Managed Care and Medicare Expenditures

1
Managed Care and Medicare Expenditures

• Health Economics Interest Group
• Seattle, WA
• Michael Chernew
• Phil DeCicca
• Robert Town
• June 24, 2006

2
Overview

• Background
• Data and Analysis Sample
• Empirical Strategy
• Results
• Tentative Conclusions

3
Background

• Investigate the existence and extent of managed
  care spillovers in Medicare
• We examine the impact of county-specific Medicare
  HMO penetration on spending of FFS Medicare
  beneficiaries
• In particular, we try to identify the impact of
  within-county changes in MC HMO penetration on
  spending

4
Background (cont)

• Existence of spillovers assumes connected
  markets
• Many pathways for spillovers
• Increased competition
• Changes in structure of delivery system
• Changes in practice patterns
• Previous work suggests spillovers exist
• Baker (1997, 1999) Bundorf et al. (2004)

5
Data and Sample Info

• Medicare Current Beneficiary Study (MCBS)
• Cost and Use Files, 1994 to 2001
• Analysis Sample
• Exclude individuals administered a Facility
  interview
• Exclude individuals enrolled in HMOs
• Exclude counties that contribute less than two
  cases per year, on average
• Yields 60,067 cases from 293 counties
• Including 2.5 with zero expenditure

6
Key Variables

• MCBS Variables
• Per-Person Total Annual Spending
• Various Broad Measures of Utilization
• Covariates including usual suspects and more
  detailed measures of health status
• County-level Variables
• Medicare HMO Penetration
• Payment Rates (AAPCC)

7
Empirical Strategy

• We Estimate Models of the Form
• Log(Spend)ictd(MCHMO)ctXßµcateict
• X depends on specification
• µ and a are County and Year effects
• dlt0 implies the existence of spillover

8
Empirical Strategy (Details)

• Two Models EstimatedShort Long
• Estimate models with and without zeroes
• Models estimated via OLS and IV
• We use the payment rate (AAPCC) and its square as
  instruments for HMO penetration
• As will see, strong relationship between payment
  rate and penetration

9
Estimates

• In general, OLS estimates practically small
• For example,
• Largest estimated effect suggests that a one
  percentage point increase in MC HMO penetration
  leads to an 0.3 percent decrease in spending by
  FFS beneficiaries
• Reduction ranges from 0.2 to 0.3 percent,
  depending on specification

10
Estimates (cont)

• OLS estimates, however, may be biased
• E.g., HMOs may enter areas based on cost growth
  or characteristics correlated with it
• Sorting into high cost growth areas would tend to
  attenuate measured spillover effects
• Sorting into low cost growth areas would tend to
  overstate the magnitude of spillovers

11
Estimates (cont)

• Overview of Remaining Estimates
• IV (First Stage)
• IV (Structural Equation)
• Utilization Models
• Sensitivity Checks
• Where are savings being generated?
• High-Use vs. Low-Use Beneficiaries

12
Estimates (cont)
Short Long
Payment Rate -0.00137 (7.76) -0.00137 (7.81)
(Payment Rate)2 0.0000167 (6.06) 0.0000167 (6.11)
--First-stage estimates strong in all specs. --Partial R2 0.14 and First-Stage F-stats 33. --Upshot Payment rate very strong predictor --First-stage estimates strong in all specs. --Partial R2 0.14 and First-Stage F-stats 33. --Upshot Payment rate very strong predictor --First-stage estimates strong in all specs. --Partial R2 0.14 and First-Stage F-stats 33. --Upshot Payment rate very strong predictor
13
Estimates (cont)
Short Long
Without Zeroes -0.0165 (3.10) -0.0138 (3.19)
With Zeroes -0.0183 (2.53) -0.0142 (2.52)
--IV estimates (d) from four separate models --Std. errors adjusted for clustering at county level --IV estimates (d) from four separate models --Std. errors adjusted for clustering at county level --IV estimates (d) from four separate models --Std. errors adjusted for clustering at county level
14
Estimates (cont)

• Interpretation
• Estimates suggest a one pct. point increase in
  HMO penetration leads to between 1.3 and 1.8
  percent reduction in spending by FFS
  beneficiaries
• (Perceived) Magnitudes
• Estimates perhaps not as large as seem when
  consider that a one pct point increase in
  penetration is off a base of 9-10 pct pts
• Many IV Diagnostics
• All suggest that IV strategy is legitimate

15
Estimates (cont)

• Next Step Estimate Utilization Models
• Here, we use broad utilization categories as
  dependent variables
• We find increases in MC HMO penetration reduce
  Inpatient and Outpatient events, especially
  on intensive margins
• Supports our spending estimates which suggest
  non-trivial spillover

16
Estimates (cont)

• Next Check Sensitivity of Main Estimates
• Est. models without CA and FL counties
• Est. models with Supplemental HI controls
• Compare effect on High vs. Low-Use
• Define high-use as FFS with 1 chronic
  condition (CC) low-use as those with no CCs.
• CCs include Diabetes, HBP, Arthritis, Heart
  Disease and Other Heart Problems
• Est. Spending Models Separately for Two Groups

17
Estimates (cont)
Without Zeroes With Zeroes
CCgt0 (High-Use) -0.0159 (3.63) -0.0179 (3.22)
CC0 (Low-Use) -0.0033 (0.24) 0.0054 (0.29)
--IV estimates from Long model reported (d) --Estimates from Short model similar --IV estimates from Long model reported (d) --Estimates from Short model similar --IV estimates from Long model reported (d) --Estimates from Short model similar
18
Estimates (cont)

• Chronic Conditions Models Details
• Results suggest main spending estimates driven by
  relatively high-use individuals
• In particular, estimates imply 1.6 to 2.3 percent
  drop in FFS spending for high-use and virtually
  no effect for those without CCs
• Perhaps not too surprising as high-use
  individuals spending 2X low-users

19
Summary

• We find evidence MC HMO penetration reduces
  spending by FFS beneficiaries
• Evidence that MC HMO penetration reduces
  utilization supports spending reductions
• Spending reductions seem to be derived from
  high-use individuals

20
THE END