Title: Measuring and Improving Quality in Managed Care: Some Statistical and Computing Issues
1Measuring and Improving Quality in Managed Care
Some Statistical and Computing Issues
- Randall K. Spoeri, Ph.D.
- Vice President
- Medical and Quality Informatics
- HIP Health Plans
- New York, NY
2Coverage
- Measurement and Quality in Managed Care
- Some Statistical and Computing Issues
- Measure Definitions
- Data Availability and Quality
- Risk Adjustment
- Analytical/Interventional Use of Results
- Predictive Modeling
- Thoughts About the Future
3About HIP Health Plans
- A managed care company with membership of over 1
million in metropolitan NYC and FL - Has a long history of over 50 years of service to
NYC residents - 2000 revenues budgeted at 2 Billion
- Has a strong commitment to quality measurement
and improvement
4Measurement and Quality in Managed Care
- Managed care has experienced rapid expansion in
recent years - With that expansion, concerns have been expressed
about the quality of care delivered - In response, quality management efforts have
relied heavily on performance measurement and
accreditation review
5Measurement and Quality in Managed Care
- The National Committee for Quality Assurance
(NCQA) has been the major provider of
accreditation reviews and performance measurement
standards - The Health Plan Employer Data and Information Set
(HEDIS) is the NCQA-sponsored collection of
standardized performance measures widely used by
Managed Care Organization (MCOs)
6Measurement and Quality in Managed Care
- With the growth of managed care and the
increasing focus on quality and performance
measurement, MCOs have begun to expand their
staff to include medical and quality informatics
personnel
7Measurement and Quality in Managed Care
- Skills of Medical and Quality Informatics (MaQI)
staff include statistics and statistical
computing, epidemiology and public health,
psychology, and economics - Demand is high for this type of individual, not
only in MCOs, but also in allied areas such as
hospital organizations and pharmaceutical
companies
8Measurement and Quality in Managed Care
- Statisticians and computing professionals can
make major contributions to the health care
industry, especially managed care - The opportunities are many, as are the challenges
9Some Statistical and Computing Issues
- Measure Definitions
- Data Availability and Quality
- Risk Adjustment
- Analytical/Interventional Use of Results
- Predictive Modeling
10Measure Definitions
- In order that comparisons can be made among
health plans, it is essential that measure
definitions be uniform, unambiguous, and precise - In the managed care world, HEDIS is the standard
set of performance measures used by MCOs
11Measure Definitions
- The recently announced HEDIS 2001 will have 52
measures across 7 domains of care Effectiveness
of Care, Access/Availability of Care,
Satisfaction with the Experience of Care, Health
Plan Stability, Use of Services, Informed Health
Care Choices, and Health Plan Descriptive
Information
12Measure Definitions
- Example measures Childhood Immunizations,
Breast Cancer Screening, Beta Blocker Treatment
After a Heart Attack, Practitioner Turnover,
Cesarean Section Rate - Developing specifications (i.e., detailed
definitions) for these measures is an arduous
task, often involving statistical issues
13Measure Definitions
- Some statistical issues include
- Measure reliability/validity
- Data source (administrative or medical record)
- Complete enumeration or sample (or hybrid)
- Sampling methodology
- Sample size
- Technical Panels were convened by NCQA to advise
on these and other issues
14Data Availability and Quality
- Encounter data are administrative data supplied
by a provider (i.e., a physician) to a health
plan on outpatient visits by the plans members - The incentives to supply these data range from
none to weak, to strong monetarily or
contractually or both
15Data Availability and Quality
- Consequently, encounter data repositories are
often incomplete and frequently of poor quality - Many physicians dont see or understand the
importance of this encounter data, and view it as
yet another disruption imposed by MCOs
16Data Availability and Quality
- The movement from paper-based systems for
submission of encounter data, to electronic ones
has promise - The ever-expanding use of encounter data for such
things as physician performance measurement
(so-called provider profiles) has begun to
garner physician attention
17Data Availability and Quality
- Federal and state mandates may also help to
improve the completeness of encounter data - But even when data repositories are complete,
there remain major questions about the quality of
the data contained therein - Encounter, claim, pharmacy, lab, and other health
care data are still of uneven quality
18Data Availability and Quality
- One approach, taken by NCQA, in connection with
the quality of HEDIS data is to require that the
complete system for this datas production be
audited (i.e., externally reviewed and verified)
structure, process and outcome
19Data Availability and Quality
- Other techniques used in data quality assurance
include the examination of source code used for
administrative data pulls, medical record reviews
to measure the accuracy of administrative data
counterparts, and medical record re-review to
assess the accuracy of primary chart abstraction
20Data Availability and Quality
- With the increasing reliance of MCOs on data and
measurement, through complex data structures like
data warehouses, the completeness and accuracy of
the information becomes paramount - Statisticians and IS/IT professionals need to
work with clinicians and administrators to
continuously improve these data systems
21Risk Adjustment
- When comparing entities such as hospitals and
physicians, it is important that an adjustment be
made for differences in the mix of patients being
treated - For example, it would be inappropriate to compare
one hospital providing mostly obstetrical care in
the suburbs to an inner city trauma center
hospital using mortality
22Risk Adjustment
- Similarly, comparing a physician seeing
predominantly young patients to another physician
seeing a large proportion of Medicare patients is
not appropriate when looking at pharmacy usage - Many of the measures used to assess hospital and
physician performance need risk adjustment, a
leveling of the field based upon differing
patient characteristics
23Risk Adjustment
- Characteristics used generally include age,
gender, diagnosis and procedure, as well as
comorbidities, for example - One commonly used approach to risk adjust
performance statistics is to use the concept of
observed and expected
24Risk Adjustment
- The observed result could be a hospitals average
length of stay for a certain surgical procedure - The expected result would be that predicted using
epidemiological or statistical methods for the
patient population undergoing the surgical
procedure at that hospital
25Risk Adjustment
- Statistical issues include methodology selection,
model calibration, and explication - Popular methods include multiple and logistic
regression and survival analytic (Bailey-Makeham) - Calibration issues often surround data
timeframe, geography, availability and quality,
breadth
26Risk Adjustment
- Explication Users of the data often become
suspicious of methods they dont understand
(e.g., complex statistical methods), even when
the methods are sound - Computing issues arise in building and handling
the enormous databases common in managed care
(e.g., encounters)
27Risk Adjustment
- Computing challenges also arise out of building
and using complex risk adjustment models, or even
in using models developed by others (i.e.,
commercial vendors) - Nevertheless, these issues and challenges provide
opportunities for statisticians, computing
professionals, and clinically prepared
individuals, working together
28Analytical/Interventional Use of Results
- The best measurements and measurement systems are
of no value unless they are followed by action
and behavior change - Some analyses prompt action and intervention by
their very naturean example follows using real
HIP data
29(No Transcript)
30Analytical/Interventional Use of Results
- Most analyses, however, require an intervention
strategy - For example, learning that an MCO has a low rate
of childhood immunizations could prompt a variety
of actions
31Analytical/Interventional Use of Results
- Example interventions
- Reminder to Primary Care Physician
- Reminder to Pediatrician
- Reminder to Parents
- Reminder stickers for medical chart
- Promotional brochures
- Educational forums
32Predictive Modeling
- Most MCOs have undertaken disease management
initiatives for high prevalence/high cost
conditions, such as asthma, diabetes, congestive
heart failure and hypertension - As part of these efforts, predictive modeling is
sometimes undertaken to identify high risk
individuals with a given disease
33Predictive Modeling
- The goals are to
- Avoid costly ER visits or hospitalizations
- Treat individuals at higher risk earlier in the
disease process, and monitor them more closely - Reduce the risk of adverse outcomes
- Improve satisfaction and quality of life
34Predictive Modeling
- Some of the methodologies used include
- Multiple Regression (Linear and Nonlinear)
- Logistic Regression
- Neural Networks
- Time Series
- Many of the disease management vendors that have
come on the scene tout predictive modeling as a
value-added benefit
35Predictive Modeling
- Many MCOs do some of this work themselves or in
partnership with disease management vendors - There is a need, however, to assure that the
statistical methodologies are being properly
utilized, and that the data used to both build
and apply the models be accurate - Results to date have been encouraging
36Thoughts About the Future
- MCOs will continue to expand their performance
measurement and improvement activities - Accreditors (e.g., NCQA), Federal and State
Agencies (e.g., HCFA and DOH/DOI), and
purchasers/employers are driving much of this
expansion
37Thoughts About the Future
- The use of statistical methods in MCO quality
management will increase - Administrative data systems must improve to
greatly reduce the need for medical chart
abstraction - Opportunities for statisticians with a bent
toward computing will abound in MCOs
38Thoughts About the Future
- The future aint what it used to be - Yogi Berra