Tip:
Highlight text to annotate it
X
This presentation is not a legal document. Official Medicare program legal
guidance is contained in the relevant statues, regulations, and rulings.
information in this presentation was correct as of the date it was recorded.
Nicole Cooney: Welcome to this video slideshow presentation on the Medicare Fee
For Service physician Feedback program and Quality and Resource Use Reports.
The content is
presented by Michael
Wrobeleski
, Senior Technical Advisor at the Center
for Medicare and Medicaid Services, Dr. Sheila Roman, Senior Medical Officer in
the Performance-Based Payment Policy Group, also at the Center for Medicare,
Peter Hickman, Senior Analyst in the
Plicy
and Data Analysis Group at the Center for Strategic Planning, Greg Pope,
Director of Healthcare
Financeing
and Payment at RTI International, and Jeff
Ballou
from
Mathematica
Policy Research Incorporated.
This presentation was recorded on December 21, 2011.
Operator
:
At this time I would like to welcome everyone to the Payment Standardization and
Risk Adjustment for the Medicare Physician Feedback and Value Modifier Program's
National Provider Call.
All lines will remain in a listen only mode until the question and answer
session. This call is being recorded and transcribed. If anyone has any
objections, you may disconnect at this time.
Thank you for your participation in today's call. I will now turn the call over
to Nicole Cooney. Thank you. Ma'am, you may begin.
Nicole Cooney
Thank you,
Holley
.
Hello. I'm Nicole Cooney from the Provider Communications Group here at CMS and
I'd like to welcome you to the Payment Standardization and Risk Adjustment for
the Medicare Physician Feedback and Value Modifier Program's National Provider
Call.
Subject matter experts are here today to discuss how and why per capita cost
measures are adjusted under these programs. We have question and answer sessions
included throughout today's call to allow time for you to provide input and ask
questions.
Before we get started, there are a few items that I'd like to cover. There is a
slide presentation for this session. If you registered for this call prior to
9:00
A.M. this morning
,
you should have received a direct link to the presentation in your e-mail. If
you did not receive this e-mail, please check your spam or junk mail folder for
an e-mail from the CMS National Provider Calls resource box.
If you did not receive that e-mail, you may download the presentation now from
the Physician Feedback Program section on the CMS website located at
http://www.cms.gov/physicianfeedbackprogram. That's all one word. At the left
side of the webpage, click on CMS Teleconferences and Events,
then
scroll down to the bottom of this page. Select the entry in the table with
today's date, which is 12/21. From this page, scroll to the bottom and select
Slide Presentation.
Let me repeat that. Go to http://www.cms.gov/physicianfeedbackprogram - all one
word. At the left side of the page, click on CMS Teleconferences and Events,
then
scroll down to the bottom. Select the entry in the table with today's date,
which is 12/21, and then scroll to the bottom and select Slide Presentation.
This call is being recorded and transcribed. An audio recording and written
transcript will be posted to the CMS Teleconferences and Events page on the
Physician Feedback Program website under the entry for today's call, which I
just referenced.
At this time, I'd like to introduce our speakers, who are subject matter experts
on today's topic. We are pleased to have with us Michael
Wroblewski
, Senior Technical Advisor in the Center for Medicare; Dr. Sheila Roman, Senior
Medical Officer in the Performance-Based Payment Policy Group, also in the
Center for Medicare; Peter Hickman, Senior Analyst in the Policy and Data
Analysis Group in the Center for Strategic Planning. We are also pleased to have
with us Greg Pope, Director of Healthcare Financing and Payment at RTI
International; as well as Jeff Ballou from
Mathematica
Policy Research Incorporated.
And now, it is my pleasure to turn the call over to our first speaker, Michael
Wroblewski
from the Center for Medicare at CMS. Michael?
Michael
Wroblewski
:
Thank you, Nicole, and welcome. The purpose of today's call - and, as I get
started here, I'm on slide two - the purpose of today's National Provider
C
all is to provide transparencies into the methodologies that CMS uses to adjust
per capita cost and resource use data in the Physician Feedback Program to
insure fair comparisons among physicians.
As you'll hear today, we have two basic methodologies that we've used. One, what
we'll call payment standardization, is to make sure that we eliminate any
geographic adjustments in rates that Medicare pays to physicians. The second
adjustment that we make to per capita cost is to take into account the health
status of individual beneficiaries. We'll call that risk adjustment.
The second purpose of the call today is to obtain stakeholder input, input into
these cost adjustment methodologies for use in the value modifier, and Dr. Roman
in her presentation will outline the physician value program, as well as our
next steps with the value modifier. And hopefully during the Q&A we'll be able
to discuss ways to further improve these cost adjustment processes.
This call is really the first in the series of calls and public events that we
will be
having
to obtain input into the Physician Feedback Program and as we develop proposals
for the value modifier. We'll be announcing details about these further events
after the New Year.
So in terms of today's agenda - I'm now on slide three - we'll have some opening
comments and background about the two programs by Dr. Roman. Peter Hickman will
talk about payment standardization, and that should last to about 1:30 or so
Eastern Time, at which point we'll then have Q&A.
Starting around 2:00 we'll have two presentations regarding adjusting cost data
for beneficiary health status. The first will be by Dr. Greg Pope, who'll talk
about the risk methodology that we've used in general, and then Dr. Jeff Ballou
will talk about how we've applied that risk methodology in our physician
feedback reports. Then we'll have about 20 minutes or so for questions and
answers about the risk methodologies. And then we'll have some closing remarks
and next steps.
So I will now turn it over to Dr. Sheila Roman to provide some background. Dr.
Roman?
Sheila Roman
:
Thank you. Good afternoon, everybody, and thank you for joining the call and
welcome to the call. I'm going to provide some background on the Physician
Feedback Program and the value-based modifier and I'm going to start on slide
five.
What is the Physician Feedback Program? Physician Feedback Program provides
comparative performance information on quality and cost of care to physicians.
It is just one part of Medicare's efforts to improve the quality and efficiency
of medical care. We do this by helping to provide meaningful and actionable
information to physicians so they can improve the care they furnish and by
changing physician reimbursement to reward value rather than volume. So, in
essence, the Physician Feedback Program is laying the groundwork for a physician
value-based purchasing program.
Turning to slide six, the program is mandated by legislation. The Physician
Resource Use Measurement and Reporting Program was created by the Medicare
Improvements for Patients and Providers Act of 2008 and was extended and
enhanced by the Affordable Care Act of 2010 and is now called the Physician
Feedback Program. Under this program, CMS produces annual
P
hysician
Q
uality and
R
esource
U
se
R
eports that I'll refer to as QRURs.
On slide seven you see a screenshot, which is the cover of our individual
physician report for this year. The QRURs, as I've alluded to, provide
comparative information so the physicians can view the clinical care their
patients receive and the cost of that care in relation to the average care and
cost of other physicians' Medicare patients. For example, CMS calculates total
annual per capita cost measures for patients attributed to a physician or
physician group practice.
Moving to slide eight, for this year we provided group reports in September to
35 large group practices that chose to participate in the
P
hysician
Q
uality
R
eporting
S
ystem by the group reporting option in 2010. And in early 2012
,
we will provide physician level QRURs to more than 20,000 individual
physicians who participated in Medicare Fee-for-Service in 2010 and practiced in
Iowa, Kansas, Missouri or Nebraska.
On slide nine, I want to talk now and focus on the
V
alue-
B
ased
P
ayment
M
odifier. The Affordable Care Act of 2010 requires that
,
under the physician fee schedule
,
Medicare begin using differential payment to physicians or groups of physicians
based upon the quality of care furnished compared with cost. This will be
applied in a budget-neutral fashion.
Th
is
V
alue-
B
ased
P
ayment
M
odifier will apply to services the physician bills under the physician fee
schedule. The types of information in the QRUR serve as the building blocks for
the information that will comprise
the
Value-Based Payment Modifier
. The
statute
also requires that the
S
ecretary applied the
V
alue-
B
ased
P
ayment
M
odifier to promote systems-based care. CMS will propose a methodology for the
V
alue-
B
ased
P
ayment
M
odifier in next year's physician fee schedule rule making, and we are using
outreach sessions
,
such as the one today and others planned for 2012
,
to help us develop these proposals.
I'm now going to move to slide 11, which is the timeline for the
Value-Based Payment Modifier
. And 2013 is the first year of note. The initial performance period for the
Value-Based Payment Modifier
is slated to begin in 2013, meaning that services provided during calendar year
2013 will be used in calculating the 2015
Value-Based Payment Modifier
. Beginning in 2015, the
Value-Based Payment Modifier
will be
phased
in over a two-year period.
In 2015, the Secretary of Health and Human Services had discretion to apply the
Value-Based Payment Modifier
to specific physicians and/or groups of physicians that he or she deems
appropriate. In 2016, the Secretary will continue his or her efforts to apply
the
Value-Based Payment Modifier
to specific physicians and/or groups of physicians that he or she deems
appropriate. And by 2017, the
Value-Based Payment Modifier
will apply to most
or all
,
physicians who submit claims under the Medicare physician fees schedule.
Moving back to slide 10, the Affordable Care Act requirements.
CMS is required to make adjustments to measures of costs in both the Physician
Feedback Program and for the
Value-Based Payment Modifier
, and these adjustments are different in geographic rates or payment
standardization and underlying health status of individual beneficiaries or risk
adjustment.
I'd like to now move to the discussion of payment standardization, and Peter
Hickman will move on with this discussion. Thank you, Peter.
Peter Hickman
:
Thank you, Sheila. I'm happy to be here this afternoon.
Just by way of background, our group - my group within CMS, the Policy and Data
Analysis Group, has used payment standardization for analytic purposes to
explore Medicare spending at the hospital (before the region) level. Our
standardized payments are being used by Institute of Medicine as part of their
study on geographic variation.
Our payment standardization has been used by researchers like the Dartmouth
Atlas Group,
by Congressional agencies, like
MedPAC
, for analytic purposes, and it was also used in our
QRUR
reports that was sent out by CMS in September.
Today I'm going to be talking about why standardization is necessary; second,
what it means with regard to Medicare payments; and then third, to provide some
simplified examples of kind of how it works; and then also be able to answer any
questions that you might have.
So I'm now on slide 13. If we - all we were trying to do was to look at the
utilization of single services, there would be no need to do any type of
standardization. You can compare inpatient space per thousand or emergency room
visits per thousand across geographic areas without making any type of
adjustments. But when you are trying to look kind of across services
,
or bundles of services, then using utilization measures alone becomes kind of
problematic, especially in the Medicare program.
For example, with regard to post acute care services, Medicare beneficiaries can
receive post acute care in the skilled nursing setting, where we pay on a per
diem basis; in a home health setting where we pay on an episode basis; or in the
IRF or LTC
where we pay on a stay basis; or they could be at home and
go to an outpatient department and receive therapy services there. So trying to
figure out what's happening with regards to
,
say
,
post acute care services using just utilization measures is very problematic.
There's also an issue of different practitioners might provide services, and if
you were looking just at utilization measures you wouldn't see that, but that
might be important for the program to know that a physician is providing a
service as opposed to a physician assistant or a nurse practitioner. Even in a
situation of the same practitioner who is providing the service, that service
possibly could be provided in the physician's office itself or in an outpatient
department, and that would be kind of important to know the kind of setting.
You could also have a service that's provided singly or potentially in
conjunction with other types of services, and that would potentially be
important to know from kind of an efficiency perspective. But if you're just
looking at utilization measures, you wouldn't see that.
So, turning to slide 14, to kind of address these issues with using utilization
measures and to be able to capture a broader picture of service use, healthcare
spending is often used as a proxy for service use. But this kind of raises its
own set of questions. One question is, well, how do you deal with differences in
wages and cost of living and cost of doing business practice expenses across
geographic areas? Do you check places folks are operating in high cost areas. Do
you want them to look inefficient relative to folks at low cost areas, and do
you need to do something about differences in wages, differences in practice
expenses?
You also have a question within the Medicare program, we have a number of
payment
- additions to our payments that basically serve kind of a social purpose and
aren't really related to the care being provided to that particular individual.
I'm thinking in terms of a hospital setting
,
the payments that we make to teaching hospitals, the payments that we make to
hospitals that have a disproportionate share of
low income
patients.
Also in the physician setting, we make additional payments to physicians who are
operating in
H
ealth
P
rofessional
S
hortage
A
reas. And, again, you know,
should those payments be included
?
I
f you include those payments, does that make the physician look less efficient
than he - than he actually is?
When you go to dollars
,
as opposed to using utilization measures, some problems do go away
.
S
o
,
the differences that you have with regards to, say, the different types of
setting for post acute care, when you're dealing with just the dollars that we
pay in those different settings
,
makes it easier to compare so you're no longer dealing with days of care or
episodes or stays, but you're dealing with program payments. Also, by using
dollars
,
you can capture differences and that result in a payment system from use of one
setting versus another, providing a service in a physician office as opposed to
an outpatient department or providing a service in ASC versus a - versus on OPD,
outpatient department.
You have another issue, which - a final issue, which is kind of how do you deal
with underlying health status
cushions
,
you know, difference in cost resulting from health status. Turning to 15, slide
15, so payment standardization is a process of adjusting Medicare allowed
charges in order to be able to make comparisons of service use within or across
geographic areas. It's normally separate and kind of a first step before you
think - go to risk adjustment, which deals with differences in allowed charges
due to variation in the beneficiary's health status, and Greg Pope will be
talking about that in a couple of minutes.
Turning to slide 16, what does standardization mean for physician payments? The
major thing it means is that we're taking out the impact of differences in
practice expense and differences in labor cost that are measured by the
geographic practice cost (indices). That's removed in a standardized world, and
so regardless of where you - where you are, that particular service, if it's
provided in the same setting, is priced at the same - at the same level. It also
- standardization also excludes payments that support larger Medicare program
goals, such as the add
-
on to physician payment in the
H
ealth
P
rofessional
S
hortage
A
reas or the differential that we have between participating physicians and
non-participating physicians.
Turning to 17, slide 17, in a standardized world we do maintain certain things.
So, for example, we maintain the difference in payments that results from the
choice of setting. So if a service is provided in a physician office versus an
outpatient department, there's difference in practice expense, and we want to -
we want to be able to capture that. Second, we also want to capture differences
resulting from who provides the service, so the current differentials and actual
payment for one, say, a physician assistant provides the service as opposed to a
physician, those differentials are maintained when we're talking about
standardized dollars.
Also, any type of other adjustment that an actual payment - say
,
for example
,
adjustments made if a procedure is provided in conjunction with other procedures
as opposed to being provided by itself, those type of adjustments are maintained
when we calculate standardized dollars.
Turning to slide 18, talking about how standardization affects payments in other
settings or other types of providers, I'd mentioned previously the medical
education payments, so in standardization
,
medical education payments, both indirect and direct, are removed;
disproportionate share payments are removed; the incremental payments that might
occur to
some
community hospitals and Medicare-dependent hospitals above and beyond the normal
DRG payments are remov
ed; there are certain rural add-
ons for inpatient rehab facilities, inpatient psych facilities. Those types of
adjustments are removed.
Going to slide 19, some additional differences in the standardized, in terms of
standardized dollars, to parallel what we're doing with removing the
GPCI
on the physician services, we remove the impact of the wage indices for the
provider payments. And also for certain providers and for certain non-labor
costs,
there's cost of living adjustments
, we don't include those type of adjustments in standardized dollars.
I
n some of our payment systems we have state fee schedules, so in a standardized
world, for standardized dollars, you would have a national payment rate as
opposed to the state fee schedules. Finally, in terms of the
types of adjustments, the outlier payments are included but outlier payments are
adjusted to reflect the wages of the particular provider.
So now, on slide 20, and this is a very, very simplified example of kind of how
payment works, normal payment works and standardized payment works. And the main
difference between the two is that we take out the geographic and practice
across
(the
indicia
)
effect, so that is removed from the formula. Otherwise, we try to mimic
everything else that happens in terms of the result of payment modifiers, so for
example the adjustments that are made for multiple procedures or the adjustments
that are made for co-surgery, all these various types of adjustments also occur
with regard to calculating standardized dollars for physician services.
Turning to slide 21, this is a simplified example for the payment for an office
or outpatient visit for an established patient. And I apologize, the Austin,
Texas, that should read office,
and Chicago should be facility. I apologize for the error on the slide.
But
,
here
,
we show the amount of payment for this particular service. The CMS - the allowed
charge. I'm sorry. And the Austin has $101.55. In Philadelphia it's $109.16.
Chicago, what's shown here is the payment for that service if it's provided in a
facility. So the facility practice expenses used in this instance, so the amount
of the - allowed is $81.94.
To contrast that with the standardized payment, again all we're doing here is
taking out the impact of
the GPCI
. The payment, if the service is provided in the office setting, would be
$102.27, and you see that's between the Austin actual allowed and the
Philadelphia actual allowed. And in the case of the service being provided in a
facility, it would have the same type of adjustment that occurs in terms of
normal payment, so in this case the payment would be $75.77.
Turning to slide 22, another example. In this case it's a procedure, a
destruction of a pre-malignant lesion. The payment in Philadelphia and Austin
reflects the payment for a single procedure, so this is the procedure just done
by
itself
. Philadelphia, we would pay $86.11; in Austin, Texas we
would pay $78.85. And Chicago shows the effect of
-
in terms
of
that service being provided with - or that procedure being performed with other
procedures where it's not the most expensive procedure. So it's discounted in
terms of our payment
,
so the payment for that destruction of a pre-malignant lesion as a second
procedure in Chicago is $43.13.
In terms of the standardized payments for
this
CPT
code, it's performed as a single procedure. It would be $79.50. Again, you see
that $79.50 is between the Austin payment of $78.85 and $86.11. And in the case
of this procedure being done with other procedures, where it's basically kind of
a second procedure, the same discounting that occurs in terms of generating
actual payments is mimicked with what regard it happens with standardized
payment, so the payment would be $39.75.
So in a standardized world, every time a physician provides this particular
code, that $79.50 would be what's credited to him or her as opposed to what the
actual amount would be, and that's kind of an equal amount across the whole
country.
Turning to slide 23, here's the example in a hospital setting. And again our
formula is a little bit more complicated here because potentially the hospital
can receive payment for
I
ME and DSH
and outlier payments. But again you see the main difference or a couple of
differences in the formula between the Medicare allowed amount and the
standardized allowed amount, first of all we're removing the impact of the wage
index. Second, we're removing
IME and DSH
payments, if they're present. And third, we're adjusting the outlier payment if
there is an outlier payment.
If you turn to slide 24, you see a numeric example. In this instance we have
three hospitals - Hospital A, located in Philadelphia; Hospital B, located in
Austin, Texas; Hospital C, located in Chicago. And this is for DRG 194, which is
simple pneumonia, and you see that the actual payments range from $5,732.25 in
Austin, so the Hospital B in Austin, to $9,199.75 to Hospital A in Philadelphia.
If you look at the columns
,
you see that one of the main reasons that the payment to Hospital A is so high
is that it has the - receives the
I
ndirect
M
edical
E
ducation payments because it's a teaching hospital. In this example
,
both Hospital B and Hospital C are not teaching hospitals, but they still
receive
D
isproportionate
S
hare of payments.
So
if you look at the bottom
,
you see what the standardized payment would be, so we remove the impact of the
wage adjustment from the operating and the capital payment. There's no payment
for IME or
DSH
, so you get a payment that's $56.69, and that would be uniform across the whole
country.
So
,
hopefully that gives you a sense of kind of why standardization is necessary,
and how we're thinking about implementing standardization in some examples. And,
with that, I'll turn it over to Nicole. Thank you very much.
Nicole Cooney
Thank you, Peter.
At this time, we would pause for just a few minutes to complete
keypad
polling so that CMS has an accurate count of the number of participants on the
line with us today. Please note
,
there may be moments of silence while we tabulate the results.
Holley
, we're ready to start the polling.
Operator
:
CMS greatly appreciates that many of you minimize the government's
teleconference expense by listening to these calls together in your office,
using only one line. Today we would like to obtain an estimate of the number of
participants in attendance to better document how many members of the provider
community are receiving this valuable information.
At this time, please use your telephone keypad and enter the number of
participants that are currently listening in. If you are the only person in the
room, enter one. If there are between two and eight of you listening in, enter
the corresponding number between two and eight. If there are nine or more of you
in the room, enter nine.
Please hold while we complete the polling. Please continue to hold while we
complete the polling. Once again, please continue to hold while we complete the
polling.
Thank you for your participation. We will now move into the Q&A session for this
call.
To ask a question, press star followed by the number one on your touchtone
phone. To remove yourself from the queue, please press the pound key. Please
state your name and organization prior to asking a question and pick up your
handset before asking your question to assure clarity.
Please note, your line will be - will remain open during the time you are asking
your question, so anything you say or any background noise will be heard in the
conference.
Nicole Cooney
And
Holley
, let me just take this time to remind everyone that this call is being recorded
and transcribed, and during this first Q&A session we'd like to hear input and
questions on the presentations we've heard so far. We do have another Q&A
session planned to hear questions and input on the second half of our agenda.
And
as
Holley
mentioned, before asking your question, please state your name and the name of
your organization. And in an effort to get to as many of your questions and
comments as possible, we ask that you limit your question or comment to just
one.
Holley
, we're ready for our first question.
Operator
:
Your first question comes from the line of William Rich.
William Rich
:
Hi. This is Bill Rich from the American Academy of Ophthalmology.
A question for Peter Hickman.
I have a little - you know, I understand standardization of cost. I'm deeply
concerned, though, about your assumption that the employed facility physician
should only be accredited - credited the work and limited practice
expense and not the 44 percent kicker that's actually part of the facility fee.
It seems to be inconsistent with your policy of looking at costs that are
associated with
choice of
facility, and this physician has chosen to work for a facility.
I appreciate your comment.
Michael
Wroblewski
:
Peter, would you like to go ahead and take - this is Michael
Wroblewski
. Peter, would you like to go ahead and take that question?
Peter Hickman
:
Sure. I mean, basically what we're trying to do is - as best as possible to
replicate what happened with regards to Medicare payments, but just to exclude
the geographic adjustors. So we're not introducing anything that doesn't happen
already within the payment systems.
William Rich
:
So indeed those individuals are paid 44 percent more than the office-based
facilities. So, again, I think that's the only assumption I think that I have
any concerns with, Peter.
Peter Hickman
Well, thank you for sharing your opinion.
Operator
:
Once again, in order to ask a question please press
*1
on your telephone keypad.
Your next question comes from the line of Douglas Carr.
Douglas Carr
:
Yes. This is Douglas Carr from Billings Clinic in Billings, Montana for Sheila
Roman. I'd like to have her expand on the quality of resource use reports that
feeds into developing the
Value-Based Payment Modifier
. I assumed that - I know this is one question, but it's all - it's all one. You
are giving t
w
o reports to
I assume
a single
TIN
, a physician group tax identification number and how does - how are the
expenses or per capita beneficiary expenditures attributed to a particular group
or to a particular physician?
Sheila Roman
:
Yes. I think you're talking about our group reports, and for our group reports,
I'll just mention that for quality, in order to participate in the
P
hysician
Q
uality
R
eporting
S
ystem
,
there were 26 common measures that all of the 35 groups reported to CMS. So we
had essentially a core measure set for quality for comparison.
For cost, we had to have two E&M visits with the - with the group in order to be
attributed to that group.
Or
a plurality of visits.
So
either two visits or a plurality of visits.
Nicole Cooney
OK.
Holley
, can we take the next question?
Operator
:
Yes, ma'am. Your next question comes from the line of Sarah Ton.
Sarah Ton
:
Yes. Hello. I'm a staff here at American Academy of Neurology, and I'm asking a
couple of things.
On slide 10, I saw that you mentioned underlying health status of individual
beneficiaries having to do with risk adjustment, and then on slide 15 again you
stated - let's see here - the payment standardization is separate from risk
adjustment, but then you say that the allowable charges due to variation and
health status.
My question is that it would help a lot if we had common language behind what
you're defining as health status
.
So
o
my understanding when health status is used, it's about quality of life, so how
the patient rep
o
rts their physical function, their mental function and their societal function,
as well as some other variables. And yet when I read risk adjustment I'm
thinking more of severity measures. So for example in stroke, adjusting for
(NIEGSF),
which is more of a clinical outcome severity adjustment rather than a health
status. So it would benefit all of us if we all understand what is being
captured on health status, or is that the way you intended to use the term
health status in terms of quality of life issues or getting into clinical
outcomes, which are adjusted by severity of illness?
Sheila Roman
:
Yes. I think your point is well made, and I think that will become clearer in
the discussion by Dr. Gregory Pope specifically on risk adjustment. In general I
would agree with you that health status refers to quality of life and I
think it will become clear that when we speak about risk adjustment we're
talking more about more of comorbidities that patients may have.
Sarah Ton
OK. Thank you. That helps.
Operator
:
At this time, there are no further questions.
Nicole Cooney
:
OK. Thank you to those of you who asked questions during our first Q&A session.
And, as a reminder, we do have another session planned to take place after the
next two presentations.
Now, I'll turn the call over to our next presenter, Mr. Greg Pope from RTI
International. Greg?
Gregory Pope
:
Thank you. This is Gregory Pope. So I'm starting on slide 25 and I'll be talking
about the CMS
H
ierarchical
C
ondition
C
ategories or CMS
HC
C risk adjustment model. I'm with RTI International, who's a contractor to CMS,
who helped develop and maintain this model, and we're a non-profit research
institution with our main offices in North Carolina's Research Triangle Park.
All right, so moving on to slide 26, an overview of the presentation. I'll start
out with why does CMS risk adjust
based
on beneficiary health status, talking about the principles of risk adjustment,
give an overview of the model and talk about included conditions and model
inputs, model structure and calibration, model performance and an example of
calculating a risk
score
with the model.
So
,
on slide 27, why does CMS risk adjust cost
?
R
isk adjustment
as a method of adjusting per capita
cost which may adjust them either up or down to account for differences in
expected costs of individuals
based on their health status. For the Physician Feedback and
V
alue
M
odifier
P
rograms, their purpose is to promote fair per capita cost comparisons by
adjusting for the health status of different beneficiary, attributed beneficiary
populations.
Risk adjustment's intended to accurately predict risk over aggregates of
individuals and not just specifically predict the cost of a particular
individual. We turn to a point later with a larger aggregations of beneficiaries
below average cost will balance out above average cost, so risk adjustment is
intended to predict the systematic portion of cost but some of the unpredictable
portions, when you aggregate over a number of beneficiaries, the above average
and below average cost, unpredictable cost will tend to average out the
systematic portion, which is what you're attempting to adjust for.
So
,
on slide 28, principles of risk adjustment. Diagnostic category should be
clinically
meaningful
and should be able to predict cost in creating an individual's clinical profile
. H
ierarchies are used to characterize the individual's illness within each disease
process as the effects of unrelated disease processes accumulate. And I'll come
back to explain in more detail later
. D
iagnostic classifications should encourage specific coding of diagnoses and not
reward coding proliferation at the same or similar codes. And providers should
not be penalized for recording additional diagnoses.
So
,
slide 29, an overview of the CMS-HCC model. The model uses beneficiary
demographic characteristics and prior year diagnoses to predict relative Part A
and Part B Medicare Fee-for-Service program payments. The CMS-HCC model does not
incorporate Medicare Part D cost
s,
which are predicted separately by the
CMS
RXHCC
model. The CMS-HCC model is prospective, meaning it uses prior year information
to predict cost and focuses on specific conditions that are important in
predicting Medicare expenditures.
Slide 30, what is the CMS-HCC model currently used for? Major use
is
is
Medicare Advantage capitation payments. It was implemented for this purpose in
2004 and
fully phased in
by 2007 for 100 percent risk adjusted payments. It's also in the final rule,
which was recently released for the
S
hared
S
avings
A
ccountable
C
are
O
rganization program to be implemented in 2012.
There are some twists as to how risk adjustment is to be used in that program.
It's not used in the same way as in the Medicare Advantage program
but it is the HCC risk
scores
are used or proposed to be used
in that
program.
And then the Medicare
P
hysician
Q
uality and
R
esource
U
se
R
eports the potential here implemented in 2009 and adjust comparisons of per
capita costs for patient health status.
So
I'm on to slide 31,
a
word about the development and maintenance of the model. The CMS-HCC model was
originally developed under contract to CMS by research
ers
at Boston University and the Research Triangle Institute, with clinical input
from Harvard Medi
c
al School physicians. The model is currently maintained by RTI under contract to
CMS, with clinical input from Harvard Medical School and other consultants to
RTI. The model is updated every year to incorporate new diagnosis code
s
and is recalibrated regularly and more recent diagnoses and expenditure data.
So
,
slide 32, in terms of what medical conditions are included in the model, the
full CMS-HCC model classifies all conditions but
,
not all conditions are used in payment and other applications of the model. The
CMS-HCC payment model includes clinically significant
,
generally
high
cost medical conditions such as cancer, heart disease, hip fractures that are -
some conditions are excluded from the payment
model
. These are those that do not predict a future cost.
For instance, appendicitis, certainly there are costs associated with that in
the year it occurs, but it's not predictive of future costs. And
,
then
,
conditions where there's a high degree of discretion or variability in
diagnosis, diagnosis coding or treatment. Symptoms codes would be an example of
this, you know, vague diagnoses would not be included.
So slide 33, model variants of the CMS-HCC model. There are actually a number of
different variants for different Medicare subpopulations or situations. For
beneficiaries entitled to Medicare by age or disability, so this would be both,
you know, elderly beneficiaries and the under 65 beneficiaries entitled by
disability.
There is a community continuing enroll
ee
model.
This would be for beneficiaries residing at the community and
wh
o have been in Medicare at least 12 months, so they have 12 months of claims
experience.
There's an institutional continuing enrollee model and then there's a new
Medicare enrollee model for beneficiaries with less than 12 months of claims
experience because that 12 months of claims experience is necessary to establish
a sufficient diagnostic profile to predict their costs. Then there are
also several variant models for
E
nd
S
tage
R
enal
D
isease beneficiaries. For the
P
hysician
Q
uality and
R
esource
U
se
R
eports, those will be using - they are using the aged, disabled community, the
new enrollee and the ESRD models.
So
slide 34, model inputs. So
,
we'll start the risk adjustors and demographic factors. So what are the risk
adjustors used by the model? They can be divided into two main classes. One is
demographic or Medicare enrollment information, insurance enrollment information
for Medicare and diagnoses. Those are part of the two main categories.
In terms of demographic factors, the model uses 24
age-sex cells
,
so an example would be a middle aged, (80 to 84). The model also uses Medicaid
dual eligible status, so if they are, you know, dual
eligible
s
for Medicaid as well as Medicare programs, so these are generally, you know,
poor beneficiaries who qualify for Medicaid as well as Medicare, so there's a
adjustment
for
the higher costs associated with these beneficiaries, which is differentiated by
sex and age versus disabled entitlement status.
Then
for disabled status for beneficiaries entitled by disability,
there's
sort of two related adjustments. There's an adjustment for beneficiaries
currently entitled by disability. So it would be the under age 65 Medicare
beneficiaries and they're separated by sex and Medicaid factors
for them
as well
S
elected diagnoses have different risk weights to affect - to reflect the
different impact of certain diagnoses on the cause for the disabled as opposed
to the aged.
Then, the second category of disabled status adjustment is for beneficiaries who
are currently entitled by age but who are originally entitled to Medicare by
disability. So
,
these are beneficiaries aged 65 or older but who originally joined the Medicare
program when they were under age 65 because of a disability. And
,
there's an adjustment for the higher cost of these beneficiaries as they age and
become older and become entitled to Medicare by age when they're older than 65.
But
,
there is still higher cost and there's a factor in the model to reflect that,
and that's differentiated by sex.
So
,
slide 35, so diagnoses. That's the other major risk adjustor used in the model
for model calibration and
F
ee-
F
or-
S
ervice applications. The diagnoses are obtained from
F
ee-
F
or-
S
ervice Medicare provider claims or bill
s
submitted to Medicare that use the
I
nternational
C
lassification of
D
iseases
V
ersion
9
C
linical
M
odification diagnosis codes and the codes will be transitioning to ICD 10 in
fiscal year 2014, so October 2013.
Diagnoses are used only from specified settings and provider. Not all diagnoses
on the Medicare claims, so the diagnoses are used from the hospital inpatient
setting, both the principal and patient diagnosis and secondary from a hospital
outpatient, from physician claims and from certain clinically trained
non-physician providers
such as clinical psychologists.
Moving to slide 36, continuing to talk about the diagnoses used
in
the model diagnoses from laboratory, radiology, home health, skilled nursing
facility and other settings are not used in the model. The number of times a
diagnosis is recorded does not affect the calculated
risk
score from the model, and the setting from which a diagnosis is reported does
not matter. So, for example, inpatient diagnoses are not weighted more heavily
than outpatient diagnoses.
Slide 37, so how does
the
model capture severity of illness? The CMS-HCC model
counts
in the most severe manifestation among related conditions and the principles
implemented through what is called disease hierarchies. So related diseases are
organized into hierarchy of severity from, you know, the most severe at the top
to the less severe at the bottom.
And they - the beneficiary - the model
counts
the most severe manifestation of a related disease for even beneficiary, for
example, if a beneficiary happened to be coded both with diabetes with
complications and diabetes without complications, only the former will be
counted. I will give some more examples in a minute.
So
,
in slide 38, how does the model handle multiple diagnoses which a Medicare
beneficiary has? The basic principle is that the model is added to across
disease hierarchies. So
,
unrelated conditions are counted cumulatively. So
,
related conditions are counted within hierarchies.
For instance, a model has a cancer hierarchy, hierarchy for heart disease, lung
disease, cerebrovascular disease, but total disease burden is measured by the
severity and disease hierarchy for related conditions.
And the cumulative
or
additive
burden
of multiple conditions across unrelated conditions.
The model also allows for disease interactions or interactive effects among
multiple conditions. For example, congestive heart failure and
C
hronic
O
bstructive
P
ulmonary
D
isease having an interactive effect beyond
their
separate
additive
effects
,
which is recognized in the model.
So slide 39, this is a graphic of the CMS-HCC model structure
.
S
pecific counts here are re
lated
to Version 12 of the model. There are a number of versions and models have been
revised over time and there's
,
really
this specific Version 12
,
which is currently being used for Medicare Managed Payment.
So
,
the starting point for the model is the 14,000 or more ICD-9 codes are coded on
provider bills submitted to Medicare. And
,
then
,
the model groups it into something we call diagnostic groups or DXGs and they're
about 800 of these in Version 12 of the model
.
Then,
there's further stage of aggregation to condition categories and there are 189
of these.
And
at that point, hierarchies are
imposed
among the related diagnoses or diagnostic categories to convert the condition
categories into Hierarchical Condition Categories or HCC
-
still 189 of those. Then
,
a subset of those 1
89
diagnostic categories are selected for the payment version of the model, which
70 conditions excluding diagnoses or codes that are thought to be, you know,
either empirically are not predictive of
significant
future cost or judged to be subject to a high degree of discretion or
variability in diagnosis or diagnostic coding or treatment.
So
,
slide 40 shows an example of the coronary artery disease hierarchy in Version
12. The top of the hierarchy
,
CC-81
,
is acute myocardial infarction. The next CC hierarchy is 8
2
unstable angina and other acute ischemic heart disease and CC-83 is angina
pectoris or myocardial infarction. And
,
then
at
the bottom of the hierarchy is CC-84 is coronary atherosclerosis, other chronic
ischemic heart disease.
So
,
if a beneficiary is coded, for example, with acute myocardial infarction, CC-81,
or I should say CC-81 is a different hierarchy here, then
,
they would be excluded from the other HCCs as would not be accounted in the
model.
So
,
slide 41 gives a clinical
vignette
to show the process of how we move from ICD-9 codes to DXGs to CCs and HCCs. So
you can see on the left side of the slide, some hypothetical ICD-9 codes that
the beneficiary might be coded with are listed. And you can see how
410.91
AMI of unspecified initial episode of care, that goes into the DXG AMI initial
episode of care. And that DXGs then mapped into the CC acute myocardial
infarction and that becomes the similar
ly
label
ed
or the same label HCC.
This beneficiary also had ICD-9 for 13.9
,
which is unspecified angina pectoris. That goes into the DXG for angina and CC,
but then that's excluded by the hierarchy because that was in the coronary
artery disease hierarchy. And since HCC 81 is abo
ve
HCC 83, only HCC 81 is counted and HCC 83 is excluded.
There's also some one diagnosis we showed to along ICD-
9
codes that are grouped into DXG for emphysema and that goes into the CC for
chronic obstructive pulmonary disease. And the HCC for chronic obstructive
pulmonary disease so then because that's in the l
u
ng hierarchies supposed in the heart hierarchy, the HCC 81 for AMI and HCC 108
for COPD would be counted additively because those are two separate hierarchies.
Then at the bottom of the slide, there are some codes for symptoms and some
relatively minor acute conditions and those conditions are classified by the
model, but they're not included in the payment versions of the model.
So if we move to slide 42, which discusses model calibration, the model is
calibrated on 100 percent Medicare Fee
-
for
-
Service data of 25 to 30 million beneficiaries. Two years of data are used in
the calibration. For example, 2009 and 2010, the first year, base year 2009 in
this example is used to accumulate a diagnostic profile from the ICD-9 diagnosis
codes. And then
that diagnostic profile
together with the demographic enrollment information
,
is used to predict Medicare payments for the second or prediction year which is
2010 in this example.
So what is predicted by the model is Medicare Program Payment, so it does
exclude the beneficiary cost sharing deductibles and co-insurance paid by the
beneficiary or supplemental insurance.
There are adjustments in the second year for partial eligibility. For example,
due to beneficiary death, but a full 12 months of year
one eligibility
is required to develop the diagnostic profile. And if the beneficiary does not
have full 12 months in year one because
,
for example
,
they're a new Medicare enrollee, they would be then allocated to the new
enrollee model that I mentioned earlier that uses demographic and enrol
l
ment data only which is available for them and they would be predicted by that
new enrollee model.
Then finally in this slide, multiple
regression
which is a statistical technique is
used to estimate the incremental cost impact of each demographic factor and
diagnostic category.
OK. So moving on to slide 43, I have a couple of slides on model performance.
The first slide is the so-called R-squared statistic
,
which is a measure of the percentage of variation in individual beneficiary
expenditures explained by the model. This is approximately 12 percent for the
CMS-HCC model which may seem relatively low, but it's important to keep in mind
that you're using prior year information to predict the following year, and much
of health expenditure variation like in a year ahead is acute or random from the
point of view of the model and is not predictable with prior year information.
But, for example, as a comparison, the R-square from a demographics only model
is about one percent, so the diagnosis base model is predicting quite a bit more
variations than the demographic model.
So if we go to slide 44, we see model performance for quintiles of predicted
expenditures and you can see that, you know, one would be the lowest quintile
and five the highest. And you can see the mean actual dollars of expenditures
ranging from 2,800 to over 19,000 and that the mean predicted dollars by the
model ranging from
about
2,500 to 19,500. And the final column of the table is a ratio of the predicted
to the actual.
So this is showing that the model when beneficiary is, you know, predicted to be
expensive or less expensive than average that on average they are in fact
predicted accurately so the model is well calibrated in the sense that when it
predicts someone to be expensive
,
on average, they will be expensive. So it's differentiating
,
sort of systematically
less expensive beneficiaries from systematically more expensive beneficiaries.
Slide 45, this is like in a model performance for predicting the average cost
associated with certain specific diagnosis. And here you can see the prediction
is actually perfect which is cheating a little bit because these ratios were
calculated on the calibration sample. If we took an independent sample it
wouldn't be quite so good. But still they'll be closed to one.
So this is showing that the model is predicting the higher average cost
associated with, you know, serious, chronic illnesses that are but important to
cost drivers. There are drivers of differences and costs among Medicare
beneficiaries.
So slide 46 is an example of a risk score calculation for specific beneficiary
for a 20 - should say 82-year-old male with prior year diagnosis of acute
myocardial infarction and
C
hronic
O
bstructive
P
ulmonary
D
isease. So the model consists of, you know, first allocating this person to
their demographic category which is male aged 80 to 84 and there's an
incremental predict
ed
cost of $4,660 associated with that age and sex group.
Then their incremental cost associated with the specific diagnoses of the acute
myocardial infarction,
H
CC-81 has an incremental cost coefficient of $2,428,
C
hronic
O
bstructive
P
ulmonary
D
isease HCC-108 is $3,129. So if you sum th
o
se three factors, you get the total predicted cost of $10,227.
The population mean cost in this example across, you know, all Medicare
beneficiaries is about $9,000. And so the risk score
,
which is the predicted cost by the mean cost
,
is
1
.130. So that risk score is interpreted to mean that
the beneficiary is predicted to
be 13
percent more expensive than the average cost Medicare beneficiary.
So that's the end of my presentation, so I'm now going to return it over to Jeff
Ballou of
Mathematica
Policy Research for the next presentation on applying risk adjustment in the
Physician Resource
Use
Reports. Thank you.
Jeffrey Ballou
Thank you, Greg, and good afternoon.
Before I get started, I would like to acknowledge the efforts of my colleague
here at
Mathematica
, senior researcher Eric
Schone
, who has led our risk adjustment modeling since we have been producing the
QRURs.
I would like to pick up where Greg left off in his presentation on slide 48. The
transition here is that we use the HCC risks scores, in particular the new
enrollee and community risks scores, as primary input into our own risk
adjustment modeling where the end result is reporting a payment standardized and
risk adjusted per capita cost statistic.
Actually
a series of
such
statistics for each physician receiving a QRUR.
So
,
the risks scores are obviously an important part of what we do. We actually do a
little additional modeling on top of that because our populations are slightly
different and the populations RTI looks at and we're interested in making sure
that the comparisons that we ultimately will make on resource
s
across physicians are as fair as possible.
So I'd like to begin by giving you a brief overview on this slide 48 here and
then talking in slightly more detail in coming slides before concluding with a
numerical example.
We start with beneficiary level
data and we need to end up with
physician level statistics. So how do we that? Well, we start with claims data
aggregate to the beneficiary level and price standardized from across total
,
basically beneficiary cost in total across Part A and Part B. And we want to
estimate the relationship between beneficiary risks scores as described
previously in these beneficiary costs.
We then want to take that estimated relationship and use it in the next step to
compute expected cost for each beneficiary, and a little bit more on that later.
We then take from each physician all of the beneficiaries we've attributed by a
separate algorithm to that physician and add up their observed costs and then
divide that sum of observed costs by the sum of expected costs that have been
calculated for that same set of beneficiaries.
And this gives us an observed to expected ratio for each physician. We then take
that ratio and translate it into a dollar format and then we use those dollar
denominated risk adjustment costs to make comparisons among
peers
.
On slide 49, I want to discuss a little bit about what gets risk adjusted. Our
focus here is on adjusting per capita, per patient, per beneficiary cost
measures for physicians to account for each beneficiary's expected costs given
the beneficiary's health status.
And mindful of the comment that was made in the earlier Q&A session, if I
reference health status in my
presentation;
I have a very specific definition in mind, a very narrow definition which is
essentially the combination of the beneficiary's risk score and whether or not
the beneficiary has ESRD, those two pieces of information taken together.
We have essentially one risk adjustment model that we apply, but we apply it to
five populations. We risk adjust all beneficiaries in our sample and then we
separately risk adjust beneficiaries with coronary artery disease, those with
COPD, those with diabetes, and those with heart failure.
And the reason for this is that we report for a given physician receiving a
QRUR, not only a risk-adjusted per capita cost number for all of their
attributed beneficiaries, but also
,
for example
,
a risk-adjusted per capita cost number for all of that physician's attributed
beneficiaries who have diabetes. And for that reason we want to make sure we
risk adjust these populations separately.
On the next slide, slide 50, who gets risk adjusted? There are some
beneficiaries who
,
in spite of having been treated by a physician
,
do not end up entering into the data that the physician received on their QRUR
that is
they're excluded. And they may be excluded for multiple reasons. But for example
part-
year beneficiaries are not eligible for attributions in 2010, which is the data
year that we're using for these reports. A part-year beneficiary is a
beneficiary who does not have a full 12 months of enro
l
lment in Medicare fee-for-service Part A and B. And there are some other reasons
why beneficiaries might be excluded.
In addition, beneficiaries with total Medicare costs in the bottom one percent
of the unadjusted cost distribution - so those with very low total standardized
Part A and Part B costs are dropped prior to doing any risk adjustment. All
remaining beneficiaries, assuming that they have a 2009 new enrollee or
community risk score
,
are included in our risk adjustment model. And a rare case where both, both
scores are provided, we default to using the new enrollee score in our model.
So on Slide 51, the question is, well, how do we estimate this relationship
between risk and cost that I referred to at the outset, and then what do we do
with it? We start by treating the data for outliers. So
,
as I've indicated on the previous slide, the very lowest cost beneficiaries are
dropped from the model. The highest cost beneficiaries are not dropped from the
model, but we are concerned about disadvantaging any physician who, by virtue of
chance, happens to be treating a beneficiary who has very high costs.
And for that reason, we take any beneficiary whose unadjusted costs are above
the 99th percentile and reset that cost to the 99th percentile. So, for example,
if one of my patients costs $150,000 on an unadjusted basis and the 99th
percentile value is $106,000, for all purposes of reporting and risk adjustment,
that beneficiary is going to look like they actually cost the lower number,
$106,000.
After we've addressed outliers in the data - again, this is for purposes of
fairness, but also to improve our subsequent model fit - we then take these 2010
costs and we use a multiple regression model - and Greg had referred to these in
his earlier presentation - and we seek to explain them based on the 2009 value
of those beneficiaries' HCC risk scores. We anticipated and find a positive
relationship - again, higher risk is consistent with higher cost, on
average. We also look at whether the beneficiary has
E
nd-
S
tage
R
enal
D
isease. Again, there is a strong and significant association between the
presence of ESRD in 2009 and costs in 2010.
And then finally, to further improve model fit, we also include the squared
value of the risk score - either the community risk score or the new enrollee
risk score, whichever happens to be applicable to a given beneficiary cost. And
so this step two here estimates the relationship between risk and cost. What is
the output of that? Well, the output is essentially a series of multipliers from
this regression model that we use to compute the expected costs from risk
factors for each beneficiary in our sample.
And so what we mean by expected costs here are, what are our best guess of what
a given beneficiary would cost if we didn't know their actual cost? Now what the
model allows us to do is to put in as that best guess an estimate of the average
cost of all beneficiaries in our sample who look like the one whose expected
cost we're considering. That is, we look at the average of all beneficiaries of
the sample who have similar risk scores and the presence or absence of ESRD.
On Slide 52, I want to tell you now how we go from risk-adjusted
beneficiary-level costs - which is
where we are at the moment - to a summary statistic that's at the physician
level. And again, as I indicated in the overview, this is going to be based on a
ratio of observed to expected costs where observed costs for a given physician
are going to be the sum of actual - that is,
pre
standardized
Part A and Part B costs for the physician's attributed beneficiaries - divided
by the sum of expected costs that have been computed for that same set of
attributed beneficiaries.
To convert the observed to
expected
ratio into dollars, we then multiply by the mean cost of all beneficiaries in
our sample, which is on the order of $11,000 and change. And then finally, it's
- I guess this is sort of the summary of risk adjustment - it's not physicians
then with low observed costs relative to their peers who look good. But after
risk adjustment, really the relevant comparison is physicians with low-observed
to expected ratio of costs by the ones who do best in peer comparisons for
resource use.
That's a summary of how we proceed, starting with the risk scores that RTI - you
know, that the CMS
HCC model outputs, and with further refinement. On Slide 53 I want to take you
back through that same process, only this time using numbers to
,
hopefully
,
help clarify and, and crystalize the ideas. And
,
so
,
on Slide 53 we have the beneficiary-level data that we start out with for four
hypothetical beneficiaries, four among many in our data. Again, the sample mean
for all beneficiaries in our data of observed cost - again, that's total Part A
plus Part B on a payment-standardized basis is $11,379.
So we have an observed cost for each beneficiary, and then we have a risk score
for each beneficiary - either the community risk score or the new enrollee risk
score. And then finally, we have information on whether each beneficiary has
E
nd-
S
tage
R
enal
D
isease or not. Now that we've got observed cost and
some risk information - risk and
ESRD - what we want to do next is figure out the expected cost of each
beneficiary based on the previously estimated relationship between risk and
cost.
So I want to show you how we'll do that for Arthur. So we're going to set aside
Arthur's observed cost - that's not going to enter into the next calculation -
and look at his community risk score and lack of
E
nd-
S
tage
R
enal
D
isease on the next slide, Slide 54, and ask how that information is used to
output an expected cost, the cost we would guess would prevail for beneficiaries
who are like Arthur in terms of having a similar community risk score and no
ESRD.
So on this slide, what you see on the left-hand side, the various rows of this
table are the different variables that are in our risk adjustment model.
The value
column are
Arthur's data. A constant applies to every beneficiary. But the other numbers
may vary from beneficiary to beneficiary. The third column is a multiplier
column. This is the output of our estimated relationship between risk and cost.
And these multipliers are going to be applied to the values of each beneficiary.
When you multiply values and the multiplier, what you end up with in the
right-most column is each variable's contribution to expected cost. So for
example, in Arthur's case, the community risks
score
1.739 when multiplied by 8,681 is going to result in a community risk score
contributing to expected cost of $15,096, and a constant in the square value of
the risk score also contribute to Arthur's expected cost. (Arthur's expected
cost can be determined then by taking all of the line, the line numbers or the
numbers in the right-most column and simply adding them up. So if we didn't know
Arthur's cost but we did know his risk score and we do know that he doesn't have
E
nd-
S
tage
R
enal
D
isease, our best guess, based on our model, is that Arthur would cost $16,768.
I'm going to take that number. I'm going to transport it to the next slide which
is the final slide, 55, and tell you that we can do the same thing, of course,
for Betty and Carol and David. And so now, in this upper panel, we have an
observed, total standardized cost for each beneficiary, and then we have an
expected cost for each beneficiary. And the final thing we have in this table on
top is the physician to whom each beneficiary is attributed by a separate
algorithm. And so, for the sake of this example, let's suppose that Dr. Smith
has two and only two patients - Arthur and Betty - who are assigned to her. And
likewise, Dr. Jones is assigned two patients - Carol and David.
Moving from the upper panel to the lower panel of this table - excuse me, of
this slide - the lower panel actually puts together the final risk-adjusted cost
statistics. So now we're down to the physician level. Dr. Smith's total observed
costs are simply going to be the sum of the observed costs of her individual
patients, Arthur and Betty, a total of $32,525. And her total expected costs are
going to be similarly computed as the sum of Arthur's expected cost and Betty's
expected cost, or $79,325.
We then take the ratio of those two numbers to get an observed to expected ratio
of .41. And for purposes of presentation in the report, we take that observed to
expected ratio and multiply it by the mean beneficiary cost for all
beneficiaries in our sample - again, that's $11,379. And so the conclusion of
this entire process is that when Dr. Smith receives her report, her overall
payment-standardized, risk-adjusted, per capita cost score for all of her
beneficiaries assigned to her will be $4,665. Now we can do the same thing
with Dr. Jones, and following the same procedure, we'll end up with a value for
Dr. Jones of somewhat over $13,000 - $13,313. And that's the process of risk
adjustment.
I've been going through this perhaps quickly and certainly very mechanically, so
I want to step back in closing and indicate that if you look at the total
observed cost column, the bottom panel, the second column from the left on this
last slide, you see that if we were to compare Drs. Smith and Jones simply on
the basis of the costs of their beneficiaries, without any adjustment other than
payment standardization, they would look quite similar. Again, though, as I
argued in the previous slide and has been explained earlier, that's not really
an apples
to apples
comparison.
By risk adjusting, we see that Dr. Smith ends up doing much better than Dr.
Jones because even though Dr. Smith had the most expensive patient in Betty in
terms of observed cost, Dr. Smith's observed costs were well below the numbers
that would have been guessed for either Arthur or Betty if we didn't actually
know their observed costs. So if Drs. Smith and Jones were graded solely on
their resource use, based on these statistics in the bottom right, right-hand
corner of the table, Dr. Smith would look better than Dr. Jones.
That is a summary of how we take the risk scores that the CMS
HCC provides to us, and ultimately use those along with beneficiary data to
arrive at per capita cost numbers that are risk adjusted and provided to
physicians receiving QRURs. I would now like to turn the presentation back over
to Nicole.
Nicole Cooney
:
Thank you, Jeff. At this point in time, we're ready to begin our final Q&A
session. Again, as a reminder, this call is being recorded and transcribed.
Before asking your question, please state your name and the name of your
organization. In an effort to get to as many of your questions as possible, we
ask that you limit your question or comment to just one.
Holley
, we're ready to start questions.
Operator
:
Thank you. We will now open the lines for our question and answer session. To
ask a question, press star, followed by the number one on your touchtone phone.
To remove yourself from the queue, please press the pound key.
Please state your name and organization prior to asking a question, and pick up
your handset before asking your question to assure clarity. Please note your
line will remain open during the time you are asking your question. So anything
you say
or any background noise
,
will be heard. Your first question comes from the line of Tracey Glenn.
Gus
Geraci
Hi, this is Dr. Gus
Geraci
from the Pennsylvania Medical Society. I'm sitting here with Tracey. I applaud
your balancing of cost and taking into account severity of the patient's
illness. But I'm not seeing how quality is taken into account, other than the -
other than cost. How, how do you - you know, you say that because expected costs
are lower, Dr. Smith did better than Dr. Jones. But what if Dr. Smith is
deliberately
under treating
patients?
Michael
Wroblewski
This is Michael
Wroblewski
. Thanks for that question. You know, the focus of today's presentation has been
on how we're adjusting costs. In future events, we'll be talking about the
additional quality measures and the attribution methodology, the models for the
value modifier. And so that would be something that would be the subject of, of
future presentations. But we are mindful of that concern. Thank you.
Dr. Gus
Geraci
Thank you.
Operator
:
The next question comes from the line of Donna Kinney.
Donna Kinney
:
This is Donna Kinney with Texas Medical Association. When I wrote comments in
the, in the fee schedule rule this year, I was questioning the issue, issues
about the risk adjustment methodology, because as we know, there are demographic
factors that heavily affect both cost and physician
compli
- and patient compliance. And I - the final rule responded to that by saying
that the HCC methodology was being modified or had been improved or expanded or
something. And yet what I heard today was that there is not any improvement in
the demographic information and the risk adjuster. Are there plans to do
something about that?
Michael
Wroblewski
Donna,
thanks for that question. I'll answer it quickly and then I'll turn it over to
Jeff for a further response. You know, the HCC model that Greg described results
in a risk-adjusted score, as you saw on his - on the last slide.
And then, as you saw during Jeff's presentation, we then do additional
adjustments for fit. And I'll let him talk about that. And then I'll come back
and ask if there are any additional risk factors that should be taken into
account, and if so, how should we go about doing them? Jeff? And then I'll come
back to that question.
Jeff Ballou
:
Right, well, thank you, Michael. There - you know, I guess I should say that
because we've been working on these reports for several years now, there's been
quite a bit of testing that has gone into what we're doing. And at various
points, you know, various risk adjusters have been considered. And there, there
may be, as you've alluded to, not only mechanical or fit-related but also
policy-related reasons for including - excluding certain adjusters.
The model that we have right now - again, the - I think what I would say is that
we do view, given the, given the data that are available to us, the inclusion of
EHCC risk scores is an important part of that model, you know, without talking
about other risk adjusters at the moment, we have experimented with including
different - I'll call them higher order terms of squared and cubed versions of
those scores. The squared term did improve fit. And so for that reason, that has
been included. Again, I guess I should let Michael speak to, you know,
additional risk adjusters that might be considered in the future. Obviously
we're, we're always interested in continuing to improve on what we're doing
currently.
Michael
Wroblewski
:
In some ways what we've done is we've used additional statistical techniques to
improve the model's fit. Are there specific factors that you would suggest, and
if so, what would - what data sources could we use that we could validate and
that would be, be able to collect?
Donna Kinney
:
I'm really not sure about the - about data sources. I'm really concerned about
data sources. But we do know that poverty and educational status and multiple
other cultural factors do have an effect on both cost and on outcomes in
quality. And one of the things that we see heavily here in Texas is, we see a
factor that's related to a history of uninsured status, so that patients who
enter Medicare eligibility by - due to death or disability, either way - enter
the program with pent-up demand, and then are very high-cost because they
have a lifetime deficiency in health care. And I don't know how you adjust for
that. But I do think it's necessary to consider it.
Michael
Wroblewski
OK. Thank you for that comment.
Operator
:
Your next question comes from the line of William Rich.
Nicole Cooney
Mr. Rich, are you on the line?
Holley
, I'm not sure - we could barely hear you there - I'm not sure if we have the
next question on the line or not.
Operator
:
Your next question comes from the line of Debra
Lansey
Debra
Lansey
Yes, hi, this is Debra
Lansey
from the American College of Physicians. I'm wondering about the payment
standardization model. A lot of it seems to be based on the ICD-9 diagnosis
codes as a place to get started. And I'm wondering what CMS has in mind for,
say, the two-year window span of claims that would span 2012 and 2013 or 2013
and 2014 when the claims are going to have a mix of ICD-9 and ICD-10 codes on
them?
Michael
Wroblewski
Debra, this is Michael
Wroblewski
. Thanks so much for that question. We're very, very mindful of the transition
to ICD-10 and are looking at ways to make sure that the payment standardization
and the risk adjustment model take account, that transition, in a way that still
ensures fair and accurate comparisons among physicians. So I guess the short
answer is, is that we're very mindful of it, and more details will be
forthcoming. But thank you.
Debra
Lansey
OK.
Operator
:
Your next question comes from the line of Barbara Hall.
Barbara Hall
:
Yes, my name is Barbara Hall. I'm with the State of Missouri Department of
Mental Health. And this whole call is way above my head, but I was told to
listen in. My question is, is this - appears to be based on what certain
physicians have elected to report. How does that affect the physicians that do
not report that? Does that not eventually affect their payments as well?
Michael
Wroblewski
:
This is Michael
Wroblewski
again. Thanks for that comment. You know, what we were trying to accomplish in
today's call was, we had been providing
resource use reports for the past several years to physicians, to select
physicians, which we - and well, our plans are to be able to produce them for
all physicians. So, so we're building off of that. In terms of your question and
how does that relate to say the value modifier, we are - this is one set of
input that we're getting. We haven't made any decisions yet in terms of what we
would do with physicians who have not yet reported, say participated in the
P
hysician
Q
uality
R
eporting
S
ystem. We are likely to have additional outreach calls and events over the next
several months before we make - to get input to proposals that we'll make during
next summer and the beginning of the rulemaking cycle for the 2013 fee schedule.
But that issue about non-reporters is one that we're very aware of, and are
looking at how to address that.
Barbara Hall
OK.
Michael
Wroblewski
But thank you for that comment.
Barbara Hall
Thank you.
Operator
:
Your next question comes from the line of Karen
Rousch
Karen
Ruschel
:
Hi, thank you for the call. My name's Karen
Ruschel
with (Pavilion) Services. I'd just like it if you could review how the
responsible physician is determined, please.
Michael
Wroblewski
:
For the, for the 2010 physician QRURs, we identified - we've provided the QRURs
to 35 groups who had participated in the group practice reporting option for
quality. And so the - we identified the 35 groups. They're self-nominated. Then
we attributed patients to those using an attribution rule in which you had to
have - the group had to have at least two (E&M) visits during the previous year,
during 2010, and that group had the plurality of visits overall. No other group
had more. Even though they may have had two, no other group had more. So that
was the attribution rule that we used for the, for the group reports.
For the individual reports that we'll be sending out in the beginning of next
year, first
qu
arter
of 2012, we have used a series of different attribution rules
and - to take, to match beneficiaries with physicians based on (E&M) visits and
total charges. And since those reports aren't out yet, I'm not going to go into
detail about those. But we are likely to do additional outreach on kind of this
attribution question once those reports come out so people can see what we've
done and then we can take comment on that.
And then in terms of how we would go forward with the value modifier, I may
sound like a little bit of broken record but we're trying to get input based on
what we've done so far
to then
make proposals next summer, but thank you.
Operator
:
Your next question comes from the line of William Rich.
William Rich
:
Hello Michael. A quick question, you didn't mention anything about the
appropriate n. I know in the rules it's something about 30. But more
importantly, what is the role of the new proposed grouper that you're looking
at? To have a higher r squared? And when would you adopt or use a selected new
grouper
i
f one me
ets your
criteria?
Michael
Wroblewski
:
Thanks for both of those questions. In terms of the n number, you know, for the
physician - the individual reports, we're actually providing information on
every physicians see or
excuse me, every patient that
i
s seen. So as well as providing what that n is. So we're trying to actually
provide a full view of all the
Medicare
Fee-for-Service
beneficiaries that a patient sees.
You know, as I said a couple of times now, we haven't made any decisions in
terms of minimum n for the value modifier. And then in terms of the
episode-based cost, you know, what we have been focusing on were - and what we
have finalized at least for the initial year, for the value-based modifier is to
use total
p
er
capita cost. And so we - the risk adjustment and the payment standardization
techniques that we've talked about today really are focusing on total
p
er
capita cost.
As you know, Medicare is - or CMS was tasked with developing an episode grouper
for - by the end of this year. We are in the process of doing that and would be
announcing you know, how we're moving forward with that again in the beginning
of next year. And at that time we would then describe how risk adjustment works
related to episode-based cost
.
William Rich
Thank you.
Michael
Wroblewski
Sure. Thank you.
Operator
:
Your next question comes from the line of Douglas Carr.
Douglas Carr
:
Yes, I think some of my questions have been answered previously but I did want
to make a comment. This is Douglas Carr from Billings Clinic, that in the
attribution that is described at
- it's
All Specialty, although it's based virility of V and M, we've had experience
with that in the original PGP.
And we note that in the shared savings program model it uses primary care
attribution. And so for a - you know, for synergy and alignment with other
programs which seems to be taken off like the shared savings plan or programs,
you use different attribution methodologies in giving feedback to group or
individual doctors, it's kind of - might be very confusing. And so I think you
need to think through that as you go forward with this physician value based
purchasing.
Additionally, the elephant in the room is Medicare Part D. And clearly in you
know, fully capitated situations, the real total cost per care, pro capita care
is the A plus
B
plus D. And you know the value equation is whether you use a combination of
that, whether you spend more money on the pharmacy and less on the hospital or
more on whatever, I think you've got to put it all together otherwise you've got
a
- you're
pressing in on the balloon and you don't have - you're really not
capturing
all of those costs, so just some comment.
Michael
Wroblewski
Thank you for both of them.
Operator
:
Your next question comes from the line of Andrea
Cornejo
Andrea
Cornejo
:
Hi, this is Andrea from TMG Health Consulting. So I'
ve got a really
basic
question
.
So if you can go back into Slide One which has the title of the presentation.
It's Payment Standardization and Risk Adjustment for the Physician's Feedback
and Value Modifier Program.
Could someone and this could really be anyone
, not direct to
anyone in particular, could someone just kind of piece together conceptually
how, how all the different presentations and their focuses kind of connect
together?
Michael
Wroblewski
Sure. This is Michael
Wroblewski
, what we were trying to do was, we have two programs, one underway, one in
development that look at how do we, for lack of a better word, profile
physicians. And one of the inputs to profiling physicians is cost data, per
capita cost.
And in order to ensure fair comparisons, we used two techniques to make sure
that we compare physicians because physicians are very interested not only in
their own performance and how they compare to their peers. And so in order to
make fair comparisons of total per capita costs, we use two methodologies to
adjust those costs.
One was the first presentation that you heard (Peter Hickman)
wh
o talked about payment standardization
t
o the Medicare payments, so we're comparing apples to apples. And then the
second, the second and third presentations really dealt with looking at the
diagnos
e
s and conditions of the underlying beneficiaries so that physicians who treat
very, very healthy beneficiaries don't automatically look much better than those
who treat beneficiaries with
lots of
conditions. So - so that's kind of, hold strings together but thank you for that
comment though.
Andrea
Cornejo
:
Yes. Thanks. And that really puts everything else and all the details kind of
much better (inaudible). Thank you.
Operator
:
Your next question comes from the line of Michael
Kichall
Michael
Kichall
Yes. Just our group was part of the
G
roup
P
ractice
R
eporting
O
ption. And we were very pleased with the report that we got from the CMS.
The concern I have though, is that the value measurements which I think I would
hope can evolve with both quality measurement and also some better risk
adjustment, my concern is with regard to specialists that don't fit the model
for these four major diseases and just are going to use the payment system for
the overall beneficiary cost, for example, anesthesiologists or pathologists.
And so when you're talking about trying to fit this value based modifier into a
payment system for all physicians, the attribution problems are going to be
humongous. And so my, my hope is that you will continue to have this progress
for groups especially groups that are going to be accountable. I think it's
going to be very difficult for individual physicians especially when getting to
those specialists that are outliers with the types of patients that they see.
So again my thanks to CMS for a good feedback report for our
G
roup
P
ractice
R
eporting
O
ption. We really appreciated it. But I think this is going to be really
difficult for attribution for other individual solo physicians especially
specialists.
Sheila Roman
Thank you very much
,
we appreciate that Tom. And we certainly are aware of the challenges of
producing value based modifier and even the QRURs at the individual level.
Additionally, we are certainly aware that certain sub-specialties and in
particular the specialties that you have alluded to, present challenges since
they are dependent on physician referrals for their patients they see rather
than patients themselves initiating visits.
We - well be working very closely and have worked closely with the medical
societies as we move forward. And as we've alluded to, we have really taken a
different approach to attribution in the individual reports and once those are
made available, we'll be holding other outreach to you to see and receive - to
see your response and receive input from you on how we've been attributing at
the individual level. So we do understand those challenges and appreciate your
comment. Thank you.
Operator
Your next question comes from the line of
Marida
Jao
Marida
Jao
Yes, hi, good afternoon. My name is
Marida
Jao
with Medical Consulting. My question is addressed to Greg
Pope
on couple slides, 33 and 34.
First one on 33, could you kindly define the difference between the community
continuing enrollee and institutional continuing enrollee? And then on slide 34,
you mentioned demographic factors that Medicaid dual eligibility status.
What about we have an Indiana population of patients that are under a state
sponsored plan but are being paid Medicare fee rates, are they being considered
in the - the statistics that are being collected? And then is this way of
modeling one to be in the future, far future? I don't know how far, going to be
applied to Medicaid as well?
Michael
Wroblewski
I can answer at least
a
third of those three questions and then I'll turn it over to Greg for the first
two in terms of the difference between community versus institutional and the
collection of data at the state level.
In terms of application to Medicaid, what this program is really looking at is
Medicare
F
ee
-F
or
-S
ervice. So that's what this is all have been focused on. It has not been focused
on Medicaid at this - up to this point. And the step does not direct us to do
so. So then I'll turn it over to Greg for the answers to the first two
questions.
Greg
Pope
:
Sure. The community versus institutional, the differentiation there is the
institutional model was applied to beneficiaries who are long term residents in
nursing facilities which is just defined as three months or more. So it include
s
like you know, a short term episode in the scaled nursing facility but a long
term three months or plus in an institution nursing facility. And essentially
everyone else would be in a community model.
As to your second question, I don't know the answer to that off hand so I think
we need more detail. And possibly I don't know if you could provide more detail
to CMS. I believe there's an email address or something. We can probably get you
an answer to that question. I don't know for sure without more - more
information on that one.
Nicole Cooney
Hi. Sure,
Greg
. This is Nicole. The email address where you can send additional detail is
qrur
, q as in quality, r as in resource, u as in use, r as
report
QRUR
@cms.hhs.gov.
Marida
Jao
OK. Thank you.
Operator
:
Your next question comes from the line of Sarah
Tonn
Sarah
Tonn
:
Hello again the staff at the American Academy and
Neu
rology. We really appreciate the time and support to regional specialties and
societies in helping to understand this complex but very good work. But I think
that when you messaged to the physicians this - it needs to be very succinct
about what this is and then how to use it.
And so I'm just going to give a little example here and you know, I'm ignorant
about how this is calculated. But if I
were
to get this or physicians were to get this and I were to read this, first of all
I'd say, "Well
what is this?"
And so I'd want to know the - you know, the clear definite that this model is
based on *** but also what the limitations of the model was and what it excludes
.
And then I would also want definitions of like what HCC new enrollee is and what
community risk score involves
So that's just one comment.
The second part of the how, so the first part is what is this, how to
- and
what it is not. And the second is how should I use this. So if I was Doctor
Smith, slide 55 I think, I would look at this and I'd say, "Gosh, I was expected
to cost $80,000 but my observed was $32,000".
So first of all, I'd want to know my patient breakdown. So then I'd go up to the
table involved and see that (inaudible) Betty. And then I'd see that Betty is
going to cost me a lot more and so my intervention plan might be x. You know,
what should, how should I, how can I interpret this? And then make a difference
so that it costs less?
So my interpretation as what I'm seeing here is Doctor S
mith
is, "Oh, I'm doing fine. I - actually with adjustment I'm at averaging $4,000
but Doctor Jones on the other side has increased to $13,000." So when I analyze
it as if I were Doctor Jones, I would look at this and say, "Gosh, you know,
last year
c
u
z
you're taking previous year experience observed - no excuse me, the expected is
previous year right?
So what's happened with Doctor Jones is the observed year at present, which I'm
not quite sure about timeframe but then my total expected based on last year's.
So last year's was less. And I should have tried to perform at 1.0 versus nine.
70 percent over expected (inaudible), so what could have happened is Doctor
Jones could have intervene
d
. With a certain, you know, to make the expected cost excuse me, lower last
year.
So do you see what I mean
?
I think it's going to be the interpretation what this is and then how to use,
how to interpret this and then act on it. So I think it's very confusing when
the feedback reports came out of the PQRS reporting.
And so this, what we want to do is get people act on this report. And so what
are we telling them to do or how do you use this stuff and so if you get
something
i
n aggregate physicians are likely to go back and say, "OK, who's costing me the
most and then you put - you actually put more heavy intervention on to that.
Bring in more med level groups so then you're costing more next time and less
until you get punished. So I think it really needs to be analyzed how these are
going to be used, expected to be used or actually are used. Thank you.
Michael
Wroblewski
:
Thank you for that comment. You know, you've summed it up
with
some
nice action things and we'll follow-up with you. Thank you.
Nicole Cooney
Holley
, we'll take one more question.
Operator
At this time there are no further questions.
Nicole Cooney
:
OK. At this time I'd like to turn the call back over to Michael
Wroblewski
for the closing.
Michael
Wroblewski
:
I just wanted to thank everyone for joining - I wanted to thank everyone for
joining us this afternoon. As I mentioned a couple times, we will be providing
details about, about future events.
Whether they're call or in-person events in the coming weeks related to the
value modifier proposals
,
as well as to the physician feedback reports. And if you have any additional
questions, that you weren't able to ask today, or you thought of since the call
has ended, please use the - the email address that Nicole gave earlier which is
QRUR
@cms.hhs.gov and we will be sure to answer those questions.
And I want to say thank you to everyone. Happy holidays.
Nicole Cooney
:
And I just wanted to add, we definitely do want to thank everyone for
participating in today's call. An audio recording and written transcript will be
posted to the physician's feedback program page on the CMS website which is
www.cms.hhs.gov/physiciansfeedbackprogram
all one word.
It's also listed on the final slide in our presentation. And the materials will
be under the CMS teleconferences and events tab listed under today's date. I'd
like to thank our speakers today, Michael
Wroblewski
Dr.
Sheila
Roman
, Peter Hickman, Greg
Pope
and Jeff Ballou for their participation. Have a great day everyone.