optimal health insurance for prevention and treatmentblogs.bu.edu › ellisrp › files › 2012 ›...
TRANSCRIPT
![Page 1: Optimal Health Insurance for Prevention and Treatmentblogs.bu.edu › ellisrp › files › 2012 › 02 › 2007-EllisManning_JHE_EJM_combined.pdfOptimal Health Insurance for Prevention](https://reader033.vdocuments.mx/reader033/viewer/2022060418/5f153817f7bcb9108f680e31/html5/thumbnails/1.jpg)
Optimal Health Insurance for Prevention and Treatment
Randall P. Ellis Department of Economics
Boston University
Willard G. ManningHarris School of Public Policy Studies
The University of Chicago
We thank Shenyi Jiang, Albert Ma,Tom McGuire, Joe Newhouse, Frank Sloan and seminar participants at Boston University and Harvard for their comments.
![Page 2: Optimal Health Insurance for Prevention and Treatmentblogs.bu.edu › ellisrp › files › 2012 › 02 › 2007-EllisManning_JHE_EJM_combined.pdfOptimal Health Insurance for Prevention](https://reader033.vdocuments.mx/reader033/viewer/2022060418/5f153817f7bcb9108f680e31/html5/thumbnails/2.jpg)
Rich literature on optimal health insurance
Zeckhauser JET (1970) is classic article describing tradeoff between risk spreading and moral hazard
Conventional approach emphasizes:
Demand responsiveness
Variance in health care spending
Degree of risk aversion
Insurance loading factor
Focus of paper is on demand side incentives, ignoring supply-side moral hazard and supply-side cost sharing
![Page 3: Optimal Health Insurance for Prevention and Treatmentblogs.bu.edu › ellisrp › files › 2012 › 02 › 2007-EllisManning_JHE_EJM_combined.pdfOptimal Health Insurance for Prevention](https://reader033.vdocuments.mx/reader033/viewer/2022060418/5f153817f7bcb9108f680e31/html5/thumbnails/3.jpg)
New questions answered in this paper
How should optimal insurance coverage be modified in the presence of
Preventive care services?
Multiple health care goods with correlated errors?
Cross price elasticities of demand?
Uncompensated losses that are correlated with health spending?
Multiple time periods when health care spending is serially
correlated over time?
![Page 4: Optimal Health Insurance for Prevention and Treatmentblogs.bu.edu › ellisrp › files › 2012 › 02 › 2007-EllisManning_JHE_EJM_combined.pdfOptimal Health Insurance for Prevention](https://reader033.vdocuments.mx/reader033/viewer/2022060418/5f153817f7bcb9108f680e31/html5/thumbnails/4.jpg)
Summary of what paper shows
1. Preventive care should be covered generously because consumers will
ignore premium savings in their private prevention decisions.
2. Health care goods that are positively correlated with other health
spending should be more generously covered
3. Health care goods that are substitutes should be covered more
generously than complements.
4. Cover services generously that are positively correlated with
uncompensated losses
5. Services that are positively serially correlated over time should be
covered more generously than those that are uncorrelated
![Page 5: Optimal Health Insurance for Prevention and Treatmentblogs.bu.edu › ellisrp › files › 2012 › 02 › 2007-EllisManning_JHE_EJM_combined.pdfOptimal Health Insurance for Prevention](https://reader033.vdocuments.mx/reader033/viewer/2022060418/5f153817f7bcb9108f680e31/html5/thumbnails/5.jpg)
Optimal insurance coverage for preventive care has received relatively little attention
Low variance suggests insurance not needed (Kenkel, AEA presentation 2007)
Demand elasticity low relative to curative care (uncertain)?
Insurance reduces value of prevention
Primary versus secondary prevention
Boundary between prevention and treatment is sometimes blurry (e.g. drugs, preventive treatment)
We define prevention activities as costly services to the consumer that reduce the probability of being sick.
![Page 6: Optimal Health Insurance for Prevention and Treatmentblogs.bu.edu › ellisrp › files › 2012 › 02 › 2007-EllisManning_JHE_EJM_combined.pdfOptimal Health Insurance for Prevention](https://reader033.vdocuments.mx/reader033/viewer/2022060418/5f153817f7bcb9108f680e31/html5/thumbnails/6.jpg)
Five types of losses modeled
Ex ante moral hazard loss from too little prevention
Ex post moral hazard loss from overinsurance
Inefficiency due to insurance loading factors
Arrow-Pratt cost of risk
Uncompensated losses due to ill health
![Page 7: Optimal Health Insurance for Prevention and Treatmentblogs.bu.edu › ellisrp › files › 2012 › 02 › 2007-EllisManning_JHE_EJM_combined.pdfOptimal Health Insurance for Prevention](https://reader033.vdocuments.mx/reader033/viewer/2022060418/5f153817f7bcb9108f680e31/html5/thumbnails/7.jpg)
Organization of paper
Literature review
Basic model – use consumer surplus measure
no prevention, homoscedastic error
Utility-based models
Prevention + one health treatment good
Key role of uncompensated health losses
Multiple health treatment goods
Multiple periods
Discussion and empirical relevance
![Page 8: Optimal Health Insurance for Prevention and Treatmentblogs.bu.edu › ellisrp › files › 2012 › 02 › 2007-EllisManning_JHE_EJM_combined.pdfOptimal Health Insurance for Prevention](https://reader033.vdocuments.mx/reader033/viewer/2022060418/5f153817f7bcb9108f680e31/html5/thumbnails/8.jpg)
Huge literature potentially related to these questions – Most of Audience in this room!
Optimal deductibles and copayments
Demand responsiveness of specific services
Optimal insurance with multiple goods
Dynamic models of health spending
Prevention
Statistical models of the distribution of health care costs over time
![Page 9: Optimal Health Insurance for Prevention and Treatmentblogs.bu.edu › ellisrp › files › 2012 › 02 › 2007-EllisManning_JHE_EJM_combined.pdfOptimal Health Insurance for Prevention](https://reader033.vdocuments.mx/reader033/viewer/2022060418/5f153817f7bcb9108f680e31/html5/thumbnails/9.jpg)
Literature is ambiguous on insurance coverage for preventive care
Low variance suggests insurance not needed (Kenkel, AEA presentation 2007)
Demand elasticity low relative to curative care (uncertain)?
Insurance reduces value of prevention (ex ante moral hazard)
Primary versus secondary prevention
Boundary between prevention and treatment is sometimes blurry (e.g. drugs, preventive treatment)
We define prevention activities as costly activities to the consumer that reduce the probability of being sick.
![Page 10: Optimal Health Insurance for Prevention and Treatmentblogs.bu.edu › ellisrp › files › 2012 › 02 › 2007-EllisManning_JHE_EJM_combined.pdfOptimal Health Insurance for Prevention](https://reader033.vdocuments.mx/reader033/viewer/2022060418/5f153817f7bcb9108f680e31/html5/thumbnails/10.jpg)
![Page 11: Optimal Health Insurance for Prevention and Treatmentblogs.bu.edu › ellisrp › files › 2012 › 02 › 2007-EllisManning_JHE_EJM_combined.pdfOptimal Health Insurance for Prevention](https://reader033.vdocuments.mx/reader033/viewer/2022060418/5f153817f7bcb9108f680e31/html5/thumbnails/11.jpg)
Basic model
One health care good X
Linear demand curve when sick of X = µX – B PX + θ
Θ = random shock affecting demand for X (also health status shock)
Marginal cost = 1
PX = cX = cost share of health care treatment
![Page 12: Optimal Health Insurance for Prevention and Treatmentblogs.bu.edu › ellisrp › files › 2012 › 02 › 2007-EllisManning_JHE_EJM_combined.pdfOptimal Health Insurance for Prevention](https://reader033.vdocuments.mx/reader033/viewer/2022060418/5f153817f7bcb9108f680e31/html5/thumbnails/12.jpg)
Adding in insurance and cost of risk
Use Arrow-Pratt approximation for welfare loss from financial risk
Cost of risk = (RA)(variance of out-of-pocket spending)/2
Assume probability of sickness = 1 - (Z)
Health spending only if sick
Insurance loading factor δ implies extra cost of using insurance funding is
= δ (1-(Z)) (1-cX) (µX – B cX)
Uncompensated loss from insurance = L
![Page 13: Optimal Health Insurance for Prevention and Treatmentblogs.bu.edu › ellisrp › files › 2012 › 02 › 2007-EllisManning_JHE_EJM_combined.pdfOptimal Health Insurance for Prevention](https://reader033.vdocuments.mx/reader033/viewer/2022060418/5f153817f7bcb9108f680e31/html5/thumbnails/13.jpg)
Ex post moral hazard loss from overinsurance = 2(1 / 2) (1 - )Xc B
Inefficiency due to insurance loading factors = (1 ) ( )X X Xc Bc
Arrow-Pratt cost of risk = 2 2 2AXR c
Uncompensated losses due to ill health = ( )L , with expectation L
Four terms in loss function to be minimized
Min
1 2 2 2(1/ 2)(1- ) (1 )( ) 2AX X X X XWL c B c Bc R c L
*2
2X
X A
B BcB B R
![Page 14: Optimal Health Insurance for Prevention and Treatmentblogs.bu.edu › ellisrp › files › 2012 › 02 › 2007-EllisManning_JHE_EJM_combined.pdfOptimal Health Insurance for Prevention](https://reader033.vdocuments.mx/reader033/viewer/2022060418/5f153817f7bcb9108f680e31/html5/thumbnails/14.jpg)
![Page 15: Optimal Health Insurance for Prevention and Treatmentblogs.bu.edu › ellisrp › files › 2012 › 02 › 2007-EllisManning_JHE_EJM_combined.pdfOptimal Health Insurance for Prevention](https://reader033.vdocuments.mx/reader033/viewer/2022060418/5f153817f7bcb9108f680e31/html5/thumbnails/15.jpg)
Results are interesting and plausible and many parameters can be calculated empirically
BUT
Does not suggest how to incorporate multiple goods
Does not expand to multiple periods well
Does not incorporate risk aversion in a fully satisfying way
Only approximates a utility maximization framework
![Page 16: Optimal Health Insurance for Prevention and Treatmentblogs.bu.edu › ellisrp › files › 2012 › 02 › 2007-EllisManning_JHE_EJM_combined.pdfOptimal Health Insurance for Prevention](https://reader033.vdocuments.mx/reader033/viewer/2022060418/5f153817f7bcb9108f680e31/html5/thumbnails/16.jpg)
Utility-based model
Three kinds of goods
X = health care goods indexed by
Y = all other consumption goods
Z = spending on preventive care
PX, PY, PZ = prices of three types of goods
π = insurance premium
I = Income
I = π +PXX + PYY + PZ Z
![Page 17: Optimal Health Insurance for Prevention and Treatmentblogs.bu.edu › ellisrp › files › 2012 › 02 › 2007-EllisManning_JHE_EJM_combined.pdfOptimal Health Insurance for Prevention](https://reader033.vdocuments.mx/reader033/viewer/2022060418/5f153817f7bcb9108f680e31/html5/thumbnails/17.jpg)
Sequence of moves
1. The insurer chooses coinsurance rates cX and cZ for treatment and
prevention.
2. The consumer chooses Z to maximize ex ante utility in Period 1 by
borrowing its cost against Period 1 income.
3. Nature decides on the consumer’s state of illness θ
4. The consumer chooses X and Y
5. If a dynamic model, then repeat steps 3 and 4 (prevention is done one
time in Period 0).
![Page 18: Optimal Health Insurance for Prevention and Treatmentblogs.bu.edu › ellisrp › files › 2012 › 02 › 2007-EllisManning_JHE_EJM_combined.pdfOptimal Health Insurance for Prevention](https://reader033.vdocuments.mx/reader033/viewer/2022060418/5f153817f7bcb9108f680e31/html5/thumbnails/18.jpg)
Expected utility framework used
E U = Eθ U(Y, H(X, θ))
Three key assumptions about utility function
1. UYH < 0
Favorable health care shocks reduce the marginal utility of income
2. X/I = 0
Zero income effects, so Consumer Surplus = Compensating Variation for health goods
3. Linear demand curves for X
X = X – B PX +
![Page 19: Optimal Health Insurance for Prevention and Treatmentblogs.bu.edu › ellisrp › files › 2012 › 02 › 2007-EllisManning_JHE_EJM_combined.pdfOptimal Health Insurance for Prevention](https://reader033.vdocuments.mx/reader033/viewer/2022060418/5f153817f7bcb9108f680e31/html5/thumbnails/19.jpg)
Assumed Effect of Income Changes on Optimal Consumption Bundles
Medical Service (Xi)
Oth
er G
oods
(Y)
Ideal ICC
Assumed ICC
![Page 20: Optimal Health Insurance for Prevention and Treatmentblogs.bu.edu › ellisrp › files › 2012 › 02 › 2007-EllisManning_JHE_EJM_combined.pdfOptimal Health Insurance for Prevention](https://reader033.vdocuments.mx/reader033/viewer/2022060418/5f153817f7bcb9108f680e31/html5/thumbnails/20.jpg)
Demand for each health service when sick:
X
y
XPX A BP
2
2
Use Roy's identity to derive risk neutral utility function when sick
( , ) S X X
Y Y Y
P PIV I P A BP P P
Introduce stochastic element to demand where ( ) with ( ) 0XA F E
![Page 21: Optimal Health Insurance for Prevention and Treatmentblogs.bu.edu › ellisrp › files › 2012 › 02 › 2007-EllisManning_JHE_EJM_combined.pdfOptimal Health Insurance for Prevention](https://reader033.vdocuments.mx/reader033/viewer/2022060418/5f153817f7bcb9108f680e31/html5/thumbnails/21.jpg)
Full model with prevention
S 1 2
1 2
X X X Z
( ) ( ) (1 ( )) ( ) ( )
where
( ) , and2
1 1 ( ) 1 1 Z
HX
Z
XX X
EV Z V J Z E V J K c L L
J I c Z
B c LK c
Z c Bc c
Prevention increases probability of being healthy
Prevention reduces treatment costs covered by premium
Prevention costs can either be paid out of pocket or covered by insurance
Benefits of health treatment are quadratic in prices, cX ,+L1
Health shocks theta affect out-of-pocket costs
Health shocks theta also affect uncompensated losses of two types: monetary and health
Insurance loading factor increases premium by proportion delta
![Page 22: Optimal Health Insurance for Prevention and Treatmentblogs.bu.edu › ellisrp › files › 2012 › 02 › 2007-EllisManning_JHE_EJM_combined.pdfOptimal Health Insurance for Prevention](https://reader033.vdocuments.mx/reader033/viewer/2022060418/5f153817f7bcb9108f680e31/html5/thumbnails/22.jpg)
Socially optimal choice of preventive care ZSOC
differs from private choice ZPRIV in that private choice ignores change in premium
SOC PRIV
SOC
S S
SOC PRIV
Compare and
'( ) '( ) = , =
if ''( ) 0 Private choice resul
Z PRIVZ
H HI I
Z Z
MCPZ Z
E V E VV E V V E V
Z Z
Z
Z Zts in too little prevention when P = MCTherefore prevention services should be subsidized
![Page 23: Optimal Health Insurance for Prevention and Treatmentblogs.bu.edu › ellisrp › files › 2012 › 02 › 2007-EllisManning_JHE_EJM_combined.pdfOptimal Health Insurance for Prevention](https://reader033.vdocuments.mx/reader033/viewer/2022060418/5f153817f7bcb9108f680e31/html5/thumbnails/23.jpg)
*
S
= 1 + 1
where
Z
Z
Z X X X
HI
P QcMC Q c Bc
Q V E V E V
Optimal cost sharing on preventive care
![Page 24: Optimal Health Insurance for Prevention and Treatmentblogs.bu.edu › ellisrp › files › 2012 › 02 › 2007-EllisManning_JHE_EJM_combined.pdfOptimal Health Insurance for Prevention](https://reader033.vdocuments.mx/reader033/viewer/2022060418/5f153817f7bcb9108f680e31/html5/thumbnails/24.jpg)
Model with multiple goods and no prevention
Demand curves for two health treatment goods
1 1 1 1 12 2 1
2 2 2 2 12 1 2
- -
X B c G cX B c G c
![Page 25: Optimal Health Insurance for Prevention and Treatmentblogs.bu.edu › ellisrp › files › 2012 › 02 › 2007-EllisManning_JHE_EJM_combined.pdfOptimal Health Insurance for Prevention](https://reader033.vdocuments.mx/reader033/viewer/2022060418/5f153817f7bcb9108f680e31/html5/thumbnails/25.jpg)
Optimal cost share One health treatment good, no insurance loading
C1* increases with B1 and
decreases with RA, 12
Confirms conventional results
* 11 2
1 1A
BcB R
![Page 26: Optimal Health Insurance for Prevention and Treatmentblogs.bu.edu › ellisrp › files › 2012 › 02 › 2007-EllisManning_JHE_EJM_combined.pdfOptimal Health Insurance for Prevention](https://reader033.vdocuments.mx/reader033/viewer/2022060418/5f153817f7bcb9108f680e31/html5/thumbnails/26.jpg)
Optimal cost share One health treatment good, Insurance loading factor >0
C1* increases with and 1
* 1 1 11 2
1 1 12 A
B BcB B R
![Page 27: Optimal Health Insurance for Prevention and Treatmentblogs.bu.edu › ellisrp › files › 2012 › 02 › 2007-EllisManning_JHE_EJM_combined.pdfOptimal Health Insurance for Prevention](https://reader033.vdocuments.mx/reader033/viewer/2022060418/5f153817f7bcb9108f680e31/html5/thumbnails/27.jpg)
S 12
1 11 1 1
2
S B LcEV E V I c c
Optimal cost share One health treatment good Uncompensated losses from health shocks
2 1* 1 1 11 2
1 1
A
A
B R LcB R
c1* decreases with uncompensated losses and can be
zero or negative at optimum
![Page 28: Optimal Health Insurance for Prevention and Treatmentblogs.bu.edu › ellisrp › files › 2012 › 02 › 2007-EllisManning_JHE_EJM_combined.pdfOptimal Health Insurance for Prevention](https://reader033.vdocuments.mx/reader033/viewer/2022060418/5f153817f7bcb9108f680e31/html5/thumbnails/28.jpg)
Optimal cost share Two health treatment goods Correlated health shocks and nonzero cross price elasticities
As limiting case, c1* should be lower for services with
positively correlated health shocks
As limiting case, c1* should be higher for complements
than substitutes
2* 12 2 12 12 12 1 2 21 2 2 2
1 1 2 2 12 12
( )( ) ( )( ) ( )( ) ( )
A A
A A A
G B R G G B R BcR B R B R G
![Page 29: Optimal Health Insurance for Prevention and Treatmentblogs.bu.edu › ellisrp › files › 2012 › 02 › 2007-EllisManning_JHE_EJM_combined.pdfOptimal Health Insurance for Prevention](https://reader033.vdocuments.mx/reader033/viewer/2022060418/5f153817f7bcb9108f680e31/html5/thumbnails/29.jpg)
Two period model = health shock serial correlation =discount factor
Services should have lower coinsurance when they are more highly serially correlated over time
The higher the savings rate adjustment for health care shocks, the less insurance is needed
If consumers discount the future more highly than the rate of interest, then there should be more insurance
1 21 1 1
*
21 1
11
11
11
A
XA
B R L sc
B R s
![Page 30: Optimal Health Insurance for Prevention and Treatmentblogs.bu.edu › ellisrp › files › 2012 › 02 › 2007-EllisManning_JHE_EJM_combined.pdfOptimal Health Insurance for Prevention](https://reader033.vdocuments.mx/reader033/viewer/2022060418/5f153817f7bcb9108f680e31/html5/thumbnails/30.jpg)
Summary of what paper shows
1. Preventive care should be covered generously because consumers will
ignore premium savings in their private prevention decisions.
2. Health care goods that are positively correlated with other health
spending should be more generously covered
3. Health care goods that are substitutes should be covered more
generously than complements.
4. Cover services generously that are positively correlated with
uncompensated losses
5. Services that are positively serially correlated over time should be
covered more generously than those that are uncorrelated