Issue Description to RRT Members (April 2011) (PDF: 125KB/13 pages)

 Protecting, maintaining and improving the health of all Minnesotans Date:
April 15, 2011
To:
Provider Peer Grouping Rapid Response Team members
From:
Katie Burns, Health Economics Program
Subject:
Quality Composite Measure Design
Thank you for participating in the Rapid Response Team. In preparation for our next
meeting, I wanted to distribute the attached memo and excel tables from Mathematica
Policy Research, Inc. The memo summarizes our planned approach for constructing the
composite quality measure and outlines several issues for which we would like your
input:

What method should be used to combine individual measures into subcomposite
categories?

How should missing data be treated within the subcomposite categories?

How should missing subcomposite category scores be treated?

How should subcomposite categories be weighted as part of the overall composite
quality score?
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 Tuesday, April 26
at 4:00 pm. Responses may be provided via email to [email protected].
MEMORANDUM
955 Massachusetts Avenue, Suite 801
Cambridge, MA 02139
Telephone (617) 491-7900
Fax (617) 491-8044
www.mathematica-mpr.com
TO:
Katie Burns, MDH
FROM:
Angela Merrill and Nyna Williams, Mathematica
SUBJECT:
Total Care Quality Composite Construction
DATE: 4/15/2011
This memo discusses the planned approach and specific issues for which we are requesting
RRT feedback in creating the quality total care composites scores for hospitals and clinics. The
composites will be made up of measure category “subcomposite” scores (e.g., prevention,
chronic disease outcomes, patient safety), as well as their performance on all individual measures
included in the subcomposites. Presenting providers with this information will give them an
overall summary of their performance, performance within certain areas to compare with their
peers, and actionable results on individual measures. There will be two peer groups in the
hospital reports: general acute hospitals and critical access hospitals (CAHs). There will be one
peer group for physician clinics, though the analysis will be limited to primary care or
multispecialty clinics offering primary care.
There are many options for combining quality measures into subcomposite measure
category scores and then into an overall composite. For example, the options include: combining
absolute versus relative scores, whether or not to impute missing data, and if not imputing, how
many measures to allow to be missing. Our proposed approach is based on our review of other
compositing approaches (e.g., AHRQ, CMS), the Provider Peer Grouping (PPG) Advisory
Group Report and meeting documents, NQF standards for composite measures, and initial
analysis of the hospital and physician data.
We also considered the following criteria in developing the approach: (1) maintaining
consistency with other reporting in Minnesota or on CMS’s Hospital Compare; and (2) trying to
be as inclusive as possible, so that we have quality measures available to compare to cost
measures for as many providers as possible. Missing data is a particular challenge since the
components of the total care composites affect a diverse set of conditions, patient populations
and providers, resulting in many providers with missing data for individual measures.
This memo is organized as follows. First, we discuss combining individual measures into the
measure category subcomposites, including how to treat missing data and minimum case size
requirements. Second, we discuss creating the overall composite. We conclude with a summary
of the proposed approach and open items. Throughout the memo, we refer to results from
preliminary analysis of our final data, including hospital measures from various sources optimal
diabetes care (ODC) and optimal vascular care (OVC) clinic data, and HEDIS group-level data.
Appendix Tables A.1 and A.2 present a summary of the proposed measures and subcomposite
groups, their source, their service period, and information on risk adjustment.
An Affirmative Action/Equal Opportunity Employer
MEMO TO: Katie Burns, MDH
FROM:
Angela Merrill and Nyna Williams, MPR
DATE:
April 15, 2011
PAGE:
2
A. CONSTRUCTION OF SUBCOMPOSITES
1. General approach for combining measures
The first step in the composite construction is to combine the measures within each category
into a “subcomposite score”. Following the PPG Advisory Group Report, the hospital quality
total care composite consists of five categories—readmission, process, patient experience,
mortality, and patient safety—and the physician quality total care composite combines three
categories—prevention, short-term (minor) acute, and chronic disease. 1 We discuss three main
options for combining the measures. For all approaches, we would first recode measures to have
the same directionality so that higher numbers would indicate a more positive outcome across all
measures.
Option 1. Combine the absolute rates for each measure: Under this option, a provider’s rates for
each performance measure in the subcomposite are combined using a simple or weighted
average. Option 1a is to standardize the rates prior to combining. A few methods to put the
scores for different measures on a standard scale include computing z-scores 2 , or taking the ratio
of the provider’s rate to the mean rate in the state (resulting in a 0 to 1 scale).
The absolute rate approach is attractive when the measures within a subcomposite (e.g.,
process measures) are on the same scale, and all providers can be compared using the same
individual measures within the category. The NQF-endorsed AHRQ composites use the absolute
score approach. This method retains distributional information of all underlying measures.
Option 1a helps when individual measures may have different distributions or when providers
report different sets of applicable measures. If providers have different sets of individual
measures with complete information, and scores are based only on those measures (e.g., missing
measures are not imputed), then using the absolute approach (without standardized scores) may
have issues with comparability across providers because one provider might report on a set of
measures that have lower average score than a set applicable to another provider. Standardizing
helps address this issue.
1
The PPG Advisory Group Report included one other subcomposite--chronic disease process—that consisted
of one asthma HEDIS measure; we have since combined this measure with chronic disease outcomes to form a
broader chronic disease subcomposite.
2
A z-score would be calculated as [(provider’s rate-mean state rate)/standard deviation in state].
MEMO TO: Katie Burns, MDH
FROM:
Angela Merrill and Nyna Williams, MPR
DATE:
April 15, 2011
PAGE:
3
Option 2. Combine percentile ranks for each measure. Under this approach, all providers are
ranked from lowest to highest on each of the individual measures in the category and then each
provider is assigned a percentile rank (0 to 100) for each individual measure. The subcomposite
score is the average (simple or weighted) of these ranks. For example, if a hospital has three
measures and is at the 90th percentile on measure 1, the 80th on measure 2 and the 85th on
measure 3, its subcomposite score would be the average percentile rank (90+80+85)/3=85.
The percentile rank approach is flexible and allows for combining diverse sets of measures
and measures that have different distributions of results. It may also be more attractive in terms
of missing measures, since it is possible to average over the measures a provider does have,
without the problem of comparing apples to oranges. However, the percentile rank approach has
limitations when there is little variation in measure results, for example “topped out” measures
where a relatively large proportion of providers all perform at or near 100%; in this case,
percentile ranking among these top performers is arbitrary, with very small differences in
measure outcomes potentially leading to large differences in ranking. Also, information on the
underlying variation in measures is lost with the percentile ranking.
Option 3. Combine absolute rates (or percentile ranks) into a “point” system. Using either
approach above, translate individual measure outcomes into scores for meeting a pre-determined
benchmark for each measure, which are then added together to form the subcomposite (e.g.,
90th percentile or above gets 10 out of 10 points).
The point system approach may work well when there are categorical measures in the
composite, which are not easily handled by the absolute score or percentile approach, and for
putting measures on the same scale. This approach is similar to the CMS Hospital Value-Based
Purchasing (HVBP) approach for achievement scores. On the other hand, we would need to
develop a point rating system and benchmarks, which could be viewed as arbitrary. This
approach also results in the loss of individual measure distributional information.
We recommend Option 1a-combining standardized scores for the measures within each
subcomposite, standardizing each measure with the z-score approach. While Option 2 may be
the most flexible approach for different sets of measures, standardizing scores will help when
providers have different sets of measures available within the subcomposite. Some of our other
recommendations below stem from this approach.
2. Treatment of missing measures
Not all providers have data for each quality measure within the subcomposite categories.
Moreover, many providers will have measures that do not meet a minimum number of cases for
sufficient reliability (discussed in next section), which we will in effect treat as missing
measures. The compositing approach must develop rules for: (1) the number of measures within
MEMO TO: Katie Burns, MDH
FROM:
Angela Merrill and Nyna Williams, MPR
DATE:
April 15, 2011
PAGE:
4
a category that a provider must have to get a subcomposite score at all; and (2) what to do about
the missing measure outcomes (e.g., leave out, impute).
In our preliminary data analysis, many providers are missing measures, even before
applying minimum case size cutoffs. There are some hospital measures where a significant
portion of hospitals have no cases for a measure, or very few cases. For example, very few
hospitals have the process or mortality measures related to cardiac procedures, since presumably
few hospitals perform these surgeries. The same issue exists for physician measures, but fewer
measures are affected. For example, some medical groups do not have data for the childhood
immunization, colorectal screening, blood pressure screening, or the asthma care measures.
Because of these missing measures, few providers have all of the measures in the
subcomposite measure categories, although the majority of providers have at least one measure
in each subcomposite category. One exception is patient experience, where 32 percent of
hospitals have no HCAHPS data. This is driven by the fact that hospitals with fewer than 500
admissions per year (predominantly critical access hospitals) do not have to report these
measures, and below we discuss removing this category from the composite score for CAHs.
Among the physician providers with any quality data, most clinics have data for both the OVC
and ODC measures.
Key to our approach to missing measures is that we are assuming that completely missing
measures are because the measure is not applicable to the provider’s population (for example,
measures that apply only to hospitals admitting a certain number of heart attack patients or
performing CABG, or physicians treating child versus adult populations or certain conditions).
While it is also possible that some missing measures could be due to selective reporting of
measures, we do not believe this is a large concern within the measure groups. Therefore, we
assume that if we summarize only the measures that a provider does have, we are not selecting
only those measures where a provider did particularly well.
Our planned approach to calculating each measure subcomposite when a provider is
missing measures is to compute a subcomposite score if they have one or more measures
within the measure category. This approach provides more inclusiveness than requiring all or
most measures to be present, and is consistent with the PPG advisory group recommendations
and the AHRQ composite approach.
Our planned approach to missing individual measures is not to impute any missing
measures for clinics or groups; however, an open item is whether to impute missing measures
for hospitals due to larger amounts of missing data, especially among CAHs. For clinics,
enough primary/multispecialty clinics appear to have quality data with which to calculate a
composite score with no imputation (over 600). Almost all clinics have the ODC and OVC
measures; however, since the groups associated with clinics have different sets of measures
applicable to them, this approach will result in clinics having different sets of HEDIS measures
MEMO TO: Katie Burns, MDH
FROM:
Angela Merrill and Nyna Williams, MPR
DATE:
April 15, 2011
PAGE:
5
that make up their subcomposite score. This approach is reasonable as long as we standardize
measures scores before combining as described above.
However, in order to have as many providers with a subcomposite category score, without
imputing subcomposites scores themselves, we are considering imputing some individual
measure scores for hospitals, but only for providers who have some cases that don’t meet the
minimum case size (presented in next section).
Options for imputing missing measures include:
Option 1: Impute missing measure by assigning the peer group mean. This is a simple version of
imputing, where a provider simply receives the appropriate peer group mean. This approach
results in all providers having a result for all measures for which they had any data. However, it
may not seem fair that the provider gets an average score when, in fact, they may be a poor
performer.
Option 2: Impute missing measure with another method. Under this approach a missing measure
could be imputed through a blended rate, a regression approach, or some other approach like
assigning median of the quartile a provider is in for the other measures in the category. A
blended approach could use a weighted average of the hospital’s own rate and the peer group
rate, where the hospital’s rate gets a higher weight the more cases they have. Using regression,
scores on non-missing measures could be used to help predict a provider’s score on missing
measures using a regression. This is a more nuanced approach that tries to take account of other
available measure performance information; however, it requires providers to have other
measures with which to impute, and is more complicated to implement and to explain and may
not be possible for providers to replicate.
We recommend imputing missing hospital measures, using the blended rate approach;
this approach uses some data from the hospital and is consistent with how MDH’s approach to
risk adjustment treats clinics with a small N in a particular payer type category for quality
reporting purposes under Minnesota Statutes 62U.02. We would apply this approach to all
hospitals (general acute and CAH), though based on initial analysis of case sizes, this would only
have an effect of allowing more CAHs to be included in the composite.
3. Minimum Case Size Requirements
We will impose a minimum case (N) restriction for each measure to ensure reliability of the
information being incorporated into the composite. Measures that do not meet the minimum N
criteria will be set to missing, and treated as described above. However, for display in provider
reports, we would indicate if a provider did not meet a minimum cutoff, and may want to also
consider showing the mean score among the providers with small N.
MEMO TO: Katie Burns, MDH
FROM:
Angela Merrill and Nyna Williams, MPR
DATE:
April 15, 2011
PAGE:
6
In many cases, the measures we received from MNCM already have minimum N
incorporated (for example, for CMS hospital 30-day measures, we do not have results if the
hospital had fewer than a certain number of cases to calculate the measures). In other instances,
there are already established minimum N for measurement reporting on Minnesota or other
websites. The CMS HVBP approach is proposing a minimum case size of 10 for all measures to
be included in their composite.
The following are the planned minimum case sizes for the composites: 3
Hospital measures:





30-day mortality and readmission outcome measures: 25 cases. These measures are
reported on Hospital Compare if the hospital has at least 25 cases.
Process of care/SCIP: 10 cases. No minimum restrictions for reporting measures on
Hospital Compare, though we recommend a minimum cutoff of 10, consistent with
hospital VBP. 4
AHRQ IQI and PSI measures: 25 cases. AHRQ uses a minimum of 3 cases to create a
rate for a hospital; however, a recent CMS reporting of these measures based on
Medicare patients only presented results for hospitals with 25 or more cases.
HCAHPS measures: 100 cases. CMS does not impose a minimum for reporting, but the
website says that results based on fewer than 100 surveys may not be reliable 5 . The
HVBP appears to exclude those with fewer than 100 cases.
Ventilator-associated bundle compliance, central line bundle compliance and SSI
rate for vaginal hysterectomy: 3 cases. The MN hospital association uses a minimum of
3 or more cases to report these measures to protect patient confidentiality.
3
It is likely that these various minimum case size thresholds were chosen in some cases based on analysis and
estimates of reliability, while others may be more arbitrary. There is some analysis supporting the use of N=25 for
the CMS 30-day outcome measures, showing that the measures have moderate levels of reliability (as measured by
interclass correlation coefficients of at least 0.40).
4
Hospital calculation of state and national averages for the individual measures uses a cutoff of 25 cases for
inclusion. The MN Hospital Quality Reports website shows the results for the Appropriate Care Model and SCIP
measures for all case sizes, but also displays a note saying that “the number of cases is too small (<25) to reliably
tell how the hospital is performing”.
5
Note that the HCAHPS data available from Hospital Compare has ranges only for surveys completed (fewer
than 100, 100-299, 300+).
MEMO TO: Katie Burns, MDH
FROM:
Angela Merrill and Nyna Williams, MPR
DATE:
April 15, 2011
PAGE:
7
Physician measures:
 OVC/ODC clinic measures: MNCM uses a minimum reporting requirement of 30 cases.
 HEDIS measures: MNCM uses a minimum reporting requirement of 30 cases per group
for the purely administrative HEDIS measures (e.g., cold and sore throat care), and a
minimum of 60 cases for hybrid HEDIS measures (childhood immunization, controlling
high blood pressure, colorectal cancer screening).
The number of providers meeting the minimum case sizes above are generally smaller than
the total number of providers with the measure, particularly for hospitals, leading to the
recommendation to impute hospital measures above. While the central line and VAP bundle
compliance and SSI rate for vaginal hysterectomy measures have a very low cutoff (three), this is
consistent with the Minnesota Hospital Association’s recommendations for publicly reporting the
data and a higher cutoff would result in dropping a significant portion of providers for these
measures.
4. Weighting the Measures within Categories, and Treatment of “Topped Out” Measures
To create the subcomposite score based on the measures and rules above, the scores need to
be combined in some way, such as through a simple or weighted average. Related to the issue of
weighting is the treatment of measures considered “topped out” when most providers have
reached a very high level of performance. For example, a simple definition of topped out is if the
75th percentile (or lower) is at a performance rate of 100%. Other definitions consider low
variation as a definition of topped out, regardless of the level of performance. 6 As seen in Table
3a, many of the measures in the proposed hospital total care composite do appear to be topped
out; for example, in the process of care domain, the AMI ACM, four SCIP measures, and two
infection measures have a performance rate of 100% at the 75th percentile. None of the
physician measures are topped out by this definition. This is problematic for the percentile
ranking approach since providers in the top 25% would be arbitrarily ranked. A standardized
score would mitigate this, though there is still little variation.
We plan to use equal weighting when combining the individual components of a
subcomposite, and not to downweight the topped out measures. Equal weighting is the simplest
approach and is preferred by the NQF and the PPG Advisory Group unless there are clear
reasons to use other weighting. We do not plan to take a more complicated weighting approach
in Year 1 of peer grouping, though this approach could be revisited in future years. We will leave
6
For example, the CMS definition of topped out is: (1) the difference between the 95th and 75the percentile is
less than 2 times the standard deviation and (2) the truncated coefficient of variation (TCV) is less than 0.1. The
TCV is equal to the coefficient of variation between the 5th and 95th percentiles.
MEMO TO: Katie Burns, MDH
FROM:
Angela Merrill and Nyna Williams, MPR
DATE:
April 15, 2011
PAGE:
8
the topped out measures in as they are; in effect these measures will contribute little to the
variance in the subcomposite score. If a percentile scoring approach were used, we would
recommend downweighting “topped out” measures in the composite.
B. CONSTRUCTION OF OVERALL TOTAL CARE COMPOSITE
Once each subcomposite score is created for a measure category, the basic steps for
combining them into a total care quality composite are to combine the subcomposite scores. Key
issues here are how to treat completely missing subcomposite scores and weighting of the
subcomposites. For the physician total care composite measures, we will also need to combine
medical group-level scores for three of the subcomposites into a clinic-level composite.
1. Treatment of Missing Subcomposites
Some providers will be missing data for an entire subcomposite because they did not report
results for the required number of measures in the category or they did not meet the minimum N
requirements for all subcomposite measures. Our proposed approach weighs inclusiveness
against the ability to compare providers on like measures. Additionally, peer grouping may
address issues such as certain types of providers missing entire categories of measures. In
preliminary data analysis we find hospitals and clinics that have no measures in certain
categories. The most frequently missing hospital category is the patient experience, though for
non-CAHs, this category is only missing for five hospitals. Missing subcomposite information is
also a concern for the physician total care quality composite, although there are no providers that
would be missing all subcomposite categories.
For the composite construction for clinics, we will require that clinics have a
subcomposite score for every measure category (that is, at least one measure meeting the
minimum case size requirement) for inclusion. However, the approach for hospital composite
construction is still open given the larger extent of missing data. As mentioned before, this
approach for clinics will allow composite construction for almost all of the
primary/multispecialty clinics. This option is consistent with the PPG Advisory Group slides
from a September 2, 2009 meeting that indicate that “a provider should have data on at least one
measure from each of the following categories to be peer grouped….” 7 However, for hospitals,
in particular CAHs, it appears that (when applying minimum N cut-offs) only two-thirds of
CAHs would have subcomposite scores for four subdomains after excluding the patient
satisfaction measure category. Another 15 percent of CAHs would have measure data for three
subcomposites.
7
http://www.health.state.mn.us/healthreform/peer/advisory090902_presentation.pdf, slide 8.
MEMO TO: Katie Burns, MDH
FROM:
Angela Merrill and Nyna Williams, MPR
DATE:
April 15, 2011
PAGE:
9
The three main options for combining the subcomposite scores include: requiring a provider
to have every subcomposite, allowing one missing subcomposite, or imputing missing
subcomposite scores.
Option 1. Do not allow any missing subcomposites (except the experience group for CAHs).
This approach is a uniform approach for all hospitals. However, with minimum N restrictions, if
we do not impute any individual measures, this approach will force the loss of the most hospitals
(about one-third of critical access hospitals). If we do impute measures with low N, about onefifth of CAHs would not receive a composite.
Option 2. Allow one missing subcomposite score and calculate the overall composite by
reassigning weights to the subcomposite scores a provider does have. For example, for CAHs,
we could redefine the composite weights to put zero weight on the missing subcomposite, and
adjust the remaining weights. This is the most inclusive approach, and might be satisfactory if
providers within a peer group are at least compared on the same subcomposites. However, it
results in different composite construction across hospitals rather than a uniform methodology.
Since hospitals could be missing different sets of measures, we would have to develop some
‘rule’ for weighting the subcomposites that a provider does have.
Option 3. Calculate the overall composite by imputing the missing subcomposite score. For
example, we could assign the peer group mean subcomposite score for the missing domain, or
the provider’s average rank in domains they do have. This is an inclusive approach and results in
uniform composite construction/weighting. However, imputing may be hard to explain if a
provider does not have a score because the measure category is not relevant to its patient
population.
We recommend requiring a provider to have all subcomposite relevant scores in order to
receive an overall composite score. For physician clinics, this approach will result in a total
quality composite for almost all primary care/multispecialty clinics. With our approach to
imputing measures for hospitals with small N recommended above, this approach will include
about 91 percent of general acute hospitals and 80 percent of critical access hospitals. (The
remaining hospitals are ones that had no data for any measure in at least one applicable
subcomposite group.) Under this approach, providers without data for at least one measure in
each subcomposite would be excluded from PPG.
3. Combining Physician Group- and Clinic-level Measures
For the physician total care quality composite, while we will have some of the Chronic
Disease measures at the clinic level, we will only have the HEDIS measures that make up the
Prevention and Short-Term Acute categories at the medical group level. Since we plan to report
at the clinic level, we will assign the same HEDIS score to each clinic within a group.. Without
MEMO TO: Katie Burns, MDH
FROM:
Angela Merrill and Nyna Williams, MPR
DATE:
April 15, 2011
PAGE:
10
data on how the patient population for each HEDIS measure is distributed among clinics in a
group, we believe this is the best approach.
4. Weighting of Subcomposite Scores
Finally, subcomposite scores are averaged to create an overall composite value, using a
weighting scheme to weight each subcomposite score. 8 The current approach from the PPG
recommendations report are “policy” weights where outcome measures receive a higher weight
than process and intermediate measures (see ‘original weight’ in table below). We recommend
keeping policy weighting (as opposed to equal weighting) but present revised weights below to
address changes in measure categories.
Recommended Weights
Measure category
HOSPITAL TOTAL CARE
Readmission
Process of care
Patient experience
Mortality
Inpatient complication
PHYSICIAN TOTAL CARE
Prevention (group-level)
Short-Term Acute (group-level)
Chronic disease outcomes (clinic)
Chronic disease process (group)
Hospital avoidance (claims)
Number
Measures
Original
Weight
Revised
Weight
3
15
7
12
6
20%
15%
15%
30%
20%
20%
15% (CAHs-30%)
15% (CAHs-0% )
30%
20%
5
3
13
0a
0b
20%
10%
25%
10%
35%
20%
20%
60%
0%
0%
a
The one measure originally in this group has been combined with moved to the Chronic disease outcomes group
to form a broader chronic disease category.
b
We will not compute claims-based outcomes measures for the total care composite, as originally proposed in the
PPG Advisory Group report; however, these measures will be included in condition-specific compositing and in
future iterations of both total care and condition-specific reports..
For hospital total care, some alternative weights are also presented to address the fact that
CAHs do not have any patient experience measures available. Because almost all CAHs are
missing the HCAHPS subcomposite, we would put 0 weight on that domain for that set of
8
If we were using an absolute score approach, then we might want to standardize each measure set to the same
scale before combining, for example, by subtracting the mean and dividing by the standard deviation.
MEMO TO: Katie Burns, MDH
FROM:
Angela Merrill and Nyna Williams, MPR
DATE:
April 15, 2011
PAGE:
11
hospitals, and using the following weights: readmission (20%); process (30%), mortality (25%),
patient safety (25%). We recommend placing higher weight on process than on readmission
measures for CAHs because small hospitals will have little variation in outcome measure results
due to the risk adjustment process for these measures that pulls rates for small hospitals to the
national rate.
Since we are not including hospital avoidance measures that were in the original PPG
Advisory Group report, the weights for the physician total care score have been adjusted, still
placing relatively more weight on the outcomes measures. We place the most weight on the OVC
and ODC measures since they are outcomes and also since the HEDIS measures will have less
variation at the clinic level since each clinic in a group will have the same score. For combining
the hospital subcomposite results, we recommend placing equal weight on the mortality and
patient safety outcomes.
C. SUMMARY AND OPEN ISSUES
The following summarizes the planned approach for constructing the total care composites
for hospitals and providers:





We will not impute missing measures or subcomposite scores for physician
clinics (see open issue for hospitals below).
We will impose minimum case size limits as described above, and treat a
measure as missing when data are available for too few patients.
We will calculate a subcomposite score when a provider has at least one measure
in a category.
We will weight measures evenly when combining, and will not downweight
topped off measures. 9
We will assign the same group-level (HEDIS) score to each clinic (that applied to
the group in 2009).
The issues for which we request RRT feedback are as follows:
1. How to combine individual measures – do we combine standardized rates for
each measure or percentile ranks?
 We recommend combining standardized scores.
9
The decision not to downweight is based on the recommendation to use standardized scores; if a percentile
approach is adopted, we would recommend downweighting topped off measures.
MEMO TO: Katie Burns, MDH
FROM:
Angela Merrill and Nyna Williams, MPR
DATE:
April 15, 2011
PAGE:
12
2. How to address missing data for hospitals (mainly, critical access hospitals) – do
we allow a hospital to be missing more than one of the relevant subcomposite
categories (and reweight when combing the ones they do have) or do we impute
missing measures (or subcomposites)? If so, how do we impute?
 Before calculating the subcomposite scores, we recommend imputing missing
individual measures for hospitals with reported data below the minimum N
criteria, and imputing by creating a blended rate of the hospital’s rate and the
peer group average rate, where the weight is based on the number of cases.
3. How to weight the final subcomposite components?
 We recommend using the proposed weights above on p. 10.