Refinements to the CMS-HCC Model For Risk Adjustment of Medicare Capitation Payments

1
Refinements to the CMS-HCC Model For Risk
Adjustment of Medicare Capitation Payments
Presented by John Kautter, Ph.D. Gregory Pope,
M.S. Eric Olmsted, Ph.D. RTI International
Contact John Kautter, PhD, jkautter_at_rti.org RTI
International is a trade name of Research
Triangle Institute.
2
History of Medicare Risk Adjustment

• Demographics (AAPCC)
• Doesnt explain cost variation
• Favorable selection gt higher program costs
• Principal inpatient diagnoses (PIP-DCG model,
  2000)
• Incentive to admit
• Penalizes plans that avoid admissions
• Inpatient and ambulatory diagnoses (2004)

3
CMS-HCC Model

• Centers for Medicare Medicaid Services (CMS)
  Hierarchical Condition Categories (HCC) model
• Prospective
• Inpatient and outpatient diagnoses w/o
  distinction
• 70 diagnostic categories (HCCs)
• Hierarchical within diseases

4
CMS-HCC Model (continued)

• Cumulative (additive) across diseases
• 6 disease interactions
• Discretionary diagnoses are excluded
• Demographic factors included
• Calibrated on 1999/2000 Medicare 5 Sample

5
CMS-HCC Model Performance

• Percentage of cost variation explained
• Age/Sex 0.8
• PIP-DCG 5.5
• CMS-HCC 10.0

6
CMS-HCC Models for Medicare Subpopulations

• Disabled
• End-stage renal disease
• Institutionalized
• New enrollees
• Secondary payer status
• Frail elderly

7
Disabled

• Over 10 of Medicare population
• Under age 65
• Model estimated separately for aged and disabled
• Overall cost patterns similar
• For 5 diagnostic categories, incremental expense
  of the disabled is higher
• 5 disease interactions for disabled in final
  CMS-HCC model

8
End-Stage Renal Disease

• About 1 of Medicare population
• Very expensive approximately 50,000/year
• 3-segment model
• Dialysis patients
• CMS-HCC model calibrated on dialysis patients
• Transplant period (3 months)
• Lump-sum payment
• Post-transplant period
• Aged/disabled CMS-HCC model w/add-on for drugs

9
Institutionalized Beneficiaries

• About 5 of Medicare population
• Costly, but less expensive than community
  residents for same diagnostic profile
• Combined CMS-HCC model
• Overpredicts costs for institutionalized
• Underpredicts costs for community frail elderly

10
Institutionalized Beneficiaries (continued)

• Different cost patterns by age and diagnosis for
  community and institutionalized
• CMS-HCC model calibrated separately on community
  and institutionalized
• Current year institutional status reported by
  nursing homes

11
New Enrollees

• Lack 12 months of base year enrollment
• Two-thirds are 65 year olds
• New enrollees versus continuing enrollees
• Much less costly at age 65
• Similar costs at other ages
• Merged new/continuing enrollee sample
• Separate cost weights for 65 year olds
• Demographic model

12
Medicare as Secondary Payer

• Beneficiaries with active employee
  employer-sponsored insurance
• Costs are lower
• Multiplier scales cost predictions down
• Multiplier is ratio of mean actual to mean
  predicted expenditures

13
Frail Elderly

• Diagnosis-based models underpredict expenditures
  for the functionally impaired
• Medicare specialty plans (e.g., PACE) serve
  functionally-impaired populations
• Frailty adjuster to better predict their costs
• Predicts costs unexplained by CMS-HCC
• Based on difficulties in ADLs
• ADLs collected from surveys or assessments

14
CMS-HCC Model Refinements

• Additional HCCs added to model
• 100 institutional sample used for institutional
  model calibration
• Changes in diagnostic classification
• 2002/2003 Medicare FFS data used for calibration
  of all models

15
Availability of Additional HCCs

• For Part D risk adjuster, plans required to
  submit diagnoses for 127 HCCs
• Additional 57 HCCs available for CMS-HCC models
  (127 70 57)

16
Adding HCCs

• Benefits
• Greater accuracy in predicting illness burden
• Rewards plans who enroll and treat beneficiaries
  with these diagnoses
• E.g., Special Needs Plans (SNPs)
• Drawbacks
• Creates greater opportunities for diagnostic
  upcoding

17
HCCs Added to CMS-HCC Model

• Available additional HCCs reviewed by project
  team to determine which were appropriate for
  payment model
• Number of HCCs increased from 70 to 101

18
Examples of HCCs Added to CMS-HCC Model

• Refined CMS-HCC Model
• HCC Community Institutional
• Type I
• Diabetes
• Mellitus 1,557 1,435
• Dementia/
• Cerebral
• Degeneration 1,576 - -
• Hypertension 388 919

19
100 Institutional Sample

• CMS-HCC institutional model calibrated on 5
  institutional sample (n 65,593)
• To increase statistical accuracy and stability,
  refined CMS-HCC institutional model calibrated
  on 100 institutional sample (n
  1,238,842)

20
Distribution of Annualized Medicare Expenditures,
2003

• 5 Community 100 Institutional
• Sample Size 1,380,978 1,238,842
• Expenditures
• Mean 6,541 11,252
• 95th Percentile 31,285 47,390
• 90th Percentile 17,682
  31,553
• Median 1,445 3,028
• 10th Percentile 56
  538
• 5th Percentile 0
  349

21
Changes in Diagnostic Classification

• Diabetes complications moved to diabetes
  hierarchy
• E.g., diabetic neuropathy moved from HCC 71
  Polyneuropathy to HCC 16 Diabetes with Neurologic
  or Other Specified Manifestation
• HCC 119 Proliferative Diabetic Retinopathy and
  Vitreous Hemorrhage deleted and most moved to HCC
  18 Diabetes with Ophthalmologic or Unspecified
  Manifestation
• Cerebral Palsy consolidated in HCC 70 Cerebral
  Palsy and Muscular Distrophy

22
Refined CMS-HCC Community and Institutional Models

• of Cost
• Variation
• Explained HCCs
• CMS-HCC
• Community 9.8 70
• Institutional 6.0 69
• Refined CMS-HCC
• Community 11.0 101
• Institutional 8.9 90

23
Refined CMS-HCC Model Performance I

• Predictive ratios, prior year expenditure
  quintiles
• Age/Sex CMS-HCC
• First 2.65 1.20
• Second 1.82 1.19
• Third 1.31 1.09
• Fourth 0.91 0.99
• Fifth 0.46 0.90

24
Refined CMS-HCC Model Performance II

• Predicted ratios by CMS-HCC predicted expenditure
  deciles
• Age/Sex CMS-HCC
• First 2.84 0.88
• Second 2.43 0.92
• Third 2.10 0.94
• Fourth 1.70 0.97
• Fifth 1.49 0.97
• Sixth 1.27 1.00
• Seventh 1.06 1.01
• Eighth 0.86 1.04
• Ninth 0.64 1.04
• Tenth 0.35 1.00

25
Conclusions

• Medicare risk adjustment has been evolving
• Demographic ? Inpatient ? All-Encounter
• (AAPCC) (PIP-DCG) (CMS-HCC)
• The refined CMS-HCC model represents a more
  comprehensive all-encounter risk adjustment model
• Increases payment accuracy for plans
• Viability of plans
• Beneficiaries access to plans