Variation in Health Care Spending: Target Decision Making, Not Geography (2013)
Chapter: Appendix D: Regression Model Specifications with "Clusters" of Predictors
Appendix D
Regression Model Specifications with “Clusters” of Predictors
Cluster 1: Partial-year enrollment (PYE), year, pharmacy benefit, age, sex, age–sex interaction (all subcontractors)
Cluster 2: Health status (All subcontractors) – BASELINE MODEL
a) PYE, year, pharmacy benefit, age, sex, age–sex interaction
b) Health status
Cluster 3: Race (All subcontractors)
a) PYE, year, pharmacy benefit, age, sex, age–sex interaction
b) Race
Cluster 4: Income (All subcontractors)
a) PYE, year, pharmacy benefit, age, sex, age–sex interaction
b) Income
Cluster 5: All common beneficiary independent variables (all subcontractors)
a) PYE, year, pharmacy benefit, age, sex, age–sex interaction
b) Health status
c) Race
d) Income
Cluster 6: Employer/insurance (Lewin and Harvard only)
a) PYE, year, pharmacy benefit, age, sex, age–sex interaction
b) Benefit generosity and payer/plan type (Harvard and Lewin only)
c) Data source (Harvard only)
d) Plan size (Lewin only)
Cluster 7: Market (all subcontractors)
a) PYE, year, pharmacy benefit, age, sex, age–sex interaction
b) Access to care, payer mix, hospital competition, supply of medical services, % population uninsured, malpractice environmental risk (all)
c) Physician competition (Harvard and Acumen)
d) Health professional mix (Lewin and Acumen)
e) Percentage of Medicare beneficiaries with supplemental insurance
f) MA/Medicaid penetration (Acumen only)
Cluster 8: Kitchen sink variables (Subcontractors included all independent variables)
All subcontractors
a) PYE, year, pharmacy benefit, age, sex, age–sex interaction
b) Race
c) Income
d) Health status and comorbidity adjuster
e) Access to care
f) Payer mix
g) Hospital competition
h) Supply of medical services
i) % population uninsured
j) Malpractice environment risk
Harvard additionally adjusted for:
• Health
behavior
• Data source
• Payer/plan type
• Benefit generosity
• Physician composition
Acumen additionally adjusted for:
• Medicare and Medicaid dual eligibility status
• Supplemental Medicare insurance
• Fee-for-service or Medicare Advantage
• Part D enrollment
• Percentage of Medicare beneficiaries with supplemental insurance
• MA/Medicaid penetration
• Physician composition
• Health professional mix
Lewin additionally adjusted for:
• Payer/plan type
• Benefit generosity
• Plan size
• Health professional mix
Cluster 9: All common independent variables only (all subcontractors)
a) PYE, year, pharmacy benefit, age, sex, age–sex interaction
b) Race
c) Income
d) Health status and comorbidity adjuster
e) Access to care
f) Payer mix
g) Hospital competition
h) Supply of medical services
i) % population uninsured
j) Malpractice environment risk
Cluster 10: Cluster 1 and Medicare only IVs (Acumen only)
a) PYE, year, pharmacy benefit, age, sex, age–sex interaction
b) Medicare and Medicaid dual eligibility status
c) Supplemental Medicare insurance
d) Part D enrollment
e) Fee for service or Medicare Advantage
