Transcript Risk Adjustment in Medicare Part D

Risk Adjustment in Medicare Part D
Prescription Drug Benefit
Open Door Forum
December 2004
Risk Adjuster Basics
• Capitated payment is adjusted according to
the expected cost of the enrollee.
• Expected cost is derived from enrollee
characteristics:
° Enrollees may be sorted, by their characteristics,
into cells that have assigned risk factors, or
° Characteristics may be assigned risk factors that
are added to produce a total risk factor
• In Part D the basic payment formula is:
Payment = standardized bid  enrollee’s risk factor
– plan’s enrollee premium
Risk Adjuster Structure
• The CMS risk adjuster produces a personspecific risk factor by summing the factors
associated with characteristics such as:
° Age/sex, e.g., male 70-74, female 70-74 ….
° Originally disabled status, e.g., male 65+ who
entered Medicare < 65 because of disability
° Presence of medical conditions, e.g., cancers,
diabetes, liver disorders, ischemic heart
diseases, heart failure, psychiatric disorders …
Risk Adjuster Structure
• The risk model is prospective: the medical
conditions from a given year are used to predict
expenditures in the next year.
• The disease groupings are clusters of related ICD9-CM diagnosis codes.
• Diagnoses from hospital inpatient, hospital
outpatient, and clinician sources are used.
• The disease groups are derived from the CMSHCC risk model used in the Medicare Advantage
program today, but have been modified to reflect
drug spending as opposed to spending on A/B
services.
Risk Adjuster Structure
• The risk factors for disease groups are
additive when the diseases are not closely
related:
° inflammatory bowel disease, AMI,
schizophrenia
• The groups may be in hierarchies when
related and their costs have a logical ranking:
° Diabetes with complications, diabetes without
complications
° Only the highest coded group in a hierarchy
counts for payment.
Risk Adjuster Structure
Additive model: factors for demographic
characteristics + factors for diagnoses
Demographics
One of :
Female age 0-34 …
Female age 95+,
Male age 0-34 …
Male age 95+
Plus factor if aged
enrollee had once been
eligible for Medicare
based on disability
Male, Orig. disab.
Female Orig. disab.
Diagnoses if present
+ HIV/AIDS
+ Opportunistic Infections
+ one of:
Diabetes w. complications
Diabetes w/o complication
+ Congestive Heart Failure
+ ….
Risk Adjuster Development
• Principal data source: Drug claims for Federal
retirees in FEHB Blue Cross Blue Shield Service
Benefit Plan + Medicaid sample
• Diagnoses from linked Medicare files
° Diagnoses from 2001 to predict costs in 2002 and
similarly from 2000 to predict 2001
• Decedents in cost years were included till death.
• A linear regression model was estimated
° Dependent variable – total spending on prescription
drugs, annualized, mix of retail and mail order
° Explanatory variables – the array of person
characteristics - age/sex, diseases
° Estimated coefficients are the incremental costs of each
condition or demographic characteristic.
Risk Adjuster Development
• In an iterative process, the disease groups
were disassembled into smaller subgroups,
and reassembled to allow empirical
estimation of costs and clinical judgment to
weigh in the development.
• The explanatory power of the model is on a
par with other drug models reported (R2=.23
for spending) and is higher than similar
models for Parts A and B.
Risk Adjuster – rough example 1 for 2006
Coded
Characteristic
Female, age 76
Diabetes, w. complications
Diabetes, uncomplicated
High cholesterol
Congestive Heart Failure
Osteoporosis
Total Annual Pred. Spending
Spending Model
Spending
Relative
Increment
Factor
$ 850
.283
1,600
.533
1,000
.333
450
.150
650
.217
500
.167
$4,050
1.350
For implementation, dollar amounts are divided by the national
mean (~ $3,000) to create relative factors that multiply base rates.
Risk Adjuster for Plan Liability
• Plans are not liable for the total spending. Standard
benefit liability is 75% of spending from $250 to
$2250, about 15% above $5100, and 0% elsewhere.
• The structure developed for modeling spending has
been re-estimated on the same data set, with the
standard benefit structure applied to each enrollee’s
spending.
• This is the Plan Liability model. (R2=.25)
• The average projected 2006 spending in the data is
about $3,000; the average plan liability from the data
is about $1,100.
Risk Adjuster – rough example 2 for 2006
Coded
Characteristic
Female, age 76
Diabetes, w. complications
Diabetes, uncomplicated
High cholesterol
Congestive Heart Failure
Osteoporosis
Total Annual Pred. Spending
Liability Model
Payment
Relative
Increment
Factor
$ 550
.500
300
.273
200
.182
150
.136
250
.227
150
.136
$1,400
1.272
For implementation, dollar amounts are divided by the national
mean (~ $1,100) to create relative factors that multiply base rates.
Risk Adjuster for Plan Liability
• Tasks to be completed
° Development of New Enrollee model for people
new to Medicare with insufficient data for risk
adjustment. This model is based on demographics.
° Study of Medicaid population to investigate:
• factors for under-65 disabled population
• adjustments for institutional population
• adjustments for low-income population
Schedule
• Late December
° Release of Bidders Data Set - 5% files of FFS beneficiaries with
imputed drug utilization
• Late January
° Posting of draft model coefficients and mappings of diagnoses to
disease groups
° Release of drug risk factors for people in 5% files
• Mid February
° 45-day notice for MA with PDP information
• Late March
° Bid and Risk Adjustment training
• April
° Release of final model, software and final risk factors for 5% file if
different from draft