admin
Fri, 02/09/2024 - 19:51
Edited Text
Consequences of Hospital Closures on Community Welfare
An Honors Thesis
by
Eli Kochersperger
California, Pennsylvania
2019
Consequences of Hospital Closures on Community Welfare
Abstract
This study investigates the economic position the Monongahela Valley hospital network assumed with the onset of deindustrialization and the extent of welfare loss
endured by these communities following their eventual closures. We survey similar analyses carried out in other rural communities, as well as those concerning the
current opioid crisis to develop our hypothesis that such closures have had acutely
detrimental impacts on average household wages. Employing data from the Centers
for Disease Control’s American Community Survey, our analysis integrates empirical
quantitative methods (in particular, multiple linear regression) with considerations of
the region’s historic political economy. The intent of carrying out such an analysis is
to relate broader economic trends in deindustrialized urban areas and rural hospital
closures to changes of overall wellbeing within these affected communities.
Keywords
Hospital Closure Monongahela Valley Welfare Medicaid Deindustrialization
Consequences of Hospital Closures on Community Welfare
Introduction
At the height of the Pennsylvania steel industry, the Monongahela River Valley contained along its shores one of the highest densities of industrial manufacturing facilities in the world, and with it a thriving population of upwardly-mobile, middle
class laborers. As these communities swelled over the immediate post-war boom, they
did so in tandem with the ascendance of a political order that viewed public capital
investment as the de rigueur policy initiative. As such, a network of hospital and
health care facilities materialized fit to be the envy of any then-industrialized society.
Once the role of heavy industry and manufacturing in the American economy waned
however, so too did political sentiments towards the continued expansion, or even
maintenance of the country’s now massive health care infrastructure.
Deindustrialization saw to the cessation of the near-entirety of the Monongahela
Valley’s manufacturing output, but as mills closed and residents were forced to adapt
to a new labor market, many communities viewed these hospitals as lifelines– both
literally and economically. The pivot towards a predominantly services-based economy included a newfound market position for hospitals as a now-major employer.
Unfortunately, the continued ageing and loss of population, compounded with stagnant income growth has pushed these lifeline institutions closer to insolvency with
each passing year. Some have resorted to downsizing or mergers with competitors to
continue operations, while others have opted to simply shutter, further populating
these communities with vacant monuments to halcyon days.
Beginning with initial investigations carried out during 1980’s, a considerable
body of research has accrued in analyzing the impacts that hospital closures have on
the economic well being of surrounding communities. While the outcomes of these
investigations have been anything but conclusive in describing the degree of influence,
professional opinions generally agree that closures do have net-negative impacts on
residents in nearby regions. Because of the pronounced uptick in closures in the past
1
Consequences of Hospital Closures on Community Welfare
decade throughout the Monongahela Valley, as well as the unique role of hospitals
within and overall economic precarity of the region, the phenomenon merits closer
inspection.
Simultaneous to these hospital closures, communities in the region have been
forced to confront their own public health crisis in the form of the opioid overdose
crisis. Due in large part to this epidemic alone, rural areas across Appalachia have
seen mortality rates rise and average lifespans plummet. Because of the enormous
costs borne by society from accidental deaths of this scale, we must consider the
implications that hospital closures and the loss of access to care have on overdose
mortality rates if we expect to compose an at all accurate model for closure-associated
economic impacts.
For the primary subject of our empirical analysis here, our research question is
thus: To what extent have hospital closures throughout the Monongahela Valley
impacted the economic well being of their host communities? Additionally, we want
to know whether there exists any observable relationship between these closures and
the frequency of accidental deaths from drug poisoning. Based on an analysis of the
relevant literature and a firm grounding in economic theory, our working hypothesis
is that household incomes in those regions which experienced hospital closures should
have measurably diminished incomes in the years of closure relative to regions which
did not. By employing a simple multiple linear regression we intend to test this
hypothesis and establish the degree to which these closures affected communities
while controlling for possible biasing influences.
Literature Review
For the quantitative component of our analysis here, our principal interest is to establish the degree to which recent hospital closures throughout that Monongahela
2
Consequences of Hospital Closures on Community Welfare
Valley region have impacted the economies of those communities they belonged to.
In consulting relevant economic literature on the subject we can conclude likely outcomes from similar analyses, as well as several necessary qualifications dictated by our
particular subjects that highlight both the complexity and novelty of this research
endeavor. Following in-line with recent trends and observations, we think it likely
that these riverfront communities have faced significant, measurable losses to overall welfare as a consequence of hospital downsizing or closure. Moreover, because of
the unique demographic makeup of these communities’ labor force and the prevailing
economic precarity therein, such welfare losses may be even more pronounced than
in those urban or rural settings previously studied.
Since at least Christianson and Faulkner (1981), modeling has shown significant
income contributions to rural communities by their hospitals. Beyond the direct
and indirect sources of income collected from hospital expenditures for local support
services and wages paid to staff, these communities also enjoy additional benefits
(most notably in the form of tax revenue) by attracting medical professionals with
incomes significantly above median values. As such, it stands to reason that any loss
of such hospital facilities would produce income losses similar to these amounts.
With respect to establishing quantitative estimates for the economic impacts of
rural hospital closures, Holmes et al. (2006) provide a comprehensive survey of efforts
made by others at accomplishing just this. Beyond the generally intuitive negative
consequences expected of such closures, Hart, Pirani, and Rosenblatt (1991) surveyed
the mayors of 130 rural towns that experienced hospital closures and queried them
of any perceived changes to overall welfare. Of those asked, more than 90 percent
believed that hospital closures had substantial negative impacts on their communities’
economic well being. Despite the adamancy expressed by those surveyed however,
initial empirical research suggested that the impact of rural hospital closures on both
short- and long-term community economic growth were trivial (Pearson and Tajalli,
3
Consequences of Hospital Closures on Community Welfare
2003; Stensland et al., 2002). The primary finding was that rural hospitals were often
so small when compared to those found in urban areas that they had miniscule- if any
influence on unemployment and income when eventually closed. Similar research has
more or less agreed with these results by failing to find any statistically-significant
relationship between hospital closures and community growth trends (Probst et al.,
1999).
It is entirely possible however, as Holmes et al. (2006) suggest, that the failure
of these studies to locate any significant, negative relationship between community
economic growth and hospital closures is more to do with methodology than their describing actual phenomenon. Most significantly, these previous analyses were carried
out using input/output methods that lacked precise datasets and failed to account
tertiary, quality-of-life contributions to growth provided by these hospitals. When the
impacts of these closures are modeled using multiple regression methods instead, we
do see considerable negative outcomes from closures. When examining data for rural
hospital closures carried out during the 1990’s in this manner, researchers observed
an average loss of 4 percent of community per-capita income, as well as an average
increase in unemployment of 1.6 percent (Holmes et al., 2006, p. 478). Importantly,
these closures do indeed have significant detrimental impacts on long-term growth,
but are largely limited to those communities which lost their sole hospital and for
which no alternative is within close proximity.
With these findings in mind, it is important to consider a few imperative distinctions between our research subjects and those rural communities studied in previous
surveys. Indeed, it is largely these very geographic and demographic differences that
warrant our analysis here. For one, most communities within the Monongahela Valley
region that have faced hospital closures, mergers, or downsizing in recent decades still
have high enough population densities to be formally considered “urban.” With the
exception of Brownsville, all school districts that fall along the Monongahela river-
4
Consequences of Hospital Closures on Community Welfare
front throughout the Allegheny, Washington, Westmoreland, and Fayette counties
have densities exceeding the state average, thus meeting the formal designation of urban areas (U.S. Census Bureau, 2010). In general, it would be unwise to extrapolate
trends observed in rural areas to describe phenomena in urban regions, but as can be
seen with some additional context, the decay experienced by these communities over
the past three decades may in fact make them exemplars of such behavior.
To be sure, at their peak, manufacturing hubs along the Monongahela River enjoyed levels of growth and prosperity quite unlike any rural areas. However, over the
most rapid period of deindustrialization in the United States, approximately 1975 to
1990, Southwestern Pennsylvania experienced some of the greatest changes to longterm employment patterns nationwide. In these few years the region lost more than
150,000 manufacturing jobs while seeing only limited gains in the service industry to
compensate (Yamatani, 1986). At its peak level of employment in the 1960’s, the primary metals manufacturing industry constituted more than 15 percent of the entire
regional workforce before diminishing to less than 5 percent by the 1990’s (Coleman, 1986). Official unemployment rates averaged at 10 percent (Biegel et al., 1989)
throughout the first-half of the 1980’s, while estimates which included discouraged
and underemployed workers placed this figure at 16 percent over the same period
(Troan, 1985). These precipitous drops in manufacturing employment did eventually
stabilize, but comparatively-high unemployment and population loss have dogged
virtually all riverfront communities within the Monongahela Valley in the decades
since (Tony, 2016). Furthermore, between population loss and an increasing trend in
Medicaid enrollments throughout the region, hospitals have been further squeezed by
diminishing revenue streams.
A major consequence of this contraction throughout the Monongahela Valley has
been a transition in demographic and labor force makeup towards patterns more typical of rural communities. In particular, as younger, able-bodied workers left the
5
Consequences of Hospital Closures on Community Welfare
region with the collapse of the steel industry, those who remained saw median community ages grow and a labor market increasingly constituted by the service sector.
Communities such as Duquesne, which had enjoyed hourly wages exceeding statewide
averages and nearly 30 percent of employment in manufacturing as late as 1980 (Biegel
et al., 1989, p. 401), now boast median household incomes only 41 percent of the
state level (U.S. Census Bureau, 2017). As incomes diminished and community age
increased, a general trend towards rural demography has occurred with respect to the
nature of community health. Diabetes, obesity, substance abuse, and heart disease- all
conditions with well-established heightened prevalence throughout rural areas- have
enjoyed significant increases in ubiquity throughout the Monongahela Valley region
since deindustrialization (Centers for Disease Control [CDC], 2013; National Center
for Health Statistics, 2016).
Changes in employment patterns in the Monongahela Valley suggest that hospital closures may produce even more pronounced consequences than in typical rural
areas. As with other rust belt communities, the transition towards a service-based
labor market has seen a significant increase in health care employment (Hobor, 2013;
Olney and Pacitti, 2017). As high-paying manufacturing jobs left the area, hospitals
represented one of the most obvious avenues for securing steady, full-time employment. Furthermore, unlike their counterparts in rural areas, hospitals throughout
the Monongahela Valley were constructed during periods of peak-population, often
exceeding 100 beds in size and offering a wide-array of specialized services. Others
have established the tendency of hospital-induced income multipliers to scale with
the number of beds (Cordes et al., 1999), so as these hospitals have been closed or
down-sized, we would therefore expect these negative shocks to economic growth to
exceed those in comparative rural areas.
Additional complicating factors include the relative difficulty of these larger hospitals to control costs, as well as their more common tendency to down-size rather
6
Consequences of Hospital Closures on Community Welfare
than cease operations entirely when facing funding issues. While the hospital networks serving rural America are plagued by an increasingly dire solvency crisis (Hsia,
Kellerman, and Shen, 2011), many of these institutions faced similar difficulties in
prior decades. To confront a previous epidemic of rural hospital closures, Congress
included language in the Balanced Budget Act of 1997 that allowed rural hospitals to
restructure as critical access hospitals (CAHs). In doing so, these hospitals were able
to considerably limit the breadth of services legally required of them and consequently
reduce their operating costs. However, the strict bed and proximity requirements for
these hospitals to be granted permission to restructure as CAHs are often such that
the conversion of larger facilities like those found throughout the Monongahela Valley
would be infeasible. Indeed, despite a prolonged and universal tendency throughout
the region for hospitals to operate at or below the margin (Pennsylvania Health Care
Cost Containment Council [PHC4], 2017a), this facility size peculiarity may explain
why there has never been such a hospital conversion throughout the region (Flex
Monitoring Team, 2004; 2018). Instead, many of these facilities have had to adopt
alternative cost-saving methods to remain solvent; most commonly, limiting services
provided with special focus paid in reorienting towards outpatient care delivery, or
merging with regional hospital networks to increase access to capital (Holmes, 2015;
PHC4, 2017b).
In carrying out our analysis here, it is not enough to construct a model that
examines total facility closure alone. Previous examinations of rural health care
networks throughout Appalachia have shown diminishing utilization of and accessibility to obstetric, dental, substance abuse, and mental care services (Stensland et
al., 2002). Within Southwestern Pennsylvania, Allegheny, Washington, Westmoreland, and Fayette counties are all considered to be underserved by dental and mental health services (Health Resources and Services Administration [HRSA], 2017). A
summary inspection of recent newspaper articles and data for medical facility closures
7
Consequences of Hospital Closures on Community Welfare
throughout the region confirm the tendency for these sorts of services in particular to
face shutdown (Snowbeck, 2000; Gough, 2017; PHC4, 2017b; Goldstein, 2018). Because the loss of access to these services has significant direct impacts on community
welfare (namely, from lost high-wage specialists and overall health from diminished
utilization), it behooves us to include these factors into our analysis.
A final consideration that may be prove to be significant and provides additional
weight to the novelty of this analysis is the phenomenon of decreasing life-expectancies
throughout the region. As detailed by the CDC, the increased frequency of substance
abuse-related deaths has had substantial influence in lowering life-expectancies nationwide, but is particularly pronounced in rural communities (2017; 2018). Pennsylvania ranked third nationally for highest frequency of drug overdose deaths, with a
rate of 44.1 per 100,000 deaths (CDC, 2018, p. 3). Within Southwestern Pennsylvania, all counties saw increases in the number of overdoses from 2013 to 2016 between
67 and 138 percent (United States Department of Agriculture, 2018, p. 5). Since this
trend has been largely observed in only the past decade and much of the literature on
hospital closures predate that, it is reasonable to suspect that the negative impacts
on community welfare observed by earlier empirical studies may be understated. The
gradual demographic shifts experienced under deindustrialized may be exacerbated
as alcoholism and opioid abuse accelerate the loss of young, able-bodied individuals,
to say nothing of the exorbitant direct costs of combating this epidemic. Researchers
have consistently demonstrated the relationship between hospital closures and increased mortality rates among those treated for drug overdose (Bazzoli et al., 2012;
Liu et al., 2014), we would expect then that within the context of an overdose epidemic that the closure of acute care facilities represent an added factor to welfare
loss over previous estimates. Additionally, while there remains considerable debate
about the extent of influence that economic precarity has on substance abuse rates
throughout rural communities (Ruhm, 2018), if such a relationship does exist we can
8
Consequences of Hospital Closures on Community Welfare
surmise the existence of feedback processes whereby hospital closures induce increased
poverty, which consequently spurs substance abuse, further diminishing community
growth, and so on and so forth. The inclusion of this phenomenon may prove difficult
within our model, it seems imperative that attempt to do so.
Data Description
All data for our empirical analysis here is sourced from the United States Census
Bureau’s American Community Survey (ACS). These extensive surveys have been
conducted nationwide since 2005 through a combination of written, phone, and inperson interviews. Because of the survey’s considerable breadth of data, number of
participants (approximately 3.5 million households each year [U.S. Census Bureau,
2019]), public availability, and the Census Bureau’s diligence in ensuring proper sampling methods, it is an ideal source from which we can draw our data. In particular,
we will look at the Household Record data set for the years 2005 to 2017, which
includes vital information on household sizes, location, and incomes.
One of the more difficult tasks in sourcing appropriate data for our analysis is
finding those that are sufficiently precise geographically. Due to federal guidelines
regarding data anonymization, the breadth of microdata of individual households
or firms for public use must be sufficiently vague to prevent identification. This
places considerable limits on the amount of data useful for our project here: since
household data points are often recorded with geographic indicators only as precise
as county of residency and not, say, postal code or municipality, there is a very real
chance that significant trends will be obscured or entirely unobservable. Consider,
for instance, poverty estimates for the Borough of Braddock in Allegheny County,
and Allegheny County itself. Survey results from the ACS put percentages of the
population living below the federal poverty level at 31.9 and 12.5 (U.S. Census Bureau,
9
Consequences of Hospital Closures on Community Welfare
2017) respectively, suggesting that the localization of our analysis to the riverfront
Monongahela Valley communities is imperative should we expect any meaningful
results. Thankfully, while the ACS lacks municipality data, it does feature a slightly
less precise Public Use Microdata Area (PUMA) code, which allows us at a minimum
to separate possible countervailing measurements from peripheral urban areas (in
particular, the cities of Washington and Pittsburgh).
Using five PUMAs from Allegheny, Fayette, Washington, and Westmoreland counties that include nearly all riverfront communities along the Monongahela, we are able
to cull from the ACS Household Record data set more than 23,000 useful household
data points for the years 2005 to 2017. By cross referencing public records (PHC4,
2017), we can identify three hospital closures that occurred within this area over this
time frame: Tara Hospital at Brownsville, in 2006; Brownsville Tri County Hospital,
in 2007; and UPMC Braddock, in 2010. In addition to these closures, we can also
identify eight hospital and medical clinic closures or restructurings, a full description
of such can be seen in Table 1.
Table 1: Hospital Closures and Mergers, 2005-2017
Closure/Merger
Year
Name
Merged
Merged
Merged
Merged
Closed
Closed
Closed
Merged
Merged
2005
2005
2005
2005
2006
2009
2010
2013
2015
Greene County Memorial Hospital
Zitelli South Ambulatory Surgical Center
Brownsville General Hospital, Inc
SemperCare Hospital of McKeesport, Inc
Tara Hospital at Brownsville
Brownsville Tri County Hospital
UPMC Braddock
Jefferson Regional Medical Center
Southwest Regional Medical Center
PUMA
4002
1807
4002
1805
4002
4002
1805
1807
4002
From the ACS we can identify our dependent variable, self-reported family income
in terms of 2017 dollars, as well as important independent control variables, number
of wage earners in family and residency PUMA. While the ACS does supply us with a
10
Consequences of Hospital Closures on Community Welfare
wide array of information on the makeup of households, explicit information for possible determinants of income are limited. For this reason, we must limit our analysis
to families and disregard non-traditional household types. By referencing the survey
years and PUMA code for each data point we can produce a series of explanatory
variables representing hospital closures and mergers. Because of its sheer size and
the rigorous sampling methodology employed by the Census Bureau in collecting this
data, we can rest assured that results are unlikely to be skewed or biased by outliers.
Indeed, fewer than one percent of surveyed families within our sample earned more
than $400,000 over this twelve year period.
Model Description
The central intent of our empirical analysis is to establish whether or not, and to
what degree hospital closures impact the economic well being of those communities
to which they belong. Our working hypothesis is that there should exist some measurable negative correlation between average family incomes and residency within a
community during or around the time of hospital closure. The method we will employ
here to test this hypothesis is a simple ordinary least squares (OLS) linear regression
that will approximate the average dollar amount that family incomes decrease when
residing near a hospital closure, while controlling for the number of household wage
earners and region. We will conduct two such analyses: one will examine the impacts
of hospital closures alone, while the other will include variables for both hospital
closures and mergers. We define our linear models as follows:
IN COM E = β0 + β1 W + β2 Ct−1 + β3 Ct + β4 Ct+1 + β5 Ct+2 + β6 P1 + . . . + β10 P5 , (1)
11
Consequences of Hospital Closures on Community Welfare
and
IN COM E = β0 + . . . + β10 P5 + β2 Mt−1 + β3 Mt + β4 Mt+1 + β5 Mt+2 ;
(2)
where the t subscripts denote year, W an integer number of wage earners in the
household, and C, P , and M dummy variables representing residence in closure area,
a specific PUMA, and merger area respectively. C and M are equal to 0 unless the
family described by the data point resides in a closure or merger PUMA; defining
additional lagging variables allows us to approximate any lingering impacts from
closures and mergers, while leading variables limit spurious outcomes in the event of
an economic downturn in the preceding year precipitating eventual closure.
One considerable complicating factor with constructing our model in this manner
is its inability to account for wage growth- a significant issue since it is effectively
our hypothesis that hospital closures should be associated with diminished economic
growth. Any negative coefficients calculated from our OLS procedure for the closure
dummy variables would necessarily be inflated as they would include the explanatory
power of both closures and relative wage loss to the period mean.
The most obvious remedy would be to simply toss our model as it is and instead
base one on percent changes in family incomes between years. Unfortunately, to adopt
this alternative method would require the compression of our data set from tens of
thousands of data points to only twelve- one for each year- thus significantly reducing
the likelihood of producing statistically-significant results. Instead, we opt to retain
our model form as previously stated but do so with the following qualification: if an
OLS procedure is able to produce significant negative beta estimates for our closure
variables; and, if these coefficient values can be shown to be significantly larger than
the difference between overall mean family income and mean family income of the
years in which closures occurred by way of a simple one-sided t-test, then we can
12
Consequences of Hospital Closures on Community Welfare
conclude that hospital closures did indeed have significant negative impacts on local
family incomes. While this means that we cannot ascribe meaning to the size of our
estimates, it does allow us to answer the central question of our hypothesis.
Supplemental Analysis
As a supplement to our primary investigation into whether hospital closures have had
any measurable impact on the economic wellbeing of their surrounding communities,
we will carry out an additional analysis to establish whether any association exists
between hospital closures and drug overdose deaths. To do so, we use annual countylevel estimates for drug poisoning overdose mortality rates from the National Center
for Health Statistics’ National Vital Statistics Center dataset (2017) for the years
2005 to 2017. Controlling for year and region, we produce the following model to
estimate the impact of closures on mortality rates:
Overdose = β0 + β1 C + β2 RF ayette + β3 RW ashington + β4 T ;
(3)
where our dependent variable, Overdose, is the mortality rate of a given county and
year, C and R variables representing county characteristics as closure area and specific county respectively, and T , the given year of datapoint. C is equal to 0 unless
the county described experienced a hospital closure within its borders at that given
year; RF ayette and RW estmoreland are equal to 0 unless datapoint describes that specific
county; and T indicates the number of years from model initiation in 2005 that observation was recorded. The intent of constructing our model in this way is to control for
latent differences that exist between counties as well as larger, supraregional trends.
Unfortunately, because no public database of mortality rates exist for our examined area over these years with geographic codes more precise than county-levels, we
must broaden the scope of our analysis to include the entire Fayette, Westmoreland,
13
Consequences of Hospital Closures on Community Welfare
and Washington counties (out of an interest of avoiding skewed results, we exclude
Allegheny county and the likely-biasing effects of the city of Pittsburgh). This lack
of precision in our data casts serious doubt on the likelihood of producing either statistically significant or meaningful outcomes, but we will proceed with doing so out
of a lack of alternative avenues for investigation. Even in the event that no such
association is observed, such results can still inform our eventual conclusions.
Empirical Analysis
Initial Results & Conclusions
After running our primary regressions for both the closure and merger models, we
produce the following results seen in table 2. In general, these results would confirm
our expectations that hospital closures are correlated to loss of family incomes. As we
can see from our closure-only model, over the twelve year period observed, residence
within a closure PUMA resulted in an average loss of $5,469 in family income during
the that same year of closure. Recalling that this coefficient is likely inflated to at least
some degree when we assume general income growth over this timeframe, we should
instead confirm that this value is significantly larger than the mean differences for
closure years and all years. Performing a one-sided t-test on our current year closure
coefficient against these mean differences, -2,682.666, yields a score comfortably below
our 5 percent critical value. Therefore, we can surmise that residency within a closure
PUMA had some demonstrable negative impact on aggregate family incomes in the
year of closure.
Our merger model largely confirms these conclusions by returning a remarkably
similar coefficient estimate for the current year closure variable. Regression results
from this model suggest that lingering effects from closures in the following year
are significantly negative as well. The impacts of hospital mergers however, appear
14
Consequences of Hospital Closures on Community Welfare
Table 2: Regression Results
Dependent variable:
Family Income
(1)
Number of Wage Earners in Family
Residence in Closure Region, Preceding Year
Residence in Closure Region, Current Year
Residence in Closure Region, +1 Year
Residence in Closure Region, +2 Year
PUMA Residence: 2002
PUMA Residence: 1805
PUMA Residence: 1807
PUMA Residence: 3900
(2)
25, 680.410∗∗∗
(434.296)
−2, 990.306
(2, 067.796)
−5, 469.000∗∗∗
(2, 050.874)
−3, 902.546∗
(2, 042.587)
−1, 251.724
(2, 084.557)
3, 682.580∗∗∗
(1, 371.196)
−4, 390.602∗∗∗
(1, 316.034)
34, 786.760∗∗∗
(1, 481.220)
−10, 520.330∗∗∗
(1, 376.106)
25, 680.960∗∗∗
(434.308)
−2, 471.620
(2, 128.765)
−5, 360.306∗∗
(2, 135.402)
−5, 187.058∗∗
(2, 120.894)
−2, 038.196
(2, 214.248)
2, 966.086∗
(1, 619.659)
−4, 801.667∗∗∗
(1, 396.780)
34, 546.090∗∗∗
(1, 499.494)
−11, 236.740∗∗∗
(1, 623.707)
−1, 718.101
(2, 521.850)
−2, 393.231
(1, 766.236)
−1, 968.265
(1, 752.287)
2, 118.504
(1, 742.499)
41, 502.550∗∗∗
(1, 497.282)
Residence in Merger Region, Preceding Year
Residence in Merger Region, Current Year
Residence in Merger Region, +1 Year
Residence in Merger Region, +2 Year
40, 786.860∗∗∗
(1, 225.520)
Constant
Observations
R2
Adjusted R2
23,517
0.182
0.182
∗
Note:
15
p<0.1;
23,517
0.183
0.182
∗∗
p<0.05;
∗∗∗
p<0.01
Consequences of Hospital Closures on Community Welfare
insignificant; therefore, there exists conclusive correlation between hospital mergers
and family incomes.
While these results are promising overall, the low Pearson coefficients of only
0.182 expose our model’s shortcomings. Without having some additional means to
account for household educational attainment, sector of employment, or years of
work experience, any estimates such as ours are going to be necessarily limited in
their explanatory power. In spite of this, we can still take solace in the model’s ability to demonstrate a statistically-well-founded negative correlation between hospital
closures and family incomes, even if the absolute degree of these impacts remains
inexact.
Further inspection of the data reveals additional corroboration for our regression
results. Plotting mean family incomes against years for all PUMAs, closure PUMAs,
and non-closure PUMAs (as seen in figure 1 ) demonstrates negative trends within
closure regions around the closure years of 2006, 2009, and 2010. While it is conceivable that these trends reflect exogenous shocks to the regions not accounted for
within our model, the relative constancy of income growth among other observed
regions would appear to affirm our conclusions.
In carrying out the regression for our supplemental analysis, we observe substantially less satisfactory outcomes. As can be seen in table 3, at a 5 percent significance
level, there is no observable correlation between hospital closures and overdose rates.
Our concerns about limited data size and lack of geographic specificity would appear
to have been well founded, but without those data we will have to simply concede
that any such relationship is presently indeterminate.
16
Consequences of Hospital Closures on Community Welfare
$95,000
$90,000
PUMAs with
No Closure
$85,000
All PUMAs
$80,000
PUMAs with
Closure
$75,000
$70,000
$65,000
2006
2008
2010
2012
2014
2016
Figure 1: Annual Mean Family Incomes by PUMA Group, 2005-2017
Overall, we can conclude the following based on the results of our multiple regression analyses. There is ample empirical evidence to suggest that in the years 2006,
2009, and 2010, hospital closures throughout the Monongahela valley produced observable declines in family incomes of those residing within the same regions as where
closures occured. While the precise estimates returned are likely different from the
actual aggregate impacts, we can demonstrate that the nature of this relationship is
significantly negative in nature, confirming our initial hypothesis. Most of the observable effects from hospital closures occur within the current year of closure, although
the results from when our model is amended to include hospital mergers suggest that
these effects persist into the following year. While hospital closures appear to have
significant impacts on the well being those who live around them, the effects of hospital mergers remain inconclusive. Lastly, while overdose deaths from drug poisonings
increased consistently over this time period, there is no demonstrable relationship
between mortality rates and hospital closures.
17
Consequences of Hospital Closures on Community Welfare
Table 3: Overdose Regression Results
Dependent variable:
Residence in Closure County
Residence in Fayette County
Residence in Washington County
Year
Constant
Observations
R2
Adjusted R2
Overdose Deaths
Logged Overdose Deaths
per 100,000 People
per 100,000 People
(1)
(2)
−1.088∗
(0.607)
0.681∗∗
(0.328)
−2.000∗∗∗
(0.312)
0.975∗∗∗
(0.038)
9.993∗∗∗
(0.330)
−0.061∗
(0.030)
0.039∗∗
(0.016)
−0.131∗∗∗
(0.015)
0.062∗∗∗
(0.002)
2.367∗∗∗
(0.016)
36
0.963
0.958
36
0.977
0.974
∗
Note:
p<0.1;
∗∗
p<0.05;
∗∗∗
p<0.01
Discussion of Results
Based on the outcomes of our analysis we can say that there is evidence to support
the claim that hospital closures throughout the Monongahela Valley between 2005
and 2017 produced significant, measurable negative impacts on household incomes.
This confirms our hypothesis and suggests that future policy initiatives aimed at
bettering health standards throughout the region must make ready access to hospitaldelivered care an imperative; to that end, such policy must necessarily confront the
solvency crisis these institutions face and its primary instigator: diminishing Medicaid
reimbursement rates.
The results of our primary analysis are largely satisfactory and provide a reasonable degree of evidence to support initial predictions regarding the impacts of hospital
18
Consequences of Hospital Closures on Community Welfare
closures, but the overall vulnerability of our model to exogenous system shocks will
require additional efforts before verification. It may be tempting to reproduce our
analysis here but with the inclusion of some additional control dataset, as many have
attempted previously, but as Holmes (2006) and others point out, there may be no
obvious analog to our examined region. Even if such a parallel community did exist
with all of the same economic peculiarities as the Monongahela Valley, there is no
assurances that that region would not bias our results with its own exogenous shock.
Instead, it may be more feasible to examine the determinants of our results piecemeal.
Perhaps the most straightforward approach would be to break down and examine the
individual components of hospitals’ influences on community well being- namely, the
utilization rates of hospital-provided care and incomes of those employed within the
health care sector.
One method of corroborating our results may be to simply examine changes in
hospital admissions in the surrounding area following closures. While our analysis
here was done so as to account for both the direct and indirect impacts of hospital
closures, diminishing admissions are likely to be the primary factor in spurring closures in the first place. Assuming that there have been no sudden, external changes
to demand of care, hospital closures should induce influxes of admissions to neighboring facilities. If, however, no such increase can be observed (and assuming any
such analysis also controls for changes in care utilization), then we can surmise that
direct health impacts to the closure community would be minimal. Alternatively,
a near-identical analysis could examine household incomes and employment rates of
individuals employed within the health care industry. If closures do indeed provoke
economic declines in these communities, the most obvious and immediately impacted
groups are going to be those whose livelihoods are dependent on them.
Other plausible analytical methods that may produce more convincing outcomes
would be to expand the breadth of examined communities to include other analogous
19
Consequences of Hospital Closures on Community Welfare
deindustrialized communities from across the country; or, reproducing our analysis
here, but with data sourced from conducting surveys ourselves so as to increase geographic precision of data points. The former method has the distinct advantage of
supplying us with a dataset considerably more impervious to localized, exogenous
shocks; and, as such a dataset would no doubt be entire orders or magnitude larger,
it may be possible control and compare closure and non-closure outcomes more comprehensively. In the latter case, increased geographic precision would allow us to
examine the impacts of closures based on households’ proximity to hospitals, rather
than residency alone. This would almost certainly present us with a more accurate
depiction of closure outcomes and would effectively eliminate biasing impacts from
nearby cities.
A final consideration that may prove to be more imperative for any future analysis
is of the significant changes to Pennsylvania’s Medicaid reimbursement policy in rural
areas. The Rural Health Model is a pilot program currently being carried out under
the auspices of the Centers for Medicare and Medicaid Services and Pennsylvania
Department of Health with the aim of introducing an Accountable Care Organization (ACO) model to rural Medicaid management. This means, among many things,
the transition away from more typical fee-for-service revenue model towards a multipayer system that stresses year-to-year revenue stability and stakeholder input in
establishing services provided (Murphy et al., 2018). In doing so, state and federal
agencies can ensure the long term success of majority-Medicaid-patient hospitals and
avoid unnecessary closures. As this program is only in its pilot stage of five participating hospitals however, it is likely that some additional closures will occur before
its universal application. In such a case, data collected from participating hospitals
may prove invaluable in establishing the efficacy of current and future ACO policy
initiatives.
The transition towards the Pennsylvania Rural Health Model also highlights the
20
Consequences of Hospital Closures on Community Welfare
subject of our secondary analysis: rural accessibility to substance abuse care services.
ACO programs initiated in other states have placed considerable emphasis on not
only establishing financial stability for rural hospitals, but in expanding the breadth
of care provided as well. Indeed, during the program’s initial rollout administrators for the Pennsylvania program noted the demand for substance abuse clinics and
therapy services as a primary motivator for pursuing this effort. Enabling greater
community planning could help address the deepening opioid crisis, both directly
by providing greater access to desperately-needed counseling services, but also indirectly by providing at least some financial means of offsetting the economic costs of
accidental deaths by drug poisonings.
While we were unable to confirm any categorical association between hospital
closures and opioid overdose deaths, we do know from previous studies (Bazzoli et
al., 2012; Liu et al., 2014) that closures elsewhere have spurred higher mortality rates
among those treated for drug poisoning. CDC estimates put the cost of accidental
drug deaths at approximately 26 percent (Florence et al., 2015) of economic costs
related to all fatal injuries, accounting for a greater share than even transportationrelated deaths. In 2014, residents of Pennsylvania were estimated to have shouldered
more than $10 billion (CDC, 2015) in medical costs and work-loss related to fatal
injuries. This suggests that even a one percent increase in overdose deaths could wreak
economic havoc on communities as small as those that make up the Monongahela
Valley. Therefore, any effort at ensuring or expanding access to emergency and drug
therapy services- as the Rural Health Model intends to do- could have considerable
impact on overall community economic well being. Naturally, continued examination
of outcomes will be necessary to establish the program’s overall effectiveness, but the
Rural Health Model represents an encouraging policy solution to counteracting the
economic and health impacts of hospital closures.
21
An Honors Thesis
by
Eli Kochersperger
California, Pennsylvania
2019
Consequences of Hospital Closures on Community Welfare
Abstract
This study investigates the economic position the Monongahela Valley hospital network assumed with the onset of deindustrialization and the extent of welfare loss
endured by these communities following their eventual closures. We survey similar analyses carried out in other rural communities, as well as those concerning the
current opioid crisis to develop our hypothesis that such closures have had acutely
detrimental impacts on average household wages. Employing data from the Centers
for Disease Control’s American Community Survey, our analysis integrates empirical
quantitative methods (in particular, multiple linear regression) with considerations of
the region’s historic political economy. The intent of carrying out such an analysis is
to relate broader economic trends in deindustrialized urban areas and rural hospital
closures to changes of overall wellbeing within these affected communities.
Keywords
Hospital Closure Monongahela Valley Welfare Medicaid Deindustrialization
Consequences of Hospital Closures on Community Welfare
Introduction
At the height of the Pennsylvania steel industry, the Monongahela River Valley contained along its shores one of the highest densities of industrial manufacturing facilities in the world, and with it a thriving population of upwardly-mobile, middle
class laborers. As these communities swelled over the immediate post-war boom, they
did so in tandem with the ascendance of a political order that viewed public capital
investment as the de rigueur policy initiative. As such, a network of hospital and
health care facilities materialized fit to be the envy of any then-industrialized society.
Once the role of heavy industry and manufacturing in the American economy waned
however, so too did political sentiments towards the continued expansion, or even
maintenance of the country’s now massive health care infrastructure.
Deindustrialization saw to the cessation of the near-entirety of the Monongahela
Valley’s manufacturing output, but as mills closed and residents were forced to adapt
to a new labor market, many communities viewed these hospitals as lifelines– both
literally and economically. The pivot towards a predominantly services-based economy included a newfound market position for hospitals as a now-major employer.
Unfortunately, the continued ageing and loss of population, compounded with stagnant income growth has pushed these lifeline institutions closer to insolvency with
each passing year. Some have resorted to downsizing or mergers with competitors to
continue operations, while others have opted to simply shutter, further populating
these communities with vacant monuments to halcyon days.
Beginning with initial investigations carried out during 1980’s, a considerable
body of research has accrued in analyzing the impacts that hospital closures have on
the economic well being of surrounding communities. While the outcomes of these
investigations have been anything but conclusive in describing the degree of influence,
professional opinions generally agree that closures do have net-negative impacts on
residents in nearby regions. Because of the pronounced uptick in closures in the past
1
Consequences of Hospital Closures on Community Welfare
decade throughout the Monongahela Valley, as well as the unique role of hospitals
within and overall economic precarity of the region, the phenomenon merits closer
inspection.
Simultaneous to these hospital closures, communities in the region have been
forced to confront their own public health crisis in the form of the opioid overdose
crisis. Due in large part to this epidemic alone, rural areas across Appalachia have
seen mortality rates rise and average lifespans plummet. Because of the enormous
costs borne by society from accidental deaths of this scale, we must consider the
implications that hospital closures and the loss of access to care have on overdose
mortality rates if we expect to compose an at all accurate model for closure-associated
economic impacts.
For the primary subject of our empirical analysis here, our research question is
thus: To what extent have hospital closures throughout the Monongahela Valley
impacted the economic well being of their host communities? Additionally, we want
to know whether there exists any observable relationship between these closures and
the frequency of accidental deaths from drug poisoning. Based on an analysis of the
relevant literature and a firm grounding in economic theory, our working hypothesis
is that household incomes in those regions which experienced hospital closures should
have measurably diminished incomes in the years of closure relative to regions which
did not. By employing a simple multiple linear regression we intend to test this
hypothesis and establish the degree to which these closures affected communities
while controlling for possible biasing influences.
Literature Review
For the quantitative component of our analysis here, our principal interest is to establish the degree to which recent hospital closures throughout that Monongahela
2
Consequences of Hospital Closures on Community Welfare
Valley region have impacted the economies of those communities they belonged to.
In consulting relevant economic literature on the subject we can conclude likely outcomes from similar analyses, as well as several necessary qualifications dictated by our
particular subjects that highlight both the complexity and novelty of this research
endeavor. Following in-line with recent trends and observations, we think it likely
that these riverfront communities have faced significant, measurable losses to overall welfare as a consequence of hospital downsizing or closure. Moreover, because of
the unique demographic makeup of these communities’ labor force and the prevailing
economic precarity therein, such welfare losses may be even more pronounced than
in those urban or rural settings previously studied.
Since at least Christianson and Faulkner (1981), modeling has shown significant
income contributions to rural communities by their hospitals. Beyond the direct
and indirect sources of income collected from hospital expenditures for local support
services and wages paid to staff, these communities also enjoy additional benefits
(most notably in the form of tax revenue) by attracting medical professionals with
incomes significantly above median values. As such, it stands to reason that any loss
of such hospital facilities would produce income losses similar to these amounts.
With respect to establishing quantitative estimates for the economic impacts of
rural hospital closures, Holmes et al. (2006) provide a comprehensive survey of efforts
made by others at accomplishing just this. Beyond the generally intuitive negative
consequences expected of such closures, Hart, Pirani, and Rosenblatt (1991) surveyed
the mayors of 130 rural towns that experienced hospital closures and queried them
of any perceived changes to overall welfare. Of those asked, more than 90 percent
believed that hospital closures had substantial negative impacts on their communities’
economic well being. Despite the adamancy expressed by those surveyed however,
initial empirical research suggested that the impact of rural hospital closures on both
short- and long-term community economic growth were trivial (Pearson and Tajalli,
3
Consequences of Hospital Closures on Community Welfare
2003; Stensland et al., 2002). The primary finding was that rural hospitals were often
so small when compared to those found in urban areas that they had miniscule- if any
influence on unemployment and income when eventually closed. Similar research has
more or less agreed with these results by failing to find any statistically-significant
relationship between hospital closures and community growth trends (Probst et al.,
1999).
It is entirely possible however, as Holmes et al. (2006) suggest, that the failure
of these studies to locate any significant, negative relationship between community
economic growth and hospital closures is more to do with methodology than their describing actual phenomenon. Most significantly, these previous analyses were carried
out using input/output methods that lacked precise datasets and failed to account
tertiary, quality-of-life contributions to growth provided by these hospitals. When the
impacts of these closures are modeled using multiple regression methods instead, we
do see considerable negative outcomes from closures. When examining data for rural
hospital closures carried out during the 1990’s in this manner, researchers observed
an average loss of 4 percent of community per-capita income, as well as an average
increase in unemployment of 1.6 percent (Holmes et al., 2006, p. 478). Importantly,
these closures do indeed have significant detrimental impacts on long-term growth,
but are largely limited to those communities which lost their sole hospital and for
which no alternative is within close proximity.
With these findings in mind, it is important to consider a few imperative distinctions between our research subjects and those rural communities studied in previous
surveys. Indeed, it is largely these very geographic and demographic differences that
warrant our analysis here. For one, most communities within the Monongahela Valley
region that have faced hospital closures, mergers, or downsizing in recent decades still
have high enough population densities to be formally considered “urban.” With the
exception of Brownsville, all school districts that fall along the Monongahela river-
4
Consequences of Hospital Closures on Community Welfare
front throughout the Allegheny, Washington, Westmoreland, and Fayette counties
have densities exceeding the state average, thus meeting the formal designation of urban areas (U.S. Census Bureau, 2010). In general, it would be unwise to extrapolate
trends observed in rural areas to describe phenomena in urban regions, but as can be
seen with some additional context, the decay experienced by these communities over
the past three decades may in fact make them exemplars of such behavior.
To be sure, at their peak, manufacturing hubs along the Monongahela River enjoyed levels of growth and prosperity quite unlike any rural areas. However, over the
most rapid period of deindustrialization in the United States, approximately 1975 to
1990, Southwestern Pennsylvania experienced some of the greatest changes to longterm employment patterns nationwide. In these few years the region lost more than
150,000 manufacturing jobs while seeing only limited gains in the service industry to
compensate (Yamatani, 1986). At its peak level of employment in the 1960’s, the primary metals manufacturing industry constituted more than 15 percent of the entire
regional workforce before diminishing to less than 5 percent by the 1990’s (Coleman, 1986). Official unemployment rates averaged at 10 percent (Biegel et al., 1989)
throughout the first-half of the 1980’s, while estimates which included discouraged
and underemployed workers placed this figure at 16 percent over the same period
(Troan, 1985). These precipitous drops in manufacturing employment did eventually
stabilize, but comparatively-high unemployment and population loss have dogged
virtually all riverfront communities within the Monongahela Valley in the decades
since (Tony, 2016). Furthermore, between population loss and an increasing trend in
Medicaid enrollments throughout the region, hospitals have been further squeezed by
diminishing revenue streams.
A major consequence of this contraction throughout the Monongahela Valley has
been a transition in demographic and labor force makeup towards patterns more typical of rural communities. In particular, as younger, able-bodied workers left the
5
Consequences of Hospital Closures on Community Welfare
region with the collapse of the steel industry, those who remained saw median community ages grow and a labor market increasingly constituted by the service sector.
Communities such as Duquesne, which had enjoyed hourly wages exceeding statewide
averages and nearly 30 percent of employment in manufacturing as late as 1980 (Biegel
et al., 1989, p. 401), now boast median household incomes only 41 percent of the
state level (U.S. Census Bureau, 2017). As incomes diminished and community age
increased, a general trend towards rural demography has occurred with respect to the
nature of community health. Diabetes, obesity, substance abuse, and heart disease- all
conditions with well-established heightened prevalence throughout rural areas- have
enjoyed significant increases in ubiquity throughout the Monongahela Valley region
since deindustrialization (Centers for Disease Control [CDC], 2013; National Center
for Health Statistics, 2016).
Changes in employment patterns in the Monongahela Valley suggest that hospital closures may produce even more pronounced consequences than in typical rural
areas. As with other rust belt communities, the transition towards a service-based
labor market has seen a significant increase in health care employment (Hobor, 2013;
Olney and Pacitti, 2017). As high-paying manufacturing jobs left the area, hospitals
represented one of the most obvious avenues for securing steady, full-time employment. Furthermore, unlike their counterparts in rural areas, hospitals throughout
the Monongahela Valley were constructed during periods of peak-population, often
exceeding 100 beds in size and offering a wide-array of specialized services. Others
have established the tendency of hospital-induced income multipliers to scale with
the number of beds (Cordes et al., 1999), so as these hospitals have been closed or
down-sized, we would therefore expect these negative shocks to economic growth to
exceed those in comparative rural areas.
Additional complicating factors include the relative difficulty of these larger hospitals to control costs, as well as their more common tendency to down-size rather
6
Consequences of Hospital Closures on Community Welfare
than cease operations entirely when facing funding issues. While the hospital networks serving rural America are plagued by an increasingly dire solvency crisis (Hsia,
Kellerman, and Shen, 2011), many of these institutions faced similar difficulties in
prior decades. To confront a previous epidemic of rural hospital closures, Congress
included language in the Balanced Budget Act of 1997 that allowed rural hospitals to
restructure as critical access hospitals (CAHs). In doing so, these hospitals were able
to considerably limit the breadth of services legally required of them and consequently
reduce their operating costs. However, the strict bed and proximity requirements for
these hospitals to be granted permission to restructure as CAHs are often such that
the conversion of larger facilities like those found throughout the Monongahela Valley
would be infeasible. Indeed, despite a prolonged and universal tendency throughout
the region for hospitals to operate at or below the margin (Pennsylvania Health Care
Cost Containment Council [PHC4], 2017a), this facility size peculiarity may explain
why there has never been such a hospital conversion throughout the region (Flex
Monitoring Team, 2004; 2018). Instead, many of these facilities have had to adopt
alternative cost-saving methods to remain solvent; most commonly, limiting services
provided with special focus paid in reorienting towards outpatient care delivery, or
merging with regional hospital networks to increase access to capital (Holmes, 2015;
PHC4, 2017b).
In carrying out our analysis here, it is not enough to construct a model that
examines total facility closure alone. Previous examinations of rural health care
networks throughout Appalachia have shown diminishing utilization of and accessibility to obstetric, dental, substance abuse, and mental care services (Stensland et
al., 2002). Within Southwestern Pennsylvania, Allegheny, Washington, Westmoreland, and Fayette counties are all considered to be underserved by dental and mental health services (Health Resources and Services Administration [HRSA], 2017). A
summary inspection of recent newspaper articles and data for medical facility closures
7
Consequences of Hospital Closures on Community Welfare
throughout the region confirm the tendency for these sorts of services in particular to
face shutdown (Snowbeck, 2000; Gough, 2017; PHC4, 2017b; Goldstein, 2018). Because the loss of access to these services has significant direct impacts on community
welfare (namely, from lost high-wage specialists and overall health from diminished
utilization), it behooves us to include these factors into our analysis.
A final consideration that may be prove to be significant and provides additional
weight to the novelty of this analysis is the phenomenon of decreasing life-expectancies
throughout the region. As detailed by the CDC, the increased frequency of substance
abuse-related deaths has had substantial influence in lowering life-expectancies nationwide, but is particularly pronounced in rural communities (2017; 2018). Pennsylvania ranked third nationally for highest frequency of drug overdose deaths, with a
rate of 44.1 per 100,000 deaths (CDC, 2018, p. 3). Within Southwestern Pennsylvania, all counties saw increases in the number of overdoses from 2013 to 2016 between
67 and 138 percent (United States Department of Agriculture, 2018, p. 5). Since this
trend has been largely observed in only the past decade and much of the literature on
hospital closures predate that, it is reasonable to suspect that the negative impacts
on community welfare observed by earlier empirical studies may be understated. The
gradual demographic shifts experienced under deindustrialized may be exacerbated
as alcoholism and opioid abuse accelerate the loss of young, able-bodied individuals,
to say nothing of the exorbitant direct costs of combating this epidemic. Researchers
have consistently demonstrated the relationship between hospital closures and increased mortality rates among those treated for drug overdose (Bazzoli et al., 2012;
Liu et al., 2014), we would expect then that within the context of an overdose epidemic that the closure of acute care facilities represent an added factor to welfare
loss over previous estimates. Additionally, while there remains considerable debate
about the extent of influence that economic precarity has on substance abuse rates
throughout rural communities (Ruhm, 2018), if such a relationship does exist we can
8
Consequences of Hospital Closures on Community Welfare
surmise the existence of feedback processes whereby hospital closures induce increased
poverty, which consequently spurs substance abuse, further diminishing community
growth, and so on and so forth. The inclusion of this phenomenon may prove difficult
within our model, it seems imperative that attempt to do so.
Data Description
All data for our empirical analysis here is sourced from the United States Census
Bureau’s American Community Survey (ACS). These extensive surveys have been
conducted nationwide since 2005 through a combination of written, phone, and inperson interviews. Because of the survey’s considerable breadth of data, number of
participants (approximately 3.5 million households each year [U.S. Census Bureau,
2019]), public availability, and the Census Bureau’s diligence in ensuring proper sampling methods, it is an ideal source from which we can draw our data. In particular,
we will look at the Household Record data set for the years 2005 to 2017, which
includes vital information on household sizes, location, and incomes.
One of the more difficult tasks in sourcing appropriate data for our analysis is
finding those that are sufficiently precise geographically. Due to federal guidelines
regarding data anonymization, the breadth of microdata of individual households
or firms for public use must be sufficiently vague to prevent identification. This
places considerable limits on the amount of data useful for our project here: since
household data points are often recorded with geographic indicators only as precise
as county of residency and not, say, postal code or municipality, there is a very real
chance that significant trends will be obscured or entirely unobservable. Consider,
for instance, poverty estimates for the Borough of Braddock in Allegheny County,
and Allegheny County itself. Survey results from the ACS put percentages of the
population living below the federal poverty level at 31.9 and 12.5 (U.S. Census Bureau,
9
Consequences of Hospital Closures on Community Welfare
2017) respectively, suggesting that the localization of our analysis to the riverfront
Monongahela Valley communities is imperative should we expect any meaningful
results. Thankfully, while the ACS lacks municipality data, it does feature a slightly
less precise Public Use Microdata Area (PUMA) code, which allows us at a minimum
to separate possible countervailing measurements from peripheral urban areas (in
particular, the cities of Washington and Pittsburgh).
Using five PUMAs from Allegheny, Fayette, Washington, and Westmoreland counties that include nearly all riverfront communities along the Monongahela, we are able
to cull from the ACS Household Record data set more than 23,000 useful household
data points for the years 2005 to 2017. By cross referencing public records (PHC4,
2017), we can identify three hospital closures that occurred within this area over this
time frame: Tara Hospital at Brownsville, in 2006; Brownsville Tri County Hospital,
in 2007; and UPMC Braddock, in 2010. In addition to these closures, we can also
identify eight hospital and medical clinic closures or restructurings, a full description
of such can be seen in Table 1.
Table 1: Hospital Closures and Mergers, 2005-2017
Closure/Merger
Year
Name
Merged
Merged
Merged
Merged
Closed
Closed
Closed
Merged
Merged
2005
2005
2005
2005
2006
2009
2010
2013
2015
Greene County Memorial Hospital
Zitelli South Ambulatory Surgical Center
Brownsville General Hospital, Inc
SemperCare Hospital of McKeesport, Inc
Tara Hospital at Brownsville
Brownsville Tri County Hospital
UPMC Braddock
Jefferson Regional Medical Center
Southwest Regional Medical Center
PUMA
4002
1807
4002
1805
4002
4002
1805
1807
4002
From the ACS we can identify our dependent variable, self-reported family income
in terms of 2017 dollars, as well as important independent control variables, number
of wage earners in family and residency PUMA. While the ACS does supply us with a
10
Consequences of Hospital Closures on Community Welfare
wide array of information on the makeup of households, explicit information for possible determinants of income are limited. For this reason, we must limit our analysis
to families and disregard non-traditional household types. By referencing the survey
years and PUMA code for each data point we can produce a series of explanatory
variables representing hospital closures and mergers. Because of its sheer size and
the rigorous sampling methodology employed by the Census Bureau in collecting this
data, we can rest assured that results are unlikely to be skewed or biased by outliers.
Indeed, fewer than one percent of surveyed families within our sample earned more
than $400,000 over this twelve year period.
Model Description
The central intent of our empirical analysis is to establish whether or not, and to
what degree hospital closures impact the economic well being of those communities
to which they belong. Our working hypothesis is that there should exist some measurable negative correlation between average family incomes and residency within a
community during or around the time of hospital closure. The method we will employ
here to test this hypothesis is a simple ordinary least squares (OLS) linear regression
that will approximate the average dollar amount that family incomes decrease when
residing near a hospital closure, while controlling for the number of household wage
earners and region. We will conduct two such analyses: one will examine the impacts
of hospital closures alone, while the other will include variables for both hospital
closures and mergers. We define our linear models as follows:
IN COM E = β0 + β1 W + β2 Ct−1 + β3 Ct + β4 Ct+1 + β5 Ct+2 + β6 P1 + . . . + β10 P5 , (1)
11
Consequences of Hospital Closures on Community Welfare
and
IN COM E = β0 + . . . + β10 P5 + β2 Mt−1 + β3 Mt + β4 Mt+1 + β5 Mt+2 ;
(2)
where the t subscripts denote year, W an integer number of wage earners in the
household, and C, P , and M dummy variables representing residence in closure area,
a specific PUMA, and merger area respectively. C and M are equal to 0 unless the
family described by the data point resides in a closure or merger PUMA; defining
additional lagging variables allows us to approximate any lingering impacts from
closures and mergers, while leading variables limit spurious outcomes in the event of
an economic downturn in the preceding year precipitating eventual closure.
One considerable complicating factor with constructing our model in this manner
is its inability to account for wage growth- a significant issue since it is effectively
our hypothesis that hospital closures should be associated with diminished economic
growth. Any negative coefficients calculated from our OLS procedure for the closure
dummy variables would necessarily be inflated as they would include the explanatory
power of both closures and relative wage loss to the period mean.
The most obvious remedy would be to simply toss our model as it is and instead
base one on percent changes in family incomes between years. Unfortunately, to adopt
this alternative method would require the compression of our data set from tens of
thousands of data points to only twelve- one for each year- thus significantly reducing
the likelihood of producing statistically-significant results. Instead, we opt to retain
our model form as previously stated but do so with the following qualification: if an
OLS procedure is able to produce significant negative beta estimates for our closure
variables; and, if these coefficient values can be shown to be significantly larger than
the difference between overall mean family income and mean family income of the
years in which closures occurred by way of a simple one-sided t-test, then we can
12
Consequences of Hospital Closures on Community Welfare
conclude that hospital closures did indeed have significant negative impacts on local
family incomes. While this means that we cannot ascribe meaning to the size of our
estimates, it does allow us to answer the central question of our hypothesis.
Supplemental Analysis
As a supplement to our primary investigation into whether hospital closures have had
any measurable impact on the economic wellbeing of their surrounding communities,
we will carry out an additional analysis to establish whether any association exists
between hospital closures and drug overdose deaths. To do so, we use annual countylevel estimates for drug poisoning overdose mortality rates from the National Center
for Health Statistics’ National Vital Statistics Center dataset (2017) for the years
2005 to 2017. Controlling for year and region, we produce the following model to
estimate the impact of closures on mortality rates:
Overdose = β0 + β1 C + β2 RF ayette + β3 RW ashington + β4 T ;
(3)
where our dependent variable, Overdose, is the mortality rate of a given county and
year, C and R variables representing county characteristics as closure area and specific county respectively, and T , the given year of datapoint. C is equal to 0 unless
the county described experienced a hospital closure within its borders at that given
year; RF ayette and RW estmoreland are equal to 0 unless datapoint describes that specific
county; and T indicates the number of years from model initiation in 2005 that observation was recorded. The intent of constructing our model in this way is to control for
latent differences that exist between counties as well as larger, supraregional trends.
Unfortunately, because no public database of mortality rates exist for our examined area over these years with geographic codes more precise than county-levels, we
must broaden the scope of our analysis to include the entire Fayette, Westmoreland,
13
Consequences of Hospital Closures on Community Welfare
and Washington counties (out of an interest of avoiding skewed results, we exclude
Allegheny county and the likely-biasing effects of the city of Pittsburgh). This lack
of precision in our data casts serious doubt on the likelihood of producing either statistically significant or meaningful outcomes, but we will proceed with doing so out
of a lack of alternative avenues for investigation. Even in the event that no such
association is observed, such results can still inform our eventual conclusions.
Empirical Analysis
Initial Results & Conclusions
After running our primary regressions for both the closure and merger models, we
produce the following results seen in table 2. In general, these results would confirm
our expectations that hospital closures are correlated to loss of family incomes. As we
can see from our closure-only model, over the twelve year period observed, residence
within a closure PUMA resulted in an average loss of $5,469 in family income during
the that same year of closure. Recalling that this coefficient is likely inflated to at least
some degree when we assume general income growth over this timeframe, we should
instead confirm that this value is significantly larger than the mean differences for
closure years and all years. Performing a one-sided t-test on our current year closure
coefficient against these mean differences, -2,682.666, yields a score comfortably below
our 5 percent critical value. Therefore, we can surmise that residency within a closure
PUMA had some demonstrable negative impact on aggregate family incomes in the
year of closure.
Our merger model largely confirms these conclusions by returning a remarkably
similar coefficient estimate for the current year closure variable. Regression results
from this model suggest that lingering effects from closures in the following year
are significantly negative as well. The impacts of hospital mergers however, appear
14
Consequences of Hospital Closures on Community Welfare
Table 2: Regression Results
Dependent variable:
Family Income
(1)
Number of Wage Earners in Family
Residence in Closure Region, Preceding Year
Residence in Closure Region, Current Year
Residence in Closure Region, +1 Year
Residence in Closure Region, +2 Year
PUMA Residence: 2002
PUMA Residence: 1805
PUMA Residence: 1807
PUMA Residence: 3900
(2)
25, 680.410∗∗∗
(434.296)
−2, 990.306
(2, 067.796)
−5, 469.000∗∗∗
(2, 050.874)
−3, 902.546∗
(2, 042.587)
−1, 251.724
(2, 084.557)
3, 682.580∗∗∗
(1, 371.196)
−4, 390.602∗∗∗
(1, 316.034)
34, 786.760∗∗∗
(1, 481.220)
−10, 520.330∗∗∗
(1, 376.106)
25, 680.960∗∗∗
(434.308)
−2, 471.620
(2, 128.765)
−5, 360.306∗∗
(2, 135.402)
−5, 187.058∗∗
(2, 120.894)
−2, 038.196
(2, 214.248)
2, 966.086∗
(1, 619.659)
−4, 801.667∗∗∗
(1, 396.780)
34, 546.090∗∗∗
(1, 499.494)
−11, 236.740∗∗∗
(1, 623.707)
−1, 718.101
(2, 521.850)
−2, 393.231
(1, 766.236)
−1, 968.265
(1, 752.287)
2, 118.504
(1, 742.499)
41, 502.550∗∗∗
(1, 497.282)
Residence in Merger Region, Preceding Year
Residence in Merger Region, Current Year
Residence in Merger Region, +1 Year
Residence in Merger Region, +2 Year
40, 786.860∗∗∗
(1, 225.520)
Constant
Observations
R2
Adjusted R2
23,517
0.182
0.182
∗
Note:
15
p<0.1;
23,517
0.183
0.182
∗∗
p<0.05;
∗∗∗
p<0.01
Consequences of Hospital Closures on Community Welfare
insignificant; therefore, there exists conclusive correlation between hospital mergers
and family incomes.
While these results are promising overall, the low Pearson coefficients of only
0.182 expose our model’s shortcomings. Without having some additional means to
account for household educational attainment, sector of employment, or years of
work experience, any estimates such as ours are going to be necessarily limited in
their explanatory power. In spite of this, we can still take solace in the model’s ability to demonstrate a statistically-well-founded negative correlation between hospital
closures and family incomes, even if the absolute degree of these impacts remains
inexact.
Further inspection of the data reveals additional corroboration for our regression
results. Plotting mean family incomes against years for all PUMAs, closure PUMAs,
and non-closure PUMAs (as seen in figure 1 ) demonstrates negative trends within
closure regions around the closure years of 2006, 2009, and 2010. While it is conceivable that these trends reflect exogenous shocks to the regions not accounted for
within our model, the relative constancy of income growth among other observed
regions would appear to affirm our conclusions.
In carrying out the regression for our supplemental analysis, we observe substantially less satisfactory outcomes. As can be seen in table 3, at a 5 percent significance
level, there is no observable correlation between hospital closures and overdose rates.
Our concerns about limited data size and lack of geographic specificity would appear
to have been well founded, but without those data we will have to simply concede
that any such relationship is presently indeterminate.
16
Consequences of Hospital Closures on Community Welfare
$95,000
$90,000
PUMAs with
No Closure
$85,000
All PUMAs
$80,000
PUMAs with
Closure
$75,000
$70,000
$65,000
2006
2008
2010
2012
2014
2016
Figure 1: Annual Mean Family Incomes by PUMA Group, 2005-2017
Overall, we can conclude the following based on the results of our multiple regression analyses. There is ample empirical evidence to suggest that in the years 2006,
2009, and 2010, hospital closures throughout the Monongahela valley produced observable declines in family incomes of those residing within the same regions as where
closures occured. While the precise estimates returned are likely different from the
actual aggregate impacts, we can demonstrate that the nature of this relationship is
significantly negative in nature, confirming our initial hypothesis. Most of the observable effects from hospital closures occur within the current year of closure, although
the results from when our model is amended to include hospital mergers suggest that
these effects persist into the following year. While hospital closures appear to have
significant impacts on the well being those who live around them, the effects of hospital mergers remain inconclusive. Lastly, while overdose deaths from drug poisonings
increased consistently over this time period, there is no demonstrable relationship
between mortality rates and hospital closures.
17
Consequences of Hospital Closures on Community Welfare
Table 3: Overdose Regression Results
Dependent variable:
Residence in Closure County
Residence in Fayette County
Residence in Washington County
Year
Constant
Observations
R2
Adjusted R2
Overdose Deaths
Logged Overdose Deaths
per 100,000 People
per 100,000 People
(1)
(2)
−1.088∗
(0.607)
0.681∗∗
(0.328)
−2.000∗∗∗
(0.312)
0.975∗∗∗
(0.038)
9.993∗∗∗
(0.330)
−0.061∗
(0.030)
0.039∗∗
(0.016)
−0.131∗∗∗
(0.015)
0.062∗∗∗
(0.002)
2.367∗∗∗
(0.016)
36
0.963
0.958
36
0.977
0.974
∗
Note:
p<0.1;
∗∗
p<0.05;
∗∗∗
p<0.01
Discussion of Results
Based on the outcomes of our analysis we can say that there is evidence to support
the claim that hospital closures throughout the Monongahela Valley between 2005
and 2017 produced significant, measurable negative impacts on household incomes.
This confirms our hypothesis and suggests that future policy initiatives aimed at
bettering health standards throughout the region must make ready access to hospitaldelivered care an imperative; to that end, such policy must necessarily confront the
solvency crisis these institutions face and its primary instigator: diminishing Medicaid
reimbursement rates.
The results of our primary analysis are largely satisfactory and provide a reasonable degree of evidence to support initial predictions regarding the impacts of hospital
18
Consequences of Hospital Closures on Community Welfare
closures, but the overall vulnerability of our model to exogenous system shocks will
require additional efforts before verification. It may be tempting to reproduce our
analysis here but with the inclusion of some additional control dataset, as many have
attempted previously, but as Holmes (2006) and others point out, there may be no
obvious analog to our examined region. Even if such a parallel community did exist
with all of the same economic peculiarities as the Monongahela Valley, there is no
assurances that that region would not bias our results with its own exogenous shock.
Instead, it may be more feasible to examine the determinants of our results piecemeal.
Perhaps the most straightforward approach would be to break down and examine the
individual components of hospitals’ influences on community well being- namely, the
utilization rates of hospital-provided care and incomes of those employed within the
health care sector.
One method of corroborating our results may be to simply examine changes in
hospital admissions in the surrounding area following closures. While our analysis
here was done so as to account for both the direct and indirect impacts of hospital
closures, diminishing admissions are likely to be the primary factor in spurring closures in the first place. Assuming that there have been no sudden, external changes
to demand of care, hospital closures should induce influxes of admissions to neighboring facilities. If, however, no such increase can be observed (and assuming any
such analysis also controls for changes in care utilization), then we can surmise that
direct health impacts to the closure community would be minimal. Alternatively,
a near-identical analysis could examine household incomes and employment rates of
individuals employed within the health care industry. If closures do indeed provoke
economic declines in these communities, the most obvious and immediately impacted
groups are going to be those whose livelihoods are dependent on them.
Other plausible analytical methods that may produce more convincing outcomes
would be to expand the breadth of examined communities to include other analogous
19
Consequences of Hospital Closures on Community Welfare
deindustrialized communities from across the country; or, reproducing our analysis
here, but with data sourced from conducting surveys ourselves so as to increase geographic precision of data points. The former method has the distinct advantage of
supplying us with a dataset considerably more impervious to localized, exogenous
shocks; and, as such a dataset would no doubt be entire orders or magnitude larger,
it may be possible control and compare closure and non-closure outcomes more comprehensively. In the latter case, increased geographic precision would allow us to
examine the impacts of closures based on households’ proximity to hospitals, rather
than residency alone. This would almost certainly present us with a more accurate
depiction of closure outcomes and would effectively eliminate biasing impacts from
nearby cities.
A final consideration that may prove to be more imperative for any future analysis
is of the significant changes to Pennsylvania’s Medicaid reimbursement policy in rural
areas. The Rural Health Model is a pilot program currently being carried out under
the auspices of the Centers for Medicare and Medicaid Services and Pennsylvania
Department of Health with the aim of introducing an Accountable Care Organization (ACO) model to rural Medicaid management. This means, among many things,
the transition away from more typical fee-for-service revenue model towards a multipayer system that stresses year-to-year revenue stability and stakeholder input in
establishing services provided (Murphy et al., 2018). In doing so, state and federal
agencies can ensure the long term success of majority-Medicaid-patient hospitals and
avoid unnecessary closures. As this program is only in its pilot stage of five participating hospitals however, it is likely that some additional closures will occur before
its universal application. In such a case, data collected from participating hospitals
may prove invaluable in establishing the efficacy of current and future ACO policy
initiatives.
The transition towards the Pennsylvania Rural Health Model also highlights the
20
Consequences of Hospital Closures on Community Welfare
subject of our secondary analysis: rural accessibility to substance abuse care services.
ACO programs initiated in other states have placed considerable emphasis on not
only establishing financial stability for rural hospitals, but in expanding the breadth
of care provided as well. Indeed, during the program’s initial rollout administrators for the Pennsylvania program noted the demand for substance abuse clinics and
therapy services as a primary motivator for pursuing this effort. Enabling greater
community planning could help address the deepening opioid crisis, both directly
by providing greater access to desperately-needed counseling services, but also indirectly by providing at least some financial means of offsetting the economic costs of
accidental deaths by drug poisonings.
While we were unable to confirm any categorical association between hospital
closures and opioid overdose deaths, we do know from previous studies (Bazzoli et
al., 2012; Liu et al., 2014) that closures elsewhere have spurred higher mortality rates
among those treated for drug poisoning. CDC estimates put the cost of accidental
drug deaths at approximately 26 percent (Florence et al., 2015) of economic costs
related to all fatal injuries, accounting for a greater share than even transportationrelated deaths. In 2014, residents of Pennsylvania were estimated to have shouldered
more than $10 billion (CDC, 2015) in medical costs and work-loss related to fatal
injuries. This suggests that even a one percent increase in overdose deaths could wreak
economic havoc on communities as small as those that make up the Monongahela
Valley. Therefore, any effort at ensuring or expanding access to emergency and drug
therapy services- as the Rural Health Model intends to do- could have considerable
impact on overall community economic well being. Naturally, continued examination
of outcomes will be necessary to establish the program’s overall effectiveness, but the
Rural Health Model represents an encouraging policy solution to counteracting the
economic and health impacts of hospital closures.
21