Safety Net Hospitals
and Minority Access to
Health Care
Gloria J. Bazzoli, Ph.D.
Virginia Commonwealth University
Lee R. Mobley, Ph.D.
Research Triangle Institute, Inc.
This research is supported by NIH Grant # 5R01HL082707-2.
1
Outline

Structure of US health safety net and the role safety net
hospitals can play in reducing racial, ethnic, and
socioeconomic health disparities

Overview of our study

Preliminary descriptive analysis for Florida:



Patient proximity to a safety net hospital and rates of particular
admissions
Location of safety net hospital closures relative to particular
patient populations
Next steps for our research
2
Structure of
the US Health Safety Net

A patchwork of health providers currently provide
care for economically disadvantaged:






public and private hospitals
private physicians
community health centers
government programs for particular patient groups (e.g.,
VA, Indian Health Service)
local health departments
US hospitals provide large share of this care:

Holahan and Hadley (Health Affairs, 2003) estimated that
66% of indigent care in 2001 was provided by hospitals
3
How Safety Net Hospitals (SNHs) Could
Help Disadvantaged Populations

Offer outreach services, either independently or with other
organizations:



Provide information about services in diverse languages and media
Establish decentralized sites of care
Hire multi-lingual staff that are culturally sensitive

Provide free or reduced fee primary care services through system of
local clinics

Collaborate with local agencies to assist in referral for social and
behavioral services

Coordinate referral and provide access to specialty services

Provide follow-up treatment and rehabilitation post-hospitalization
4
But Are SNHs Effective?

SNHs are most often located in neighborhoods where
the poor and racial or ethnic minorities reside.

However, prior studies found that SNH presence in an
area had:
 minor effects on access to care for uninsured
 little to no effect on health disparities

A problem with prior studies is that measures of SNH
availability and capacity are crude:
 presence of SNH in an individual’s county
 number of beds or ER visits at the SNHs in a county
5
Motivation for Our NIH
Research Project

Assess how location of uninsured patients vis-à-vis SNH affects
care patterns:





Do patients who live closer to a SNH have better access to
health services than those farther away?
Is proximity to a SNH especially important to racial or ethnic
minority access to care?
What happens to the uninsured when a SNH exits a community
through closure or ownership conversion?
Are there regional clusters of minority individuals who are
especially affected when a SNH exits?
Assess strategies that communities have taken to lessen the
detrimental effects of SNH exit on disadvantaged populations
6
Motivation for our NIH
Research Project

In 1990, about 1,600 (32%) of 5,000 US community hospitals were
SNHs, based on definition developed by Darrell Gaskin*

By 2000, 121 (7%) SNHs closed and 269 (17%) experienced an
ownership conversion that could affect their mission

Although SNH changes seem small in number, these facilities may
be very important locally:
o
o
o
For example, the closure of Milwaukee County Hospital in Wisconsin.
The hospital had 12,568 admissions in 1990, of which 2,302 were Medicaid.
Overall, the hospital provided 9.2% of total hospital admissions and 22.7% of
all Medicaid admissions in the city of Milwaukee.
*Specifically, Gaskin and his colleagues defined SNHs as all public hospitals and those
private, nonprofit hospitals with a large Medicaid patient share (specifically greater than mean
plus 1 standard deviation for NFPs in the hospital’s state).
7
Focus of Our Study

Our study examines SNHs in 5 states:
 Arizona, California, Florida, New York, Wisconsin
 Specifically examine hospital discharge data in AHRQ SID

Study period:
 Early year in 1990s (base period) and 2000 or 2003 (ending
period)

Study sample:
 Patients residing in zip codes within given distances to a SNH
 Primarily interested in those who are uninsured
 Within this group, distinguish non-Hispanic white, non-Hispanic
black, and Hispanic individuals
8
SNH Changes in Our 5 States
Over Time
Total
Community
Hospitals:
Base Year
Total SNHs:
Base Year
SNH
Closures
SNH
Ownership
Conversions
Arizona
56
11
0
3
California
423
115
9
20
Florida
201
38
3
8
New York
221
53
1
3
Wisconsin
125
28
3
4
9
Primary Access Measures of
Interest

Examining hospital discharge data to identify specific types of
hospitalizations:

Ambulatory care sensitive conditions (ACSCs) that could have been
avoided if adequate primary and preventive services were present:
o

Specialized services for which patients require hi-tech hospital services
and specialty physician involvement, referred to as referral sensitive
conditions (RSCs):
o

angina, asthma, acute diabetic events, gastroenteritis, hypertension
CABG surgery, PTCA, hip/knee joint replacement, organ/bone marrow
transplantation
Marker conditions (MCs), which require immediate urgent care, and
are thus not likely to be access sensitive:
o
AMI, hip fracture, appendicitis
10
Working
Hypotheses/Expectations

Presence of nearby SNH for uninsured:





reduces ACSC admissions (relative to MC) if SNH is facilitating
access to primary care
increases RSC admissions (relative to MC) if SNH is providing or
facilitating access to specialty services
effects should be more pronounced the closer an individual is to
a SNH
potentially these effects are stronger for racial or ethnic minorities
if their access to other sources of care is especially weak
Exit of a SNH through closure or ownership conversion could
eliminate these beneficial effects
11
Preliminary Descriptive
Analysis

For patients residing in low income zip codes (median family income <
250% of FPL):



identified travel distance to the hospitals at which they received care based
on patient and hospital zip code centroids
established near, moderate, and far travel distance thresholds for urban and
rural hospitals in each state
Next turned to SNHs in a state for its base year:





identified distance to nearest SNH for each patient zip code in the state
applied near, moderate, and far thresholds to these distances
for a given SNH and all its associated near, moderate, and far zip codes,
selected hospital discharges of all uninsured individuals in these zip codes
these discharges were not only from the SNH but also from any other
hospital at which patient received care
calculated rate of ACSC to MC discharges and RSC to MC discharges for
uninsured non-Hispanic white, non-Hispanic black, and Hispanic individuals
within each distance category
12
ACSC Preliminary Findings for
Florida: Large Metro Areas
Ratio of ACSC to MC admissions for uninsured, 1992
12
10
8
6
4
2
0
White
Black
Hispanic
Miles to Nearest SNH:
13
0-3 miles
3-8.2 miles
8.2-22 miles
RSC Preliminary Findings for
Florida: Large Metro Areas
Ratio of RSC to MC admissions for uninsured, 1992
1
0.8
0.6
0.4
0.2
0
White
Black
Hispanic
Miles to Nearest SNH:
14
0-3 miles
3-8.2 miles
8.2-22 miles
Overall Preliminary Findings
for Florida: Large Metro Areas

Simple descriptive data raise many questions:


For uninsured non-Hispanic whites and Hispanic individuals, why
do ACSC rates decline as they are farther from a SNH?
For uninsured non-Hispanic blacks:
o
o
o
o
what is contributing to higher ACSC rate relative to other
racial/ethnic groups for each distance category?
unlike other groups, the ACSC rate increases with distance;
although consistent with what we expected, what effect will
distance have after we control for other factors?
what is contributing to lower RSC rates relative to other
racial/ethnic groups?
why is there such a large drop in RSC rates for uninsured blacks
in the 3 miles or greater categories relative to 0-3 miles?
15
Planned Multivariate Analysis

Estimate a logistic model examining probability of an
ACSC (or RSC) discharge relative to a MC discharge at
the person-level of observation, with the form:
log(
Pi , ACSC
Pi , MC
)
 1 X i   2 HS i   3 SL i   4 ( X i * SL i )   5TI i   6Yi   i
where Xi are patient characteristics (age, gender, race/ethnicity, payer);
HSi are health system characteristics (SNH and FQHC capacity, availability
of physician and non-physician personnel); TIi are indicators of travel
impedance or barriers (land use indicators, commuting patterns to work);
SLi are indicators of distance to SNHs; and Yi is a year indicator.
16
SNH Change Over Time in
Florida

As noted, we are also interested in SNH change
over study period: closures and conversions that
could affect hospital mission

Some of the changes in Florida during our study
period caught our attention:

As expected, SNHs in base period were primarily located in
areas with high rates of poverty and minority populations

However, three SNHs closed and their local socioeconomic
profiles had distinctive features.
17
SNH Change Over Time in
Florida: 2000 Demographics
CLOSED SNHs
% Non%
% in
Total
Hispanic Hispanic Poverty Population
Black
Everglades Memorial
Hospital, Pahokee, FL
56.1%
29.5%
32.0%
5,985
Fish Memorial Hospital,
De Land, FL
19.2%
8.7%
19%
20,904
Polk General Hospital,
Bartow, FL
28.4%
8.1%
13.1%
15,340
Overall FL statistics
14.6%
16.8%
12.5%
16 million
18
SNH Change Over Time in
Florida

For Everglades Memorial Hospital:



a nearby public hospital (10 miles away) converted to forprofit 2 years after the Everglades closure
the nearest remaining SNH was 44 miles away in Palm
Beach, FL.
For Fish Memorial Hospital:



a church-affiliated health system acquired this hospital and
a second public hospital located in De Land
Fish Memorial was closed and a replacement facility was
built by the system in Orange City, FL (19 miles away)
a church-affiliated hospital remains in De Land but loss of
Fish Memorial reduced SNH capacity by one-third
19
% Non-Hispanic Black 2000 by
Census Tract: Volusia County
New Fish Memorial
Hospital
Original Fish
Memorial Hospital
20
% Poverty 2000 by Census
Tract: Volusia County
New Fish Memorial
Hospital
Original Fish
Memorial Hospital
21
Planned Multivariate Analysis

Do a pre-event/post-event analysis in which we match areas
where SNH changes occurred to a similar one where SNH
changes are not present.

Criteria for matching treatment and comparison hospitals (our
wish list):
 In same state
 In same type of area (rural, small metro, large metro)
 Similar SNH hospital market structure in base year in terms of
numbers, ownership, and bed size
 Similar non-SNH structure ownership status in base year and
similar changes over time
 No overlap in patient zip codes
 Similar socio-demographic characteristics
 Similar patterns of patient travel distances
22
Planned Multivariate Analysis

A separate patient-level analysis for each individual matched
area:

In Florida:
o
o

Matching De Land where 1 public SNH closed and one converted
from public to NFP with Titusville, FL where one public SNH remained
operational with no ownership change
Matching Bartow, FL and Pahokee, FL to Titusville, FL in two
separate analyses.
In Wisconsin:
o
Matching Milwaukee with its 2 SNHs that closed (Milwaukee County
Hospital and Northwest General Hospital) with Madison, WI where
University of Wisconsin Hospital remained operational.
23
Planned Multivariate Analysis

Estimate a patient-level hospital choice model for
ACSC (RSC) where individual i’s expected utility of
hospital j (Vij):
Vij = 1Zij + 2(Xi*Zij) + 3(TIi*Zij) + 4(SLi*Zij) + 5(SLi*Xi *Zij) + ij
and assume that the error is Type I extreme value to obtain the
McFadden conditional logit model :
Pij 
exp( Vij )

exp( Vij )
j _ in _ choice _ set
In this case, SLi will be a series of dummy variables indicating distance
to a closed or converted SNH. If no closure/conversion, they all equal 0.
24
Planned Multivariate Analysis

The SLi do not enter the model directly (as a main effect) because, like
patient characteristics in Xi, they are not a characteristic of a given
hospital j

The interaction of SLi with elements of Zij will reveal many things.




Travel distance to each hospital option j is included in Zij, and its interaction
with SLi will reveal if patients are less deterred by travel distance to obtain
hospital care when a nearby SNH exits.
We will also include the three-way interaction of SLi, travel distance, and
patient race/ethnicity to assess if there are racial/ethnic differences in
response to travel distance when SNHs exit.
Zij will also contain an indicator of whether a given hospital in a patient’s
choice set is a SNH. Its interaction with SLi will indicate if patients are more
likely to choose a SNH after the loss of a nearby SNH.
A three-way interaction of the above with patient race/ethnicity will reveal if
minority individuals are more apt to seek out another SNH relative to
comparable non-minority individuals.
25
Planned Multivariate Analysis

We also plan to estimate a much simpler
patient-level model where actual travel
distance to a hospital is the dependent
variable.

Specifically:
log (dij*) = 1 + 2Xi +3Zij + 4HSi + 5TIi + 6SLi + 7 (SLi*Xi )+ 8Yi + ij
26
Desired Outcomes from
Analysis

Our proposed analysis seeks to understand how
patterns of access are affected by:

Geographic location of uninsured relative to SNHs,
assuming those nearest to an SNH achieve greatest
benefit

Safety net hospital contractions that occur in a market,
assuming these might negate any positive benefits derived
from nearness to a SNH

Differential effects based on patient race or ethnicity
27
Desired Outcomes from
Analysis

We also plan to develop measures of accessibility using the
conditional logit analysis (namely use estimated coefficients to
measure log value of denominator in the model before and after
SNH closure or conversion).



We examine values of this variable across individuals to see if
there are regional clusters of individuals distinguished by race or
ethnicity that are especially affected by SNH exit
We will use this and other measures to identify communities with
the greatest access impediments following safety net hospital
closure or conversion.
We then identify a few sites for in-depth case studies of
communities in which access effects were small versus those
where they were larger
28
Descargar

The Impact of the Health System and Organizational …