University of Southern California Dissertations and Theses

Three essays on estimating the effects of government programs and policies on health care among disadvantaged population

(USC Thesis Other)

Three essays on estimating the effects of government programs and policies on health care among disadvantaged population

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

THREE ESSAYS ON ESTIMATING THE EFFECTS OF GOVERNMENT PROGRAMS
AND POLICIES ON HEALTH CARE AMONG DISADVANTAGED POPULATION

by

Tianyi Lu

A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree

DOCTOR OF PHILOSOPHY
(HEALTH ECONOMICS)

August 2019
I

TABLE OF CONTENTS
Chapter 1: Introduction ............................................................................................................... 1
Chapter 2: The Impact of China’s New Rural Pension Scheme on Health Care Access and
Outcomes ................................................................................................................................... 4
1. Introduction .......................................................................................................................... 4
2. Methods................................................................................................................................ 7
2.1 Research Design ........................................................................................................... 7
2.2 Data and Sample ........................................................................................................... 8
2.3 Measure ........................................................................................................................ 8
2.4 Statistical Analysis ..................................................................................................... 10
3. Results ................................................................................................................................ 11
4. Discussion .......................................................................................................................... 14
References ................................................................................................................................. 17
Chapter 3: Disparities in health care coverage and access by English language proficiency,
2009-2016 ................................................................................................................................. 29
1. Introduction ........................................................................................................................ 29
2. Methods.............................................................................................................................. 30
2.1 Study Design............................................................................................................... 30
2.2 Data and Study Population ......................................................................................... 31
2.3 Study Variables........................................................................................................... 31
2.4 Statistical Analysis ..................................................................................................... 32
3. Results ................................................................................................................................ 33
4. Discussion .......................................................................................................................... 35
References ................................................................................................................................. 37
Chapter 4: Medicaid Eligibility Expansions May Address Gaps In Access To Diabetes
Medications ................................................................................................................................. 46
1. Introduction ........................................................................................................................ 46
2. Study Data and Methods .................................................................................................... 48
2.1 Study Design............................................................................................................... 48
2.2 Data ............................................................................................................................. 49
2.3 Statistical Analysis ..................................................................................................... 49
2.4 Limitations .................................................................................................................. 51
3. Results ................................................................................................................................ 51
4. Discussion .......................................................................................................................... 53
5. Conclusion ......................................................................................................................... 56
References ................................................................................................................................. 57
Appendix ................................................................................................................................... 66
Chapter 5: Conclusion .............................................................................................................. 84

1

Chapter 1: Introduction
This dissertation consists of three essays that investigated the effects of government programs
and policies on health care among disadvantaged population. I examined different populations
in different policy contexts.
In the first essay, I examined the effect of the New Rural Pension Scheme (NRPS) on income,
health care use, and health among older adults in rural China. China is home to 1 in 4 people
aged 60 and older worldwide. The majority live in rural China, where poverty and low access to
health care can impede healthy aging. In 2009, with the goal of reducing poverty among rural
older adults, the Chinese government launched the New Rural Pension Scheme (NRPS) for
adults aged 60 and older. I used data from the China Health and Retirement Longitudinal Study
to assess the level of pre-existing poverty and uptake of NRPS at the province level, and used a
regression discontinuity design to assess the impacts of NRPS eligibility at age 60. I repeated
the analysis before and after expansions in catastrophic medical insurance in 2012. In 2011,
77% of adults in rural China aged 58-59 reported no current source of income. NRPS eligibility
at age 60 increased income by 85 yuan per month (USD $12), and increased inpatient care use
by 10 percentage points, prior to the expansion of catastrophic medical insurance in 2012. I did
not detect an effect of NRPS eligibility on outpatient care or health outcomes. NRPS uptake was
less than 25% in multiple high-poverty provinces. The findings indicated that the NRPS and
catastrophic medical insurance programs increased access to inpatient care among older adults
in rural China. Improved targeting of the poor may be important to increase NRPS program
impact.
2

In the second essay, I described changes in insurance coverage and access to health care by
English language proficiency over 2006-2016. In the United States, people with limited English
proficiency (LEP) disproportionately experience gaps in health insurance coverage and access to
care. The Patient Protection and Affordable Care Act (ACA) of 2010 included reforms that
could improve these outcomes for people with limited English proficiency. Data used in this
study was from the Medical Expenditure Panel Survey over 2006-2016. I used regression
models to estimate changes in coverage and access before and after 2010 for adults with high vs.
limited English proficiency, adjusting for respondents’ socio-economic status, demographic
characteristics, and health care needs. I used difference-in-differences regression models to
assess adjusted changes in disparities by English proficiency after 2010. Gains in health
insurance coverage after 2010 were significant for adults with high English proficiency and
adults with limited English proficiency (2.5 percentage points, p<0.000, and 5.0 percentage
points, p=0.003, respectively); disparities by English proficiency did not significantly change.
Adults with LEP showed significantly larger improvements than adults with high English
proficiency in access to care, including having a usual source of care (3.3 percentage points,
p=0.022), access to needed medical care and dental care (2.2 percentage points, p=0.001, and
3.1 percentage points, p=0.001, respectively). Findings remained similar when we used nearest-
neighbor propensity-score matching to balance the characteristics of respondents. The findings
in this study indicated that disparities in health care access by English proficiency declined after
2010, the year of passage of the ACA.
The third essay examined the impacts of Medicaid eligibility expansion on access to prescription
medications for diabetes. We analyzed IQVIA’s Xponent data on over 96 million prescription
fills for diabetes medications using Medicaid insurance from January 2008 to December 2015.
3

Using a differences-in-differences design, we find that Medicaid eligibility expansions were
associated with thirty additional Medicaid diabetes prescriptions filled per 1,000 population in
2014 and 2015, relative to states that did not expand Medicaid eligibility. Age groups with
higher prevalence of diabetes exhibited larger increases. The increase in prescription fills grew
significantly over time. Overall, fills for insulin and for newer medications increased by 40
percent and 39 percent, respectively. Our findings suggest that Medicaid eligibility expansions
may address gaps in access to diabetes medications, with increasing effects over time.

4

Chapter 2: The Impact of China’s New Rural Pension Scheme on Health Care Access
and Outcomes
1. Introduction
China is home to one in four people aged 60 or older worldwide, and is one of the most rapidly
aging countries in the world.
1
Fertility has declined sharply since the 1970s, implying that the
proportion of people aged 60 or older will expand from 15% in 2015 (200 million people) to
34% in 2050 (450 million people).
1
Given the diminishing number of younger adults available
to provide care for each older adult, boosting the health and independence of older adults in
China is crucial.
2

According to 2010 population census, 55.9% of older adults in China live in rural areas, where
low incomes and low access to inpatient health care can impede healthy aging.
3
In 2008, only
about 10% of older adults in rural China were covered by old-age pension schemes, in contrast
to higher rates among older adults in urban areas and former government workers.
2
Older adults
in rural China also face higher out-of-pocket medical spending than older adults in urban areas
because of the less generous health insurance that covers them, the New Cooperative Medical
Scheme (NCMS).
4,5
Urban employees with government-sponsored insurance have 34% cost-
sharing for inpatient care costs, compared with a median of 75% cost-sharing for rural residents
under NCMS.
5
Rural older adults are more likely than urban older adults to drop out of inpatient
treatment due to financial difficulties,
4,5
and suffer from elevated rates of undiagnosed or
untreated conditions,
6,7
as well as elevated rates of mortality from chronic disease.
7

In 2009, in response to concerns about poverty and lower quality of life among rural older
adults, the Chinese government introduced the New Rural Pension Scheme (NRPS). The
ultimate goal of the program was to reduce poverty among older adults in rural areas by
5

providing a guaranteed basic pension at 60 years of age.
8
NRPS consists of two components, the
basic pension and personal account. The basic pension component is primarily sponsored by the
central government. Its initial benefit in 2009 was 55 yuan per month, which is close to the
poverty line in rural China.
9
Local governments are encouraged to make additional contributions
from their own revenues. Enrollees can contribute 100-500 yuan annually and are required to
contribute for at least 15 years to be eligible for additional pension income from a personal
account at age 60.
10

By providing a new source of income, NRPS program could plausibly increase the uptake of
needed health care that was previously unaffordable for rural older adults. The impact of NRPS
income on health care use could be moderated by subsequent policy changes. To protect
families from medical impoverishment by reducing cost sharing, China launched catastrophic
medical insurance starting in 2012, which was available in rural areas free of charge. The
introduction of this supplemental coverage increased reimbursement rates for care in rural China
to 50% or higher.
11
We hypothesize that the impact of NRPS on access to health care was higher
before the introduction of supplemental coverage.
Despite the dearth of evidence specific to China, international evidence suggests that the
impacts of supplemental income on health care use may be important. Broader studies of the
associations between income and use of health care on the country-level find increasing health
care expenditures as income increases,
12
while studies using natural experiments such as oil
price shocks,
13
a Social Security benefit notch,
14
and lottery wins
15
have suggested that
supplemental income can increase the use of pharmaceuticals and health care visits. These
studies of natural experiments focus on developed countries, whereas the majority of people
foregoing needed health care live in developing countries. Studies of pension programs in
6

Mexico found that support for low-income adults increased doctor visits and the purchase of
pharmaceuticals,
16
whereas in Colombia, pension program reduced hospitalizations among poor
older men in Colombia.
17
The mixed impacts of income on health care use across settings
suggest that the extent to which these effects are generalizable is not known.
The importance of China to trends in healthy aging worldwide and the substantial public
investment in NRPS suggest the importance of independent evaluations of the program’s
impacts. Currently, little is known about the impacts of NRPS on health care use, a key
component of healthy aging that was previously unaffordable to many rural older adults in
China. Previous studies of NRPS have focused on its impact NRPS on income, expenditures,
migration decisions, and reported mental and physical health rather than the use of health
care.
9,18
Similarly, little is known about the impacts of supplemental catastrophic insurance on
use of health care.
We used data from three waves of the China Health and Retirement Longitudinal Study
(CHARLS), a nationally representative sample of Chinese residents aged 45 and older and their
spouses, to examine the impact of China’s New Rural Pension Scheme (NRPS) on access to
health care among older adults in rural China, and interactions with catastrophic medical
insurance. Specifically, we document the level of pre-existing poverty and uptake of NRPS at
the province level, and assess the impact of becoming eligible for NRPS at age 60 on
respondents’ income, use of health care, and health outcomes including disability. We
additionally document how these impacts change after the introduction of catastrophic medical
insurance. This independent assessment contributes to our understanding of the impacts of
large-scale safety net expansions on healthy aging in rural China.
7

2. Methods
2.1 Research Design
Two elements of NRPS policy are crucial to the research design used in this paper. First, people
who were aged 60 or older at the time of introduction of the program could directly receive the
basic pension income without any contribution. NRPS program was envisioned as a
contributory pension scheme for future generations, but people who were already age-eligible at
the time of program roll-out had had no opportunity to contribute.
19
Therefore, people above
aged 60 at the time of program introduction could receive the basic pension income for free,
whereas people below age 60 could not. Second, NRPS was not introduced to all areas of rural
China simultaneously. In 2009, 320 (11%) pilot counties were selected to implement this
program. 450 (16%) and 1076 (39%) counties were added in the following two years, and the
program reached nationwide coverage by the end of 2012.
We used a regression discontinuity design which analyzed receipt of NRPS as a discontinuous
function of age. Regression discontinuity designs are frequently used to analyze the impacts of
policies on health care use and health outcomes.
20
The assumptions underlying the analysis are
that people do not manipulate their age to gain entry to the program, and that the probability of
receiving pension income increased discontinuously at age 60 whereas other factors were
continuous around this threshold. The plausibility of both assumptions was directly assessed in
our statistical analysis. Our research was approved by the University of Southern California
Institutional Review Board.
8

2.2 Data and Sample
We used individual-level data from three waves of the China Health and Retirement
Longitudinal Study (CHARLS). CHARLS is a nationally representative sample of Chinese
residents aged 45 and older and their spouses. The sample includes 17,708 respondents from
more than 10,000 households in 450 villages/urban communities, 150 counties/districts and 28
provinces of China. The baseline survey was conducted in 2011 and the follow-up surveys were
in 2013 and 2015. CHARLS is harmonized with the Health and Retirement Study (HRS) family
of studies. The survey collects a wide range of information on demographics and family
structure, health status and functioning, health insurance and health care, income and
consumption, work, retirement and pension.21
Sample selection issues were minimal. We included people who were rural residents (i.e., had
rural hukou according to the Chinese household registration system) in 2011-2015. People in the
baseline survey of 2011 and people in 2013 and 2015 were analyzed separately because in the
follow-up two years, people with NCMS were automatically enrolled in the newly implemented
catastrophic medical insurance. We focused on people whose ages were close to the cutoff point
of age 60 and the number of observations were determined by the selected bandwidth.
2.3 Measure
Our main outcome of interest was the use of health care and whether the use was in a general
hospital. In China, top-tier general hospitals in cities provide comprehensive health services
with high quality, whereas village clinics and township hospitals are the main primary care
provider.
22
Patients consider general hospitals as a costlier but higher quality source of inpatient
and outpatient care.
22
We constructed five binary variables indicating the use of any outpatient
9

care, the use of any inpatient care, and using outpatient or inpatient care in a general hospital.
Secondary outcomes of interest included self-reported health status, a binary variable that
indicated an individual had one or more chronic conditions, the number of chronic conditions,
and a binary variable that indicated an individual had any difficulties with activities of daily
living (ADL) such as walking and bathing. We measured income (whether the person had any
personal income last year and log of last year’s personal income) as an outcome of interest to
check that the pension income was received. Finally, to document the level of pre-existing
poverty and uptake of NRPS at the province level, we calculated the proportions of rural adults
aged 45-59 who reported having no income in 2011 and rural adults aged 60 and older who
reported receiving NRPS income in 2015 by province.
The exposure variables of interest were NRPS implementation at village level and age-eligibility
for the pension benefit. The CHARLS data include a survey of the committee office in each
village which provide information on social policies implemented.
21
Treated villages were
defined as villages that had implemented NRPS in 2011; all other villages were considered to be
control villages. In the follow-up surveys of 2013 and 2015, all villages became treated villages
as the policy reached nationwide coverage by the end of 2012. We also constructed an indicator
variable of being age-eligible (over age 60) for the program.
We extracted individual-level factors for use in multivariate modeling and balance checks.
These variables included age, gender, marital status, spouse living in the household or not,
education attainment (primary school or below, middle school, high school, some college,
college or above), self-reported health status varying from 1 (excellent) to 6 (very poor) and
indicators of having one or more chronic conditions, disabilities, and difficulties with activities
of daily living (ADL).
10

2.4 Statistical Analysis
To identify the casual effect of additional pension income on the use of health care among rural
older adults, we employed a regression discontinuity design, a method widely used in health
care research.
20
We limited our data to a small window around age 60 to compare health care
utilization for individuals aged above vs. below 60, and used a data-driven bandwidth selection
procedure based on mean squared-error.
23
This procedure suggests an optimal bandwidth of 6
years for our data. In robustness checks, we repeated the analysis using 3, 4, 5 or 10 years as the
bandwidth.
To implement the analysis, we used a local linear specification with a triangle kernel, which
places higher weight on observations closer to age 60.
24
Robust bias-corrected standard errors
were clustered at village level. We also estimated the following OLS model for person i in
village j:
𝑌 𝑖𝑗
= 𝜇 + 𝛾 1
𝑃𝑜𝑠𝑡 60
𝑖𝑗
+ 𝛾 2
𝑃𝑜𝑠𝑡 60
𝑖𝑗
× ( 𝐴𝑔𝑒 𝑖𝑗
− 60 ) + 𝛾 3
𝑈𝑛𝑑𝑒𝑟 60
𝑖𝑗
× ( 𝐴𝑔𝑒 𝑖𝑗
− 60 ) + 𝛽𝑋
𝑖𝑗
+ 𝜃𝑉
𝑗 + 𝘀 𝑖𝑗

where Yij indicates the outcome variables of health care use. Post60ij indicates that the individual
is older than 60 and therefore age-eligible for pension income while Under60ij indicates that the
individual is younger than 60. Ageij denotes an individual’s age in quarters. We allowed age
trend terms to vary above vs. below the cutoff. Xij is a vector of control variables. The model
also adjusted for village-level control variables Vj, including village-level averages of personal
and household income and number of children in household and village-level average
proportions of people with high school diploma and people with one or more chronic conditions.
The coefficient of interest is 1, which captures discontinuities at age 60. Heteroscedasticity
11

robust standard errors were clustered at village level to account for the repeated sampling of
villages across waves of the survey, possible sharing of income across people within a village,
and village-level variation in NRPS program.
We conducted tests of the key assumptions underlying our analytic strategy. The first
assumption was that people did not manipulate their age to gain entry to the program. We used a
McCrary density test to check for heaping of the age variable near the cutoff of 60, which would
be evidence of age manipulation.
25
The second assumption was that the probability of receiving
NRPS increased discontinuously at age 60 whereas other determinants of our outcomes of
interest were continuous at age 60. If this assumption holds, there would be no discontinuity in
receipt of NRPS or our outcomes of interest at age 60 in the absence of the program. We
assessed the plausibility of this assumption by applying the regression discontinuity analysis
villages that had not yet received the program in 2011.
3. Results
Figures 1 and 2 depict trends in income and the proportion of people who receiving pension
income just above vs. below age 60, in villages with and without NRPS program. The figures
use data from 2011, when NRPS was not yet implemented in all villages. NRPS program receipt
jumped from nearly zero percent to between 30 to 40 percent at age 60 in villages with NRPS
program, suggesting that the program was indeed delivered to those aged 60 years and older and
not to younger people. This is the variation used in our analysis. Receipt of NRPS remained
near zero and income remained stable above and below age 60 in villages that did not yet have
the program in 2011, suggesting the validity of our placebo check.
12

Table 1 shows descriptive statistics for respondents aged 55-59, just a few years younger than
the age cutoff for NRPS, in villages with and without NRPS program in 2011. There were no
significant differences between individuals in treated and control villages for most of these
predetermined characteristics such as gender, marital status, and educational attainment.
Compared to their counterparts in villages without NRPS, people aged 55-59 living in villages
with NRPS were less likely to have a high school diploma and had fewer children living in the
household. We found significant differences in personal and household annual income between
treated and control villages, indicating that NRPS was implemented first in wealthier locations.
These unbalanced characteristics were controlled for in our regression analysis.
Figure 3 depicts the proportion of rural adults ages 45-59 reporting no income, who may have
little chance to save for retirement; Figure 4 depicts the proportion of rural adults receiving any
pension income from NRPS. Xinjiang, Gansu, Yunnan, and Guizhou had high poverty rates and
higher NRPS program uptake, whereas Guangxi, Inner Mongolia, and Heilongjiang had high
rates of poverty but very low rates of NRPS program receipt.
Table 2 reports the results for the regression discontinuity design. Column 2 and 3 display our
findings in 2011, one year before the implementation of catastrophic medical insurance, and
Column 4 and 5 show the findings in 2013 and 2015, after the launch of the policy. These data
are displayed alongside the baseline means for people aged 58-59 in treated villages in Column
1. We found that the probability of having any personal income increased significantly by 39 to
50 percentage points at age 60, representing a 170% to 217% increase compared with people
aged 58-59 in villages with NRPS.
13

The use of inpatient care increased significantly at age 60 in 2011, including a 10 percentage-
point (200%) increase in the probability of any inpatient care last year and a 5 percentage-point
(14%) increase in the probability of using inpatient care in a general hospital. The coefficients
on inpatient care were not statistically significant when we used data after 2012, indicating that
when a more generous reimbursement plan for inpatient care became available, the effects of
pension income on inpatient care use were no longer statistically significant. We did not find
any significant effects on outpatient care, self-reported health, chronic conditions, or disability
as measured by activities of daily living.
Multiple robustness checks supported the plausibility of the assumptions underlying our
analysis. Our first key assumption was no evidence for manipulation of age in villages where
NRPS program was implemented. Consistent with this assumption, a manipulation test showed
that we could not reject the null hypothesis of no difference in the number of people reporting
ages just above vs. just below age 60 (T=-1.2251, p=0.2206 and T=-0.2033, p=0.8389 for the
2011 and 2013-2015 data waves, respectively). As evidence for our second analytic assumption
that our outcomes would have been continuous at age 60 in the absence of the program, we
found no additional change in our outcomes of interest at age 60 in villages without NRPS.
These data are shown in Columns 6 and 7 of Table 2. Finally, Table 3 shows that our results are
not sensitive to the bandwidth of data used around age 60. The coefficients on inpatient care use
in 2011 remain qualitatively unchanged and statistically significant at a range of bandwidths and
model specifications.
14

4. Discussion
This study examines the impacts of the New Rural Pension Scheme (NRPS), a pension program
for the estimated 99.3 million older adults in rural China, on the use of health care.
3
We
employed a regression discontinuity design strategy that leveraged the onset of pension
eligibility at age 60 and the slow roll-out of the program across rural China. We found increases
in pension income at age 60 in rural villages with NRPS but not in villages without NRPS,
suggesting the validity of our analytic strategy to obtain causal effects.
We found that NRPS increased the income of older people in rural areas of China and improved
access to inpatient care. Based on CHARLS individual-level data, in 2011, the average monthly
pension benefit was 85 Chinese yuan, accounting for 55% of the monthly average income
among people aged 60-64 who received the pension. NRPS eligibility improved uptake of
needed inpatient care by 10 percentage points in 2011 and induced rural older adults to switch to
higher-quality providers for their care. Finally, NRPS only significantly increased health care
use prior to the launch of catastrophic health insurance coverage, suggesting that the availability
of catastrophic medical insurance shaped how NRPS income was spent.
The importance of our findings is underscored by the substantial burden of inpatient care costs
and medical impoverishment in rural China prior to NRPS. In 2008, just prior to the onset of
NRPS, the average inpatient expenditures accounted for 56% of per capita annual income
among people covered by the New Cooperative Medical Scheme (NCMS), the government
sponsored health insurance available in rural China.
7
About half of rural respondents in
CHARLS reported that more than 20% of total personal expenditure was spent on out-of-pocket
payments for inpatient care.
5

15

Our finding that NRPS income increased the affordability of inpatient is plausible, given the
large magnitude of NRPS’s impact on poverty. Overall, NRPS significantly reduced the
incidence of poverty by 20-35 percent in rural China.
26
While we found significant impacts of
NRPS on use of inpatient care, we did not find any effect on use of outpatient care. This may
reflect the fact that NCMS coverage for outpatient care was relatively generous compared to
coverage for inpatient care.
27

Several provinces with high poverty such as Guangxi, Inner Mongolia, and Heilongjiang had
low program coverage at age 60, with fewer than a quarter of rural adults aged 60 and older
receiving income from NRPS. Methods to increase NRPS coverage in these provinces should be
considered to enhance the program’s impact among vulnerable older adults.
Our study has limitations. We found that the effects of NRPS on the use of inpatient care were
stronger when access to catastrophic medical insurance was unavailable, i.e., prior to 2012.
However, we cannot rule out the possibility that other changes in 2012 aside from the expansion
of catastrophic medical insurance account for this change in the impacts of NRPS. Additionally,
our conclusions are limited based on the self-reported nature of the CHARLS data. While we
were unable to detect an impact of NRPS on self-reported health and disability, there might be
impacts on physical health we were unable to measure.
In conclusion, our study provides new evidence that an unconditional cash transfer to rural older
adults of on average 85 Chinese yuan per month (USD $12) improved uptake of inpatient care
by about 10 percentage points when generous health insurance was unavailable. These findings
suggest that income transfers can help improve access to care for vulnerable older adults in an
otherwise unequal system. In China, inequality between rural and urban areas exists in the
16

health insurance system, allocation of health care resources, catastrophic health expenditure, and
health services utilization.
28,29
Alleviating the financial burden of medical care can provide
much-needed support for healthy aging in rural China.
30

17

References
1. United Nations. World Population Prospects: The 2017 Revision, Medium Variant [Internet].
United Nations, Department of Economic and Social Affairs, Population Division; 2017
[cited 2019 Apr 15]. Available from:
https://population.un.org/wpp/Download/Standard/Population/
2. Cai F, Giles J, O’Keefe P, Wang D. The Elderly and Old Age Support in Rural China :
Challenges and Prospects [Internet]. Washington, DC: World Bank; 2012 [cited 2018 Sep
18]. Available from: https://openknowledge.worldbank.org/handle/10986/2249
3. National Bureau of Statistics of China. Tabulation on the 2010 population census of the
People’s Republic of China [Internet]. 2010. Available from:
http://www.stats.gov.cn/english/Statisticaldata/CensusData/rkpc2010/indexch.htm
4. Jian W, Chan KY, Reidpath DD, Xu L. China’s rural-urban care gap shrank for chronic
disease patients, but inequities persist. Health Aff Proj Hope. 2010 Dec;29(12):2189–96.
5. Zhang C, Lei X, Strauss J, Zhao Y. Health Insurance and Health Care among the Mid-Aged
and Older Chinese: Evidence from the National Baseline Survey of CHARLS. Health Econ.
2017 Apr;26(4):431–49.
6. Xu Y, Wang L, He J, Bi Y, Li M, Wang T, et al. Prevalence and Control of Diabetes in
Chinese Adults. JAMA. 2013 Sep 4;310(9):948–59.
7. Li J, Shi L, Li S, Xu L, Qin W, Wang H. Urban-rural disparities in hypertension prevalence,
detection, and medication use among Chinese Adults from 1993 to 2011. Int J Equity Health
[Internet]. 2017 Mar 14;16. Available from:
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5348878/
8. Shen C, Williamson JB. China’s new rural pension scheme: can it be improved? Int J Sociol
Soc Policy. 2010;30(5/6):239–250.
9. Cheng L, Liu H, Zhang Y, Zhao Z. The heterogeneous impact of pension income on elderly
living arrangements: evidence from China’s new rural pension scheme. J Popul Econ. 2018
18

Jan 1;31(1):155–92.
10. Chen X, Eggleston K, Sun A. The impact of social pensions on intergenerational
relationships: Comparative evidence from China. J Econ Ageing. 2018 Nov 1;12:225–35.
11. Li H, Jiang L. Catastrophic medical insurance in China. The Lancet. 2017 Oct
14;390(10104):1724–5.
12. Wagstaff A, van Doorslaer E. Measuring and Testing for Inequity in the Delivery of Health
Care. J Hum Resour. 2000;35(4):716–33.
13. Acemoglu D, Finkelstein A, Notowidigdo MJ. Income and Health Spending: Evidence from
Oil Price Shocks. Rev Econ Stat. 2012 Jul 17;95(4):1079–95.
14. Moran JR, Simon KI. Income and the Use of Prescription Drugs by the Elderly: Evidence
from the Notch Cohorts. J Hum Resour. 2006;41(2):411–32.
15. Cheng TC, Costa-Font J, Powdthavee N. Do You Have to Win It to Fix It? A Longitudinal
Study of Lottery Winners and Their Health-Care Demand. Am J Health Econ. 2017 Aug
22;4(1):26–50.
16. Aguila E, Kapteyn A, Smith JP. Effects of income supplementation on health of the poor
elderly: the case of Mexico. Proc Natl Acad Sci U S A. 2015 Jan 6;112(1):70–5.
17. Hessel P, Avendano M, Rodríguez-Castelán C, Pfutze T. Social Pension Income Associated
With Small Improvements In Self-Reported Health Of Poor Older Men In Colombia. Health
Aff Proj Hope. 2018 Mar;37(3):456–63.
18. Chen X, Wang T, Busch SH. Does money relieve depression? Evidence from social pension
expansions in China. Soc Sci Med. 2019 Jan 1;220:411–20.
19. State Council. Guidance Opinion on Developing Pilot Projects for the New Rural Social
Pension Insurance Plan [Internet]. 2009. Available from: http://www.gov.cn/zwgk/2009-
09/04/content_1409216.htm
20. Venkataramani AS, Bor J, Jena AB. Regression discontinuity designs in healthcare research.
19

BMJ. 2016 Mar 14;352:i1216.
21. Zhao Y, Hu Y, Smith JP, Strauss J, Yang G. Cohort Profile: The China Health and Retirement
Longitudinal Study (CHARLS). Int J Epidemiol. 2014;43(1):61–68.
22. Barber SL, Yao L. Development and status of health insurance systems in China. Int J Health
Plann Manage. 2011 Oct 1;26(4):339–56.
23. Calonico S, Cattaneo MD, Titiunik R. Robust Nonparametric Confidence Intervals for
Regression-Discontinuity Designs. Econometrica. 2014;82(6):2295–2326.
24. Calonico S, Cattaneo M, Titiunik R. Robust data-driven inference in the regression-
discontinuity design. Stata J. 2014;14(4):909–46.
25. McCrary J. Manipulation of the running variable in the regression discontinuity design: A
density test. J Econom. 2008 Feb;142(2):698–714.
26. Zhang C, Giles J, Zhao Y. A Policy Evaluation of China’s New Rural Pension Program:
Income, Poverty, Expenditure, Subjective Well-Being and Labor Supply. China Econ Q.
2014;14(1):203–30.
27. Babiarz KS, Miller G, Yi H, Zhang L, Rozelle S. China’s New Cooperative Medical Scheme
Improved Finances Of Township Health Centers But Not The Number Of Patients Served.
Health Aff (Millwood). 2012 May 1;31(5):1065–74.
28. Anand S, Fan VY, Zhang J, Zhang L, Ke Y, Dong Z, et al. China’s human resources for
health: quantity, quality, and distribution. The Lancet. 2008 Nov 15;372(9651):1774–81.
29. Liu M, Zhang Q, Lu M, Kwon C-S, Quan H. Rural and urban disparity in health services
utilization in China. Med Care. 2007 Aug;45(8):767–74.
30. Gu D, Zhang Z, Zeng Y. Access to healthcare services makes a difference in healthy
longevity among older Chinese adults. Soc Sci Med. 2009 Jan 1;68(2):210–9.

20

Figure 1 Receipt of NRPS pension income prior to and after NRPS eligibility at age 60

This graph shows the proportion of people who received NRPS pension income last year in treated and control
villages, i.e., rural villages with vs. without NRPS program. The x-axis is age, which ranges from 50 to 70. Age 60
is the age at which people are eligible to receive NRPS benefits. The scatterplots are overlaid with local linear lines
fit separately for treated and control villages, above and below this age cutoff. The data source is CHARLS 2011,
prior to the national implementation of NRPS.

21

Figure 2 Personal income prior to and after NRPS eligibility at age 60

This graph shows the proportion of people who report any source of income in the past year in treated and control
villages, i.e., rural villages with vs. without NRPS program. The x-axis is age, which ranges from 50 to 70. Age 60
is the age at which people are eligible to receive NRPS benefits. The scatterplots are overlaid with local linear lines
fit separately for treated and control villages, above and below this age cutoff. The data source is CHARLS 2011,
prior to the national implementation of NRPS.

22

Figure 3 Rural adults aged 45-59 reporting no income last year

This graph shows the proportion of rural people aged 45-59 who reported having zero income last year at the
province level. The data source is CHARLS 2011, prior to the national implementation of NRPS.

23

Figure 4 Rural adults aged over 60 receiving NRPS pension income

This graph shows the proportion of rural people aged over 60 who reported receiving any pension income from
NRPS at the province level. The data source is CHARLS 2015.

24

Table 1 Summary statistics at baseline (age 55-59, just prior to NRPS eligibility)
People aged 55-59
Variables Full Sample Rural villages with
NRPS
Rural villages without
NRPS
Demographic
Characteristics

Age 57.031 (0.035) 57.028 (0.047) 56.995 (0.039)
Female 0.509 (0.012) 0.507 (0.017) 0.51 (0.014)
Married 0.914 (0.010) 0.924 (0.010) 0.916 (0.008)
Education: Primary
school
0.774 (0.011) 0.801 (0.013) 0.774 (0.012)
Education: Middle
school
0.169 (0.011) 0.155 (0.012) 0.167 (0.010)
Education: High school 0.05 (0.004) 0.038 (0.006) 0.055 (0.006)**
Education: Some college 0.005 (0.002) 0.005 (0.002) 0.003 (0.001)
Education: College or
above
0.002 (0.001) 0 (0.000) 0.001 (0.001)
Personal Income 3075.612 (204.840) 3443.067 (348.035) 2443.645 (231.868)**
Household Income 20307.27 (2382.779) 19291.38 (1155.447) 14499.64 (731.962)***
Family Structure
Household size 3.092 (0.041) 3 (0.062) 3.099 (0.055)
No of Children 2.397 (0.024) 2.255 (0.036) 2.523 (0.031)***
No of Children living in
the household
0.794 (0.017) 0.715 (0.026) 0.823 (0.024)***
Health Status
Self-rated health status:
Excellent
0.006 (0.002) 0.008 (0.003) 0.005 (0.002)
Self-rated health status:
Very good
0.072 (0.005) 0.075 (0.009) 0.064 (0.007)
Self-rated health status:
Good
0.176 (0.011) 0.178 (0.013) 0.157 (0.010)
Self-rated health status:
Fair
0.438 (0.012) 0.438 (0.016) 0.437 (0.014)
Self-rated health status:
Poor
0.279 (0.011) 0.272 (0.015) 0.304 (0.013)
25

Self-rated health status:
Very poor
0.028 (0.003) 0.03 (0.005) 0.03 (0.005)
Having one or more
chronic conditions
0.67 (0.012) 0.675 (0.016) 0.678 (0.013)
Number of chronic
conditions
1.282 (0.03) 1.288 (0.045) 1.335 (0.038)
Having any difficulties
with activities of daily
living (ADL) 0.422 (0.011) 0.438 (0.016) 0.447 (0.014)
Current smoker 0.296 (0.01) 0.297 (0.015) 0.315 (0.013)
Health Care
Had outpatient care 0.191 (0.011) 0.185 (0.013) 0.195 (0.011)
Outpatient care in a
general hospital 0.051 (0.009) 0.041 (0.007) 0.045 (0.006)
Had inpatient care 0.079 (0.010) 0.058 (0.008) 0.083 (0.008)**
Inpatient care in a
general hospital 0.041 (0.009) 0.023 (0.005) 0.041 (0.006)**
*** p<0.01, ** p<0.05, * p<0.1
Notes: Weighted baseline characteristics were estimated using data from CHARLS 2011. We tested for significant
differences between people aged 55-59 in treated and control villages using t-tests. Standard errors are in
parenthesis.

26

Table 2 Estimated impacts of NRPS eligibility at age 60 on health and health care
utilization: A regression discontinuity analysis

(1) (2) (3) (4) (5) (6) (7)
Mean
at age
58-59,
treated
villages
, 2011-
2015
Impact of NRPS, before
catastrophic insurance
introduced
Impact of NRPS, after
catastrophic insurance
introduced
Placebo tests: Rural
villages without NRPS
OLS Local linear
regression
OLS Local
linear
regression
OLS Local
linear
regression
Personal
income
Any income
last year
0.23
0.475*** 0.505*** 0.413*** 0.394*** -0.0253 -0.0100
(0.0592) (0.0590) (0.0271) (0.0291) (0.0429) (0.0461)
N=1,922 N=1,893 N=5,457 N=5,396 N=2,756 N=2,733
Income (log)
1.98
3.133*** 3.398*** 2.865*** 2.717*** -0.354 -0.200
(0.457) (0.435) (0.197) (0.215) (0.362) (0.397)
N=1,922 N=1,893 N=5,456 N=5,395 N=2,756 N=2,733
Any catastrophic health
expenditure last year (more
than 30% personal income)
0.05
0.0560* 0.0470 0.00343 0.00459 -0.0362 -0.0113
(0.0311) (0.0346) (0.0215) (0.0223) (0.0262) (0.0299)
N=1,930 N=1,901 N=5,476 N=5,414 N=2,770 N=2,747
NCMS coverage
0.93
-0.0136 -0.00819 0.0256 0.0216 0.0189 0.0232
(0.0274) (0.0342) (0.0206) (0.0242) (0.0247) (0.0303)
N=1,928 N=1,899 N=5,476 N=5,415 N=2,767 N=2,744
Self-
reported
health
status
Excellent/Ver
y good/Good
0.28
-0.0328 -0.0347 0.0263 0.0277 -0.0122 0.0117
(0.0461) (0.0552) (0.0297) (0.0337) (0.0376) (0.0428)
N=1,930 N=1,901 N=5,422 N=5,361 N=2,767 N=2,744
Fair/Poor/Ver
y poor
0.72
0.0328 0.0347 -0.0263 -0.0277 0.0122 -0.0117
(0.0461) (0.0552) (0.0297) (0.0337) (0.0376) (0.0428)
N=1,930 N=1,901 N=5,422 N=5,361 N=2,767 N=2,744
Chronic
condition
s
Having one or
more
conditions
0.65
0.00948 0.0148 0.00203 -0.00936 -0.0096 -0.0069
(0.0479) (0.0594) (0.0324) (0.0352) (0.0440) (0.0542)
N=1,930 N=1,901 N=5,476 N=5,414 N=2,770 N=2,747
Number of
conditions
1.28
0.149 0.168 0.0369 0.0550 -0.0621 -0.0729
(0.167) (0.179) (0.0969) (0.105) (0.126) (0.159)
N=1,930 N=1,901 N=5,476 N=5,414 N=2,772 N=2,749
0.49 -0.0692 -0.103 0.0383 0.0267 0.00944 -0.0135
27

Having any difficulties with
activities of daily living
(ADL)
(0.0519) (0.0652) (0.0373) (0.0395) (0.0497) (0.0537)
N=1,930 N=1,901 N=5,476 N=5,414 N=2,772 N=2,742
Outpatien
t care
Any care last
month
0.18
0.0424 0.0365 -0.00876 0.0124 -0.0414 -0.0643
(0.0464) (0.0501) (0.0269) (0.0309) (0.0382) (0.0413)
N=1,928 N=1,899 N=5,451 N=5,390 N=2,765 N=2,742
Source of
care: general
hospital
0.03
0.0417** 0.0237 0.0292* 0.0360* -0.0013 -0.0114
(0.0206) (0.0240) (0.0161) (0.0185) (0.0205) (0.0234)
N=1,928 N=1,899 N=5,464 N=5,403 N=2,767 N=2,744
Number of
visits last
month
0.39
0.123 0.136 -0.00972 -0.00536 -0.0849 -0.0349
(0.133) (0.130) (0.103) (0.121) (0.129) (0.131)
N=1,928 N=1,899 N=5,464 N=5,403 N=2,767 N=2,744
Inpatient
care
Any care in
the past year
0.05
0.103*** 0.0944** 0.00866 0.00307 -0.0045 -0.0161
(0.0328) (0.0398) (0.0229) (0.0255) (0.0283) (0.0348)
N=1,928 N=1,899 N=5,454 N=5,393 N=2,765 N=2,742
Source of
care: general
hospital
0.08
0.0529** 0.0263 0.00706 0.0105 -0.0057 -0.0137
(0.0217) (0.0238) (0.0182) (0.0194) (0.0201) (0.0253)
N=1,928 N=1,899 N=5,464 N=5,403 N=2,767 N=2,744
Manipula
tion test
statistics
2011 data t=-1.2251 (p=0.2206)
2013-2015 data t=-0.2033 (p=0.8389)
*** p<0.01, ** p<0.05, * p<0.1
Notes: The models used in column (2), (4) and (6) were estimated using ordinary least squares. The models used in
column (3), (5) and (7) were estimated using local linear regression with a triangle kernel. Bandwidth was 6 years.
In column (2) and (3), only people living in treated villages in 2011 were included in the analysis, whereas in
column (4) and (5), people living in treated villages in 2011 and people in 2013 and 2015 were pooled together for
the analysis. The sample used for placebo tests, in column (6) and (7), included only people living in control
villages in 2011. Robust standard errors clustered at the village level are in parentheses.

28

Table 3 Robustness tests of the impact of NRPS eligibility at age 60 on income and health
care use: Changes in data bandwidth and model type

OLS Local linear regression
(1) (2) (3) (4) (5) (6) (7) (8)
±3 ±4 ±5 ±10 ±3 ±4 ±5 ±10
Personal
income
Any
income
last year
0.527*** 0.502*** 0.508*** 0.453*** 0.604*** 0.525*** 0.512*** 0.458***
(0.106) (0.0740) (0.0637) (0.0446) (0.117) (0.0800) (0.0652) (0.0448)
N=837 N=1,231 N=1,598 N=3,078 N=816 N=1,197 N=1,578 N=3,059
Income
(log)
3.590*** 3.359*** 3.440*** 2.986*** 4.079*** 3.554*** 3.450*** 3.031***
(0.809) (0.548) (0.492) (0.343) (0.881) (0.596) (0.481) (0.337)
N=837 N=1,231 N=1,598 N=3,078 N=816 N=1,197 N=1,578 N=3,059
Inpatient
care
Any care
in the past
year
0.120* 0.113** 0.0803** 0.068*** 0.134* 0.126** 0.101** 0.084***
(0.0751) (0.0456) (0.0387) (0.0206) (0.0795) (0.0571) (0.0456) (0.0267)
N=839 N=1,234 N=1,601 N=3,085 N=818 N=1,200 N=1,581 N=3,066
*** p<0.01, ** p<0.05, * p<0.1
Notes: The models used in column (1)-(4) were estimated using ordinary least squares, whereas the models used in
column (5)-(8) were estimated using local linear regression with a triangle kernel RD. ±3, ±4, ±5, ±10 denote
bandwidths set to 3, 4, 5 or 10 years. Robust standard errors clustered at the village level are in parentheses.

29

Chapter 3: Disparities in health care coverage and access by English language proficiency, 2009-2016
1. Introduction
In 2011, 8.8 percent of the United States population (25.3 million people) had limited English
proficiency (LEP), reporting that they speak a language other than English at home and they
speak English less than “very well”.
1
These figures are rapidly growing as the United States
population becomes increasingly diverse.
2
LEP has been widely documented as a barrier to
health care in the United States. People with LEP experience difficulties in obtaining health
insurance coverage,
3,4
accessing health care services,
5–11
receiving good quality care with high
patient satisfaction,
12–14
communicating with the health care provider,
15–19
using preventive
health care, such as cancer screening and influenza vaccinations,
7,20–24
and achieving medication
and treatment adherence.
25–28
People with LEP also experience worse health outcomes than
those with high English proficiency. They are more likely to report poor self-rated health status
and psychological distress.
5,7,29
They have higher odds of undiagnosed or uncontrolled
hypertension, poor glycemic control and asthma control.
28,30,31
LEP patients also have high risk
for unplanned emergency room (ER) visits,
32,33
prolonged hospital length of stay,
34,35
frequent
hospital readmission,
36
and serious adverse effects.
13,37,38

The Patient Protection and Affordable Care Act (ACA) of 2010 was designed to expand health
insurance coverage to Americans who were previously uninsured, improve access to care, and
advance health equity. It included numerous provisions that aimed at improving coverage and
access among disadvantaged populations, such as people with LEP. For example, the
nondiscrimination provision of the ACA asserts that any health programs and activities that
receive federal financial assistance must provide meaningful access to each individual with LEP
who may require assistance.
39
This requirement supplements prior legislation such as the Civil
Rights Act of 1965, the Executive Order 13166 of 2000 which prohibit discrimination by
national origin and set standards for providing meaningful health care access.
40
The ACA also
supported navigator programs and community health centers and loan-repayment programs to
encourage providers to locate in health professional shortage areas. These provisions could
facilitate health insurance enrollment and access to care among disadvantaged patients in
underserved areas, including people with LEP.
41
Additionally, the ACA revisited the National
30

Culturally and Linguistically Appropriate Service (CLAS) standards to underscore the
importance of implementing culturally and linguistically appropriate services that helped
eliminate health care disparities by English language proficiency.
40,42
These various ACA
provisions were implemented over time, with the earliest starting shortly after the law’s passage
in 2010, and could plausibly have a positive impact on health insurance coverage, access to, and
affordability of care among individuals with LEP.
40

Given that addressing barriers to care and promoting health equity were goals of the ACA, the
impacts of the ACA on this population with particularly low health care access and poor health
outcomes merits study. One study assessed the association between the ACA and patient-
provider communication by English language proficiency. It indicated that disparities in patient-
provider communication by English language proficiency narrowed but persisted after 2010.
40

However, little is known about how health insurance coverage and access to health care changed
among individuals with LEP after the ACA.
To address this gap, the objective of this study was to assess whether the ACA was associated
with improvements in insurance coverage and access to care for adults with limited English
proficiency. We also sought to assess changes in the disparities in coverage and access by
English language proficiency. To achieve these objectives, we used 2006-2016 data from the
nationally representative Medical Expenditure Panel Survey (MEPS). MEPS is available to
participants in either English or Spanish, and provides interpretation services to participants who
prefer to use other languages.
2. Methods
2.1 Study Design
Our study was designed to test two hypotheses. First, we hypothesized that individuals with
limited English proficiency, as an under-served population, would experience improvements in
coverage and access to care after the ACA. To test this hypothesis, we used multivariable
regression model to compare coverage and access before vs. after 2010, adjusting for a number
of potential confounders. These models were conducted separately for adults with high vs.
limited English proficiency.
31

Second, we hypothesized that pre-existing disparities by English proficiency would decline after
the ACA – that is, that gains in coverage and access under the ACA would be significantly
larger for individuals with limited English proficiency than for individuals with high English
proficiency. To test this hypothesis, we used a multivariable difference-in-differences regression
model to compare the changes in gaps in health insurance coverage and access to care between
individuals with high vs. limited English proficiency (first difference) before vs. after 2010
(second difference), after adjustment for a number of potential confounders. In a robustness
check, we used nearest-neighbor propensity-score matching to balance the high vs. limited
English proficiency participants on a range of demographic, socio-economic, and health related
characteristics.
2.2 Data and Study Population
We used data from the Medical Expenditure Panel Survey (MEPS), which provides nationally
representative estimates for the U.S. civilian noninstitutionalized population. Our study included
data from the annual cross-sectional MEPS surveys over 2006 through 2016. Overall response
rates over these years ranged from 46.0% to 59.3%.
43
The study was approved by the University
of Southern California Institutional Review Board.
The ACA was passed in 2010. While some provisions were implemented in later years, other
provisions were implemented immediately. Accordingly, 2006-2009 was considered the pre-
ACA period and 2010-2016 was considered the post-ACA period.
Our study focused on nonelderly (18-64 years) U.S.-born adults and foreign-born adults who
have lived in the country for more than five years. Following previous studies of the ACA, we
selected this sample because the ACA’s provisions were designed to benefit U.S. citizen and
lawful non-citizens, and because people aged younger than 18 or older than 65 were less
impacted by ACA coverage expansions.
44

2.3 Study Variables
Our main predictor variable was limited English proficiency (LEP). We considered people who
have LEP if they reported speaking a language other than English at home and reported
speaking English less than “very well.” This strategy has been used by United States Census
32

Bureau and the American Community Survey (ACS) to identify people with limited English-
speaking ability.
45

The outcomes of interest were measures of health insurance coverage and access to care. The
health insurance coverage measure was a binary variable indicating whether the participant had
any health insurance coverage in the past 12 months. Measures of access to care included binary
variables that indicated whether the respondent had a usual source of health care and whether
the respondent needed necessary care (medical, dental, or preventive care) but was unable to
receive it.
The covariates used in multivariable modeling included information on respondents’ gender, age
group (age 18-24, 25-34, 35-44, 45-54 and 55-64), race (non-Hispanic white, non-Hispanic
black, non-Hispanic Asian, or Hispanic), marital status, educational level (less than high school
degree, high school degree, college degree, or advanced degree), household income (income less
than vs. above 138% federal poverty level, a cutoff relevant to eligibility for Medicaid insurance
under the ACA), employment, region of residence in the U.S. (Northeast, Midwest, South and
West), U.S.-born citizenship, self-reported health (good or excellent health, vs. fair or poor
health), and reporting any diagnosed chronic conditions. Categorical variables with three or
more categories were modeled using multiple binary variables.
2.4 Statistical Analysis
To compare the changes on the absolute scale in health insurance coverage and access to health
care between the pre- and post-ACA periods, we estimated multivariable linear regression
models separately for adults with high English proficiency and adults with limited English
proficiency. These models adjusted for the respondent’s gender, age group, race and ethnicity,
marital status, educational levels, household income, employment, U.S.-born citizenship, region
of residence, self-reported health, and diagnosed chronic conditions as specified above.
To estimate whether the preexisting disparities in coverage and access to care by English
proficiency diminished after the ACA, we used a difference-in-differences model. The
coefficient of interest in this model was an interaction term between an indicator of the post-
ACA period (i.e., 2010 or later) and an indicator of limited English proficiency. Specifically, we
estimated the following model:
33

𝑌 𝑖𝑡
= 𝜇 + 𝛽 1
𝐿𝐸𝑃 𝑖𝑡
+ 𝛾 1
( 𝐿𝐸𝑃 𝑖𝑡
× 𝑃𝑜𝑠𝑡𝐴𝐶𝐴 𝑖𝑡
) + 𝜃𝑌 𝑒𝑎 𝑟 𝑡 + 𝛿𝑋
𝑖𝑡
+ 𝘀 𝑖𝑡

where i indexes individual, t year, and 𝛾 1
was the coefficient of interest. Yit denotes the outcome
variables as noted above. LEPit is an indicator variable that equals 1 if the respondent had
limited English proficiency. LEPit×PostACAit is an interaction term between an indicator of
years after 2010 and limited English proficiency. Yeart controls for year fixed effects, which
adjust for secular trends. Xit is the set of control variables defined above.
Heteroscedasticity robust standard errors were used to account for heteroscedasticity in the
linear probability models we estimated. Analyses also incorporated used survey weights to
account for the survey design of the MEPS.
People with high vs. low English proficiency may differ in many important characteristics other
than English proficiency. In a robustness check, we matched respondents with high vs. limited
English proficiency on propensity scores calculated using logit model using the patient-level
characteristics Xit defined above using a nearest-neighbor matching procedure. The goal of using
this matching method was to limit potential confounding by balancing the respondents with high
vs. limited English proficiency on measured demographic factors, socio-economic status, and
variables related to health care need.
3. Results
Table 1 shows the characteristics of individuals with limited vs. high English proficiency, in the
MEPS data from 2006 to 2009. Compared to English proficient adults, adults with LEP were
significantly older (average age was 40.2 vs. 30.2 years among LEP vs. English proficient
adults, respectively), more likely to not have finished high school (87.5% vs. 66.4%), more
likely to live in a low-income household (41.7% vs. 29.3%), more likely to be married (52.7%
vs. 28.5%) and less likely to be employed (55.1% vs. 60.0%). Adults with LEP were
significantly more likely than English proficient adults to report fair or poor health status
(19.6% vs. 9.9%), but less likely to report having any chronic conditions (53.4% vs. 68.3%).
This finding matches prior analyses suggesting people with LEP are more likely to be
undiagnosed for their prevalent chronic conditions.
31,46,47

34

Table 2 reports the baseline means of the outcomes between individuals with high vs. limited
English proficiency, and the regression results for the single difference and difference-in-
differences models in the unmatched sample.
The first column of Table 2 shows the baseline data from prior to the ACA, in 2006-2009. These
data illustrate the gaps in coverage and access to care by English proficiency prior to the passage
of the ACA. 50.1% of respondents with low English proficiency had health insurance coverage
during the past 12 months, compared with 77.9% of respondents with higher English
proficiency. 10.1% of respondents with low English proficiency forewent necessary health care,
compared with only 5.4% of respondents with high English proficiency. The gap in access to
care by language proficiency was largest for dental care: 7.4% of respondents with low English
proficiency skipped necessary health care, compared with 3.7% of respondents with high
English proficiency.
The second column of Table 2 shows the adjusted changes in insurance coverage and access to
care after 2010. These data show significant improvements after 2010, particularly among
respondents with LEP. Insurance coverage increased by 5 percentage points (p=0.003) after
2010 for respondents with LEP. Access to care also improved for respondents with LEP after
2010: the probability of foregoing any necessary health care declined by 3.7 percentage points
(p<0.000), of foregoing necessary medical, dental, and preventive care declined by 2.3, 2.6, and
0.8 percentage points, respectively (p<0.000, p<0.000, p<0.000 respectively), and the
probability of having a usual source of care increased by 5.4 percentage points (p<0.000) after
2010. Respondents with high English proficiency experienced no significant changes in access
to care, and experienced an increase in insurance coverage by 2.5 percentage points (p<0.000)
after 2010.
Reflecting these disproportionate gains in health care access among respondents with LEP,
disparities in access to health care by English language proficiency significantly declined after
2010. These data are shown in the third column of Table 2. Respondents with LEP showed
larger declines after 2010 than respondents with high English proficiency in foregoing necessary
medical care (2.1 percentage points, p=0.001), necessary dental care (3.1 percentage points,
p=0.001), or any necessary care (4.4 percentage points, p<0.000). Respondents with LEP also
showed larger increases after 2010 in having a usual source of care (3.3 percentage points,
35

p=0.022). All estimates remained similar when propensity-score matching was used to balance
the included respondents with limited vs. high English proficiency on demographic
characteristics, socio-economic variables, and self-reported health; see Table 3.
4. Discussion
This study used data from the Medical Expenditure Panel Survey (MEPS) to document changes
in health insurance coverage and access to care among adults with high vs. limited English
proficiency before vs. after the ACA. We focused on people with limited English proficiency
because they are an underserved population with high uninsured rates, low level of access to
care, low rates of receiving needed health care and poorer health outcomes than people with
high English proficiency.
We found improved access to care among individuals with limited English proficiency after the
ACA, including improvements access to necessary health care and having a usual source of
care; we additionally documented significant reductions in disparities in these outcomes by
English proficiency. Our analysis adjusted for a number of potential confounders, and our
findings are robust to the use of matching to balance the high vs. limited English proficiency
samples on demographic, socio-economic, and health related variables.
These data add to the growing evidence that socioeconomic gaps in access to care diminished
after enactment of the ACA.
48–51
A number of provisions of the ACA were plausibly relevant to
people with limited English proficiency, including the provision of free, in-person help with
shopping for coverage in the new Health Insurance Marketplaces from navigators or certified
enrollment counselors; requirements that health care providers provide translation as necessary
for people with limited English proficiency; and provisions of additional funding for health care
providers who relocated to underserved areas.
A substantial share of the U.S. immigrant population resides in states such as Florida and Texas
that did not elect to expand Medicaid eligibility to non-disabled low-income adults, an ACA
policy that was deemed optional for the states by the Supreme Court.
52
These same states also
enacted strict requirements that diminished the capabilities of patient navigators and outreach
workers who were commissioned by the ACA to help patients enroll in coverage.
53
These local
36

variations may contribute to our finding that, even as gaps in access to care by English
proficiency closed after the ACA, gaps in coverage by English proficiency did not.
The study had several limitations. Firstly, geographic identifiers are not available in MEPS
public use files. This analysis only controlled for four U.S. regions (Northeast, Midwest, South
and West) and did not include state-level fixed effects. Future studies could take advantage of
the state-level information to examine cross-state variation and explore the effects of the ACA
on reducing health care disparities by English proficiency in states that expanded Medicaid
eligibility. Additionally, our analysis was based on the respondent’s recall of insurance
coverage, access to and use of health care, and it might be subject to recall bias. Finally, MEPS
did not distinguish citizens, non-citizens, and undocumented immigrants. Even though we
limited our sample to U.S.-born citizens and foreign-born people who have lived in the U.S. for
more than five years, we were unable to exclude undocumented immigrants from our analysis,
who were not eligible for Medicaid or insurance plans through the Marketplace. Nonetheless,
our matched analysis was more successful in restricting the sample to people with limited vs.
high English proficiency who were similar in demographic and socio-economic characteristics,
and the findings from the matched and unmatched analyses were similar.
Despite these limitations, our study provides new evidence of recent improvements in health
care access and reduction in disparities in health care access by English proficiency among U.S.
population, and documents improvements in coverage for people with high and low English
proficiency. It offers important insights into the changes that occurred for this important and
growing population under the ACA.

37

References
1. Ryan C. Language use in the United States: 2011. Am Community Surv Rep. 2013;22:1–16.
2. Zong J, Batalova J. The Limited English Proficient Population in the United States [Internet].
Washington, DC: Migration Policy Institute; 2015. Available from:
http://curry.virginia.edu/uploads/resourceLibrary/MPI_Report_2015.pdf
3. Caesar LG. English Proficiency and Access to Health Insurance in Hispanics Who Are
Elderly: Implications for Adequate Health Care. Hisp J Behav Sci. 2006 Feb 1;28(1):143–
52.
4. Gonzales G. State estimates of limited English proficiency (LEP) by health insurance status.
The State Health Access Data Assistance Center (SHADAC); 2014.
5. Ponce NA, Hays RD, Cunningham WE. Linguistic Disparities in Health Care Access and
Health Status Among Older Adults. J Gen Intern Med. 2006 Jul;21(7):786–91.
6. Cheng EM, Chen A, Cunningham W. Primary Language and Receipt of Recommended
Health Care Among Hispanics in the United States. J Gen Intern Med. 2007 Nov;22(Suppl
2):283–8.
7. DuBard CA, Gizlice Z. Language Spoken and Differences in Health Status, Access to Care,
and Receipt of Preventive Services Among US Hispanics. Am J Public Health.
2008 Nov;98(11):2021–8.
8. Brach C, Chevarley FM. Demographics and health care access and utilization of limited-
English-proficient and English-proficient Hispanics. Agency for Healthcare Research and
Quality; 2008.
9. Shi L, Lebrun LA, Tsai J. The influence of English proficiency on access to care. Ethn Health.
2009 Dec;14(6):625–42.
10. Smith D. Health Care Disparities for Persons with Limited English Proficiency:
Relationships from the 2006 Medical Expenditure Panel Survey (MEPS). J Health Disparities
Res Pract [Internet]. 2012 Apr 17;3(3). Available from:
https://digitalscholarship.unlv.edu/jhdrp/vol3/iss3/4
11. Lebrun LA. Effects of length of stay and language proficiency on health care experiences
among immigrants in Canada and the United States. Soc Sci Med 1982. 2012
Apr;74(7):1062–72.
12. Weech-Maldonado R, Morales LS, Elliott M, Spritzer K, Marshall G, Hays RD.
Race/Ethnicity, Language, and Patients’ Assessments of Care in Medicaid Managed Care.
Health Serv Res. 2003 Jun;38(3):789–808.
13. Divi C, Koss RG, Schmaltz SP, Loeb JM. Language proficiency and adverse events in US
38

hospitals: a pilot study. Int J Qual Health Care. 2007 Apr 1;19(2):60–7.
14. Pippins JR, Alegría M, Haas JS. Association Between Language Proficiency and the Quality
of Primary Care Among A National Sample of Insured Latinos. Med Care. 2007
Nov;45(11):1020–5.
15. Morales LS, Cunningham WE, Brown JA, Liu H, Hays RD. Are Latinos Less Satisfied with
Communication by Health Care Providers? J Gen Intern Med. 1999 Jul;14(7):409–17.
16. Wilson E, Chen AH, Grumbach K, Wang F, Fernandez A. Effects of Limited English
Proficiency and Physician Language on Health Care Comprehension. J Gen Intern Med. 2005
Sep;20(9):800–6.
17. Flores G. Language barriers to health care in the United States. N Engl J Med. 2006 Jul
20;355(3):229–31.
18. Lopez-Quintero C, Berry EM, Neumark Y. Limited English Proficiency Is a Barrier to
Receipt of Advice about Physical Activity and Diet among Hispanics with Chronic Diseases
in the United States. J Am Diet Assoc. 2009;109(10):1769–1774.
19. Karliner LS, Auerbach A, Nápoles A, Schillinger D, Nickleach D, Pérez-Stable EJ. Language
Barriers and Understanding of Hospital Discharge Instructions. Med Care. 2012
Apr;50(4):283–9.
20. Woloshin S, Schwartz LM, Katz SJ, Welch HG. Is Language a Barrier to the Use of
Preventive Services? J Gen Intern Med. 1997 Aug;12(8):472–7.
21. Jacobs EA, Karavolos K, Rathouz PJ, Ferris TG, Powell LH. Limited English proficiency
and breast and cervical cancer screening in a multiethnic population. Am J Public Health.
2005 Aug;95(8):1410–6.
22. Diaz JA, Roberts MB, Goldman RE, Weitzen S, Eaton CB. Effect of Language on Colorectal
Cancer Screening Among Latinos and Non-Latinos. Cancer Epidemiol Biomark Prev Publ
Am Assoc Cancer Res Cosponsored Am Soc Prev Oncol. 2008 Aug;17(8):2169–73.
23. Wallace SP, Gutiérrez VF, Castañeda X. Access to preventive services for adults of Mexican
origin. J Immigr Minor Health. 2008 Aug;10(4):363–71.
24. Pearson WS, Zhao G, Ford ES. An Analysis of Language as a Barrier to Receiving Influenza
Vaccinations among an Elderly Hispanic Population in the United States. Adv Prev Med
[Internet]. 2011;2011. Available from: http://dx.doi.org/10.4061/2011/298787
25. Masland MC, Kang SH, Ma Y. Association between limited English proficiency and
understanding prescription labels among five ethnic groups in California. Ethn Health. 2011
Apr;16(2):125–44.
26. Wisnivesky JP, Krauskopf K, Wolf MS, Wilson EAH, Sofianou A, Martynenko M, et al. The
association between language proficiency and outcomes of elderly patients with asthma. Ann
39

Allergy Asthma Immunol. 2012 Sep 1;109(3):179–84.
27. Moreno G, Lin EH, Chang E, Johnson RL, Berthoud H, Solomon CC, et al. Disparities in the
Use of Internet and Telephone Medication Refills among Linguistically Diverse Patients. J
Gen Intern Med. 2016 Mar;31(3):282–8.
28. Fernandez A, Schillinger D, Warton EM, Adler N, Moffet HH, Schenker Y, et al. Language
barriers, physician-patient language concordance, and glycemic control among insured
latinos with diabetes: The diabetes study of Northern California (DISTANCE). J Gen Intern
Med. 2011;26(2):170–176.
29. Kim G, Aguado Loi CX, Chiriboga DA, Jang Y, Parmelee P, Allen RS. Limited English
proficiency as a barrier to mental health service use: a study of Latino and Asian immigrants
with psychiatric disorders. J Psychiatr Res. 2011 Jan;45(1):104–10.
30. Wisnivesky JP, Kattan M, Evans D, Leventhal H, Musumeci-Szabó TJ, McGinn T, et al.
Assessing the relationship between language proficiency and asthma morbidity among inner-
city asthmatics. Med Care. 2009 Feb;47(2):243–9.
31. Kim EJ, Kim T, Paasche-Orlow MK, Rose AJ, Hanchate AD. Disparities in Hypertension
Associated with Limited English Proficiency. J Gen Intern Med. 2017 Jun;32(6):632–9.
32. Njeru JW, St. Sauver JL, Jacobson DJ, Ebbert JO, Takahashi PY, Fan C, et al. Emergency
department and inpatient health care utilization among patients who require interpreter
services. BMC Health Serv Res. 2015 May 29;15(1):214.
33. Ngai KM, Grudzen CR, Lee R, Tong VY, Richardson LD, Fernandez A. The Association
between Limited English Proficiency and Unplanned Emergency Department Revisit within
72 hours. Ann Emerg Med. 2016 Aug;68(2):213–21.
34. John-Baptiste A, Naglie G, Tomlinson G, Alibhai SMH, Etchells E, Cheung A, et al. The
Effect of English Language Proficiency on Length of Stay and In-hospital Mortality. J Gen
Intern Med. 2004 Mar;19(3):221–8.
35. López L, Rodriguez F, Huerta D, Soukup J, Hicks L. Use of interpreters by physicians for
hospitalized limited English proficient patients and its impact on patient outcomes. J Gen
Intern Med. 2015 Jun;30(6):783–9.
36. Lindholm M, Hargraves JL, Ferguson WJ, Reed G. Professional language interpretation and
inpatient length of stay and readmission rates. J Gen Intern Med. 2012 Oct;27(10):1294–9.
37. Hines AL, Andrews RM, Moy E, Barrett ML, Coffey RM. Disparities in Rates of Inpatient
Mortality and Adverse Events: Race/Ethnicity and Language as Independent Contributors.
Int J Environ Res Public Health. 2014 Dec;11(12):13017–34.
38. Tang EW, Go J, Kwok A, Leung B, Lauck S, Wong ST, et al. The relationship between
language proficiency and surgical length of stay following cardiac bypass surgery. Eur J
Cardiovasc Nurs. 2016 Oct 1;15(6):438–46.
40

39. U.S. Government. Affordable Care Act § 1557. 2010.
40. Berdahl TA, Kirby JB. Patient-Provider Communication Disparities by Limited English
Proficiency (LEP): Trends from the US Medical Expenditure Panel Survey, 2006-2015. J
Gen Intern Med. 2018 Dec 3;
41. Somers S, Mahadevan R. Health Literacy Implications of the Affordable Care Act. Center
for Health Care Strategies; 2010.
42. National CLAS Standards: Fact Sheet [Internet]. Available from:
https://www.thinkculturalhealth.hhs.gov/pdfs/NationalCLASStandardsFactSheet.pdf
43. Agency for Healthcare Research and Quality. MEPS-HC Response Rates by Panel. 2018.
44. Centers for Medicare & Medicaid Service. Eligibility for non-citizens in Medicaid and CHIP
[Internet]. 2014 Nov. Available from: https://www.medicaid.gov/medicaid/outreach-and-
enrollment/downloads/overview-of-eligibility-for-non-citizens-in-medicaid-and-chip.pdf
45. Bureau UC. About Language Use in the U.S. Population [Internet]. [cited 2019 Jun 19].
Available from: https://www.census.gov/topics/population/language-use/about.html
46. Johnson HM, Thorpe CT, Bartels CM, Schumacher JR, Palta M, Pandhi N, et al.
Undiagnosed hypertension among young adults with regular primary care use. J Hypertens.
2014 Jan;32(1):65–74.
47. Eamranond PP, Patel KV, Legedza ATR, Marcantonio ER, Leveille SG. The association of
language with prevalence of undiagnosed hypertension among older Mexican Americans.
Ethn Dis. 2007;17(4):699–706.
48. Sommers BD, Gunja MZ, Finegold K, Musco T. Changes in Self-reported Insurance
Coverage, Access to Care, and Health Under the Affordable Care Act. JAMA. 2015 Jul
28;314(4):366–74.
49. Wherry LR, Miller S. Early Coverage, Access, Utilization, and Health Effects Associated
With the Affordable Care Act Medicaid Expansions: A Quasi-experimental Study. Ann
Intern Med. 2016 Jun 21;164(12):795.
50. Miller S, Wherry LR. Health and Access to Care during the First 2 Years of the ACA
Medicaid Expansions. N Engl J Med. 2017 Mar 9;376(10):947–56.
51. Chen J, Vargas-Bustamante A, Mortensen K, Ortega AN. Racial and ethnic disparities in
health care access and utilization under the affordable care act. Med Care. 2016;54(2):140–
146.
52. Aug 01 P, 2012. A Guide to the Supreme Court’s Decision on the ACA’s Medicaid
Expansion [Internet]. The Henry J. Kaiser Family Foundation. 2012 [cited 2019 Jun 19].
Available from: https://www.kff.org/health-reform/issue-brief/a-guide-to-the-supreme-
courts-decision/
41

53. Pollitz K, Tolbert J, Sep 24 MDP, 2018. Data Note: Further Reductions in Navigator Funding
for Federal Marketplace States [Internet]. The Henry J. Kaiser Family Foundation. 2018
[cited 2019 Jun 19]. Available from: https://www.kff.org/health-reform/issue-brief/data-
note-further-reductions-in-navigator-funding-for-federal-marketplace-states/

42

Table 1 Baseline characteristics: Participants with limited vs. high English proficiency
Variable Participants with limited
English proficiency
Participants with high
English proficiency
Sample size 16,870 182,601
Average age in years 40.241 (0.66979) 30.184 (0.2129)***
Proportion with the following characteristics (%):
Female 51.3 (0.893) 50.9 (0.291)
Education: less than high school 87.5 (0.815) 66.4 (0.513)***
Education: high school diploma 20.3 (0.804) 23.9 (0.369)***
Education: college degree 10.6 (0.703) 27.1 (0.416)***
Education: advanced degree after college 1.9 (0.309) 6.5 (0.248)***
Low-income household 41.7 (1.891) 29.3 (0.625)***
Married 52.7 (1.422) 28.5 (0.386)***
Employed 55.1 (1.433) 60.0 (0.46)***
Self-reported health: poor or fair 19.6 (0.889) 9.9 (0.199)***
Self-reported health: excellent, very good or
good
80.4 (0.889)
90.1
(0.199)***
Any diagnosed chronic conditions 53.4 (1.444) 68.3 (0.331)***
Region: South 36.5 (3.537) 41.6 (1.220)
Region: Northeast 13.6 (1.372) 16.1 (0.708)
Region: West 41.5 (2.755) 28.6 (1.201)***
Region: Midwest 8.3 (1.138) 13.7 (0.683)***
Data source: Medical Expenditure Panel Survey (MEPS) 2006-2009
Standard are errors in parentheses.
We tested for significant differences between the groups using t-tests. Stars denote the significance of the
differences between participants with vs. without LEP: *** p<0.01, ** p<0.05, * p<0.1
43

Table 2 Changes in health insurance coverage and access to health care for adults aged 18-
64 after the ACA, by English language proficiency
Outcome Baseline Mean Change in outcomes
after the ACA
Change in disparities
after the ACA
Has health insurance
High English proficiency
0.778
0.0254***
(0.00315)
Reference
p value 0.000
Limited English
proficiency
0.499
0.0502***
(0.0165)
0.0166
(0.0164)
p value 0.003 0.354
Foregoing any necessary care
High English proficiency
0.055
-0.00191
(0.00259)
Reference
p value 0.463
Limited English
proficiency
0.097
-0.0374***
(0.00783)
-0.0444***
(0.00974)
p value 0.000 0.000
Forgoing any necessary medical care
High English proficiency
0.023
-0.00245
(0.00152)
Reference
p value 0.108
Limited English
proficiency
0.044
-0.0227***
(0.00540)
-0.0217***
(0.00566)
p value 0.000 0.001
Forgoing any necessary dental care
High English proficiency
0.038
-0.00228
(0.00212)
Reference
p value 0.284
Limited English
proficiency
0.07
-0.0264***
(0.00732)
-0.0308***
(0.00881)
p value 0.000 0.001
Forgoing any necessary preventive care
High English proficiency 0.012 -0.00140 Reference
44

(0.00123)
p value 0.258
Limited English
proficiency
0.02
-0.00848***
(0.00320)
-0.0103
(0.00421)
p value 0.009 0.017
Usual source of care
High English proficiency
0.721
0.000731
(0.00431)
Reference
p value 0.865
Limited English
proficiency
0.519
0.0536***
(0.0124)
0.0333***
(0.0147)
p value 0.000 0.022
Data source: Medical Expenditure Panel Survey (MEPS) 2006-2016. The pre-ACA period is 2006-2009, and the
post-ACA period is 2010-2016. The changes in outcomes and disparities (columns 2 and 3) are adjusted for the
geographic, socio-economic, demographic, and health related characteristics listed in the text. All analyses account
for the MEPS sampling scheme using sample weights. Robust standard errors are in parentheses.
*** p<0.01, ** p<0.05, * p<0.1

45

Table 3 Changes in health care disparities after the ACA by English language proficiency,
after propensity score matching
Variable Coefficient SE p value
Has insurance coverage 0.0117 (0.0124) 0.345
Access to care
Forgoing any necessary care -0.0286*** (0.00731) 0.000
Forgoing any necessary medical care -0.0171*** (0.00486) 0.000
Forgoing any necessary dental care -0.0104* (0.00613) 0.090
Forgoing any necessary preventive care -0.00648* (0.00376) 0.085
Has usual source of care 0.0406*** (0.0127) 0.001
Data source: Medical Expenditure Panel Survey (MEPS) 2006-2016
This replicates the analysis shown in column 3 of Table 2, except the exposure group (people with limited English
proficiency) and control group (people with high English proficiency) are matched using nearest-neighbor
propensity score matching. Propensity scores are constructed using the geographic, socio-economic, demographic,
and health related characteristics listed in the text. All analyses account for the MEPS sampling scheme using
sample weights. Robust standard errors are in parentheses.
*** p<0.01, ** p<0.05, * p<0.1.

46

Chapter 4: Medicaid Eligibility Expansions May Address Gaps In Access To Diabetes
Medications
1. Introduction
Expanding access to prescription medications for diabetes is critical for improving US
population health. Diabetes is one of the top ten causes of death and is a risk factor for heart
disease, the top cause of death.
1–3
Many of the complications of diabetes can be prevented by
the appropriate application of glucose-lowering drugs.
4–8
Yet not all people with diabetes
receive the medications they need.
9–11
High out-of-pocket spending contributes to treatment
nonadherence among patients with diabetes.
12–15
Uninsured people with diabetes may have
difficulty obtaining needed care and often show elevated risk of poor glycemic control.
16–18

The average per patient cost of diabetes medications has risen in recent years, in part because of
the increasing use of newer medications.
19,20
The number of medication categories for blood
glucose control has swelled from three to eleven since the early 1990s. In 2013 the mean
expenditure per patient for newer insulin analogues was almost double that for older
formulations; likewise, the mean expenditure per patient for newer oral antihyperglycemic
medications was almost double that for older oral medications.
21

Despite their higher cost, these newer medications can carry important health benefits. For
example, rapid- and long-acting insulin analogues provide equivalent glycemic control to large-
dose conventional insulin therapy but with significantly less hypoglycemia in a non–intensive
care setting.
22–24
Likewise, extended-release metformin is more effective than conventional
formulations in improving glyco-metabolic control and lipid profile with a convenient dosing
schedule.
25,26
Finally, glucagon-like peptide-1 (GLP-1) receptor agonists and sodium-glucose
cotransporter-2 (SGLT-2) inhibitors have shown favorable effects on rates of hypoglycemia and
47

body weight, which in turn shape patients’ risk of cardiovascular events and mortality.
27–31
As a
result, providing access to these costly newer medications could improve patients’ health.
The Affordable Care Act (ACA) originally required that all states expand eligibility for
Medicaid to all adults with incomes below 138 percent of the federal poverty level. However,
the US Supreme Court ruled that Medicaid expansion would be voluntary for the states.
32,33

Ultimately, twenty-five states expanded Medicaid eligibility in January 2014, and twenty-nine
states and the District of Columbia did so in either 2014 or 2015.
These expansions were associated with an increase in the number of Medicaid prescriptions per
enrollee and a drop in cost-related prescription nonadherence.
34–36
They also improved access to
primary care among newly insured patients, which translated into increased health care use.
37–41

A prior study showed increases in the numbers of diabetes prescriptions filled using Medicaid
insurance after Medicaid eligibility expansions, and did not crowd-out other types of insurance
to a great extent during the first fifteen months.
42
Notably, diabetes prescriptions increased more
than those in any other clinical category considered in that study. That study did not provide
estimates by age and sex, as our study does. Another study measured the increases in
prescribing of diabetes prescription drugs among people with one or more chronic conditions
who gained Medicaid coverage in the period January 2012–December 2014.
34
This study
focused on changes in prescription drug use over time among patients who had already filled
prescriptions at baseline.
We are not aware of any studies that measure the additional Medicaid diabetes prescriptions
filled during the first twenty-four months of the Medicaid eligibility expansions, or that report
how the expansions affected the use of specific classes of diabetes medications. A study of
48

specific drug classes would help define the health benefits associated with the expansions. A
substantial increase in the use of newer medications would imply that the expansions helped
resolve the slow diffusion of innovation to low-income patients with diabetes, possibly
improving their health.
43,44

Improving access to diabetes medications, including newer ones, has the potential to influence
the health of people living with diabetes for decades to come.
8
Therefore, to inform ongoing
policy discussions about expanding Medicaid eligibility in additional states or rolling back the
expansions in other states, we assessed the impact of the expansions on the use of diabetes
prescription medications filled using Medicaid insurance during the first twenty-four months
after the expansions. We present estimates of the changes in diabetes prescriptions filled after
the expansions by type of medication and patients’ sex and age. We used the estimates by age to
conduct multiple checks of the data. Finally, we conducted a trend analysis to examine whether
the changes associated with the expansions grew over time. Our study contributes to the
literature on the implications of the expansions for patients with chronic conditions.
2. Study Data and Methods
2.1 Study Design
We used a quasi-experimental difference-in-differences design to distinguish changes in
diabetes prescription fills related to Medicaid eligibility expansions from temporal trends.
Specifically, trends in diabetes prescription fills before versus after the expansions (the first
difference) were compared in states with versus those without such expansions (the second
difference). The pre-intervention period was January 2008–December 2013, and the post-
intervention period was January 2014–December 2015. We defined expansion states as those
49

twenty-nine states (and the District of Columbia) that expanded Medicaid eligibility in 2014 or
2015, and we classified the other twenty-one states as nonexpansion or control states. Appendix
A1 provides additional details on the classification of states.
2.2 Data
We measured fills of diabetes prescriptions using a large and representative administrative data
set, the IQVIA Xponent data. The data captured prescription fills in all fifty states and the
District of Columbia over eight years, including more than ninety-six million diabetes
prescription fills for patients ages 20–64 paid for by Medicaid insurance. We tabulated these
data by year, quarter, and state, as well as by patients’ age group and sex. We combined these
data with intercensal population data estimates,
45
and quarterly unemployment rates for each
state from the Bureau of Labor Statistics.
46
We also used data for 2013–14 from the National
Health and Nutrition Examination Survey to calculate the prevalence of diabetes by age group.
47

2.3 Statistical Analysis
We used the difference-in-differences method to model changes in diabetes prescriptions filled
using Medicaid insurance after Medicaid eligibility expansions for states with versus those
without such expansions. To account for the fact that the number of prescriptions increases
along with the population, we used negative binomial models in which current population was
the exposure variable.
45
We report the effects associated with the expansions as changes in
prescription fills per 1,000 population per year (that is, average marginal effects per 1,000
population). Appendix A2A provides additional details.
Our outcomes of interest were prescription fills for metformin (extended-release and regular), a
first-line treatment for non-insulin-dependent type 2 diabetes; insulin (rapid- and long-acting
50

insulin analogues and regular insulin), a treatment for type 1 and insulin-dependent type 2
diabetes; three classes of newer oral medications (dipeptidyl peptidase [DPP]-4 inhibitors, GLP-
1 agonists, and SGLT-2 inhibitors); and all other classes of diabetes medications. Prescriptions
for all other drug classes were grouped together since they are not first-line agents, are not
newer drugs, and were not used as frequently as other classes. We also analyzed the total
numbers of diabetes prescriptions filled.
We clustered standard errors at the state level to account for the state-level nature of Medicaid
eligibility expansions. We addressed possible residual confounding by adjusting for year by
quarter indicator variables, state indicator variables, the age and sex of the person filling the
prescription, and quarterly state-level unemployment rates.
We also conducted specification checks such as testing for parallel trends in Medicaid expansion
versus nonexpansion states before the expansions, using linear and nonlinear specifications. We
conducted a number of robustness checks. These included stratifying the data by age to compare
changes in Medicaid diabetes prescription fills after Medicaid eligibility expansions with age-
specific diabetes prevalence, estimating changes separately for 2014 and 2015, and examining
whether quarterly changes after Medicaid expansions grew over time. We also examined
whether the gap in Medicaid diabetes prescription fills between residents of expansion states
and those of nonexpansion states shrank as expected once patients became eligible for Medicare
at age sixty-five, omitted data from before 2011, and excluded states that expanded Medicaid
eligibility before or after January 2014. Appendix A2A provides additional details on these
analyses.
All analyses were performed using Stata/MP, version 14.
51

2.4 Limitations
Our study had several limitations. First, we were not able to track people over time. Instead, we
analyzed data at the age-sex-state level across different time periods.
Second, patients’ race and ethnicity were not reported in the IQVIA data and therefore these
variables were not included in our analyses.
Third, approximately 15 percent of retail pharmacies did not share their prescription fills data
with IQVIA. Missing data were imputed by IQVIA using validated methods.
48
Fourth, nonretail
prescriptions and mail-order prescriptions were outside the sampling frame. If Medicaid
eligibility expansions also increased prescription fills at federally qualified health centers or
mail-order prescription fills, our estimates would be underestimates of the total effect.
Fifth, we evaluated the association between Medicaid eligibility expansions and diabetes
prescription fills in states with expanded Medicaid eligibility. Our findings might not be
generalizable to a nationwide expansion of Medicaid eligibility.
Finally, ours was an observational study, and we therefore cannot rule out the possibility that
other changes also accounted for or contributed to our results.
3. Results
States that did and those that did not expand Medicaid eligibility during 2014–15 appeared
similar in population-level characteristics in 2010, the year of the ACA’s passage (Table 1).
Prescription fills for diabetes medications showed a slightly increasing trend before the ACA, in
both expansion and non-expansion states (Figure 1). Appendix A3 shows the trends by
medication class. For each outcome and age group, we could not reject the null hypothesis that
52

trends in our outcomes of interest were similar in these two groups of states before 2014; see
Appendix A4B and A4C. Additionally, an analysis of the annual gap in prescription fills
between expansion and non-expansion states showed a flat trend before 2014 and a break in that
trend in 2014 (appendix A4D). In 2014–15 Medicaid eligibility expansions were associated with
increases of thirty Medicaid prescription fills for diabetes medications per 1,000 population
among adults ages 20–64 (Table 2). We observed larger increases in fills in 2015 than in 2014.
When we divided the quarterly increase in fills into an intercept and a slope, we found that the
slope was positive and significant—which indicates that the changes after Medicaid eligibility
expansions grew over time (appendix A2D).
Newer medications (rapid- and long-acting insulin analogues, extended-release metformin,
DPP-4 inhibitors, GLP-1 agonists, and SGLT-2 inhibitors) accounted for about one-third of the
increase in prescriptions (Table 2). This likely represented an increase in patients’ uptake of
newer medications, compared to patients without insurance. In the IQVIA data in 2013, 15
percent of diabetes prescriptions filled using cash were newer medications, compared with 35
percent, 37 percent, and 38 percent filled using Medicare, private insurance, and Medicaid,
respectively. The lower uptake of newer medications among uninsured patients is consistent
with the fact that these medications required substantially higher out-of-pocket spending at the
time of the Medicaid eligibility expansions.
21,49

All findings remained qualitatively unchanged in additional robustness checks, when we
eliminated states that expanded Medicaid eligibility before 2014, eliminated all states that
expanded Medicaid eligibility in months other than January 2014, or omitted data from before
2011. Results of these analyses are in appendix A2E and A2F.
53

The relationship between Medicaid eligibility expansions and Medicaid prescription fills for
diabetes medications declined dramatically after patients reached age sixty-five, as expected.
The increase in prescription fills was 82 percent smaller for people ages 65–69 than for people
ages 60–64, despite the fact that the two groups had identical diabetes prevalence (Figure 4).
Among people younger than age sixty-five, those in age groups with a higher prevalence of
diabetes experienced larger increases in treatment (Figure 4 and appendix A5). The correlation
between diabetes prevalence and changes in Medicaid prescription fills for diabetes medications
among people ages 20–64 was 0.98 (p < 0.01). The results (shown in appendix A6)
45
were
similar when, as a robustness check, we included only patients with diagnosed diabetes in the
NHANES analysis.
4. Discussion
This study analyzed the associations between Medicaid eligibility expansions and Medicaid
prescription fills for diabetes medications by patients’ age and sex and by medication category.
We used a large, representative administrative data set that captured over ninety-six million
Medicaid prescription fills for diabetes medications in retail outlets in the period January 2008–
December 2015. The analysis accounted for changes in population and many possible
confounders. Our results imply that an average of thirty additional Medicaid prescriptions for
diabetes medications were filled annually per 1,000 population in states that expanded Medicaid
eligibility.
Age groups with higher prevalence of diabetes before the ACA, such as people ages 55–59,
showed larger increases in diabetes prescription fills after Medicaid eligibility expansions. In
addition, increases in fills after the expansions declined dramatically among people ages sixty-
54

five and older. Because Medicaid is a payer of last resort, eligibility for Medicaid was expected
to have a smaller impact among patients who were also eligible for Medicare.
We found that annual prescription fills for insulin and metformin using Medicaid insurance each
increased by approximately 40 percent after Medicaid eligibility expansions. In the period
2002–13, insulin’s mean price rose 197 percent—growth faster than that of any other drug class
used to treat diabetes.
21,50
Estimated insulin spending per patient more than tripled, from
$231.48 in 2002 to $736.09 in 2013.
21
Patients without insurance would have been exposed to
the full costs of insulin. Gaining Medicaid insurance would have significantly reduced out-of-
pocket spending for insulin for previously uninsured patients, thereby facilitating uptake of the
medication.
Furthermore, the sizable increase in the use of metformin suggests that many of the newly
treated patients may have had recent onset of diabetes. This finding echoes those of previous
analyses that linked Medicaid eligibility expansions with increased diabetes diagnoses.
51–54
Indeed, the drug class with the largest relative increase after the expansions (52 percent) was
extended-release metformin, a reformulation of the first-line medication for type 2 diabetes.
More broadly, our data suggest that Medicaid eligibility expansions were associated with
increased prescription fills for newer diabetes medications. This is important because these
medications carry higher costs than the older formulations do but provide benefits such as
reduced risk of hypoglycemia and reduced side effects.
22,23,26
Newer medications accounted for
about one-third of the increase in Medicaid diabetes prescriptions after the expansions, in line
with prior Medicaid prescribing patterns in expansion states. This represents an increase in
patients’ use of newer medications compared with uninsured patients.
55

Our findings point to the possible health effects of Medicaid eligibility expansions. In the past,
changes in cost sharing for diabetes medications have been associated with changes in health
outcomes for patients with diabetes.
55,56
An analysis by the Centers for Disease Control and
Prevention found that each additional treated patient with diabetes can lead to a reduction of
$4,330 (in 1997 US dollars, equivalent to about $6,394 in 2017 US dollars) in inpatient care
costs because of prevented hospital admissions.
57
These figures may underestimate the current
health effects of treatment, given that improved treatment regimens are now available.
29,58

Indeed, a decline in diabetes-related hospitalizations was observed shortly after Medicaid
eligibility expansions in states with high baseline uninsured populations.
59

We found that the changes in the use of diabetes medications associated with Medicaid
eligibility expansions increased over time. Ausmita Ghosh and coauthors reported a 24 percent
increase in Medicaid diabetes medications through the first quarter of 2015.
42
We found a 33
percent increase by the end of 2014, within the confidence intervals implied by the standard
errors reported in that study, and a 49 percent increase by the end of 2015. The increasing gap
over time between states that did and did not expand Medicaid is apparent from Figure 1 and is
significant, as shown in appendix A2D.
Our study had a number of strengths. First, by using administrative data on prescription fills, we
avoided issues of patient self-report bias. Second, these data provided a sufficient sample size to
examine the treatment of specific conditions by patients’ demographic characteristics and type
of medication. Third, because these data were collected as prescriptions were filled, our data
were timely and provided eight months of additional follow-up, compared to existing studies.
56

Fourth, although states with Medicaid eligibility expansions may differ from other states in
some respects, population-level factors that differ between the groups do not bias the results in a
difference-in-differences analysis as long as trends between the groups would have remained
parallel in the absence of an intervention. We presented several analyses indicating parallel
trends before the expansions, which provides evidence in support of this assumption.
Finally, we adjusted for state and year by quarter indicator variables, patient age and sex, and
quarterly changes in unemployment on the state level to address residual confounding.
5. Conclusion
This study provides policy makers with new information about the potential benefits of
continuing financial support for expansions of Medicaid eligibility. Our findings by drug class
suggest that these expansions helped address some of the gaps in access to newer medications
for low-income patients. An increase in access to newer medications may have important health
effects, because the use of these medications has been linked with improved diabetes control
and reduced symptoms in both clinical trials and observational data.
22–28
Furthermore, over a
third of the additional diabetes Medicaid prescriptions associated with Medicaid eligibility
expansions were for metformin, the first-line oral medication to treat diabetes that is not yet
insulin dependent. Improvements in population health that are attributable to improved access to
diabetes treatment, including the timely treatment of early-stage disease, could also justify some
of the cost of expanding Medicaid. Finally, our study provides new evidence that the increases
in treatment associated with Medicaid eligibility expansions can grow over time.

57

References
1. Danaei G, Ding EL, Mozaffarian D, Taylor B, Rehm J, Murray CJL, et al. The Preventable
Causes of Death in the United States: Comparative Risk Assessment of Dietary, Lifestyle,
and Metabolic Risk Factors. PLOS Med. 2009 Apr 28;6(4):e1000058.
2. Heron M. National Vital Statistics Reports Deaths: Leading Causes for 2010. National Vital
Statistics Reports; 2013. Report No.: 6.
3. Murray CJL, Abraham J, Ali MK, Alvarado M, Atkinson C, Baddour LM, et al. The State of
US Health, 1990-2010: Burden of Diseases, Injuries, and Risk Factors. JAMA. 2013 Aug
14;310(6):591–606.
4. Farley TA, Dalal MA, Mostashari F, Frieden TR. Deaths Preventable in the U.S. by
Improvements in Use of Clinical Preventive Services. Am J Prev Med. 2010 Jun 1;38(6):600–
9.
5. Huang ES, Meigs JB, Singer DE. The effect of interventions to prevent cardiovascular disease
in patients with type 2 diabetes mellitus. Am J Med. 2001 Dec 1;111(8):633–42.
6. Kelly TN. Systematic Review: Glucose Control and Cardiovascular Disease in Type 2
Diabetes. Ann Intern Med. 2009 Sep 15;151(6):394.
7. Ray KK, Seshasai SRK, Wijesuriya S, Sivakumaran R, Nethercott S, Preiss D, et al. Effect
of intensive control of glucose on cardiovascular outcomes and death in patients with diabetes
mellitus: a meta-analysis of randomised controlled trials. The Lancet. 2009 May
23;373(9677):1765–72.
8. Group UPDS (UKPDS). Intensive blood-glucose control with sulphonylureas or insulin
compared with conventional treatment and risk of complications in patients with type 2
diabetes (UKPDS 33). The Lancet. 1998 Sep 12;352(9131):837–53.
9. Hill SC, Miller GE, Sing M. Adults with Diagnosed and Untreated Diabetes: Who Are They?
How Can We Reach Them? J Health Care Poor Underserved. 2011 Nov 13;22(4):1221–38.
10. Gakidou E, Mallinger L, Abbott-Klafter J, Guerrero R, Villalpando S, Ridaura RL, et al.
Management of diabetes and associated cardiovascular risk factors in seven countries: a
comparison of data from national health examination surveys. Bull World Health Organ.
2011 Mar 1;89(3):172–83.
11. Roehrig C, Daly M. Prevalence Trends For Three Common Medical Conditions: Treated And
Untreated. Health Aff (Millwood). 2015 Aug 1;34(8):1320–3.
12. Goldman DP, Joyce GF, Escarce JJ. Pharmacy benefits and the use of drugs by the
chronically ill. JAMA [Internet]. 2004;291(19). Available from:
http://archneur.jamanetwork.com/article.aspx?articleid=198761
58

13. Solomon MD, Goldman DP, Joyce GF, Escarce JJ. Cost Sharing and the Initiation of Drug
Therapy for the Chronically Ill. Arch Intern Med. 2009 Apr 27;169(8):740–8.
14. Kurlander JE, Kerr EA, Krein S, Heisler M, Piette JD. Cost-Related Nonadherence to
Medications Among Patients With Diabetes and Chronic Pain. Diabetes Care. 2009 Dec
1;32(12):2143–8.
15. Karter AJ, Parker MM, Solomon MD, Lyles CR, Adams AS, Moffet HH, et al. Effect of Out-
of-Pocket Cost on Medication Initiation, Adherence, and Persistence among Patients with
Type 2 Diabetes: The Diabetes Study of Northern California (DISTANCE). Health Serv Res.
:n/a-n/a.
16. Rhee MK, Cook CB, Dunbar VG, Panayioto RM, Berkowitz KJ, Boyd B, et al. Limited
Health Care Access Impairs Glycemic Control in Low Income Urban African Americans
With Type 2 Diabetes. J Health Care Poor Underserved. 2005 Nov 23;16(4):734–46.
17. Arch G Mainous III. Diabetes management in the USA and England: comparative analysis
of national surveys. J R Soc Med. 2006 Sep;99(9):463.
18. Zhang X, McKeever Bullard K, Gregg EW, Beckles GL, Williams DE, Barker LE, et al.
Access to health care and control of ABCS of diabetes. Diabetes Care. 2012;35(7):1566–
1571.
19. Alexander GC, Sehgal NL, Moloney RM, Stafford RS. National Trends in Treatment of Type
2 Diabetes Mellitus, 1994–2007. Arch Intern Med. 2008 Oct 27;168(19):2088–94.
20. White JR. A Brief History of the Development of Diabetes Medications. Diabetes Spectr Publ
Am Diabetes Assoc. 2014 May;27(2):82–6.
21. Hua X, Carvalho N, Tew M, Huang ES, Herman WH, Clarke P. Expenditures and Prices of
Antihyperglycemic Medications in the United States: 2002-2013. JAMA. 2016 Apr
5;315(13):1400–2.
22. Rosenstock J, Dailey G, Massi-Benedetti M, Fritsche A, Lin Z, Salzman A. Reduced
hypoglycemia risk with insulin glargine: a meta-analysis comparing insulin glargine with
human NPH insulin in type 2 diabetes. Diabetes Care. 2005 Apr;28(4):950–5.
23. Umpierrez GE, Latif K, Stoever J, Cuervo R, Park L, X. Freire A, et al. Efficacy of
subcutaneous insulin lispro versus continuous intravenous regular insulin for the treatment
of patients with diabetic ketoacidosis. Am J Med. 2004 Sep 1;117(5):291–6.
24. Wei W, Buysman E, Grabner M, Xie L, Brekke L, Ke X, et al. A real-world study of treatment
patterns and outcomes in US managed-care patients with type 2 Diabetes initiating injectable
therapies. Diabetes Obes Metab. 2017;19(3):375–86.
25. Derosa G, D’Angelo A, Romano D, Maffioli P. Effects of metformin extended release
compared to immediate release formula on glycemic control and glycemic variability in
patients with type 2 diabetes. Drug Des Devel Ther. 2017 May 16;11:1481–8.
59

26. Schwartz S, Fonseca V, Berner B, Cramer M, Chiang Y-K, Lewin A. Efficacy, Tolerability,
and Safety of a Novel Once-Daily Extended-Release Metformin in Patients With Type 2
Diabetes. Diabetes Care. 2006 Apr 1;29(4):759–64.
27. Marso SP, Daniels GH, Brown-Frandsen K, Kristensen P, Mann JFE, Nauck MA, et al.
Liraglutide and Cardiovascular Outcomes in Type 2 Diabetes. N Engl J Med. 2016 Jul
28;375(4):311–22.
28. Zinman B, Wanner C, Lachin JM, Fitchett D, Bluhmki E, Hantel S, et al. Empagliflozin,
Cardiovascular Outcomes, and Mortality in Type 2 Diabetes. N Engl J Med. 2015 Nov
26;373(22):2117–28.
29. Chao EC, Henry RR. SGLT2 inhibition--a novel strategy for diabetes treatment. Nat Rev
Drug Discov. 2010 Jul;9(7):551–9.
30. Ingelfinger JR, Rosen CJ. Cardiac and Renovascular Complications in Type 2 Diabetes--Is
There Hope? N Engl J Med. 2016 Jul 28;375(4):380–2.
31. Meier JJ. GLP-1 receptor agonists for individualized treatment of type 2 diabetes mellitus.
Nat Rev Endocrinol. 2012 Dec;8(12):728–42.
32. Rosenbaum S, Westmoreland TM. The Supreme Court’s Surprising Decision On The
Medicaid Expansion: How Will The Federal Government And States Proceed? Health Aff
(Millwood). 2012 Aug 1;31(8):1663–72.
33. Decker SL, Kostova D, Kenney GM, Long SK. Health Status, Risk Factors, and Medical
Conditions Among Persons Enrolled in Medicaid vs Uninsured Low-Income Adults
Potentially Eligible for Medicaid Under the Affordable Care Act. JAMA. 2013 Jun
26;309(24):2579–86.
34. Mulcahy AW, Eibner C, Finegold K. Gaining Coverage Through Medicaid Or Private
Insurance Increased Prescription Use And Lowered Out-Of-Pocket Spending. Health Aff
(Millwood). 2016 Sep 1;35(9):1725–33.
35. Kennedy J, Wood EG. Medication Costs and Adherence of Treatment Before and After the
Affordable Care Act: 1999–2015. Am J Public Health. 2016 Aug 23;106(10):1804–7.
36. Wen H, Borders TF, Druss BG. Number Of Medicaid Prescriptions Grew, Drug Spending
Was Steady In Medicaid Expansion States. Health Aff (Millwood). 2016 Sep 1;35(9):1604–
7.
37. Sommers B, Gunja M, Finegold K, Musco T. Changes in Self-reported Insurance Coverage,
Access to Care, and Health Under the Affordable Care Act. JAMA. 2015;314(4):366.
38. Polsky D, Candon M, Saloner B, Wissoker D, Hempstead K, Kenney GM, et al. Changes in
Primary Care Access Between 2012 and 2016 for New Patients With Medicaid and Private
Coverage. JAMA Intern Med. 2017 Apr 1;177(4):588–90.
60

39. Decker SL. Two-thirds of primary care physicians accepted new Medicaid patients in 2011-
12: a baseline to measure future acceptance rates. Health Aff Proj Hope. 2013
Jul;32(7):1183–7.
40. Polsky D, Richards M, Basseyn S, Wissoker D, Kenney GM, Zuckerman S, et al.
Appointment Availability after Increases in Medicaid Payments for Primary Care. N Engl J
Med. 2015 Feb 5;372(6):537–45.
41. Sommers BD, Blendon RJ, Orav EJ, Epstein AM. Changes in Utilization and Health Among
Low-Income Adults After Medicaid Expansion or Expanded Private Insurance. JAMA Intern
Med. 2016 Oct 1;176(10):1501–9.
42. Ghosh A, Simon K, Sommers BD. The Effect of State Medicaid Expansions on Prescription
Drug Use: Evidence from the Affordable Care Act. NBER Work Pap Ser [Internet].
2017;(23044). Available from: http://www.nber.org/papers/w23044%0ANATIONAL
43. Barua S, Greenwald R, Grebely J, Dore GJ, Swan T, Taylor LE. Restrictions for Medicaid
Reimbursement of Sofosbuvir for the Treatment of Hepatitis C Virus Infection in the United
States. Ann Intern Med. 2015 Aug 4;163(3):215.
44. Canary LA, Klevens RM, Holmberg SD. Limited Access to New Hepatitis C Virus Treatment
Under State Medicaid Programs. Ann Intern Med. 2015 Aug 4;163(3):226–8.
45. Bridged-Race Population Estimates 1990-2015 Request [Internet]. [cited 2019 Jun 19].
Available from: https://wonder.cdc.gov/bridged-race-v2015.html
46. Local Area Unemployment Statistics Home Page [Internet]. [cited 2017 Jun 28]. Available
from: https://www.bls.gov/lau/
47. Centers for Disease Control and Prevention. National Health and Nutrition Examination
Survey. Atlanta, GA;
48. Belongia EA, Sullivan BJ, Chyou PH, Madagame E, Reed KD, Schwartz B. A community
intervention trial to promote judicious antibiotic use and reduce penicillin-resistant
Streptococcus pneumoniae carriage in children. Pediatrics. 2001 Sep;108(3):575–83.
49. McEwen LN, Casagrande SS, Kuo S, Herman WH. Why Are Diabetes Medications So
Expensive and What Can Be Done to Control Their Cost? Curr Diab Rep. 2017 Sep
1;17(9):71.
50. Greene JA, Riggs KR. Why is there no generic insulin? Historical origins of a modern
problem. N Engl J Med. 2015 Mar 19;372(12):1171–5.
51. Baicker K, Taubman SL, Allen HL, Bernstein M, Gruber JH, Newhouse JP, et al. The Oregon
Experiment — Effects of Medicaid on Clinical Outcomes. N Engl J Med. 2013 May
2;368(18):1713–22.
52. Wherry LR, Miller S. Early Coverage, Access, Utilization, and Health Effects Associated
61

With the Affordable Care Act Medicaid Expansions: A Quasi-experimental Study. Ann
Intern Med. 2016 Jun 21;164(12):795.
53. Myerson R, Laiteerapong N. The Affordable Care Act and Diabetes Diagnosis and Care:
Exploring the Potential Impacts. Curr Diab Rep. 2016;16(4):27.
54. Kaufman HW, Chen Z, Fonseca VA, McPhaul MJ. Surge in Newly Identified Diabetes
Among Medicaid Patients in 2014 Within Medicaid Expansion States Under the Affordable
Care Act. Diabetes Care [Internet]. 2015; Available from:
http://care.diabetesjournals.org/content/early/2015/03/19/dc14-2334.abstract
55. Chandra A, Gruber J, McKnight R. Patient Cost-Sharing and Hospitalization Offsets in the
Elderly. Am Econ Rev. 2010;100(1):193–213.
56. DP G, GF J, Y Z. Prescription drug cost sharing: Associations with medication and medical
utilization and spending and health. JAMA. 2007;298(1):61–69.
57. CDC Diabetes Cost-effectiveness Group. Cost-effectiveness of intensive glycemic control,
intensified hypertension control, and serum cholesterol level reduction for type 2 diabetes.
JAMA. 2002 May 15;287(19):2542–51.
58. Levin PA, Zhou S, Gill J, Wei W. Health Outcomes Associated with Initiation of Basal
Insulin After 1, 2, or ≥ 3 Oral Antidiabetes Drug(s) Among Managed Care Patients with Type
2 Diabetes. J Manag Care Spec Pharm. 2015 Dec 1;21(12):1172–82.
59. Freedman S, Nikpay S, Carroll A, Simon K. Changes in inpatient payer-mix and
hospitalizations following Medicaid expansion: Evidence from all-capture hospital discharge
data. PLoS ONE. 2017;12(9):1–9.

62

Figure 1 Diabetes prescription fills using Medicaid insurance per 1,000 population ages 20-
64, in states that did and did not expand eligibility for Medicaid
Authors' analysis of data from IQVIA.
Note: Most of the states that expanded Medicaid eligibility among low-income, nondisabled adults in 2014 and
2015 did so in the first quarter of 2014
63

Figure 2 Prevalence of diabetes in 2013-2014 and increase in Medicaid diabetes
prescription fills per 1,000 population in 2014-15 associated with Medicaid eligibility
expansions, by age groups (years)
Authors' analysis of data from IQVIA and the National Health and Nutrition Examination Survey.
Note: 2013-14 are the years just before and during Medicaid eligibility expansions, for most of the twenty-nine
states (and the District of Columbia) that had expanded eligibility for Medicaid by the end of our sample period.

64

Table 1 Average characteristics in 2010 of states that did and those that did not expand
eligibility for Medicaid under the Affordable Care Act
Characteristic Nonexpansion states Expansion states
p value of
difference
Prevalence of diagnosed diabetes
7.96% 9.60% 0.06
Mortality per 100,000 people
829.57 826.60 0.93
Population
6,161,336 5,887,206 0.89
Male
49.20% 49.46% 0.25
Hispanic
11.53 9.11 0.40
Black
10.2 12.55 0.46
Older than age 65
13.53 12.84 0.15
Authors’ analysis of data from the Census Bureau, the Centers for Disease Control and Prevention, the National
Center for Health Statistics, and the National Vital Statistics System. Note: Twenty-nine states and the District of
Columbia expanded Medicaid eligibility by the end of our sample period.
65

Table 2 Additional increases in annual Medicaid diabetes prescription fills per 1,000 population ages 20–64 associated with
Medicaid eligibility expansions during 2014-2015
Difference-in-differences estimate
Baseline fills Average annual change, 2014–15 Change in 2014 Change in 2015
Increase 95% CI Increase 95% CI Increase 95% CI
All patients 73.05 29.93 (21.40, 38.45) 24.21 (17.4,
31.39)
35.93 (25.62, 46.25)
Men 58.73 23.98 (17.42, 30.54) 19.19 (13.66,
24.72)
29.4 (21.70, 37.20)
Women 87.22 31.52 (21.77, 41.26) 25.89 (17.66,
34.12)
37.40 (25.63, 49.17)
Most common types of diabetes medications
Insulin and insulin
analogues
23.21 9.35 (6.59, 12.11) 7.73 (5.44,
10.20)
11.4 (7.69, 14.39)
Metformin 29.13 12.17 (8.71, 15.63) 9.88 (6.95,
12.81)
14.58 (10.39, 18.77)
Newer diabetes medications
Rapid-acting insulin
analogs
6.13 2.65 (1.87, 3.42) 2.18 (1.55, 2.80) 3.14 (2.18, 4.10)
Long-acting insulin
analogs
12.1 4.54 (3.16, 5.92) 3.79 (2.66, 4.63) 5.32 (3.62, 7.10)
Extended-release
metformin
3.05 1.59 (1.10, 2.70) 1.24 (0.80, 1.67) 1.95 (1.38. 2.53)
DPP-4 inhibitors,
GLP-1 agonists, and
SGLT-2 inhibitors
5.16 1.51 (1.6, 1.97) 1.13 (0.71, 1.54) 1.91 (1.37, 2.45)
Authors’ analysis of data from IQVIA. NOTES Twenty-nine states and the District of Columbia expanded eligibility for Medicaid by the end of our sample
period. Baseline fills are those in 2013, measured in states that subsequently expanded eligibility for Medicaid. Difference-in-differences estimates were adjusted
for year by quarter indicator variables, state indicator variables, patient’s age group and sex, and quarterly state-level unemployment rates. All changes were
statistically significant with p < 0.01. 95% confidence intervals are in parentheses. DPP is dipeptidyl peptidase. GLP is glucagon-like peptide. SGLT is sodium-
glucose cotransporter.

66

Appendix
Appendix A1. List of treatment and control states
Treatment Control
States that expanded
Medicaid eligibility on
1/1/2014
Arkansas
Colorado
Illinois
Iowa
Kentucky
Maryland
Massachusetts
Nevada
New Mexico
North Dakota
Ohio
Oregon
Rhode Island
Vermont
West Virginia
Alabama
Florida
Georgia
Idaho
Kansas
Louisiana
1

Maine
Mississippi
Missouri
Montana
2

Nebraska
North Carolina
Oklahoma
South Carolina
South Dakota
Tennessee
Texas
Utah
Virginia
Wisconsin
3

Wyoming

States that also expanded
Medicaid eligibility to
low-income adults prior
to 2014
Arizona
4

California
5

Connecticut
Delaware
District of Columbia Hawaii
Minnesota
New Jersey
New York
6

1
Louisiana adopted the Medicaid eligibility expansion on 7/1/2016.
2
Montana adopted the Medicaid eligibility expansion on 1/1/2016.
3
Wisconsin covers adults up to 100% FPL in Medicaid but did not adopt the ACA Medicaid eligibility expansion.
4
Since 2000, Arizona offered Medicaid-equivalent benefits to childless adults with incomes below 100% FPL
through a Section 1115 waiver program. The state closed the program to new enrollees in July 2011.
5
California adopted the Low-Income Health Program, which was an elective program implemented at the county
level. Eligibility varied by county up to 200% FPL.
6
Delaware, Hawaii, New York, and Vermont expanded Medicaid eligibility up to 100% FPL before 2014.
67

Washington

States that expanded
Medicaid after 1/1/2014
7

Alaska
Indiana
Michigan
New Hampshire
Pennsylvania
Source: Kaiser Family Foundation data
8,9
and Centers for Medicare & Medicaid Services data
10

7
Medicaid eligibility expansions became effective on 4/1/2014 for Michigan and on 8/15/2014 for New Hampshire.
Pennsylvania expanded Medicaid eligibility on 1/1/2015, Indiana on 2/1/2015 and Alaska on 9/1/2015.
8
Kaiser Family Foundation, States Getting a Jump Start on Health Reform’s Medicaid Expansion. Available from:
https://www.kff.org/health-reform/issue-brief/states-getting-a-jump-start-on-health/
9
Kaiser Family Foundation, Medicaid Income Eligibility Limits for Other Non-Disabled Adults, 2011-2018.
Available from: https://www.kff.org/medicaid/state-indicator/medicaid-income-eligibility-limits-for-other-non-
disabled-adults/
10
Centers for Medicare & Medicaid Services, Lessons from Early Medicaid Expansions Under Health Reform:
Interviews with Medicaid Officials. Available from:
https://www.cms.gov/mmrr/Downloads/MMRR2013_003_04_a02.pdf
68

Appendix A2. Analyses of changes in diabetes prescription fills
A. Description of all analyses
Main analysis
We conducted the differences-in-differences analysis using negative binomial regression
models, which are appropriate for count data with a known exposed population. The models
took the following form:
ln ( Y
it as
) = μ + γ
1
( Expa nsi onState
i
× PostEx pa nsi on
it
) + θ Ti me
t
+ π Sta te
i
+ δAg e Grou p
a
+ ω Gen der
s
+ βSta te Um p
it
+ ln ( Popul at ion
it as
)
where i indexes state, t year, a age group, s gender. θ is a vector of time (year by quarter) fixed
effects, and π is a vector of state fixed effects. Genders and AgeGroupa control for gender and
age group effects. StateUmpit denotes the quarterly state-level unemployment rates.
Populationitas is the exposure variable in the negative binomial model and has a coefficient of 1.

Yitas indicates the outcome variables. Our primary outcome was total diabetes prescriptions
filled. Secondary outcomes included prescription fills for metformin (total, and broken down
into extended release vs. regular metformin); insulin (total, and two newer categories: rapid-
acting, long-acting insulin analogs); three classes of newer oral medications (DPP4 inhibitors,
GLP1 agonists, and SGLT2 inhibitors), and all other classes of diabetes medications.

The coefficient of interest is 1, which captures the additional change in diabetes prescription
fills after Medicaid eligibility expansions. ExpansionStatei× PostExpansionit is an interaction
term between ExpansionStatei and PostExpansionit. ExpansionStatei is a binary variable
indicating whether the state expanded Medicaid in our sample period. The non-interacted
variable is omitted from the model due to collinearity with the state fixed effects.

For states that expanded Medicaid eligibility after January 2014, PostExpansionit was set to 0 in
the quarters before that state’s Medicaid eligibility expansion. Therefore, the interaction term is
equal to 1 for states that expanded Medicaid eligibility in the post-expansion period, and 0 for
all others.

The regression used Huber-White robust standard errors clustered at the state level.

Subgroup analyses depicted in Figure 4 of the main text use the same model with samples of
people from different age groups. AgeGroupa, the age group fixed effect, was excluded in the
subgroup regressions.

69

Calculating the quantity of interest:
Negative binomial regression coefficients are not readily interpretable. The adjusted differences-
in-differences estimates presented in Table 2 were calculated as average marginal effects of the
coefficient of interest, applied to a population of size 1,000. In our case, average marginal
effects were calculated by applying the regression coefficients to the data from all states and
computing the average change in the predicted outcomes across all states, for a population of
size 1,000, when the Medicaid eligibility expansions predictor variable is set to 1 rather than set
to 0. We used the margins command in Stata to conduct this calculation.

Supplemental analyses

Diabetes prevalence and changes in diabetes prescription fills

We additionally examined whether the prevalence of diabetes by age group was associated with
changes in prescription fills after Medicaid eligibility expansions. To accomplish this, we first
stratified the models to obtain the number of additional prescriptions filled after Medicaid
eligibility expansions per 1000 population for each age group. We then calculated the
prevalence of diabetes using NHANES biomarker and survey data from 2013-2014.
Specifically, in NHANES, participants who had been told by a doctor that he/she had diabetes,
who had Hemoglobin A1c (HbA1c) above 6.5%, or who had a fasting glucose test (FGT) above
7.0 mmol/L were considered to have diabetes. (In a robustness check, we only included
diagnosed diabetes.) We estimated the proportion of diabetes patients for each age group,
incorporating survey design variables to account for the complex, multi-stage sampling design
of NHANES survey.

Changes over time

We used two methods to assess whether the change in diabetes prescription fills after Medicaid
eligibility expansions grew over time. In one analysis, we estimated the change separately for
2014 vs. 2015 by interacting the variable for current Medicaid eligibility expansions with
indicators for these two years.

In a second analysis, we added an interaction term between current Medicaid eligibility
expansion and years since January 2014. (Because we used quarterly data, years were measured
in fractions.) The coefficient on this term gave the annual slope of the change after Medicaid
eligibility expansions. A positive and statistically significant slope term would therefore imply
that the effects associated with Medicaid eligibility expansions grew over time.

Additional robustness checks

The validity of the differences-in-differences method rests on the assumption that in the absence
of the policy intervention of interest, trends in states with different policy interventions would
have remained parallel. Although this assumption is not testable, parallel trends prior to the
policy intervention provide evidence of the assumption’s plausibility. As such, we tested for
parallel trends in Medicaid eligibility expansion vs. non-expansion states prior to Medicaid
70

eligibility expansions for each model presented. We additionally conducted a pre-trend analysis
that used quarterly indicator variables to check for non-linearity in the trend prior to 2014.
Finally, we assessed the robustness of our findings to the length of the pre-period by omitting
data from prior to 2011 in a supplemental analysis.
Using the data stratified by age group, we additionally examined whether the gap between
residents of expansion and non-expansion states shrank as expected once patients become age-
eligible for Medicare at age 65. We would expect Medicaid eligibility expansions to have a
smaller effect in the population that is age-eligible for Medicare, because Medicaid is the payer
of last resort.

In two additional checks, we excluded states that expanded Medicaid prior to or after January
2014. We expected that expanding Medicaid eligibility would have a smaller impact if some
non-disabled low-income adults were already eligible. Additionally, we expected that if the
effect of expanding Medicaid eligibility grows over time, removing late-expanding states would
increase the observed effects.

71

B. Raw regression coefficients: Changes in Medicaid prescription fills by type of diabetes
medication
Negative binomial regression coefficients are not readily interpretable. However, for
completeness, we present the raw regression coefficients corresponding to the adjusted
differences-in-differences estimates (average marginal effects) presented in Table 2 and Figure
2 of the main text.
Outcomes

Variables
All drugs Insulin All
metformi
n
Rapid-
acting
insulin
analog
Long-acting
insulin
analog
Extended
release
metformi
n
DPP4 and
SGLT2
inhibitor,
GLP1
agonist
State currently
expanding
Medicaid
eligibility
0.518*** 0.489*** 0.556*** 0.538*** 0.483*** 0.583*** 0.462***
(0.0707) (0.0693) (0.0754) (0.0759) (0.0706) (0.0855) (0.0667)
State-level
unemployment
-0.00251 -0.00581 0.00313 -0.00469 0.000589 -0.0113 -0.0245
(0.0252) (0.0252) (0.0274) (0.0300) (0.0265) (0.0286) (0.0281)
State fixed effect yes yes yes yes yes yes yes
Time fixed effect yes yes yes yes yes yes yes
Gender fixed
effect
yes yes yes yes yes yes yes
Age group fixed
effect
yes yes yes yes yes yes yes
Population
exposure term
yes yes yes yes yes yes yes
Constant -5.588*** -6.142*** -7.031*** -7.236*** -7.589*** -8.481*** -10.37***
(0.159) (0.157) (0.170) (0.179) (0.162) (0.168) (0.199)
Observations 26,112 26,112 26,112 26,112 26,112 26,112 26,112
Pseudo R-squared 0.126 0.120 0.142 0.121 0.148 0.147 0.176
Note:
* denotes significant at 0.05 level.
** denotes significant at 0.01 level.
Robust standard errors are in parentheses.
One state was omitted from the state fixed effects to avoid perfect collinearity with the constant.

72

C. Raw regression coefficients: Changes in Medicaid prescription fills by age group
Age

Variables
20-25 26-29 30-34 35-44 45-54 55-59 60-64
State
currently
expanding
Medicaid
eligibility
0.550*** 0.544*** 0.509*** 0.464*** 0.568*** 0.577*** 0.596***
(0.0784) (0.0826) (0.0797) (0.0830) (0.0831) (0.0797) (0.0710)
State-level
unemployment
-0.0206 0.00275 0.00515 0.00244 -0.00490 -0.0117 -0.0166
(0.0248) (0.0261) (0.0246) (0.0284) (0.0297) (0.0298) (0.0283)
State fixed
effect
yes yes yes Yes yes yes yes
Time fixed
effect
yes yes yes Yes yes yes yes
Gender fixed
effect
yes yes yes Yes yes yes yes
Population
exposure term
yes yes yes Yes yes yes yes
Constant -5.506*** -5.538*** -4.958*** -4.658*** -4.082*** -3.588*** -3.357***
(0.145) (0.159) (0.148) (0.168) (0.177) (0.178) (0.169)
Observations 3,264 3,264 3,264 3,264 3,264 3,264 3,264
Pseudo R-
squared 0.141 0.141 0.135 0.124 0.128 0.133 0.139
Note:
* denotes significant at 0.05 level. ** denotes significant at 0.01 level.
Robust standard errors are in parentheses.
One state was omitted from the state fixed effects to avoid perfect collinearity.

73

D. Dividing changes after Medicaid eligibility expansions into slope and intercept, to assess
changes in the size of the effect over time
Additional change in annual Medicaid diabetes prescription fills per 1,000 population after
Medicaid eligibility expansions in states that expanded Medicaid eligibility, age 20-64
Change after Medicaid eligibility expansions: Difference in differences
estimate

Intercept

Slope
All 22.39** (14.99 to 29.79) 12.92** (8.10 to 17.74)
By sex
Men 17.79** (11.96 to 23.62) 10.74** (6.94 to 14.55)
Women 23.74** (15.42 to 32.6) 12.99** (7.48 to 18.50)
Most common types of medication
Insulin and insulin analogs 7.29** (4.92 to 9.66) 3.80** (2.23 to 5.36)
Metformin 9.14** (6.8 to 12.19) 5.25** (3.24 to 7.25)
Newer diabetes medications only
Rapid-acting insulin analogs 2.8** (1.43 to 2.73) 1.12** (0.63 to 1.60)
Long-acting insulin analogs 3.58** (2.40 to 4.76) 1.78** (0.96 to 2.60)
Extended release metformin 1.14** (0.70 to 1.57) 0.82** (0.54 to 1.9)
DPP4 inhibitors, GLP1 agonists, and
SGLT2 inhibitors
0.99** (0.56 to 1.42) 0.82** (0.49 to 1.14)
Source: Authors’ analysis of IQVIA data. Estimates are adjusted for year by quarter indicator variables, state
indicator variables, patient’s age group and gender as well as current state-level unemployment rates.
95% confidence intervals are in parentheses.
* denotes significant at 0.05 level. ** denotes significant at 0.01 level.

74

E. Analyses excluding early expansion states, or excluding states that expanded at times
other than January 2014
Change after Medicaid eligibility expansions:
Difference in differences estimate

Excluding early
expansion states

Excluding states that
expanded at times other
than January 2014
All 29.21** (19.40 to 39.1) 33.36** (22.37 to 44.35)
By sex
Men 22.82** (15.93 to
29.71)
25.75** (18.5 to 33.45)
Women 30.41** (18.49 to
42.34)
34.89** (21.60 to 48.17)
Most common types
of medication

Insulin and insulin
analogs
9.64** (6.31 to 12.97) 11.8** (7.41 to 14.76)
Metformin 11.53** (7.67 to 15.40) 13.7** (8.69 to 17.45)
Newer diabetes
medications only

Rapid-acting insulin
analogs
2.79** (1.80 to 3.79) 3.16** (2.4 to 4.28)
Long-acting insulin
analogs
4.69** (3.7 to 6.32) 5.38** (3.59 to 7.18)
Extended release
metformin
1.42** (0.90 to 1.95) 1.62** (0.99 to 2.25)
DPP4 inhibitors,
GLP1 agonists, and
SGLT2 inhibitors
1.27** (0.81 to 1.73) 1.43** (0.90 to 1.95)
Source: Authors’ analysis of IQVIA data. Estimates are adjusted for year by quarter indicator variables, state
indicator variables, patient’s age group and gender as well as current state-level unemployment rates.
95% confidence intervals are in parentheses.
* denotes significant at 0.05 level. ** denotes significant at 0.01 level.

75

F. Analysis with shorter pre-intervention period (excluding data from before 2011)
Additional change in annual Medicaid diabetes prescription fills per 1,000 population after
Medicaid eligibility expansions in states that expanded Medicaid eligibility, age 20-64
Change after Medicaid eligibility expansions: Difference in differences
estimate

Average annual
change

Change in 2014

Change in 2015
All 29.25** (20.97 to
37.53)
23.45** (16.61 to
30.29)
35.41** (25.22 to
45.60)
By sex
Men 23.92** (17.46 to
30.39)
18.97** (13.64 to
24.29)
29.21** (21.19 to
37.24)
Women 30.67** (21.21 to
40.13)
24.99** (17.10 to
32.88)
36.66** (25.9 to
48.23)
Most common
types of
medication

Insulin and insulin
analogs
9.12** (6.43 to
11.80)
7.46** (5.29 to 9.63) 10.87** (7.54 to
14.20)
Metformin 12.38** (8.90 to
15.87)
9.98** (7.6 to 12.89) 14.95** (10.67 to
19.22)
Newer diabetes
medications only

Rapid-acting
insulin analogs
2.63** (1.85 to 3.41) 2.14** (1.53 to 2.76) 3.14** (2.16 to 4.13)
Long-acting
insulin analogs
4.59** (3.19 to 5.99) 3.80** (2.67 to 4.92) 5.43** (3.68 to 7.18)
Extended release
metformin
1.63** (1.14 to 2.12) 1.26** (0.83 to 1.68) 2.3** (1.44 to 2.62)
DPP4 inhibitors,
GLP1 agonists,
and SGLT2
inhibitors
1.53** (1.3 to 2.4) 1.12** (0.68 to 1.56) 1.96** (1.35 to 2.58)
Source: Authors’ analysis of IQVIA data. Estimates are adjusted for year by quarter indicator variables, state
indicator variables, patient’s age group and gender as well as current state-level unemployment rates.
95% confidence intervals are in parentheses.
* denotes significant at 0.05 level. ** denotes significant at 0.01 level.
76

Appendix A3. Prescription fills for patients aged 20-64 using Medicaid insurance, in states
with vs. without Medicaid eligibility expansions by each medication category and age
group
A. Insulin and metformin

77

B. Newer diabetes medications

Source: Authors’ analysis of IQVIA data.
The vertical lines indicate the first quarter of 2014, which was the time of the Medicaid eligibility expansions for
most expansion states.

78

Appendix A4. Tests of parallel trends in Medicaid expansion vs. non-expansion states
prior to the Medicaid eligibility expansions
A. From regression equations to tests of parallel trends
We conducted the differences-in-differences analysis using negative binomial regression
models, which are appropriate for count data with a known exposed population. The models
took the following form:
ln ( Y
it as
) = μ + γ
1
Ti me
t
+ γ
2
( Ti me
t
× Expa nsionState
i
) + π Sta te
i
+ δAg eGroup
a
+ ω Gen der
s
+ βSta te Ump
it
+ ln ( Popul at ion
it as
)
where i indexes state, t time (combination of quarter and year), a age group, s gender. π is a
vector of state fixed effects. Genders and AgeGroupa control for gender and age group effects.
StateUmpit denotes the quarterly state-level unemployment rates. Populationitas is the exposure
variable in the negative binomial model and has a coefficient of 1.

Yitas indicates the outcome variables. Our primary outcome was total diabetes prescriptions
filled. Secondary outcomes included prescription fills for metformin (total, and broken down
into extended release vs. regular metformin); insulin (total, and two newer categories: rapid-
acting, long-acting insulin analogs); three classes of newer oral medications (DPP4 inhibitors,
GLP1 agonists, and SGLT2 inhibitors), and all other classes of diabetes medications.

Timet×ExpansionStatei is an interaction term between Timet and ExpansionStatei.
ExpansionStatei is a binary variable indicating whether the state expanded Medicaid. The non-
interacted variable is omitted from the model due to collinearity with the state fixed effects.
Timet is a continuous variable indicating the annual and quarterly time periods. The regression
used Huber-White robust standard errors clustered at the state level.

Subgroup tests used the same model with samples of people from different age groups.
AgeGroupa, the age group fixed effect, was excluded in the subgroup regressions.

Our test of interest was:
Null hypothesis: 2,=0
Alternate hypothesis: 2,≠0

This tests for common trends in Medicaid expansion vs. non-expansion states in 2008-2013
prior to Medicaid eligibility expansions. All models for tests for common trends only used data
from prior to 2014.

B. Testing common trends in diabetes prescription fills prior to Medicaid eligibility
expansions: adults aged 20-64
79

These tests assess common pre-trends for the models used in Table 2 of the main text.

Difference in prior trends between
expansion and non-expansion states
a
p-value
All -3.8 0.51
By sex
Men -1.5 0.79
Women -5.9 0.3
By type of medication
Insulin 4.2 0.32
Rapid-acting insulin 3.4 0.48
Long-acting insulin 4.7 0.4
Pre-mixed insulin 2.8 0.76
Metformin -6.7 0.42
Extended release metformin 6.2 0.49
Other metformin -7.9 0.4
DPP4 inhibitors, GLP1 agonists, and
SGLT2 inhibitors
12 0.22
All other diabetes medications -8.9 0.25
a
Data are adjusted for year by quarter indicator variables, state indicator variables, patient’s age group and gender
as well as current state-level unemployment rates.
* denotes significant at 0.05 level. ** denotes significant at 0.01 level.

80

C. Testing common trends in diabetes prescription fills prior to Medicaid eligibility
expansions: by age group
These tests assess common pre-trends for the models used to create Figure 2 of the main text.
Difference in prior trends
between expansion and non-
expansion states
a
p-value
Age 20-25 2.5 0.67
Age 26-29 -7.5 0.25
Age 30-34 -2.4 0.7
Age 35-44 -10.7 0.11
Age 45-54 -8.3 0.2
Age 55-59 -1 0.88
Age 60-64 -6.4 0.39
Age 65-69 1.6 0.8
a
Data are adjusted for year by quarter indicator variables, state indicator variables, patient’s age group and gender
as well as current state-level unemployment rates.
* denotes significant at 0.05 level. ** denotes significant at 0.01 level.

81

D. Adjusted gap in prescription fills between Medicaid expansion and non-expansion
states per 1,000 population, quarterly 2008-2015

Data are adjusted for year by quarter indicator variables, state indicator variables, patient’s age group and gender as
well as quarterly state-level unemployment rates.

10 15 20 25 30
2008 2010 2012 2014 2016
Quarter
Estimate 95% Confidence Interval
82

Appendix A5 Prevalence of diabetes just prior to and during Medicaid eligibility
expansions
Age group Prevalence of diabetes per 100 population in
2013-2014
20-25 0.56 (95% CI 0.06 to 1.06)
26-29 1.23 (95% CI 0.09 to 2.38)
30-34 3.79 (95% CI 1.67 to 5.91)
35-44 7.67 (95% CI 5.77 to 9.58)
45-54 12.99 (95% CI 10.44 to 5.531)
55-59 18.67 (95% CI 13.86 to 23.49)
60-64 24.93 (95% CI 19.52 to 30.33)
65-69 25.11 (95% CI 19.18 to 31.05)
Source: Authors’ analysis of 2013-2014 NHANES data. Patients were categorized as having prevalent diabetes if
they self-reported prior diagnosis of diabetes by a health professional or showed signs of undiagnosed diabetes, i.e.,
no self-reported diagnosis of diabetes but biomarker meeting diabetes diagnostic criteria.

83

Appendix A6. Prevalence of diagnosed diabetes and changes in annual Medicaid
prescription fills after Medicaid eligibility expansions, by age group

84

Chapter 5: Conclusion
This dissertation evaluated different government programs or policies that aimed at improving
health and health care among different disadvantaged populations. In the first paper, I focused
on older adults in rural china, who have low income and little access to health care. In China,
there was no formal safety net for rural older adults before 2009. They faced higher out-of-
pocket medical spending due to the less generous health insurance system in rural China, the
New Cooperative Medical Scheme (NCMS). This paper provided new evidence that the New
Rural Pension Scheme (NRPS), a nationwide old-age pension program in rural China, increased
the income of older adults in rural areas of China and improved their access to inpatient care.
The program helped alleviate the financial burden of inpatient care among rural older adults and
provided financial support for healthy aging in China.
In the second and third paper, I examined the impacts of the Patient Protection and Affordable
Care Act (ACA). The ACA was designed to expand health insurance coverage to Americans
who were previously uninsured, improve access to care, and advance health equity. It included
numerous provisions that would potentially increase insurance coverage and improve health
care access among disadvantaged populations, such as people with limited English proficiency
(LEP). Besides that, 29 states and the District of Columbia expanded Medicaid eligibility among
all adults with incomes below 138 percent federal poverty level under the ACA. The ACA
Medicaid eligibility expansions would improve access to diabetes medications among low-
income people. In the second paper, it provided policy makers new information about recent
improvements in health coverage and health care access and reduction in disparities in health
care access by English proficiency among U.S. population after the ACA. In the third paper, it
provided evidence of increases in access to diabetes medications associated with the ACA
85

Medicaid eligibility expansions. The findings added to the growing literature that
socioeconomic gaps in access to health care and medications has been reduced since the
enactment of the ACA. Future studies could take a further step forward and document potential
health benefits among disadvantaged populations associated with the ACA.

Abstract (if available)

Abstract This dissertation consists of three essays that investigated the effects of government programs and policies on health care among disadvantaged population. I examined different populations in different policy contexts. ❧ In the first essay, I examined the effect of the New Rural Pension Scheme (NRPS) on income, health care use, and health among older adults in rural China. China is home to 1 in 4 people aged 60 and older worldwide. The majority live in rural China, where poverty and low access to health care can impede healthy aging. In 2009, with the goal of reducing poverty among rural older adults, the Chinese government launched the New Rural Pension Scheme (NRPS) for adults aged 60 and older. I used data from the China Health and Retirement Longitudinal Study to assess the level of pre-existing poverty and uptake of NRPS at the province level, and used a regression discontinuity design to assess the impacts of NRPS eligibility at age 60. I repeated the analysis before and after expansions in catastrophic medical insurance in 2012. In 2011, 77% of adults in rural China aged 58-59 reported no current source of income. NRPS eligibility at age 60 increased income by 85 yuan per month (USD $12), and increased inpatient care use by 10 percentage points, prior to the expansion of catastrophic medical insurance in 2012. I did not detect an effect of NRPS eligibility on outpatient care or health outcomes. NRPS uptake was less than 25% in multiple high-poverty provinces. The findings indicated that the NRPS and catastrophic medical insurance programs increased access to inpatient care among older adults in rural China. Improved targeting of the poor may be important to increase NRPS program impact. ❧ In the second essay, I described changes in insurance coverage and access to health care by English language proficiency over 2006-2016. In the United States, people with limited English proficiency (LEP) disproportionately experience gaps in health insurance coverage and access to care. The Patient Protection and Affordable Care Act (ACA) of 2010 included reforms that could improve these outcomes for people with limited English proficiency. Data used in this study was from the Medical Expenditure Panel Survey over 2006-2016. I used regression models to estimate changes in coverage and access before and after 2010 for adults with high vs. limited English proficiency, adjusting for respondents’ socio-economic status, demographic characteristics, and health care needs. I used difference-in-differences regression models to assess adjusted changes in disparities by English proficiency after 2010. Gains in health insurance coverage after 2010 were significant for adults with high English proficiency and adults with limited English proficiency (2.5 percentage points, p<0.000, and 5.0 percentage points, p=0.003, respectively)

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Three essays on the evaluation of long-term care insurance policies

PDF

Delivering better care for children with special health care needs: analyses of patient-centered medical home and types of insurance

PDF

Effects of a formulary expansion on the use of atypical antipsychotics and health care services by patients with schizophrenia in the California Medicaid Program

PDF

The value of novel antihyperlipidemic treatments in the U.S. healthcare system: Reducing the burden of cardiovascular diseases and filling the gap of low adherence in statins

PDF

Long-term impacts of childhood adversity on health and human capital

PDF

Economic, clinical, and behavioral outcomes from medical and pharmaceutical treatments

PDF

Three essays in health economics

PDF

Essays on health insurance programs and policies

PDF

Health care utilization and spending of the U.S. aging population

PDF

Essays on the use of microsimulation for health and economic policy analysis

PDF

Three essays on economics of early life health in developing countries

PDF

Essays on development and health economics: social media and education policy

PDF

Essays on macroeconomics of health and labor

PDF

Social determinants of physiological health and mortality in China

PDF

The influence of child day care on cognitive and psychosocial functioning in adolescence

PDF

Essays on health and aging with focus on the spillover of human capital

PDF

Inter professional education and practice in the health care setting: an innovative model using human simulation learning

PDF

Racial/ethnic differences in colorectal cancer patient experiences, health care utilization and their association with mortality: findings from the SEER-CAHPS data

PDF

A series of longitudinal analyses of patient reported outcomes to further the understanding of care-management of comorbid diabetes and depression in a safety-net healthcare system

PDF

Acculturation team-based clinical program: pilot program to address acculturative stress and mental health in the Latino community

Asset Metadata

Creator Lu, Tianyi (author)

Core Title Three essays on estimating the effects of government programs and policies on health care among disadvantaged population

School School of Pharmacy

Degree Doctor of Philosophy

Degree Program Health Economics

Publication Date 08/15/2019

Defense Date 05/28/2019

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag Affordable Care Act,China,Health Economics,Health policy,Medicaid expansion,New Rural Pension Scheme,OAI-PMH Harvest

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Myerson, Rebecca (committee chair), Joyce, Geoffrey (committee member), Romley, John (committee member)

Creator Email tianyi.econ@gmail.com,tianyil@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c89-216044

Unique identifier UC11662942

Identifier etd-LuTianyi-7665.pdf (filename),usctheses-c89-216044 (legacy record id)

Legacy Identifier etd-LuTianyi-7665.pdf

Dmrecord 216044

Document Type Dissertation

Format application/pdf (imt)

Rights Lu, Tianyi

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA