Risk Adjustment Models for Medicare Part D Capitation Payments Modeling John Kautter Melvin J. Ingber Gregory C. Pope Sara Freeman RTI International AcademyHealth 2011 RTI International is a trade name of Research Triangle Institute. www.rti.org
Credits Funding Source: Centers for Medicare & Medicaid Services (CMS) Melissa Evans, Project Officer 2
Medicare Payments to Part D Plans The Medicare program contracts with drug plans to provide Part D covered prescription drugs to Medicare beneficiaries enrolled in the plans For each beneficiary the plan enrolls, Medicare pays the plan a monthly capitation payment The capitation payment is risk adjusted to account for the health status of the beneficiary Risk adjustment is important because it allows Medicare to pay plans more accurately 3
Simplified Example of Risk Adjusted Capitation Payment Suppose capitation payment for a beneficiary with average health status is $100 If the beneficiary is expected to be twice as costly as average, then his/her risk score is 2.0, and risk adjusted capitation payment is $100 x 2.0 = $200 If the beneficiary is expected to be half as costly as average, then his/her risk score is 0.5, and risk adjusted capitation payment is $100 x 0.5 = $50 4
Where Do Risk Scores Come From? First, a prospective risk adjustment model is estimated Regression model Dependent (left hand side) variable is beneficiaries prescription drug expenditures in Year 2 that the plan is liable for, i.e., plan liability expenditures Independent (right hand side) variables are beneficiaries demographics and Year 1 diagnoses Diagnoses are grouped into disease clusters called Rx-HCCs Low income subsidy and nursing home residence status are used as well Second, a given beneficiary s characteristics are used in conjunction with the estimated risk adjustment model to calculate his/her risk score 5
Simplified Example of Risk Score Calculation Beneficiary is male, age 77, with the chronic conditions congestive heart failure, diabetes with complications, and chronic obstructive pulmonary disease Coefficients from prospective risk adjustment model are: Male age 77 = $500 CHF = $200 Diabetes w/ comp = $300 COPD = $200 Beneficiary s predicted expenditures are $1,200 Average expenditures for all beneficiaries are $1,000 Beneficiary s risk score = $1,200 / $1,000 = 1.2 6
Modeling Data, Variables, Sample 2007 2008 100% Medicare administrative data Part D prescription drug event (PDE) claims files Common Medicare Environment (CME) Health plan management system (HPMS) Variables 2008 annualized plan liability expenditures 78 Rx Hierarchical Condition Categories (Rx-HCCs) 24 age/sex categories 2 originally disabled categories Part D low income subsidy status Long term institutional (nursing home) status 7
Modeling Data, Variables, Sample (cont.) 100% of Medicare aged, disabled, and ESRD beneficiaries meeting the following criteria: Enrolled in a Part D standalone PDP plan for at least one month in 2008 Continuously enrolled in traditional fee-for-service Medicare in 2007 Other criteria 8
Combined Part D Risk Adjustment Model Sample includes: Long-term institutional and community-residing Low-income and non-low income Aged (age>=65) and Non-Aged (age<65) N = 14,224,301 Mean expenditures = $1,147 R2 = 0.2640 Age/sex coefficients range from $450 to $800 Higher for Non-Aged Rx-HCC median coefficient is $196 Range from $13 to $2,688 9
Issue 1 Long Term Institutional LTI beneficiaries comprise 7.1% of the combined sample LTI beneficiaries 57% more expensive than community beneficiaries ($1,748 vs. $1,146) Combined model under-predicts plan liability expenditures for LTI beneficiaries by 12% Higher drug expenditures for LTI residents given disease burden could result from a number of factors: Greater drug adherence Higher drug prices for specially packaged drugs 10
Issue 1 Long-term Institutional (cont.) Examined risk adjustment models estimated separately on LTI and community samples LTI age/sex coefficients higher than community $1,000 - $1,400 higher for <65 age categories $700 - $1,100 higher for >=65 age categories LTI Rx-HCC coefficients tend to be lower than community 70 out of 78 are lower (90%) One possible reason is drug therapies for community beneficiaries with same condition could be more intensive and newer and more expensive 11
Selected Rx-HCC Coefficients from Separate Community and LTI Models Rx-HCC Community LTI Difference HIV/AIDS $2,706 $2,004 $703 Diabetes with complications 291 171 120 Alzheimer`s disease 416 12 404 Schizophrenia 944 466 478 Congestive heart failure Chronic obstructive pulmonary disease and asthma 197 111 86 228 128 99 12
Issue 2 Low Income Subsidy LIS beneficiaries comprise 45% of the community sample LIS beneficiaries 30% more expensive than non-lis ($1,274 vs. $977) Community model under-predicts plan liability expenditures for LIS beneficiaries by 3% Higher drug expenditures for LIS beneficiaries given disease burden could result from various factors: Induced demand resulting from lower cost sharing that LIS beneficiaries are entitled do 13
Issue 2 Low Income Subsidy (cont.) Examined risk adjustment models estimated separately on LIS and non-lis samples LIS age/sex coefficients higher than non-lis $100 - $200 higher for <65 age categories About same for >=65 age categories LIS Rx-HCC coefficients tend to be higher than non-lis 65 out of 78 are higher (83%) One possible reason is drug therapies for LIS beneficiaries with same condition could include more brand name drugs 14
Selected Rx-HCC Coefficients from Separate Low-Income and Non-Low Income Models Non-Low Rx-HCC Low Income Income Difference Chronic myeloid $2,419 $1,870 $550 leukemia Bipolar disorders 683 414 269 Autism 715 278 437 Multiple sclerosis 1,319 688 632 Cystic fibrosis 1,205 282 924 Dialysis status 449 245 204 15
Issue 3 Non-Aged (Age<65) Low Income Subsidy Sample Non-Aged comprise 55% of sample Non-Aged 24% more expensive than Aged ($1,426 vs. $977) Separate Non-Aged and Aged Models Rx-HCCs tend to be higher for Non-Aged (59 out of 78, or 78%) Non-Low Income Subsidy Sample Non-Aged beneficiaries comprise 8% of sample Non-Aged 13% more expensive than Aged ($1,096 vs. $967) Separate Non-Aged and Aged Models Rx-HCCs tend to be higher for Non-Aged (53 out of 78, or 68%) 16
Part D Risk Adjustment Models Implemented in 2011 Community, non-lowincome, age 65 Community, non-lowincome, age < 65 Community, low income, age 65 Community, low income, age < 65 Institutional Sample size 6,901,504 568,164 3,404,054 2,773,904 1,012,548 Mean plan liability expenditures $967 $1,096 $1,147 $1,426 $1,762 R-Squared 0.1951 0.1580 0.2608 0.2898 0.1274 17
Ongoing Research Research activities are ongoing to improve the Part D risk adjustment model Use of information on a beneficiary s prescription drug utilization (therapeutic classes) to predict plan liability expenditures Tradeoff between predictive accuracy and potential gaming Use of information on a beneficiary s disease interactions to predict plan liability expenditures Contraindications, side effects Recalibration of model on more recent Part D prescription drug event data 18
For More Information John Kautter jkautter@rti.org 781-434-1723 www.rti.org 19