APPENDIX

Supplement to: Medicaid Expansion, Mental Health, and Access to Care Among Childless Adults With and Without Chronic Conditions

Tyler N.A. Winkelman and Virginia W. Chang

PAGE 2 Section 1. Details on Regression Specification

Section 2. Pre-ACA Trend Comparisons Between Expansion and Non-Expansion States (Parallel Trends Assumption)

PAGES 2-3Section 3.Trends Among Elderly Americans With and Without Chronic Conditions

PAGE 3 Section 4. Trends Among Childless Adults With and Without Chronic Conditions

Section 5.Trends in Chronic Condition Prevalence Among Childless Adults

PAGE 4eTable 1. Pre-ACA Trends Among Childless Adults With Chronic Conditions

PAGE 5eTable 2. Pre-ACA Trends Among Childless Adults Without Chronic Conditions

PAGE 6eTable 3.Changes Following Medicaid Expansion Among Elderly Adults With Chronic Conditions

PAGE 7eTable 4.Changes Following Medicaid Expansion Among Elderly Adults Without Chronic Conditions

PAGE 8 eTable 5. Changes Following Medicaid Expansion Among Elderly Adults (Ages 65 and Older) With Chronic Conditions Compared to Non-Elderly Adults (Ages 18-64) With Chronic Conditions

PAGES9-10eFigure 1. Trends Among Childless Adults With Chronic Conditions

PAGES 11-12eFigure 2. Trends Among Childless Adults Without Chronic Conditions

PAGE 13eTable 6. Changes in Chronic Condition Prevalence Following Medicaid Expansion Among Childless Adults

Section 1. Details on Regression Specification

For each outcome (Y) we estimated the following multivariable linear regression model:

Yisq = 0 + 1Medicaidsq + 2UnemploymentRates+ xXisq + s + q + isq Equation (1)

Where i indicates an individual survey respondent, s indicates the state, and q indicates the interview quarter. In our model, Xisq indicates several individual characteristics including: age, race/ethnicity, sex, education, and marital status. State-level indicator variables are represented by sand capture time-invariant differences between expansion and non-expansion states. Quarter-year indicators are represented by qand capture the underlying, secular time trend. Medicaidsq equaled1 if an individual lived in a state that had expanded Medicaid by the first of the month when the interview was conducted and 0 for all other observations. It is essentially an interaction between two indicators: one for expansion vs. non-expansion state and one for pre- vs. post-expansion. The main effects for these indicators are subsumed by the state- and time-level fixed effects.The coefficient 1 provides an estimate of the adjusted difference between Medicaid expansion and non-expansion states for the change in outcome Y from the pre-ACA period to the post-ACA period.2 captures annual state-level unemployment rates. We used survey weights for all models and accounted for clustering at the state level.

Section 2. Pre-ACA Trend Comparison Between Expansion and Non-Expansion States (Parallel Trends Assumption)

Yisq = 0 + 1Timet + 2Expansions*Timet+ 3UnemploymentRates + xXisq + s + isq Equation (2)

The key assumption in a difference-in-differences analysis is that trends in the control (Non-Medicaid expansion states) and experimental (Medicaid expansion states) groups are similar before implementation of a given policy (e.g., Medicaid). Therefore, we examined differences in trends between expansion and non-expansion states in the years prior to implementation of the ACA (2011-2013). Any significant differences in pre-ACA trends would be identified if the coefficient for2 was statistically significant. Timet represents a linear quarter-year time trend between 2011 and 2013. Other variables are described in Section 1. We used survey weights for all models and accounted for clustering at the state level. Results of our sensitivity analysis for individuals with and without chronic conditions can be found in eTable 1 and eTable 2, respectively.

Section 3. Trends Among Elderly Americans With and Without Chronic Conditions

To determine whether trends in health outcomes and access to care could be influenced by factors other than the ACA, we repeatedall analyses using Equation (1) for adults aged 65 and older. If factors other than the ACA were responsible for changes in health and access, we would anticipate similar trends among elderly individuals. Results among elderly adults with and without chronic conditions can be found in eTable 3 and eTable 4, respectively, and were not substantively similar to changes among non-elderly adults.

We next examined whether trends in access to care measures and health outcomes between elderly and non-elderly adults with chronic conditions were statistically different by conducting a difference-in-difference-in-differences analysis using the following model:

Yisq = 0 + 1Post-Expansionq*NonElderlyi +2NonElderlyi*ExpansionStates + 3Post-Expansionq* ExpansionStates + 4Post-Expansionq* ExpansionStates*NonElderlyi + 5UnemploymentRates+ xXisq + s + q + isq Equation (3)

Where i indicates an individual survey respondent, s indicates the state, and q indicates the interview quarter. Xisq indicates several individual characteristics including: age, race/ethnicity, sex, education, and marital status. Of note, our age variable was a six-category variable with the last category indicating individuals aged 65 and older, which captures the direct effect of NonElderly. State-level indicator variables are represented by s and capture time-invariant differences between expansion and non-expansion states and the direct effects of ExpansionState. Quarter-year indicators are represented by q and capture the underlying, secular time trend and the direct effects of Post-Expansion. 2 captures annual state-level unemployment rates. We used survey weights for all models, accounted for clustering at the state level, and limited our analysis to childless adults with chronic conditions.

β4 is the primary coefficient of interest and measures the relative change in a given Y after Medicaid expansion among non-elderly adults (18-64) in expansion states versus non-expansion states, net of the change in Y among elderly adults (65 and older). β3 measures the relative change in Y after Medicaid expansion among elderly adults in expansion states, compared to elderly adults in non-expansion states. Results for each outcome are presented in eTable 5. Changes in self-reported general health, poor mental health days, depression, disability, coverage, and seeking care due to cost among non-elderlyadults in expansion vs. non-expansion states were significantly different than trends among elderly adults in expansion vs. non-expansion states (P<.05). Disability became significant in this model due to a large increase (1.6 percentage points [95% CI, -0.7, 3.8]) in disability among elderly adults in expansion states relative to non-expansion states.

Section 4. Trends Among Childless Adults With and Without Chronic Conditions

We qualitatively analyzed unadjusted time trends for each outcome as an additional check of the Parallel Trends Assumption. Figures for each outcome are included below (eFigure 1 and eFigure 2).

Section 5. Trends in Chronic Condition Prevalence Among Childless Adults

To determine whether Medicaid expansion was associated with changes in chronic condition prevalence, we used Equation (1) where Y was a dichotomous variable indicating the presence of 0 or 1 chronic condition. Medicaid expansion was not statistically significantly associated with changes in diagnosis of a chronic condition in both unadjusted and adjusted models (eTable 6).

eTable 1. Pre-ACA Trends Among Childless Adults With Chronic Conditions

OUTCOME / Differential Trend in Medicaid Expansion vs.
Non-Expansion States
Coefficient* / 95% CI / P value
Fair/poor self-reported health / 0.0001 / -0.005, 0.005 / .98
Total days in poor health(days per month) / 0.04 / -0.09, 0.16 / .54
Days limited by poor health(days per month) / 0.03 / -0.09, 0.15 / .62
Poor physical health (days per month) / 0.02 / -0.10, 0.14 / .74
Poor mental health (days per month) / -0.0004 / -0.12, 0.12 / .99
Disabled / -0.01% / -0.44, 0.43 / .98
Depression / 0.23% / -0.37, 0.84 / .44
Health insurance coverage / 0.55% / -0.02, 1.12 / .06
Didn’t seek care due to cost / -0.43% / -0.76, -0.10 / .01
Check-up within last year / 0.05% / -0.44, 0.55 / .83
Personal doctor / 0.63% / 0.17, 1.1 / .01

*Coefficient represents 2from Equation (2).

NOTES A negative coefficient indicates a decrease in states that expanded Medicaid relative to states that did not expand Medicaid, whereas a positive coefficient indicates a relative increase in Medicaid expansion states.Estimates are adjusted for state fixed-effects, linear time trend, age, sex, race/ethnicity, education, marital status, and annual state unemployment rate.

eTable 2. Pre-ACA Trends Among Childless Adults Without Chronic Conditions

OUTCOME / Differential Trend in Medicaid Expansion vs.
Non-Expansion States
Coefficient* / 95% CI / P value
Self-reported health (1=poor, 5=excellent) / -0.002 / -0.007,0.003 / .37
Total days in poor health (days per month) / -0.07 / -0.23, 1.0 / .42
Days limited by poor health (days per month) / -0.07 / -0.18, 0.03 / .15
Poor physical health (days per month) / -0.03 / -0.17, 0.12 / .73
Poor mental health (days per month) / -0.06 / -0.14, 0.02 / .11
Disabled / -0.01% / -0.30, 0.29 / .97
Depression / -0.19% / -0.54, 0.16 / .29
Health insurance coverage / 0.23% / -0.18, 0.63 / .27
Didn’t seek care due to cost / -0.20% / -0.88, 0.47 / .55
Check-up within last year / -0.43% / -0.75, -0.12 / .01
Personal doctor / 0.07% / -0.37, 0.51 / .75

*Coefficient represents 2from Equation (2).

NOTES A negative coefficient indicates a decrease in states that expanded Medicaid relative to states that did not expand Medicaid, whereas a positive coefficient indicates a relative increase in Medicaid expansion states. Estimates are adjusted for state fixed-effects, linear time trend, age, sex, race/ethnicity, education, marital status, and annual state unemployment rate.

eTable 3.Changes Following Medicaid ExpansionAmong Elderly Adults With Chronic Conditions

OUTCOME / Adjusted difference-in-differences estimate
Coefficient* / 95% CI / P value
Self-reported health (1=poor, 5=excellent) / -0.03 / -0.07,0.02 / 0.19
Total days in poor health (days per month) / -0.26 / -0.83, 0.30 / 0.35
Days limited by poor health (days per month) / -0.01 / -0.63, 0.62 / 0.98
Poor physical health (days per month) / -0.20 / -0.69, 0.30 / 0.42
Poor mental health (days per month) / -0.15 / -0.66, 0.36 / 0.56
Disabled / 0.01 / -0.01, 0.04 / 0.34
Depression / 0.008 / -0.01, 0.03 / 0.39
Health insurance coverage / -0.001 / -0.01, 0.01 / 0.85
Didn’t seek care due to cost / -0.005 / -0.02, 0.01 / 0.51
Check-up within last year / 0.003 / -0.02, 0.02 / 0.76
Personal doctor / 0.01 / -0.004, 0.03 / 0.13

*Coefficient represents 1from Equation (1).

NOTES A negative coefficient indicates a decrease in states that expanded Medicaid relative to states that did not expand Medicaid, whereas a positive coefficient indicates a relative increase in Medicaid expansion states.Estimates are adjusted for state and quarter-year fixed effects, age, sex, race/ethnicity, education, marital status, and annual state unemployment rate.

eTable 4.Changes Following Medicaid Expansion Among Elderly Adults Without Chronic Conditions

OUTCOME / Adjusted difference-in-differences estimate
Coefficient * / 95% CI / P value
Self-reported health (1=poor, 5=excellent) / -0.08 / -0.21,0.05 / 0.21
Total days in poor health (days per month) / 1.08 / 0.26, 1.91 / 0.01
Days limited by poor health (days per month) / 0.21 / -0.45, 0.88 / 0.52
Poor physical health (days per month) / 0.63 / -0.15, 1.40 / 0.11
Poor mental health (days per month) / 0.45 / -0.30, 1.20 / 0.23
Disabled / 0.01 / -0.03, 0.05 / 0.60
Depression / 0.02 / -0.01, 0.05 / 0.17
Health insurance coverage / -0.004 / -0.03, 0.02 / 0.76
Didn’t seek care due to cost / -0.0002 / -0.04, 0.04 / 0.99
Check-up within last year / 0.01 / -0.03, 0.06 / 0.53
Personal doctor / 0.05 / 0.01, 0.09 / 0.01

*Coefficient represents 1from Equation (1).

NOTES A negative coefficient indicates a decrease in states that expanded Medicaid relative to states that did not expand Medicaid, whereas a positive coefficient indicates a relative increase in Medicaid expansion states. Estimates are adjusted for state and quarter-year fixed effects, age, sex, race/ethnicity, education, marital status, and annual state unemployment rate.

eTable 5. Changes Following Medicaid Expansion Among Non-Elderly Adults (18-64) With Chronic Conditions Compared to Elderly Adults (65 and older) With Chronic Conditions

OUTCOME / Adjusted difference-in-difference-indifferences estimate
Coefficient* / 95% CI / P value
Self-reported health (1=poor, 5=excellent) / 0.08 / 0.01, 0.15 / 0.03
Total days in poor health (days per month) / -0.65 / -1.4, 0.11 / 0.09
Days limited by poor health (days per month) / -0.70 / -1.6, 0.23 / 0.14
Poor physical health (days per month) / 0.07 / -0.80, 0.95 / 0.87
Poor mental health (days per month) / -0.82 / -1.4, -0.21 / 0.01
Disabled / -0.03 / -0.06, -0.001 / 0.04
Depression / -0.04 / -0.08, -0.01 / 0.02
Health insurance coverage / 0.09 / 0.03, 0.14 / 0.003
Didn’t seek care due to cost / -0.03 / -0.06, -0.001 / 0.04
Check-up within last year / 0.03 / -0.004, 0.07 / 0.08
Personal doctor / 0.009 / -0.02, 0.04 / 0.59

*Coefficient represents 4from Equation (3)

NOTES A negative coefficient indicates a decrease among non-elderly adults in Medicaid expansion states relative to non-elderly adults in non-expansion states, net changes among the elderly in expansion states relative to non-expansion states, whereas a positive coefficient indicates a relative increase among non-elderly adults in Medicaid expansion states, net changes among the elderly in expansion states. Estimates are adjusted for state and quarter-year fixed effects, age, sex, race/ethnicity, education, marital status, and annual state unemployment rate.

eFigure 1.Trends Among Childless Adults With Chronic Conditions

eFigure 1.Trends Among Childless Adults With Chronic Conditions (cont.)

eFigure 2.Trends Among Childless Adults Without Chronic Conditions

eFigure 2.Trends Among Childless Adults Without Chronic Conditions (cont.)

1

eTable 6. Changes in Chronic Condition Prevalence Following Medicaid ExpansionAmong Childless Adults
Unadjusted pre-ACA prevalence / Unadjusted post-ACA prevalence / Unadjusted
difference-in-differences / P value / Adjusted
difference-in- differencesa / Pvalue
Medicaid Expansion State / 48.5 / 50.4 / -0.55
(-2.4,1.3) / .55 / -0.87
(-2.5,0.75) / .29
Non-Medicaid Expansion State / 52.7 / 55.1
aAdjusted analyses accountfor state and quarter-year fixed effects, age, sex, race/ethnicity, education, marital status, and annual state unemployment rate
NOTE A negative coefficient (i.e., unadjusted and adjusted difference-in-differences estimates) indicates a decrease in states that expanded Medicaid relative to states that did not expand Medicaid, whereas a positive coefficient indicates a relative increase in Medicaid expansion states.

1