Spatial Econometric Model of Healthcare Spending Garen Evans MISSISSIPPI STATE UNIVERSITY LOCAL!
Background Health Care spending as Percentage of GSP
Health Care Spending • Hospitals • Professional Services • Long Term Care • home health care, nursing homes • Personal Medical Supplies • durables, drugs, supplies • Other
U.S. Personal Healthcare Spending* * Millions of 2004 dollars
Local Health Care Spending? • National • Personal health care spending • Sector detail • Hospitals, home health care, etc. • State • Place-based • Residence-based • County ?
County-Level Spending • Usage: • Quantify importance of health care in small economies • Often combined with input-output analysis. • Leverage interest in local health care • eg., Critical Care Access Hospital designation • Gauge effectiveness of healthcare policy as an economic engine • Test global hypotheses
County-level Spending • Non-structural approach • Product of LPC-adjusted state per-capita spending and local population • Patient-origin analysis • National benchmarks • Trade area capture • Structural approach • Identify factors related to health care spending
Health Care Spending • Factors that affect spending: • Demographic • Population distributions • Socioeconomic • Income • Market-related • Physician concentration • Policy • Managed care
Demographic • Age 65+ tend to use six times the healthcare compared to younger persons • Martin, 2005 • At least one chronic condition by age 70 • Neese, 2002 • Out-of-pocket spending for chronic conditions varies with age • Hwang, 2001
Socioeconomic • Higher growth in per-capita income leads to growth in per-capita private spending. • Smith, 1998 • Almost 18% of per-capita spending due to income growth. • Peden, 1995 • Spending for children in poverty was 14% higher than average. • Holahan, 2001
Market Factors • Uninsured spend less than those with Medicaid • Holahan, 2001 • High physician concentration generates higher levels of spending • Martin, 2002 • Large provider networks exert leverage over insurers when negotiating prices. • Brudevold, 2004
Policy factors • High levels of enrollment in HMOs reduces spending growth • Staines, 1993; Cutler, 1997. • Medicaid managed care enrollment not a significant predictor of Medicaid expenditures. (Only state per capita income and regional differences were significant predictors of Medicaid costs. ) • Weech-Maldonado, 1995
Objectives • Develop local spending model. • Counties in Mississippi • Cross-sectional • Examine relationship of factors associated with healthcare spending. • Explore space.
Data • Health Spending Impact Model (HSIM) • County-level health care spending estimates • Based on state-level per-capita spending • Local Purchase Coefficients • Hospitals • Physicians, Dentists, et al. • Long Term Care • Medical Supplies • Other
Statewide Spending Population 2.9 million Hospital Care $7.3 billion Per-Capita $2,517
Local Hospital Spending 52.2% of Oktibbeha County residents received hospital care in other counties. LPC is 47.8% or… $1,202 per-capita Pop 42,454 Total: $51 million
Percentage of residents discharged from local hospital Mean: 41.2% Std Dev.: 27.6%
County-level per-capita spending for health care Mean: $3,576 Max: $5,189 Min: $956 11 < 1 SD (13%) 16 > 1 SD (19.5%)
Data • Socioeconomic/Demographic • Per-capita income – Woods and Poole • Poverty rate - Small Area Income & Poverty Estimates; US Census. • Market • Hospital – MSDH Report on Hospitals • Diabetes (mortality) – MSDH Vital Statistics • Insurance • Small Area Health Insurance Estimates (SAHIE; US Census) 2001
Spatial clustering can occur in behavioral risk factors and outcomes Mobley, 2006. Spatial lag can lead to biased and inconsistent estimators Anselin, 2006 Spatial Weights
Summary Statistics PCI: $000 COVER: % not covered by health insurance HOSP: dummy (1=hospital) POVRTY: Percentage of population at below 100% poverty rate. DIABET: mortality per 100,000 population LSPC: local spending per capita, $000 RHO1: rook-based spatial weights RHO2: queen-based spatial weights
#1 BASELINE MODEL LSPC = f(PCI, COVER, POVRTY, DIABET, HOSP) + - - + + #2 SPATIAL LAG MODEL (ROOK-BASED WEIGHTS) LSPC = f(PCI, COVER, POVRTY, DIABET, HOSP, RHO1) + #3 SPATIAL LAG MODEL (QUEEN-BASED WEIGHTS) LSPC = f(PCI, COVER, POVRTY, DIABET, HOSP, RHO2) + Models
Rook-based Queen-based LSPC Moran Scatterplots
LSPC, rook LSPC, queen Local Indicators of Spatial Association (LISA)
Summary • Per-capita income, presence of hospital, poverty rate, and insurance coverage help explain local per-capita spending for healthcare services. • Space matters in the analysis of healthcare spending
Summary • Space is significant, but does not appear to be substantial… • 1.94% of variation in the rook model. • 2.63% of variation in the queen model. • Negative Rho implies dissimilarity in neighboring areas.
Working paper and presentation is online: http://giwiganz.com/garen/NARSC07 Garen Evans firstname.lastname@example.org 662-325-2750