EFFECTS OF LONG-TERM CARE SETTING ON SPOUSAL HEALTH OUTCOMES Jing - PowerPoint PPT Presentation

EFFECTS OF LONG-TERM CARE SETTING ON SPOUSAL HEALTH OUTCOMES Jing Dong, IMPAQ International Harold Pollack, University of Chicago R. Tamara Konetzka, University of Chicago

BACKGROUND EXPANSION IN NONINSTITUTIONAL LTC • LTC and LTC setting • A large expansion in noninstitutional care. Public financing of LTC has been shifting away from nursing home to home and community-based services (HCBS). • Elderly people generally prefer HCBS to institutional care • Less expensive for users with less intensive care needs • It is hoped that the shift to HCBS will help individuals to get high-quality services at lower costs

BACKGROUND KNOWLEDGE GAP AND RESEARCH QUESTIONS • Family members usually participate in LTC decisions and are inevitably affected by the decision to use different care • Prior studies have mainly focused on the preferences and well-being of LTC users in different care settings • Little is known about whether and how HCBS (versus nursing home) use may affect health outcomes for family members of care recipients • Spouses play an important role in providing informal care and making LTC decisions – important to understand the impact of different care settings on their health

BACKGROUND KNOWLEDGE GAP AND RESEARCH QUESTIONS • Hypotheses: • HCBS may inevitably place greater informal caregiving burden on the spouses, and informal care leads to worse physical and mental health outcomes for caregivers • An altruistic spouse may gain internal satisfaction and have better mental health outcomes from providing more care, from supporting the care recipients to stay in their preferred LTC setting, and from living with the care recipients • We would expect HCBS to have negative impact on physical health but unclear impact on mental health for spouses of care recipients

BACKGROUND SIGNIFICANCE AND CONTRIBUTION • To inform policy makers about the potential costs and benefits of HCBS expansion for spouses of HCBS recipients and help them to design programs to better support spouses: • Describe the characteristics of HCBS and nursing home users and their spouses • Examine the causal impact of HCBS (versus nursing home) on physical and mental health outcomes for spouses of care recipients

METHODS DATA • Health and Retirement Study (HRS) (1996-2012) • Longitudinal study that surveys a nationally representative sample of adults 50+ and their spouses every 2 years since 1992 • Information on SES, health, insurance, and medical expenditures • Cross-Wave Geographic Information file (restricted) • State, county and zip code information • Area Health Resource File and Census Data • County-level nursing home bed supply (IV)

METHODS SAMPLE HRS (207,816 observations/ 37,319 individuals/ 23,373 households) Keep only 1996-2012 175,522 observations/ 35,044 individuals/ 22,163 households Keep if spouses are in 2 consecutive waves and have a partner 88,672 observations/ 22,065 individuals/ 11,345 households Keep if care recipients use only HCBS or only nursing home in the 2 nd wave. Drop if spouses lived in a different county in the 1 st wave 8,789 observations/ 6,031 individuals/ 4,757 households

METHODS VARIABLES • Spousal health outcomes • Physical health: (1) good self-rated health; (2) need help with Activities of Daily Living (ADL); (3) need help with Instrumental Activities of Daily Living (IADL); (4) onset of five common chronic conditions • Mental health: (1) onset of diagnosed psychiatric problems; (2) CESD >=3 • Treatment • HCBS (versus nursing home) use by care recipient • Covariates • Spouse SES, household wealth, family, spouse health insurance, spouse health at previous wave, care recipient health, and year and state FEs

METHODS CONCERNS WITH NAÏVE IDENTIFICATION STRATEGY • Selection bias • In many cases, the care setting is chosen by care recipients and/or their spouses and may, therefore, be correlated with factors that also affect spousal health. • Failure to control for all confounders may result in selection bias • Reverse causality • Spousal health may reversely affect LTC decisions

METHODS SOLUTION: INSTRUMENTAL VARIABLE (IV) Potential threats to IV IV Assumptions Strengths exogeneity and solutions (1) People move for desired LTC settings – drop spouses who moved 2 years before LTC use (1) IV should predict HCBS use County-level, less likely to (2) Nursing home demand County-level number of be correlated with induced supply – nursing home beds per (2) IV should not directly individual-level unobserved average 0.8% annual 1,000 people 65+ affect spousal health or confounders change in IV other unobserved confounders (3) IV is correlated with county-level variables – balance check: observables are balanced

METHODS TWO-STAGE RESIDUAL INCLUSION MODEL (2SRI) • First stage model 𝑚𝑝𝑕𝑗𝑢 𝑄 𝐼𝐷𝐶𝑇 𝑠,𝑢 = 1 = 𝛽 0 + 𝛽 1 𝐽𝑊 𝑠,𝑢 + 𝛽 2 𝐼 𝑡,𝑢−1 + 𝛽 3 𝑌 𝑡,𝑢 + 𝛽 4 𝑌 𝑠,𝑢 + 𝑍𝑓𝑏𝑠 𝑢 + 𝑇𝑢𝑏𝑢𝑓 𝑡,𝑢 • Second stage model 𝑚𝑝𝑕𝑗𝑢 𝑄 𝐼 𝑡,𝑢 = 1 = 𝛾 0 + 𝛾 1 𝐼𝐷𝐶𝑇 𝑠,𝑢 + 𝛾 2 𝐼 𝑡,𝑢−1 + 𝛾 3 𝑌 𝑡,𝑢 + 𝛾 4 𝑌 𝑠,𝑢 + 𝑍𝑓𝑏𝑠 𝑢 + 𝑇𝑢𝑏𝑢𝑓 𝑡,𝑢 + ෞ 𝑠 𝑡,𝑢 • HCBS= whether care recipient used HCBS • IV= county-level number of skilled nursing home beds per 1000 people 65+ • H s,t-1 = a set of previous health variables for spouse • X= a set of control variables • Yeart and State= year and state fixed effects • 𝑗𝑢 = response residuals from the first stage model 𝑠 ෞ • Standard errors are clustered at individual level • Bootstrap procedure for both stages with 500 iterations

RESULTS SAMPLE CHARACTERISTICS Independent variables Nursing Home (N=781) HCBS (N=3,553) P value Spouse SES Age (Mean) 75.7(10.3) 68.1(10.0) <0.001 Female (%) 54.2 49.4 0.029 Race (%) 0.049 White 81.8 85.6 Black 13.2 10.7 Other races 5.0 3.7 Hispanic (%) 9.9 7.6 0.045 Education (%) <0.001 Less than HS 34.6 18.4 GED 3.9 4.3 High school 30.9 30.1 Some college 17.1 22.2 College and above 13.6 25.1 Retired (%) 66.2 52.6 <0.001 Household wealth Log total financial assets ($) 7.3(5.0) 8.3(4.9) <0.001 Log total income ($) 10.4(1.1) 10.9(1.0) <0.001 Family Number of children (%) <0.001 0 6.3 3.4 1 11.2 8.2 2 23.3 26.5 3+ 59.2 62.0

RESULTS SAMPLE CHARACTERISTICS Independent variables Nursing Home (N=781) HCBS (N=3,553) P value Spouse health insurance Uninsured (%) 2.9 4.3 0.125 Has LTCI (%) 13.8 14.4 0.694 Spouse health at t-1 Lagged # diagnosed disorder (Mean) 2.2(1.5) 1.9(1.4) <0.001 Lagged # mobility tasks cannot do (Mean) 1.5(1.6) 1.0(1.4) <0.001 Lagged any psychiatric problems (%) 21.7 15.6 <0.001 Lagged any pain problems (%) 38.4 32.7 0.006 LTC user health # diagnosed disorder (Mean) 3.1(1.5) 2.6(1.5) <0.001 # mobility tasks cannot do (Mean) 3.0(1.9) 1.7(1.7) <0.001 Any psychiatric problems (%) 31.1 22.6 <0.001 Any pain problems (%) 40.3 49.2 <0.001 The comparisons between the two treatment arms are calculated based on simple two-sample t-tests or chi-squared test. Standard deviations in parentheses.

RESULTS MARGINAL EFFECTS OF HCBS ON SPOUSE PHYSICAL HEALTH (BASE MODELS) (1) (2) (3) (4) 5 Common Model Good Health Any ADLs Any IADLs Conditions IV models Marginal effect -0.073** 0.014 0.015 0.008 Bootstrap S.E. (0.031) (0.027) (0.025) (0.031) First-stage F statistics 11.6 11.6 11.6 11.6 Mean of dependent var 0.724 0.171 0.164 0.157 Observations 8,775 8,777 8,777 8,758 p<0.1; ** p<0.05; *** p<0.01. Standard errors in parentheses

RESULTS MARGINAL EFFECTS OF HCBS ON SPOUSE PHYSICAL HEALTH (T+1 MODELS) (1) (2) (3) (4) 5 Common Model Good Health Any ADLs Any IADLs Conditions IV models Marginal effect -0.082* 0.040 0.058* 0.081** Bootstrap S.E. (0.043) (0.032) (0.031) (0.040) First-stage F statistics 8.8 8.5 8.5 13.1 Mean of dependent var 0.716 0.176 0.167 0.124 Observations 6,422 6,425 6,424 6,456 p<0.1; ** p<0.05; *** p<0.01. Standard errors in parentheses

RESULTS MARGINAL EFFECTS OF HCBS ON SPOUSE MENTAL HEALTH (BASE MODELS) (5) (6) Psychiatric Problems CESD>=3 Model IV models Marginal effect -0.019 -0.154*** Bootstrap S.E. (0.023) (0.038) First-stage F statistics 11.8 8.9 Mean of dependent var 0.024 0.210 Observations 8,651 7,954 p<0.1; ** p<0.05; *** p<0.01. Standard errors in parentheses

RESULTS MARGINAL EFFECTS OF HCBS ON SPOUSE PHYSICAL HEALTH (T+1 MODELS) (5) (6) Psychiatric Problems CESD>=3 Model IV models Marginal effect 0.013 -0.027 Bootstrap S.E. (0.014) (0.043) First-stage F statistics 8.7 7.3 Mean of dependent var 0.024 0.217 Observations 6,311 5,851 p<0.1; ** p<0.05; *** p<0.01. Standard errors in parentheses