Health Insurance, Medical Care, and Health Outcomes:

A Model of Elderly Health Dynamics

Zhou Yang, Emory University

Donna B. Gilleskie, Univ of North Carolina

Edward C. Norton, Univ of MichiganJournal of Human Resources 44(1): 48-108, 2009

November 16, 2010UNC School of Nursing

As individuals, what do we know?• The U.S. spends a lot on medical care.

• Most elderly are covered by Medicare (parts A and B).

• Elderly may choose Medicare’s managed care plan (part C).

• Many of the elderly have supplemental health insurance.

• Medicare, generally, did not cover prescription drugs.

• The Medicare Prescription Drug Improvement and Modernization Act has made drug coverage an option for the elderly (part D).

As economists, what do we know?• Third-party coverage of medical care expenses leads to

increased demand for covered services.

• Prescription drug coverage leads to greater consumption of prescription drugs.

• Increased prescription drug use reduces mortality (and morbidity).

• Differences in the cost-sharing characteristics of coverage for different types of medical care can affect consumption behavior.

• Differences in the effectiveness of different types of medical care can affect consumption behavior.

Can we predict the long-run impact of Rx coverage?

Yes, but what we don’t want to do is:

• ignore the endogeneity of insurance selection

• consider the effect of drug coverage on drug expenditures only

• measure the effect of prescription drug use on mortality only

• fail to model changes in health over time

• evaluate outcomes in a static setting

• ignore unobserved individual heterogeneity likely to influence behavior in several dimensions

The Big Picture

Prescription Drugs

Physician Services, Hospitalization

Health: Morbidity, Mortality

Supplemental Insurance,

Rx Coverage

Age

HealthSudden death: “extreme” health shock but no functional decline

Terminal Illness: good functional health then health shock and certain decline in function

Frailty: no health shock(s) or serious chronic condition, but slow decline in function

Entry-re-entry: chronic condition(s) associated with multiple health shocks and expected decline in function

Typical Patterns of Health Decline among the Elderly

JAMA 289(18), 2003

A Preview of our Main Findings

A change from Medicare with no drug coverageto a plan that covers prescription drugs reveals that:

• Drug expenditures over 5 years increase between 7 and 27%.

• Survival rates increase 1-2%. But the distribution of functional status among survivors shifts toward worse health.

• Marginal survivors spend significantly more than individuals who would have survived anyway.

• There is some contemporaneous reallocation of consumption (a cross-price effect), but changes in consumption are largely driven by changes in health and survival as people age.

Model of behavior of individuals age 65+

It , Jt St At, Bt, Dt Et+1, Ft+1

beginning of age t

beginning of age t+1

insurance and drug coverage

health shock

medical care demand

health production

Ωt= (Et, Ft,

At-1, Bt-1, Dt-1, Xt,

ZIt, ZH

t, ZMt )

Ωt+1= (Et+1, Ft+1,

At, Bt, Dt, Xt+1,

ZIt+1, ZH

t+1, ZMt+1 )

And we model the set of structural equations jointly, allowing unobserved components to be correlated

Empirical Model

It , Jt St At, Bt, Dt Et+1, Ft+1

beginning of t

beginning of t+1

insurance and drug coverage

health shock

medical care demand

health production

Multinomial logit:

Medicare only (parts A and B) ( 8%) Medicaid dual coverage (12%) Private plan supplement (64%) Medicare managed care plan (part C) (16%)

Logit: Rx coverage (63%)

(conditional on private or Part C plan)

Empirical Model

It , Jt Skt At, Bt, Dt Et+1, Ft+1

beginning of t

beginning of t+1

insurance and drug coverage

health shock(s)

medical care demand

health production

Separate logits:

Heart/stroke event (ICD-9 390-439) in period t (24.5 %)

Respiratory event (ICD-9 480-496) in period t ( 4.8 %)

Cancer event (ICD-9 140-209) in period t ( 5.7 %)

Empirical Model

It , Jt Skt At, Bt, Dt Et+1, Ft+1

beginning of t

beginning of t+1

insurance and drug coverage

health shock(s)

medical care demand

health production

Separate logit for any use and OLS log expenditures conditional on any:

Hospital use and expenditures in period t (20 % and $13,057)

Physician service use and expenditures in period t (84 % and $2,013)

Prescription drug use and expenditures in period t (90 % and $980)

Empirical Model

It , Jt Skt At, Bt, Dt Ek

t+1, Ft+1

beginning of t

beginning of t+1

insurance and drug coverage

health shock(s)

medical care demand

health:ever had chronic

condition k , functional status

Multinomial logit for functional status entering period t+1:

Not disabled (no ADL or IADLs) (58%) Moderately disabled (IADL or <3 ADLs) (28%) Severely disabled (3 or more ADLs) (10%) Dead ( 5%)

Indicator for having ever had a chronic condition entering period t+1:

Heart/stroke (47%) Respiratory (15%) Cancer (19%) Diabetes (20%)

Ekt+1 = Ek

t + Skt

Empirical Model

It , Jt Skt At, Bt, Dt Et+1, Ft+1

beginning of t

beginning of t+1

insurance and drug coverage

health shock(s)

medical care demand

health production

Jt= J(Et, Ft, At-1, Bt-1, Dt-1, Xt, ZIt, ZH

t, ZMt, t , uJ

t)

Skt= S(Et, Ft, Xt, ZH

t, ukt), k = 1, 2, 3

At= A(ItJt, St, Et, Ft, At-1, Bt-1, Dt-1, Xt, ZMt, t , uA

t)

Ft+1= F(Et, Ft , St, At, Bt, Dt, Xt, uft)

It= I(Et, Ft, At-1, Bt-1, Dt-1, Xt, ZIt, ZH

t, ZMt, t , ui

t)

Bt= B(ItJt, St, Et, Ft, At-1, Bt-1, Dt-1, Xt, ZMt, t , uB

t)

Dt= D(ItJt, St, Et, Ft, At-1, Bt-1, Dt-1, Xt, ZMt, t , uD

t)

Unobserved Heterogeneity Specification

• Permanent: risk aversion or attitude toward medical care use

• Time-varying: unmodeled health shocks or natural rate of deterioration

uet = ρe μ + ωe νt + εe

t

where uet is the unobserved component for equation e decomposed into

• permanent heterogeneity factor μ with factor loading ρe

• time-varying heterogeneity factor νt with factor loading ωe

• iid component εet

distributed N(0,σ2e) for continuous equations and

Extreme Value for dichotomous/polychotomous outcomes

Features of our Empirical Model Suggested by Theory

• Supplemental insurance coverage is chosen at the beginning of the period before observing health shocks, but with knowledge of one’s functional status, chronic conditions, and, most importantly, unobserved individual characteristics entering the period.

Features of our Empirical Model Suggested by Theory

• Permanent and time-varying unobserved individual characteristics affect annual demand for all three types of medical care.

Adverse selection

Features of our Empirical Model Suggested by Theory

• Health transitions are a function of medical care input allocations and health shocks during the year. (Grossman)

Adverse selectionJointly estimated demand

Features of our Empirical Model Suggested by Theory

• Previous medical care use may alter the utility of medical care consumption today; hence, lagged use affects current expenditures directly as well as indirectly through health transitions.

Adverse selectionJointly estimated demandDynamic health production

Features of our Empirical Model Suggested by Theory

Adverse selectionJointly estimated demandDynamic health productionDynamic demand for medical care

Medicare Current Beneficiary Survey (MCBS) Sample

• Survey and Event files from 1992-2001

• Overlapping samples followed from 2 to 5 years

• Exclude individuals ever in a nursing home

• Attrition due to death and sample design

• Sample: 25,935 men and women; 76,321 person-year obs

Actual and Simulated Annual Mortality Rate, by Age

Actual and Simulated Prescription Drug Expenditures, by Age and Death

Actual and Simulated Physician Services Expenditures, by Age and Death

Actual and Simulated Hospital Expenditures, by Age and Death

Simulations• Start everyone off with a particular type of health insurance

– Medicare only– Dual coverage by Medicaid– Private supplement without Rx coverage– Private supplement with Rx coverage– Medicare managed care (part C) without Rx coverage– Medicare managed care (part C) with Rx coverage

• Simulate behavior for 5 years

• Examine expenditures and health outcomes over 5 years

• Examine expenditures of 5-year survivors

Five-year Simulations – with unobserved heterogeneity

Five-year Simulations – without unobserved heterogeneity

Five-year Simulations – with unobserved heterogeneity

22.5

10.6

4.8

10.7

Sole Survivors vs. Marginal Survivors

Rx expenditures triple or quadruple

} With increases here, too

Increases in expenditures are 3.5 to 5.5 times larger

Take home message…so far• Methodologically, we have built and estimated a comprehensive

dynamic model of health behavior of the elderly as they age.

• Substantively, our model allows us to examine the effects of health insurance extensions (Rx coverage) not simply on prescription drug use but also on other types of care, as well as the impacts of this altered demand on health outcomes and subsequent behavior over time.

• Increases in Rx coverage increase short-run demand for drugs, as well as other types of care. Mortality rates decline, but functional status of survivors is worse. Hence, total expenditures increase over a 5-year period.

Why might nursing care matter?

• Clearly it might affect health outcomes, conditional on endogenous inputs – affects marginal product of health input– but only the hospital care input

• Might it affect demand for care?– consumers care about price (budget constraint)– but preferences might also depend on quality

Or better, where would it enter the model?

Identification in the set of dynamic equations

• Exogeneity of some explanatory variables conditional on the unobserved heterogeneity– theoretically-relevant exogenous supply-side variables – lagged values of exogenous (both ind and ss)– lagged values of endogenous variables

• Exogenous variables, in the reduced-form initial condition equations, that are excluded from the dynamic structural equations

• Specification and covariance structure of the permanent and time-varying unobserved individual heterogeneity

• Functional form of the equations

What next?• We lack good data at the individual level

– on outcomes– on inputs– at reasonable intervals– for a large sample of representative people

• We lack a theory that considers the effects of both price and quality on demand for medical care and health production.

Five-year Simulations – with unobserved heterogeneity

Five-year Simulations – without unobserved heterogeneity

Unobserved Heterogeneity Distribution

Actual and Simulated Prescription Drug Use and Expenditures, by Age

Actual and Simulated Hospital Use and Expenditures, by Age

Actual and Simulated Physician Services Use and Expenditures, by Age