HEALTH POLICY/ORIGINAL RESEARCH
Increased Use of the Emergency Department After Health Care
Reform in Massachusetts
Peter B. Smulowitz, MD, MPH*; James O’Malley, PhD; Xiaowen Yang, PhD; Bruce E. Landon, MD, MBA
*Corresponding Author. E-mail: [email protected].
Study objective: With implementation of the Patient Protection and Affordable Care Act, 30 million individuals are
predicted to gain access to health insurance. The experience in Massachusetts, which implemented a similar reform
beginning in 2006, should provide important lessons about the effect of health care reform on emergency department
(ED) utilization. Our objective is to understand the extent to which Massachusetts health care reform was associated
with changes in ED utilization.
Methods: We compared changes in ED utilization at the population level for individuals from areas of the state that
were affected minimally by health care reform with those from areas that were affected the most, as well as for those
younger than 65 years and aged 65 years or older. We used a difference-in-differences identification strategy to
compare rates of ED visits in the prereform period, during the reform, and in the postreform period. Because we did not
have population-level data on insurance status, we estimated area-level insurance rates by using the percentage of
actual visits made during each period by individuals with insurance.
Results: We studied 13.3 million ED visits during 2004 to 2009. Increasing insurance coverage in Massachusetts was
associated with increasing use of the ED; these results were consistent across all specifications, including the younger
than 65 years versus aged 65 years or older comparison. Depending on the model used, the implementation of health
care reform was estimated to result in an increase in ED visits per year of between 0.2% and 1.2% within reform and
0.2% and 2.2% postreform compared with the prereform period.
Conclusion: The implementation of health care reform in Massachusetts was associated with a small but consistent
increase in the use of the ED across the state. Whether this was due to the elimination of financial barriers to seeking
care in the ED, a persistent shortage in access to primary care for those with insurance, or some other cause is not
entirely clear and will need to be addressed in future research. [Ann Emerg Med. 2014;64:107-115.]
Please see page 108 for the Editor’s Capsule Summary of this article.
A feedback survey is available with each research article published on the Web at www.annemergmed.com.
A podcast for this article is available at www.annemergmed.com.
0196-0644/$-see front matter
Copyright © 2014 by the American College of Emergency Physicians.
http://dx.doi.org/10.1016/j.annemergmed.2014.02.011
SEE EDITORIAL, P. 116.
INTRODUCTION
Background and Importance
The full implementation of the Patient Protection and
Affordable Care Act of 2010, including both the individual
mandate to purchase insurance and Medicaid expansions set to
begin in 2014, will extend health insurance to a projected 30
million people by 2019.1 The act was modeled after similar
health care reform legislation that Massachusetts implemented
beginning in 2006. The combination of an individual mandate,
employer assessments, insurance market reforms, and expansion
of publicly subsidized insurance products was designed primarily
to reduce the number of Massachusetts residents without health
Volume 64, no. 2 : August 2014
insurance. Recent estimates show that about 1.9% of
Massachusetts residents are uninsured2 compared with about
10% before the enactment of health care reform and 16%
nationally.3
A looming question for policymakers and administrators is
how health care reform will affect the use of health care services
and, in particular, patterns of seeking emergency care. The
answer to this question may have a profound effect on state
and delivery system planning in advance of the widespread
increases in insurance coverage that will begin in 2014. The
Massachusetts experience provides a unique opportunity to
study how implementation of health insurance reform on which
the Patient Protection and Affordable Care Act was modeled
affected patterns of emergency department (ED) use on a
population level. To date, evaluations of the effect of health care
Annals of Emergency Medicine 107
Smulowitz et al
Increased Use of Massachusetts Emergency Departments After Health Care Reform
Editor’s Capsule Summary
What is already known on this topic
The effect of gaining health insurance on emergency
department (ED) use is not clear.
What question this study addressed
How does ED use change after mandated health
insurance coverage?
What this study adds to our knowledge
ED visits increased after Massachusetts health care
reform. The increase was greater in areas of
Massachusetts with larger increases in insurance
coverage.
How this is relevant to clinical practice
We should expect ED visit rates to increase as the
Patient Protection and Affordable Care Act expands
subsidized health insurance to millions of low-income
Americans.
reform on ED visits in Massachusetts have revealed conflicting
results.4-6
Goals of This Investigation
In this study, we examine changes in ED use that resulted
from the implementation of Massachusetts health care reform,
using data from a statewide database capturing all ED visits,
including those that resulted in hospital admissions or
observation stays. Our study design takes advantage of the
differential effect of health care reform across the state by
comparing changes in ED care at the population level for
individuals from areas of the state that were affected the most by
health care reform with those from areas that were affected the
least. Thus, our method allows us to make stronger inferences
about the effect of health care reform on use as opposed to simply
tracking use trends over time or making comparisons with states
that might differ in other fundamental ways. In addition, we
examine changes in use for the population aged 65 years and
older, which, because of Medicare coverage, should not have
been affected by Massachusetts health care reform. Our major
limitation is the use of the frequency of ED visits by the
uninsured as an estimate of the true frequency of uninsured
individuals in an area. We use this estimate because we did not
have access to population-level data on the percentage of
individuals without insurance. We compensated for this by using
several different modeling strategies to test the robustness of our
results. Because attaining insurance typically is associated with
increased use of services and because of constraints on access to
alternative sites of care outside of the ED, we hypothesized that
health care reform would be associated with increased ED use.
108 Annals of Emergency Medicine
MATERIALS AND METHODS
In 2006, an act providing access to affordable, quality,
accountable health care put in place in Massachusetts a series of
Medicaid expansions, subsidized insurance offerings, insurance
market reforms, safety net alterations, and individual and
employer mandates to purchase or provide health insurance. The
various components of the reform, implemented in stages
between 2006 and 2008, have been previously described in
detail4 and are depicted in Figure 1. More details can be found in
Appendix E1 (available online at http://www.annemergmed.com).
We analyzed data from 3 periods: prereform (October 1,
2004, through September 30, 2006), within reform (October
1, 2006, through September 30, 2007), and postreform (October
1, 2007, through September 30, 2009). Although the
postreform period technically includes the last quarter before
the implementation of the full individual penalty, we used
this structure to maintain consistent seasons in each group.
Data Collection and Processing
We obtained data from the state of Massachusetts that included
all patients treated and discharged from the ED (ED discharge
database), as well as databases on observation and full hospital
admissions that could be used to identify those admissions that
came from the ED. Thus, across the 3 data sources, we captured
information on all ED visits in the state during the period
spanning October 1, 2004, to September 30, 2009, that are
submitted annually from 69 acute care hospitals, accounting for
approximately 2 million annual outpatient ED visits, 850,000
inpatient admissions, and 150,000 observation stays. Each of the
3 databases includes deidentified detailed information on the
demographic characteristics of the patient (including age, sex,
and race/ethnicity), payer type, and 5-digit zip code of residence.
We included patients of all ages in the study.
There is no extant data source that can be used to measure the
insurance status of every resident of the state. In place of a direct
measure, we estimated area-level rates of insurance for each zip
code during each period (first by quarter and then for each of
the 3 periods of study) by calculating the percentage of visits by
the insured for actual visits during the relevant period in each
of the zip codes in the state (there are technically 566 zip codes
but only 502 with a population greater than zero) and then
calculated the change in insurance coverage from the first to the
last period. We included 500 of these zip codes in our
analyses because a few of them had such small populations that
no ED visits originated from these areas.
We categorized patients as being self-pay, “free care” (in the
uncompensated care pool or health safety net), Commonwealth
Care, Medicaid, Medicare, or commercial insurance. We then
created 2 larger groupings of coverage as “uninsured” for the
self-pay category, and “insured” for all else because free care
patients have coverage for services even though only at an
individual hospital or community health center. Because we
estimated the “rate” of uninsured in each zip code from the actual
percentage of uninsured visiting the ED, our calculations of
Volume 64, no. 2 : August 2014
Smulowitz et al
Increased Use of Massachusetts Emergency Departments After Health Care Reform
Figure 1. Timeline of Massachusetts health reform and study periods.
changes in the percentage of insurance coverage in an area could
have been confounded by changes in use that might have
resulted from health care reform. Because visits over time might
thus be endogenous with insurance status in an area, we
separately calculated the baseline percentage of insurance, using
ED visits from the prereform period as a sensitivity analysis
because this measure would not be affected by changes in use
over time that might have resulted from health care reform.
We categorized patients throughout all the years as white,
black, Asian, or other and calculated the proportions in
each category for each zip code for each year of the study.
Unfortunately, the inconsistent coding practices for race made it
difficult to accurately report the percentage of Hispanic patients
with ED visits over time.
Primary Data Analysis
We used 2 specifications based on a difference-in-differences
identification strategy to directly examine the influence of health
care reform on ED utilization. A difference-in-differences
approach is meant to compare differences between the before and
after periods (“differences”) for treatment and control groups, as
opposed to simply measuring the difference in one group before
and after an intervention or between different groups after an
intervention.7 This approach assumes that the differences in the
proportion of ED visits by the insured across the different study
periods is due to the health care reform and not other reasons,
which are differenced by the inclusion of contemporaneous
control areas within the state that were less affected by health care
reform.
Our first approach regresses the number of visits per zip code
(measured quarterly) on the change in insurance rate from the
Volume 64, no. 2 : August 2014
first to the last period within the zip code (and separately using
the baseline prereform percentage of visits by the insured),
controlling for patient demographics and time of year. The key
coefficients are the interaction effects between the change in
insurance (or baseline insurance as a sensitivity analysis) and
period, which represent the extent to which the increase in the
number of visits in the within-reform and postreform periods was
greater in zip codes that experienced larger increases in insurance,
or in the case of the sensitivity analysis, to which any change in
visits is related to a higher percentage of visits by the uninsured in
the prereform period.
The second difference-in-differences specification tests
whether ED utilization for the population younger than 65 years
in the state was differentially affected across the periods of health
care reform compared with the population aged 65 years or older,
almost all of whom were covered by Medicare throughout the
study period.
Finally, as a robustness check, we evaluated the extent that
visits increased within a zip code as a function of changes in
the insurance rate over time. We accounted for common
variation in the number of visits across time (eg, seasonal
effects, secular trends) through the inclusion of month
dummies. This model tests whether changes in the insurance
rate that resulted from health care reform (ignoring the period
of health care reform) are related to changes in the rate of visits
within a zip code.
A more detailed discussion of the approach to our statistical
models is presented in Appendix E2 (available online at http://
www.annemergmed.com). All models assume that unmeasured
characteristics of the population in different geographic areas
of the state remained relatively constant over time and, more
Annals of Emergency Medicine 109
Smulowitz et al
Increased Use of Massachusetts Emergency Departments After Health Care Reform
Table 1. Characteristics of visits for each period of health care reform.*
<65 Years
Characteristics, Health Care
Reform Period
Total visits
Insured
Uninsured
Visits by uninsured
Visit disposition
Discharged
Insured
Uninsured
Observation
Insured
Uninsured
Admitted
Insured
Uninsured
Median age (IQR), y
Insured
Uninsured
Women
Insured
Uninsured
Race
White
Insured
Uninsured
Black
Insured
Uninsured
Asian
Insured
Uninsured
Hispanic/other
Insured
Uninsured
Primary insurance type
Commercial
Medicare
Medicaid
Uncompensated care pool/health safety net
Commonwealth Care
Missing
Uninsured
‡65 Years
Baseline
(2 Years)
During Reform
(1 Year)
Postreform
(2 Years)
3,910,524
372,678
9.5
2,001,535
147,769
7.4
4,225,596
239,715
5.7
90.2
96.6
91.1
97.1
90.8
97.3
2.3
0.7
2.3
0.8
2.7
0.9
7.5
2.7
6.6
2.2
6.5
1.8
31 (28)
28 (18)
31 (28)
28 (17)
31 (27)
28 (18)
41.3
52.1
44.0
52.4
40.3
52.5
62.7
76.7
59.9
73.9
64.2
75.4
12.0
11.0
12.1
11.2
13.9
11.5
1.4
1.7
1.2
1.7
1.5
1.9
23.9
18.6
26.9
21.7
20.4
16.7
54.0
7.0
23.6
6.7
0
0.01
8.7
51.6
7.4
27.6
6.5
0.1
0.02
6.9
49.3
7.9
29.7
4.9
2.8
0.1
5.4
Baseline
During
Reform
Postreform
921,311
444,180
919,667
1.2
1.0
0.8
56.1
58.0
57.0
4.7
5.3
6.8
39.3
36.8
36.2
78 (13)
78 (13)
78 (13)
59.7
59.4
59.2
88.9
83.9
85.9
4.0
4.1
4.4
1.0
1.0
1.1
6.1
11.0
8.5
9.3
88.5
1.1
0
0
0.01
0.4
10.1
87.7
1.4
0
0
0.01
0.2
10.6
87.0
1.6
0
0.1
0.04
0.1
*Data are percentage unless otherwise indicated.
important, that there were no shifts in places of residence as a
direct result of health care reform. The expected number of ED
visits was modeled with a Poisson distribution. Additionally, we
include fixed effects for zip code and quarter to account for timeinvariant confounders and the general population trend over
time. Our quarterly time-trend dummies account for year effects,
seasonal effects, and their interaction.
In all our models, we incorporated data from 500 of the 566
zip codes in the state; we removed zip codes where no ED visits
originated from that area. Because we use a Poisson regression
model, the model automatically weights the contribution of each
zip code such that larger zip codes (those with a higher expected
110 Annals of Emergency Medicine
number of visits) contribute more to the fitted regression model
than smaller ones.
All analyses were performed with SAS (version 9.1.3; SAS
Institute, Inc., Cary, NC). The study was approved by the
institutional review board at Beth Israel Deaconess Medical Center.
RESULTS
For the population younger than 65 years, we studied 4.3
million visits in the 2 years before health care reform, 2.2 million
visits during the health care reform period, and 4.5 million
visits during the post–health care reform period (Table 1). The
Volume 64, no. 2 : August 2014
Smulowitz et al
Increased Use of Massachusetts Emergency Departments After Health Care Reform
Table 2. Percentage of ED visits by the uninsured.
Period, %
Quartile
Pre
During
Post
1
2
3
4
9.25
7.11
5.9
5.15
7.03
5.59
4.8
4.53
4.86
4.52
4.17
4.37
population aged 65 years and older included an additional
921,311 visits in the prereform period, 444,180 visits during
health care reform, and 919,667 visits in the postreform period.
Overall ED visits by the uninsured younger than 65 years
decreased from 9.5% of overall visits before health care reform to
5.7% of visits afterward, whereas the percentage of ED visits by
the uninsured for the population aged 65 years and older
remained steady, at about 1%. Free care visits for the population
younger than 65 years decreased from 6.7% before reform to
3.8% after. Five hundred zip codes were included in our analysis.
From the prereform to within-reform periods, 462 zip codes
experienced a decrease in the percentage of visits by the
uninsured and 38 zip codes experienced an increase. From the
prereform to postreform periods, 484 zip codes experienced a
decrease in the percentage of visits by the uninsured and 16 zip
codes experienced an increase.
Table 1 shows the demographic characteristics of the
population, as well as visits for each type of ED visit (admission,
observation, and discharge) across the periods of health care
reform. The median age of patients younger than 65 years was 31
years, and approximately half were women across all study years.
The demographic characteristics of those visiting the ED
remained substantively unchanged over time. Unfortunately,
changes in coding practices for race made it difficult to properly
report the percentage of Hispanic patients with ED visits over
time. Table 1 also demonstrates a breakdown of visits for all 3
periods of the study by type of insurance. This table in particular
illustrates the visits by individuals covered by the new publicly
subsidized insurance product referred to as Commonwealth Care,
in addition to visits related to the expansion of Medicaid.
Table 2 is presented to demonstrate the percentage of ED
visits by the uninsured for each of the 4 quartiles of change in
insurance during the 3 periods of health care reform. Results for
the modeling estimation strategies are presented in Table 3. Our
difference-in-differences analyses based on the change in
insurance rate reveal a positive effect on ED utilization both
during and after reform (within-reform estimate 0.32, P¼.03;
postreform estimate 0.57, P¼.008), suggesting that areas of the
state with greater increases in insurance as a result of health care
reform had greater increases in ED utilization. This translates to a
1.2% increase (95% confidence interval [CI] 0.2% to 2.3%) in
visits per year during the within-reform period and a 2.2%
increase (95% CI 0.6% to 3.8%) during the postreform period.
The analysis comparing use for the populations younger than
65 years versus aged 65 years and older also revealed a positive
effect of health care reform on overall ED utilization among the
population younger than 65 years for both the within-reform
(estimate 0.05; P<.001) and postreform periods (estimate 0.04;
P<.001). This would result in a 0.2% increase (95% CI 0.1% to
0.2%) in visits per year during the within-reform period and a
0.2% increase (95% CI 0.1% to 0.2%) in the postreform period.
The results of the fixed-effect models and the sensitivity analysis
(Table 3) also are consistent with the results of the primary
difference-in-differences models.
Figure 2 demonstrates the average number of visits across all
zip codes, using the raw data, adjusted only for season. This
figure must be interpreted with caution because it does not
examine the change in visits in areas according to area changes
in the percentage of visits by the uninsured. Figure 3 shows ED
Table 3. Results of the primary model of ED utilization and sensitivity analyses.*
Results
Modeling Approach
Model based on changes in the percentage of insurance coverage over time
across zip codes (comparing prereform to during reform and postreform
periods)
Same as above but using baseline percentage of visits by individuals with
insurance
Model comparing visits for the <65 y and 65 y over time (comparing
prereform to during reform, and postreform periods)
Sensitivity analysis: effect of insurance level on use within a zip code over
time‡
Estimate
(Regression
Coefficient)
P Value
Increase in Additional
Visits per Year From Health
Care Reform (95% CI), %†
During reform: 0.32
Postreform: 0.57
During reform: .03
Postreform: .008
During reform: 1.2 (0.2–2.3)
Postreform: 2.2 (0.6–3.8)
During reform: 0.27
Postreform: 0.45
During reform: 0.05
Postreform: 0.04
0.52
During reform: <.001
Postreform: <.001
During reform: <.001
Postreform: <.001
<.001
During reform: 1.0 (0.5–1.6)
Postreform: 1.7 (0.9–2.5)
During reform: 0.2 (0.1–0.2)
Postreform: 0.2 (0.1–0.2)
Increase of 2.0 (1.0–3.0)
*Controls for person and place characteristics.
†
Computed as percentage change¼100[EXP(estimate0.038)–1], where 3.8% is the absolute increase in the percentage of visits by individuals with insurance from the
prereform to postreform periods. The resulting quantity is the percentage change in ED visits, given a 3.8% change in the corresponding predictor.
‡
Model accounts for common variation in the number of visits across time (eg, seasonal effects, secular trends) through the inclusion of month dummies and includes fixed
effects for zip code to account for time-invariant features of each zip code. Tests whether changes in the percentage of insurance coverage that resulted from health care reform
(ignoring the period of health care reform) are related to the number of visits in a zip code. Model also controls for patient and place characteristics measured at the zip code level.
Volume 64, no. 2 : August 2014
Annals of Emergency Medicine 111
Smulowitz et al
Increased Use of Massachusetts Emergency Departments After Health Care Reform
Figure 2. Average number of visits per quarter across all zip codes.*
visits over time (by quarter) for the 4 quartiles of zip codes,
ranging from the quartile that was most affected by health care
reform to the quartile of zip codes that showed the least change in
insurance coverage. Visually, it is apparent that the quartile
including the zip codes with the largest change in insurance had a
relative increase in visits compared with the quartile with the least
change in insurance. On the other hand, there is no clear pattern
of increase for the population older than 65 years.
LIMITATIONS
There are several limitations to our analyses. The ideal way to
study this issue outside the context of a randomized controlled
trial would be to obtain insurance coverage data on a defined
cohort of patients and compare those who obtain health
insurance with those who either did not obtain health insurance
or already had health insurance, using a difference-in-differences
approach. Such data, however, are unavailable. The statewide
Figure 3. Change in ED visits (estimated from the second regression model) for younger than 65 years (left panel) and aged 65
years or older (right panel), indexed by quarter, comparing the 4 quartiles of zip codes after removing seasonal variation and
standardizing to an index of 100 in quarter 1 (October 2004).
112 Annals of Emergency Medicine
Volume 64, no. 2 : August 2014
Smulowitz et al
Increased Use of Massachusetts Emergency Departments After Health Care Reform
data set of ED visits that we used is a next best approximation. A
major limitation of this approach is the use of the frequency of
ED visits by the uninsured as an estimate of the overall
percentage of uninsured in the overall area population. We used
this estimate because there is no extant source of population-level
data on the percentage of individuals without insurance. This
potentially means that our calculations of changes in rates of
insurance coverage in an area could have been confounded by
changes in use that might have resulted from health care reform.
We compensated for this by using several different modeling
strategies. In addition, we found consistent results in models by
using the baseline insurance rates, which would not have the
same limitation. Furthermore, the difference-in-difference
analysis using the population older than 65 years as a control
group had similar results.
An additional limitation is the use of a visit-based analysis
(deidentified data not accounting for individual person or
repeated visits) rather than a person-based analysis. Using
ED visits as opposed to individual persons may introduce the
potential bias of determining population-based insurance
rates based on those with frequent use of the ED. Our
sensitivity analysis based on the prereform uninsurance rate,
however, addresses this concern and demonstrated similar
results.
We also could not explicitly control for population-level
factors in each zip code, other than by relying on information
from patients who used the ED. However, it is unlikely that
migration patterns within the state would have been driven by
health care reform, or that any migration patterns would explain
our results. Although the rate of increase of the Massachusetts
population between 2005 (6,398,743) and 2010 (6,646,144) of
3.8% slightly exceeds the growth in ED visits observed in our
study, there would have to be substantial growth centered
specifically in the areas of increased insurance for population
growth to solely explain our findings. Furthermore, as
mentioned above, the consistency in results of our 3 models is
reassuring in that our results are not explained by trends in
population growth.
We also did not focus specifically on use by individuals with
Medicaid, which has been studied extensively elsewhere.8-10 Our
analyses focus on the larger effect of increasing health insurance
coverage—temporally associated with health care reform—on
utilization of the ED. However, our results are generally
consistent with the most recent results from the Oregon
Medicaid experiment, in which gaining insurance was associated
with increases in ED utilization.10
Finally, our analyses are likely to underestimate the effect of
health care reform nationally because of the baseline high rate of
insurance coverage in Massachusetts. Even before health care
reform implementation, Massachusetts had a robust safety net
system in place, and most patients had health insurance. For
instance, our data show that the percentage of insured visits to
the ED increased by just 3.8% from the prereform to the
postreform period. Thus, the magnitude of the effect on ED
visits is no doubt limited.
Volume 64, no. 2 : August 2014
DISCUSSION
To our knowledge, this is the first study using complete data
on statewide ED visits to demonstrate that the increase in
insurance coverage associated with health care reform in
Massachusetts resulted in a small but measurable increase in ED
visits across the state. Not unexpectedly, these findings were
concentrated in areas of the state that were most affected by
health care reform and were not present for the population that
was covered by Medicare for the entire study period. Our
findings also are robust and consistent across several different
specifications. These results carry significant implications for our
understanding of the effect of insurance on the use of emergency
care and for states planning for the impending effect of national
health care reform.
In a previous study using data from selected Massachusetts
hospitals, health care reform was associated with a small decrease
in low-urgency ED visits but an increase in admissions for
ambulatory care–sensitive conditions, perhaps indicating that the
effect of health care reform was overshadowed by limitations in
access to primary care.4 Chen et al5 recently demonstrated that
although the rate of overall ED visits in Massachusetts continued
to increase after health care reform, the rate of ED use and
admissions from the ED in Massachusetts did not differ
substantially from that of nearby states during the same period.
That study, however, was underpowered to detect changes of the
magnitude we observed and could not account for state-specific
policies that might have differentially affected growth rates within
a state. In addition, recent survey data from the Urban Institute
and the Blue Cross Foundation reported a small decrease in selfreported ED utilization from 2006 to 2010, but these findings
are limited by small sample sizes and possible nonresponse
bias.11,12 Finally, Miller6 studied the effect of health care reform
on all ED discharges (ie, did not include observation visits or
admissions) according to county-level prereform uninsurance
rates and found a decrease in lower-urgency ED visits. This study
has a number of potential limitations, most notably evaluating
only ED discharges. Our study improves on these earlier studies
by using comprehensive data on all ED visits in the state and by
comparing areas or populations within Massachusetts for which
we would expect more or less effect from health care reform. As a
robustness test, we also examined these same trends using a
model comparing use with overall coverage during the 20
quarters, without respect to the phase of health care reform,
which also yielded similar results.
Both economic theory and previous empirical evidence
suggest that having health insurance may increase use of certain
health care resources, including the ED, and, conversely, that
greater financial barriers such as increased copayments may result
in reduced use of both necessary and unnecessary care.13 Recent
evidence from the Oregon Medicaid expansion further supports
this link between insurance and greater overall health care use.9,10
The connection between health insurance and ED utilization
likely is related to a number of factors, including financial barriers
to care, universally available ED care as mandated by the federal
Emergency Medical Treatment and Labor Act, and access to
Annals of Emergency Medicine 113
Smulowitz et al
Increased Use of Massachusetts Emergency Departments After Health Care Reform
primary care. In this context, our results are as expected:
increasing the rate of insurance may have reduced financial
barriers to care while minimally affecting barriers to accessing
timely care outside of the ED.
Thus, our results are consistent with the view that although
extending health insurance through expansions of programs
such as Medicaid and Commonwealth Care (publicly subsidized
insurance products) may reduce some of the financial barriers to
accessing a regular source of care, other nonfinancial barriers
continue to limit overall access to care. In addition, access to
primary care for individuals who already had insurance might
have been affected by the increased number of newly insured,
leading to increased use of the ED among this population. The
reality is that primary care access remains limited in
Massachusetts14,15 and across the United States, and in most
states the problem is far grimmer than in Massachusetts. Indeed,
the past decade has witnessed an increase in the prevalence of
barriers to timely primary care and an associated increase in ED
utilization in other settings.16 Consequently, our study suggests
that other states should be prepared for equal or greater influxes
of patients into the ED after health care reform is fully
implemented. Before making any definitive conclusions about
the potential effect of health care reform in other states,
however, we will need additional data in Massachusetts about
primary care and other health care use after health care reform
in addition to data from other states as health care reform is
implemented.
This study also should further weaken the long-held notion
that high use of the ED is being driven mainly by the uninsured.
Other recent data confirm that the percentage of ED visits made
by the uninsured is proportional to the rate of uninsured in
the population, though the reasons for ED visits differ between
the insured and uninsured.17 The drivers of ED use are
multifactorial, encompassing enabling factors such as
transportation, 24-hour availability, the ability to take time off
from work or find suitable child care, distance to the nearest ED
or primary care provider,18 and the perceived efficiency,
technologic expertise, and quality of ED care,19,20 increasing
awareness of emergency conditions, and the increasing
commodification of health care that accompanies the diminished
role and availability of primary care.21 Health system reforms will
need to be mindful of each of these factors when care systems are
designed for the future.
In summary, in this study of health care reform
implementation in Massachusetts we find that increases in
health insurance coverage were associated with small but
consistent increases in overall ED utilization. Additional study
after the implementation of the Patient Protection and
Affordable Care Act will be important. Although there is no
definitive explanation for these findings, previous work suggests
there are several potential explanations, including but not
limited to pent-up demand before obtaining health insurance
coverage, decreased economic barriers to seeking emergency
care, the lack of access to primary care, or a combination of
these factors.
114 Annals of Emergency Medicine
Supervising editor: Melissa L. McCarthy, ScD
Author affiliations: From the Department of Emergency Medicine
(Smulowitz) and Division of General Medicine and Primary Care
(Landon), Beth Israel Deaconess Medical Center, Boston, MA; the
Dartmouth Institute for Health Policy and Clinical Practice Geisel
School of Medicine at Dartmouth, Hanover, NH (O’Malley); the
Department of Health Care Policy, Harvard Medical School, Boston,
MA (Landon); and the Department of Economics, Massachusetts
Institute of Technology, Boston, MA (Yang).
Author contributions: PBS, JO, and BEL conceived the study and
obtained research funding and data. PBS and XY were responsible
for collecting and managing the data. XY analyzed the data. JO
provided primary statistical advice. PBS drafted the article, and all
authors contributed substantially to its revision. PBS takes
responsibility for the paper as a whole.
Funding and support: By Annals policy, all authors are required to
disclose any and all commercial, financial, and other relationships
in any way related to the subject of this article as per ICMJE conflict
of interest guidelines (see www.icmje.org). The authors have stated
that no such relationships exist and provided the following details:
Dr. Smulowitz received a grant from the Charles A. King Trust
Postdoctoral Research Fellow Program and the Eleanor and Miles
Shore Fellowship Program for Scholars in Medicine.
Publication dates: Received for publication August 19, 2013.
Revisions received November 20, 2013; January 16, 2014; and
January 29, 2014. Accepted for publication February 7, 2014.
Available online March 20, 2014.
REFERENCES
1. Congressional Budget Office. Estimates for the insurance coverage
provisions of the Affordable Care Act updated for the recent Supreme
Court decision. July 2012. Available at: http://cbo.gov/sites/default/
files/cbofiles/attachments/43472-07-24-2012-CoverageEstimates.
pdf. Accessed March 2, 2014.
2. Massachusetts Health Connector. Health reform facts and figures. Fall
2012. Available at: https://www.mahealthconnector.org/portal/
binary/com.epicentric.contentmanagement.servlet.ContentDelivery
Servlet/Health%2520Care%2520Reform/Facts%2520and%2520
Figures/Facts%2520and%2520Figures.pdf. Accessed March 2, 2014.
3. US Census Bureau. Current population survey, 2009. Available at:
http://www.census.gov/hhes/www/cpstables/032010/health/h06_
000.htm. Accessed March 2, 2014.
4. Smulowitz PB, Lipton R, Wharam J, et al. Emergency department
utilization after the implementation of Massachusetts health reform.
Ann Emerg Med. 2011;58:225-234.
5. Chen C, Scheffler G, Chandra A. Massachusetts’ health care
reform and emergency department utilization. N Engl J Med. 2011;7:
1-3.
6. Miller S. The effect of insurance on emergency room visits: an analysis
of the 2006 Massachusetts health reform. J Public Econ. 2012;96:
893-908.
7. Wooldridge JM. Econometric Analysis of Cross-Sectional and Panel
Data. Cambridge, MA: MIT Press; 2002.
8. Baicker K, Finkelstein A. The effects of Medicaid coverage—learning
from the Oregon experiment. N Engl J Med. 2011;365:683-685.
9. Finkelstein A, Taubman S, Wright B, et al. The Oregon health insurance
experiment: evidence from the first year. Q J Econ.
2012;127:1057-1106.
Volume 64, no. 2 : August 2014
Smulowitz et al
Increased Use of Massachusetts Emergency Departments After Health Care Reform
10. Taubman SL, Allen HL, Wright BJ, et al. Medicaid increases emergencydepartment use: evidence from Oregon’s health insurance experiment.
Science. 2014;343:263-268.
11. Long S, Stockley K, Dahlen H. Health reform in Massachusetts: an
update as of Fall 2010. Available at: http://bluecrossmafoundation.
org/publication/health-reform-massachusetts-update-fall-2010.
Accessed May 25, 2013.
12. Long SK, Stockley K, Dahlen H. Massachusetts health reforms:
uninsurance remains low, self-reported health status improves as
state prepares to tackle costs. Health Aff (Millwood). 2012;31:
444-451.
13. Manning W, Newhouse J, Duan N, et al. Health Insurance and the
Demand for Medical Care. Evidence From a Randomized Experiment.
Santa Monica, CA: RAND Corp; 1988. Report R-3476-HHS. ISBN
0-8330-0864-1. Available at: http://www.rand.org/pubs/reports/
2005/R3476.pdf. Accessed March 2, 2014.
14. Massachusetts Medical Society Physician Workforce Study. Available
at: http://www.massmed.org/AM/Template.cfm?Section¼MMS_
15.
16.
17.
18.
19.
20.
21.
News_Releases&TEMPLATE¼/CM/ContentDisplay.cfm&CONTENTID¼
31511. Accessed September 2009.
Long S, Masi P. Access and affordability: an update on health reform in
Massachusetts. Health Aff (Millwood). 2009;28:578-587.
Cheung PT, Wiler JL, Ginde AA. Changes in barriers to primary care and
emergency department utilization. Arch Intern Med. 2011;171:
1397-1399.
Carlson JN, Menegazzi JJ, Callaway CW. Magnitude of national ED visits and
resource utilization by the uninsured. Am J Emerg Med. 2013;31:722-726.
Lowe RA, Schull M. On easy solutions. Ann Emerg Med. 2011;58:235-238.
Stockwell M, Findley S, Irigoyen M, et al. Change in parental reasons
for use of an urban pediatric emergency department in the past
decade. Pediatr Emerg Care. 2010;26:181-185.
Northington W, Brice J, Zou B. Use of an emergency department by
nonurgent patients. Am J Emerg Med. 2005;23:131-137.
McKinlay J, Marceau L. When there is no doctor: reasons for the
disappearance of primary care physicians in the US during the early
21st century. Soc Sci Med. 208;67:1481–1491.
Change of Shift: Collected Essays
By
Anna Olson, MD
This free iBook, available for iPAD users in the iBooks store, includes the top
selections from over three decades of essays, poetry, and personal perspectives
published in Annals. Some of these essays have been used as the basis of
episodes of the former TV show, “ER.” Download your copy today. Available for
free on the iTunes Store.
Volume 64, no. 2 : August 2014
Annals of Emergency Medicine 115
Increased Use of Massachusetts Emergency Departments After Health Care Reform
APPENDIX E1.
Description of Massachusetts health care reform.
Overall, approximately 70% of the newly insured in
Massachusetts obtained coverage through expansion of
MassHealth, the state’s Medicaid program (including an
extension of children’s eligibility from 200% to 300% of the
federal poverty level), and enrollment in Commonwealth Care, a
publicly subsidized option for individuals with low income who
do not qualify for Medicaid. On October 1, 2006,
Commonwealth Care became available to eligible individuals
with incomes under 100% of the federal poverty line. Also, an
employer assessment of $295 per employee was put into place for
employers with more than 11 employees who failed to offer “fair
and reasonable” coverage. Expanded Commonwealth Care with
sliding subsidies for individuals with incomes between 100% and
300% of federal poverty line became available on January 1,
2007. Many of the individuals who enrolled in both fully
subsidized and partially subsidized Commonwealth Care plans
were previously uninsured but covered for services at specified
individual hospitals and community health centers through the
115.e1 Annals of Emergency Medicine
Smulowitz et al
Uncompensated Care Pool, a system of reimbursement to
individual hospitals and community health centers that provided
coverage to lower-income and underinsured people, but only at a
specific institution. Under health care reform, the
Uncompensated Care Pool was replaced with a smaller Health
Safety Net Fund providing care mostly to undocumented
immigrants and underinsured lower-income individuals for
whom insurance was deemed not “affordable.”
In April of 2007, the Commonwealth Connector, a quasi
public health insurance exchange, began offering unsubsidized
coverage for small businesses and individuals with incomes in
excess of 300% of the federal poverty line (Commonwealth
Choice plans). The Connector also manages the Commonwealth
Care plans offered by 4 existing Medicaid managed care plans.
Beginning in July 2007, the individual mandate to purchase went
into effect, subjecting individuals without insurance to a penalty
initially equal to loss of the personal income tax exemption: about
$218 for an individual and $437 for a family. Starting in 2008,
the tax penalty increased to half the monthly cost of the lowest
available premium for which an individual would have qualified
through the connector (equal to $912 in 2008).
Volume 64, no. 2 : August 2014
Smulowitz et al
Increased Use of Massachusetts Emergency Departments After Health Care Reform
APPENDIX E2.
Detailed description of statistical models.
We use 3 general approaches based on a difference-indifferences identification strategy analysis to directly examine the
influence of health care reform on ED utilization and 1 repeatedmeasurements approach to examine the indirect influence of
health care reform on ED utilization through the percentage
insured. Thus, 4 types of models were estimated.
A crucial assumption of all models is that the size and
composition of the population in different geographic areas of
the state remained relatively constant over time and, more
important, that there were no shifts in places of residence as a
direct result of health care reform, which would lead to biased
estimates.
In each analysis, we include fixed effects for zip code and
quarter to account for variation between zip codes and across
time that is unrelated to health care reform. Although we fit
fixed-effect models for zip code that incorporate adjustments to
the standard errors using generalized estimating equations,
when these models have a nonindependent working correlation,
the parameter estimates differ from those under the fixed-effect
model with no adjustment for clustering. Therefore, as a
sensitivity analysis we ran the models with an independent as
opposed to an AR(1) or other nonindependence correlation
structure to ensure that the assumed correlation structure did
not sway the results substantially. Main effects for person and
place characteristics (percentage of male patients, percentage in
various racial groups, median age, etc) related to use were
controlled for to account for differences in patient
characteristics across zip codes and across time (see main text).
Our quarterly time-trend dummies account for year effects,
seasonal effects, and their interaction.
Because the outcome is a count, it is reasonable to assume
they are from a Poisson distribution with a mean whose log
depends linearly on the predictors. Because ED visits are nested
within zip codes, we account for the grouped error structure,
using generalized estimating equations with appropriately chosen
working correlation matrices.
MODELS RELATING THE NUMBER OF VISITS TO
BASELINE OR CHANGE IN INSURANCE
We hypothesized that the effect of health care reform during
our 3 periods of study (before, during, and after health care
reform) would differentially affect individuals from areas of the
state that were most likely to be affected by health care reform (ie,
zip codes with higher rates of uninsurance before reform or where
we observed the largest changes in insurance coverage from the
prereform to the postreform period according to actual ED visits)
versus those from areas of the state least likely to be affected by
Volume 64, no. 2 : August 2014
health care reform (ie, areas of the state with high insurance
coverage levels throughout the entire period). We computed the
change in percentage visits by the uninsured from the preperiod
to the postperiod and the average percentage of visits by the
uninsured in the preperiod and for each zip code (which
identifies areas of the state most likely to be affected by health
care reform). We also defined dummy variables for the withinreform and postreform periods. Under both specifications, we
assume an autoregressive correlation structure to account for
possible latent correlation proportional to the longitudinal
closeness of the observations.
We first fit a marginal Poisson regression model in which the
linear predictor or systematic component of the model has the
form:
logðmit Þ ¼ li þ gt þ bT1 x it þ b2 DInsi InHR it þ b3 DInsi PostHR it:
(1)
Under (1) the quantities expðb2 Þ and expðb3 Þ represent the
effects on the number of visits during the within-reform and
postreform periods, respectively, because of a 1-percentage-point
increase in the percentage of insured patients from the baseline
period to the postperiod. These effects represent the extent to
which the increase in the number of visits in the within-reform
and postreform periods was greater in zip codes whose percentage
insured increased the most.
An alternative form of this model uses the preperiod
percentage of insured instead of the change in insurance:
logðmit Þ ¼ li þ gt þ bT1 xit þ b2 PreUninsi InHR it
þ b3 PreUninsi PostHR it;
(2)
wheremit ¼ E½Yit jx it is the expected number of visits, Yit is the
observed number of visits, and x it is a vector of aggregate patient
covariates (case-mix variables) in zip code i during quarter t, li is
the effect of zip-code i, gt is the effect of time period t, PreUninsi
is the percentage uninsured in the pre–health-reform period in
zip code i, andInHR it andPostHR it are dummy indicator
variables for the within-reform and the postreform periods,
respectively.
The key coefficients in model (2) are the regression
coefficients of the interaction terms, b2 and b3 . The quantities
expðb2 Þ and expðb3 Þ are the relative increases in the number of
visits during the within-reform and the postreform periods,
respectively, because of a 1-percentage-point increase in the
percentage of uninsured patients in the preperiod. (The main
effect of percentage insured in the preperiod is absorbed in the
zip code fixed effect, so is not explicitly represented.) The
rationale is that zip codes with low preinsurance rates should
experience the greatest increase in visits because their insured rate
is subject to the most change from health care reform.
Annals of Emergency Medicine 115.e2
Increased Use of Massachusetts Emergency Departments After Health Care Reform
DIFFERENCE-IN-DIFFERENCE ANALYSIS
EXPLOITING PATIENT AGE GROUP
The third difference-in-differences approach does not
involve frequencies of insurance. Instead, it tests whether the
population younger than 65 years was differentially affected
compared with the population older than 65 years in the
during-reform and postreform periods relative to the
prereform period. We hypothesized that health care reform
would preferentially affect the population younger than 65
years because almost all residents aged 65 years or older were
covered by Medicare throughout the study period. In this
analysis, we analyze separate counts of visits for the
populations younger and older than 65 years in a single
model; that is, we can think of the dependent variable
for each zip code quarter as being a bivariate count. Therefore,
in this analysis the trend for the population older than 65
years acts as a control across all periods, enabling the
difference-in-difference effect of the population younger than
65 years in the within-reform and postreform periods
compared with the prereform period to be attributed to health
care reform. The model has the general form
logðmit Þ ¼ li þ gt þ bT1 x it þ b2 AgeU65i þ b3 AgeU65i InHR it
(3)
þ b4 AgeU65i PostHR it;
where is a binary variable that takes the value 1 for the population
younger than 65 years and 0 for the population older than 65
years. Unlike the zip code insurance measures used in models
(1) and (2), varies within each zip code and so its main effect is
estimable and should be accounted for to extract the modification
of its effect because of health care reform.
The effects of interest under (3) are b3 and b4 . Their
exponentials equal the relative increase in the number of visits
during the within-reform and postreform periods among
individuals younger than 65 years old compared with those older
than 65 years in the same zip code. If the hypothesis that health
care reform affects only individuals younger than 65 years is true,
115.e3 Annals of Emergency Medicine
Smulowitz et al
then we would expect both b3 and b4 to be positive. Because of
the nonoptimal means we have to estimate zip code insurance
rate, the model in (3) approach might be viewed as our most
robust difference-in-difference approach.
Because separate repeated observations were made on the
populations younger and older than 65 years, we treated zip code
by patient type (<65 years, >65 years) as the clustering variable.
Surprisingly, we encountered difficulties estimating the model
with generalized estimating equations with an autoregressive
working correlation structure. Therefore, we used an
independent working correlation; this enables consistent
estimation but by not directly accounting for latent serial
correlation may compromise some precision.
MODEL THAT ESTIMATES DIRECT EFFECT OF
HEALTH INSURANCE ON VISITS
The fourth approach evaluates the extent that more visits
occur in the same zip code during quarters of higher insurance
compared with quarters of lower insurance. As for models (1)
to (3), it accounts for differences in the size of zip codes
through the inclusion of zip code dummies (fixed effects) and
common variation in the number of visits across time (eg,
seasonal effects, secular trends) through the inclusion of
month dummies. The systematic component of the model has
the general form:
logðmit Þ ¼ li þ gt þ bT1 xit þ b2 Insit;
(4)
The key predictor in (4) is the quarterly contemporaneous
insurance rate, Insit ; the exponential of its effect, expðb2 Þ , is the
relative increase in the expected number of visits associated with a
1-percentage-point increase in insurance that quarter. Because zip
codes act as their own controls, in (4) b2 tests whether percentage
insured (considered a consequence of health care reform) is
associated with the number of visits. We estimated the model
with generalized estimating equations with an autoregressive
working correlation matrix.
Volume 64, no. 2 : August 2014