Supplemental Nutrition Assistance Program (SNAP) Participation and Healthcare Expenditures Among Low-Income Adults Seth A. Berkowitz, MD MPH; Hilary K. Seligman, MD MAS; Joseph Rigdon, PhD; James B. Meigs, MD MPH; Sanjay Basu MD PhD
Disclosures • I have no conflicts of interest to report
Background • Food insecurity increased healthcare costs – ~$77 billion annually in excess costs
SNAP • Supplemental Nutrition Assistance Program – Formerly the Food Stamp Program –Nation’s largest anti -hunger effort – Known to reduce food insecurity
Research Question • Is SNAP participation associated with lower subsequent healthcare costs among low-income adults? – Hypothesis: SNAP participation is associated with lower healthcare expenditures, compared with eligible non- participants
Data Source • 2011 National Health Interview Survey (NHIS) – Linked to 2012-2013 Medical Expenditure Panel Survey (MEPS) • Sample: – All adult (age > 18) MEPS participants with household income below 200% FPL in NHIS
Does SNAP reduce healthcare costs? • Difficult question to answer as enrollment in SNAP can’t be randomized – People who enroll in SNAP often sicker than those who are eligible but do not enroll – May have other unmeasured factors that influence relationship between SNAP and health
SNAP • Supplemental Nutrition Assistance Program – Federally-set eligibility criteria – Administered by the states • States have broad leeway in determining enrollment practices
‘Near/far matching’ • ‘Near/far matching’ combines elements of propensity score and instrumental variables approaches
‘Near/far matching’ • Uses algorithm to simultaneously match: – participants who are alike in relevant measured characteristics (‘near’) – unalike in the amount of ‘encouragement’ into the intervention they received (‘far’)
Method • For ‘near/far’: – Instruments are state level variations in SNAP enrollment • Online application, broad-based categorical eligibility, simplified reporting
Sensitivity Analyses • ‘Standard’ Regression Adjustment – GLM with gamma distribution to account for properties of healthcare expenditures as outcome • Augmented Inverse Probability Weighting (AIPW)
SNAP and Healthcare Expenditures • Treatment: – Receipt of SNAP in 2011 • Outcome: – Total expenditures in 2012-2013 – All costs converted to 2015 dollars and annualized
SNAP and Healthcare Expenditures • Covariates: – Age – Gender – Race/ethnicity – Education – Income – Health Insurance – Rural vs. Urban – Region – Disability – Death during study period – Comorbidity – State Medicare Spending
‘ Near/Far’: Testing Instruments • Instrument predicts SNAP enrollment: OR 4.98, p < .0001 • Instrument is strong: First stage F = 44.2 • Instrument does not predict other indicators of state ‘generosity’ – Instrument not correlated with state Medicaid spending: • r = 0.11 (p=0.46) – Instrument not correlated with TANF benefit levels: • r = 0.11 (p= 0.44)
Selected Demographics No SNAP SNAP P N=2,558 N=1,889 Age, y 45 40 <.0001 Female, % 52 59 <.0001 Race/Ethnicity, % <.0001 Non-Hispanic White 53 43 Non-Hispanic Black 12 26 Hispanic 27 27 Asian/multi-/other 8 4 Income, % <.0001 <100% FPL a 32 63 100-150% FPL 29 24 151-200% FPL 39 13 Insurance, % <.0001 Private 30 15 Medicare 18 7 Other Public 15 45 Uninsured 37 33
Post-Match Demographics Discouraged Encouraged No SNAP SNAP N=2,558 N=1,889 N=1838 N=1838 Age, y 45 40 41 40 Female, % 52 59 59 59 Race/Ethnicity, % Non-Hispanic 53 43 22 21 White Non-Hispanic 12 26 26 26 Black Hispanic 27 27 45 46 Asian/multi-/other 8 4 7 7 Income, % <100% FPL a 32 63 60 60 100-150% FPL 29 24 24 24 151-200% FPL 39 13 16 15 Insurance, % Private 30 15 19 19 Medicare 18 7 9 8 Other Public 15 45 30 30 Uninsured 37 33 42 43
Expenditure Estimates Annualized P Annualized 95% Estimated Confidence Difference Expenditures Interval SNAP $2,116 $1,143 to $ -5,160 0.04 $3,089 No SNAP $7,276 $85 to -- $14,638 Estimates from 2SRI GLM/Gamma 2 nd stage regression adjusted for: age, age squared, gender, race/ethnicity, education, income, rural residence, Insurance, region, disability, death in study period, state spending, and comorbidity CI: bias corrected bootstrap (500 replications) Estimated expenditures in 2015 dollars.
Comparing Estimates
Limitations • Single assessment of SNAP receipt • IV assumptions • ‘near/far’ estimates LATE; standard regression and AIPW estimate ATE
Conclusions/Implications • SNAP consistently associated with reduced expenditures – Exact amounts vary somewhat by analysis specification • SNAP costs borne by Federal Gov’t – Medicaid shared between states and Fed • May be role for ‘linkage’ interventions that help folks enroll in SNAP
Thank you! This project was supported with a grant from the University of Kentucky Center for Poverty Research through funding by the U.S. Department of Agriculture, Economic Research Service and the Food and Nutrition Service, Agreement Number 58-5000-3-0066. Seth A. Berkowitz's role in the research reported in this publication was supported by the National Institute Of Diabetes And Digestive And Kidney Diseases of the National Institutes of Health under Award Number K23DK109200. The opinions and conclusions expressed herein are solely those of the author(s) and should not be construed as representing the opinions or policies of the sponsoring agencies.
Thank you! Co-authors! (Hilary Seligman, Joseph Rigdon, James Meigs, Sanjay Basu) Questions: SABerkowitz@partners.org “The homeopathists use gold as medicine, but they do not give it in large doses enough” – Samuel Butler, 1884
Results – ‘Standard’ Regression Model Parameters Expenditure Estimates β (95% CI) p-value Annualized 95% Annualized Estimated Confidence Difference Expenditures Interval SNAP -.2767 (-.5199 P=0.026 $4,379.49 $3,257.17 to $-1,396.09 to -.0335) $6,188.82 No SNAP ref -- $5,775.59 $3,557.33 to -- $7,993.84 Estimates from GLM/Gamma regression adjusted for: age, age squared, gender, race/ethnicity, education, income, rural residence, Insurance, region, disability, death in study period, and comorbidity Estimated expenditures in 2015 dollars.
‘Near/Far’: Testing Instruments Test Result Overidentifying Sargan p = 0.3070 Basmann p = 0.3099 SNAP OR 4.98, p < .0001 First-Stage F 44.2 0.11 (p= 0.44) TANF Medicaid P=0.80
Results – ‘AIPW’ Expenditure Estimates Annualized Difference 95% Confidence Interval SNAP $ -855.75 $ -1952.67 to $ -88.34 No SNAP -- -- Estimates from probit/linear AIPW adjusted for: age, age squared, gender, race/ethnicity, education, income, rural residence, Insurance, region, disability, death in study period, and comorbidity CI: bias corrected bootstrap (500 replications) Estimated expenditures in 2015 dollars.
Cycle of Food Insecurity & Chronic Disease FOOD INSECURITY HOUSEHOLD INCOME SPENDING COPING TRADEOFFS STRATEGIES: Dietary Quality STRESS Eating Behaviors Bandwidth HEALTH CARE EXPENDITURES CHRONIC DISEASE EMPLOYABILITY
Why does food insecurity result in poor health? Dietary Intake Food Stress Poor Insecurity Self-Efficacy Health Bandwidth Competing Demands Binge-Fast Cycles Employability Stability
SNAP and Healthcare Costs • Conceptual Model – Short-term: SNAP may provide/free up resources to attend to chronic disease management – Long-run: May help maintain health/prevent illness – Given time-frame of current data, focused on short-term effects
Comorbidity No SNAP SNAP P N=2,558 N=1,889 Disabled 10.16 22.70 <.0001 Obesity 31.08 37.59 0.0088 Hypertension 36.21 39.74 0.0625 Heart Disease 17.17 17.89 0.6351 Diabetes 9.98 11.99 0.0983 Stroke 5.07 6.39 0.2383 Arthritis 29.66 30.56 0.6960 COPD 2.73 4.68 0.0423
Results – ‘Near/far’ Model Parameters Expenditure Estimates β (95% CI) p-value Annualized 95% Annualized Estimated Confidence Difference Expenditures Interval SNAP -1.24 P=0.043 $2115.79 $1142.52 to $ -5,160.17 (-3.03 to -.06) $3,089.05 No SNAP ref -- $7275.95 $85.78 to -- $14,637.69 Estimates from 2SRI GLM/Gamma 2 nd stage regression adjusted for: age, age squared, gender, race/ethnicity, education, income, rural residence, Insurance, region, disability, death in study period, state spending, and comorbidity CI: bias corrected bootstrap (500 replications) Estimated expenditures in 2015 dollars.
Recommend
More recommend
