Written Responses from RRT Members (PDF: 853KB/8 pages)

Protecting, maintaining and improving the health of all Minnesotans
Date:
March 3, 2011
To:
Provider Peer Grouping Rapid Response Team members
From:
Katie Burns, Health Economics Program
Subject:
Risk Adjustment
Thank you for participating in the Rapid Response Team. In preparation for our next
meeting, I wanted to distribute the attached memo from Mathematica Policy Research,
Inc. The memo is related to risk adjustment and outlines several issues for which we
would like your input:
•
What risk factors should be used to capture severity of illness?
•
Should regression or standardization methods be used?
•
Over what time period should patient history be assessed?
•
How should variations in the amount of diagnostic information be managed?
•
What risk factors other than diagnosis-based factors should be included?
We will review the memo during our meeting to ensure you have an opportunity to
clarify your understanding of the issues and to ask questions.
Response deadline: We will need your feedback on these issues by Monday, March
14 at 4:00 pm. Responses may be provided via email to [email protected].
MN Council of Health Plans: Sue Knudson
Thank you for the opportunity to provide input on the risk adjustment options, feedback on the
Mathematica recommendations and to provide other input based on our experience with these
methods and tools in our community.
Overall, we agree with the recommendation of using the indirect standardization approach.
This is an effective method to handle the fact that not all providers see patients with an array of
all conditions. We further support the use of the bootstrapping technique to validate the
model.
While we agree overall with the methodology, it is complicated and not replicable at the
individual plan level if future consistency in approaches is desired. To that end, during the
testing process, we recommend that more simple and straightforward methods also be tested
and preferentially used if they produce equal results. Some risk adjustment approaches are
fairly standard in the community today and have proved effective in peer grouping and
transparency applications. Key examples of consistent approaches are use ACGs instead of
ADGs to calculate risk adjustment rates and attribution of patients to a single provider vs.
proportionally to more than one (acknowledging that different methods may be used to
attribute to a single provider).
We do have some specific input and questions on details of the methodology as outlined below.
Provider Methodology
Attribution:
Page two of the memo refers to the use of a proportional attribution method indicating:
“Physician costs will include all costs incurred by their patients over a certain time period, though
costs of patients who are attributed to more than one physician will be attributed
proportionately to all of the different physicians.” It is not outlined or contemplated in the
methodology memo but it seems critical to the risk adjustment application because it would
also need to proportionally distribute risk score to align with the dollars for an overall cost of
care measurement. How will the method account for the need to synchronize with the
proportional attribution rule?
Truncation:
We recommend truncation be included in the methodology and be consistently applied from
weight development through to risk adjustment application. We recommend this because best
practice in risk adjustment calls for bidirectional consistency - that is, both the study and the
cohort group need the same exclusions and there also needs to be consistency with the weight
development and ACG application method and process. A truncation approach is preferred over
removing the entire members spend.
The Society of Actuaries studied truncation and reports a $100,000 limit maximizes the Rsquared of the risk adjustment model and reduces exposure to large cost claimants due to how
risk adjustment models under-predict high cost cases (SOA Study 2007 p. 49, Table A-5.3).
Winkleman, Ross; Mehmud, Syed. A Comparative Analysis of Claims-Based Tools for Health Risk
Assessment. Society of Actuaries April 20, 2007. http://www.soa.org/files/pdf/risk-assessmentc.pdf
Approach to standardizing pricing:
Page two of the memo references risk adjustment applying to two cost measures. The first of
which is total cost based on “standardized prices”. What is the approach to standardize pricing?
This is important to understand because the end result of vetting the peer grouping
methodology one component at a time could be compromised when decisions on each
component could be and often are related to one another. The unintended circumstance could
be an efficient provider looking quite the opposite if the standardized pricing model is not
sensitive to the same service being done at variable pricing levels based on place of service (e.g.
an outpatient procedure routinely performed at a more cost effective surgery center should be
priced on a standard schedule lower than the same procedure when performed at a more costly
acute outpatient hospital setting).
What risk factors should be used to capture severity of illness?
ADG’s are the right indicator of risk to use if there is a desire to customize below ACG cell level –
it is the same foundation of the ACG cell based approach that we use predominately in the
community. We would recommend age and gender are included in the model as those are not
defined by the ADG cells and that interactions are considered to accurately get at patient
complexity
Should regression or standardization methods be used?
See the excerpt from the ACG manual below on this topic – this comes from a section on rate
setting and capitation which is very similar to peer grouping concept for which we are building
methodology. The Mathematica memo is correct in that building a customized model using
ADGs and capturing the interactions between those ADGs in the model will likely produce a
stronger model than using the ACG cells. That said, some key application questions in this
approach are:
• How frequently is the model rebuilt? If the model is rebuilt each year, then a regression
model is subject to changes and fluctuations (see the excerpt below).
• What are the approaches to handling the potentially small N’s within specific
combinations of risk factors?
• Will an ADG regression model result in provider effectiveness being lost due to a model
that is too tight?
Page 6-2 of the ACG Applications Guide (Version 9.0 December 2009)
Over what time period should patient history be assessed?
We agree with the recommendation to use a concurrent adjustment. Further, we discourage
future consideration of the prospective view because the explanatory power is lost and perhaps
introducing a concern about “efficient care” getting swallowed up by the risk adjuster. This may
occur with a model that is built too tightly using the concurrent approach as well, such as the
ADG regression model outlined. In lieu of this approach, consideration of an ACG cell based
approach would be more appropriate.
A cost per month measure of resource use must consider the enrollment length of a person.
Pregnancy/Birth is a good example of a short term enrollee and a long term enrollee having the
same annual costs, but very different PMPMs for use in peer grouping. Although longitudinal
history can be built across payers, this scenario may still occur for new residents or newly
insured individuals. Once an adjustment for enrollment time is made, how short or long the
minimum requirement is less important.
How should variations in the amount of diagnostic information be managed?
This is one of the single biggest issues when it comes to risk adjustment application using a
multi-payer database. The principle of equivalence must be adhered to ensure accurate results
but the best manner to achieve this is not by capping the use of codes. Instead, we suggest the
source data include the complete and uncapped diagnoses coding and that transparency in
collection rates by payer source be made available. We discourage the use of “completion
adjustment” processes that may complete data but produce manufactured results that may not
be valid.
What risk factors other than diagnosis-based factors should be included?
Payer type is a meaningful way to get at socioeconomic status and lends itself useful to
adjusting for such differences. Another acceptable and widely used option is to report peer
grouping results by payer type (e.g. commercial, Medicaid, Medicare) which may be more useful
to consumers.
We agree with the intent of the service mix adjustment process but will await further
explanation on the actual primary care index adjustment process as the memo indicates more
detail on this approach will be released in the future.
Hospital Methodology:
Again, while we don’t necessarily disagree with the approach, it is potentially over-complicated
and not replicable at the individual plan level if future consistency in approaches is desired. To
that end, during the testing process, we recommend that more simple and straightforward
methods also be tested and preferentially used if they produce equal results. For example,
would use of APR-DRGs in a case mix adjusted cost analysis yield similar results using a less
complex method than to risk adjustment using ADGs.
We agree adjustments should be made for primary payer type and service mix. That said, we
also suggest on top of the adjustment applied to peer grouped costs, reconsideration of
exclusion of quaternary services offered by a very limited number of hospitals such as burn and
trauma because the notion of peer grouping does not lend itself to these specialized services.
Including all services and overly adjusting could result in “over-fitting” the risk adjustment as
noted in the ACG application guide (see excerpt above).
Lastly, we recommend consideration of truncation of hospital claims such as using average
length of stay at the APR-DRG level or by implementing a cost threshold level. The testing
process could be used to identify the most effective method of truncation.
MN Medical Association: Janet Silversmith
MN Business Partnership: Beth McMullen
No response submitted.
AARP: Michelle Kimball
No response submitted.
DHS: Marie Zimmerman
No response submitted.
MN Hospital Association: Mark Sonneborn
No response submitted.