Longitudinal change in spending and mortality
Understanding Changes
in Local Public Health Spending
Glen Mays, PhD, MPH
Department of Health Policy and Management
University of Arkansas for Medical Sciences
•Half of all gains attributable to medical spending
•$36,300 per year of life gained
•What can we say about public health spending?
Cutler et al. NEJM 2006
Geographic variation in spending and mortality
Some research questions of interest…
interest…
X
Medical spending varies by a
factor of more than 2 across
local areas
X
How does public health spending vary across
communities and change over time?
X
Medicare enrollees in highspending regions receive more
care but do not experience
lower mortality
X
Are changes in spending associated with
changes in population health outcomes?
X
What is the value of public health spending?
What can we say about public
health spending?
Fisher et al. Annals 2003
Change in Local Public Health Spending,
19931993-2005
15%
15%
Variation in Local Public Health Spending
10%
Fraction of Agencies
65%
5%
Fraction of Agencies
5%
10%
Gini = 0.472
35%
$0
$50
$100
$150
Expenditures per capita, 2005
$200
0
0
X
–$50
–$40 –$30 –$20
–$10
$0
$10
$20
$30
$40
$50
Change in Per Capita Spending (Current Dollars)
Example: crosscross-sectional association
between PH spending and mortality
The problem with public health spending
X
Public health spending/capita
Heart disease mortality
Local funding sources often dependent on local
economic conditions that may also influence health
Public health spending may be correlated with other
resources that influence health
Sources of Local Public Health Agency Revenue, 2005
Fees
6%
Medicare
2%
Other
12%
Local
28%
120
200
195
80
190
185
60
180
40
175
20
170
165
0
Medicaid
9%
Federal
direct
7%
205
100
Quintile 1
Federal
pass-thru
13%
State direct
23%
Quintile 2
7000
Medicare spending per recipient
6800
80
6600
60
6400
40
6200
20
6000
0
Medical spending/person ($) . Public health spending/capita ($) .
Public health spending/capita
5800
Quintile 3
Quintile 4
Quintile 5
Quintiles of public health spending/capita
Financial and institutional data collected on the
national population of local public health agencies
(N≈3000) in 1993, 1997, and 2005
X
Residual state spending estimates from US
Census of Governments
X
Residual federal spending estimates from
Consolidated Federal Funding Report
X
Community characteristics obtained from Census
and Area Resource File (ARF)
Unmeasured
economic
distress
+
Mortality
PH spending
+
Approaches
Unmeasured
disease burden,
risk
+
1.
Cross-sectional regression: control for observable confounders
2.
Fixed effects: also control for time-invariant, unmeasured
differences between communities
3.
IV: use exogenous sources of variation in spending
4.
Discriminate between causes of death amenable vs. nonamendable to PH intervention
Data used in empirical work
X
Quintile 5
Addressing the problem with spending
_
7200
Quintile 2
Quintile 4
Quintile of public health spending/capita
120
Quintile 1
Quintile 3
NACCHO 2005
Example: crosscross-sectional association
between PH spending and Medicare spending
100
Deaths per 100,000
X
Federal & state funding sources often targeted to
communities based in part on disease burden, risk, need
Public health spending/capita
X
Analytical approach
X
Dependent variables
– Age-adjusted mortality rates, conditions sensitive to
public health interventions (infant mortality, heart
disease, cancer, diabetes, influenza)
– Counterfactual mortality rates (alzheimer’s,
unintentional injuries)
X
Independent variables of interest
– Local spending per capita, all sources
– Residual state spending per capita
(funds not passed thru to local agencies)
– Direct federal spending per capita
Analytical approach: IV estimation
X
Identify exogenous sources of variation in
spending, unrelated to outcomes
– Governance structures: local boards of health
– Centralized state-local PH administration
X
Analytical approach
Controls for unmeasured factors that jointly
influence spending and outcomes
Governance
Unmeasured
economic
distress
Other Variables Used in the Models
X
Agency characteristics: type of government jurisdiction,
scope of services offered, governance, state-local
administration
X
Community characteristics: population size, rural-urban,
poverty, education, age distributions, physicians per capita,
CHC funding per low income
X
State characteristics: Private insurance coverage, Medicaid
coverage, state fixed effects
Mortality
PH spending
Unmeasured
disease burden,
risk
Factors associated with local public health
spending
Multivariate estimates of association
between spending and mortality
Cross-sectional
model
Variable
Coefficient
0.145**
(0.099, 0.196)
Centralized structure (1=Yes)
-0.234**
(-0.364, -0.102)
Population size (log)
-0.136***
(-0.168, -0.103)
Income per capita (log)
0.196**
(0.001, 0.392)
0.234**
(0.032, 0.436)
Outcome
Elasticity
Elasticity
St. Err.
Elasticity
0.0234
0.0192
-0.6854
0.2668 ***
-0.0103
0.0040 **
-0.3216
0.1600 **
0.0187
-0.0487
0.0174 ***
-0.1439
0.0605 **
0.0029 *
-0.0075
0.0240
-0.1131
0.0566 **
0.0200 **
-0.0275
0.0107 **
-0.0252
0.0362
0.0516
Heart disease
-0.0003
0.0051
Diabetes
0.0323
Cancer
0.0048
-0.0400
0.0181 **
0.0024
0.0075
0.0032
0.0047
0.0051
0.0472
Injury
0.0007
0.0083
0.0004
0.0031
0.0013
0.0086
*p<0.10
Conclusions
X
Local public health spending varies widely
across communities
X
Governance and administrative structures
appear influential in spending decisions
– Local governing boards
– Decentralized administrative structures
Growth in spending is associated with reductions
in mortality from leading preventable causes of
death
St. Err.
Alzheimer’s
***p<0.01
Hierarchical logistic regression estimates controlling for community-level and state-level characteristics
X
St. Err.
Infant mortality
Influenza
**p<0.05
IV model
95% CI
Local board of health (1=Yes)
Local tax burden (% of income)
Fixed-effects
model
**p<0.05
***p<0.01
Limitations
X
Aggregate spending measures
– Average effects
– Role of allocation decisions?
X
Mortality – distal measures with long
incubation periods
X
Confounding—unmeasured factors tightly
correlated with public health spending?