of the event to claim your credit. You will
have 14 days to claim your credit for participating in this webinar. Please
claim your CLE by September, 27th 2018. Upon exiting today’s webinar,
AHRQ is also fielding a brief evaluation and we hope that you will be able to
complete this survey to share your feedback with us. Thank you all very much
for participating in this webinar today and we hope you will join us for future
learning opportunities.

Author:

Leave a Reply

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