Models
and Analyses Based on Geographic Data
Given
the practical and methodological shortcomings
evident in the analyses using facility-level
data, the project team shifted its attention
to models based only on county-level data
that were nationally available and frequently
updated. This shift seemed justified theoretically
as well, because the inability of a facility
to recruit and retain RNs in a county
with sufficient overall supply of RNs
may be a result of organizational culture
rather than a genuine shortage. Limiting
analyses to easily obtainable county level
data seemed to serve these ends better
than further pursuit of a model incorporating
facility-level data.
A.
Limitations and Challenges
There are limitations and challenges to
a method based solely on geographic factors.
For one, patterns of RN employment and
health service utilization often transcend
county (and state) lines. Knowing where
RNs and patients live does not necessary
tell researchers where services were provided
or received, and thus where shortages
actually existed.
Furthermore, the use of county-level
data can mask large differences in facilities
within counties. This is particularly
true in the largest metropolitan counties.
For example, New York County (Manhattan)
may not meet the criteria for worst county-level
RN shortage, but this ignores the fact
that some facilities within Manhattan
have a much harder time recruiting RNs
than others (e.g., public facilities,
those located in neighborhoods perceived
as unsafe). Geography-based methodologies
also may not adequately account for special
circumstances specific to facilities.
Regardless of whether a facility is in
a large county or not, it may have extenuating
circumstances. There may be adequate numbers
of RNs in the county, for example, but
it may still be difficult to recruit RNs
to work with the homeless.
Supplementing geography-based models
with other procedures can minimize some
of these limitations. Primary care Health
Professional Shortage Areas (HPSAs) are
currently designated based on geography-level
characteristics, on facility-level characteristics,
or on service to special populations.
A similar tiered process could be developed
for nursing shortage designations. Geographic
designations could also be supplemented
with an application process that allows
facilities to submit facility-specific
data. Special rules could be established
to address sub-county variations in large
urban areas (e.g., certain facilities
in counties with population greater than
one million—public, in a HPSA, or in a
high-poverty Census tract—might automatically
qualify).
One thing that emerged clearly in the
analyses of facility-level data is that
certain types of facilities were disadvantaged
in the competition for RNs relative to
others. The current methodology for awarding
nursing loan repayment funds is based
on categories of facilities, and this
could be preserved so that certain types
of facilities continue to receive preference,
but in combination with geographic designations.
Geographic designations could also be
combined with facility type, in recognition
of the fact that certain types of facilities
(e.g., long-term care) may face greater
disadvantages than others (e.g., hospitals).
Facilities located in shortage counties
could be given priority based on facility
type, or conversely, facilities within
priority categories (e.g., disproportionate
share hospitals, community health centers)
could be given priority designations based
on county-level shortages.
An application procedure would allow
facilities that feel they have been unfairly
disadvantaged by a county-level designation
to submit facility-level data to document
their situation. This would ease the burden
on HRSA because most designations would
be based on geography, but would provide
facilities with special circumstances
with an opportunity to qualify.
B.
Measuring RN Supply at the County Level
The counts of RNs by county were taken
from the 2000 U.S. Census long-form data,
which is a 1-in-6 sample of the U.S. population.
These data gave RNs by county of residence,
not employment, and were less accurate
when the actual number of RNs in the county
was low (due to sampling error), but this
was probably the best source available
for county-level counts of RNs nationally.
In larger counties, the sample size should
be sufficiently accurate. But in smaller
counties, sampling error could have the
effect of either undercounting or overcounting
RNs. One person in the sample represents,
on average, six people. If a small county
has 102 RNs, theoretically one would expect
17 to be selected by the Census sample.
If only 13 were in fact selected, the
county would appear to have only 78 RNs,
and might inappropriately qualify as a
shortage county. On the other hand, if
20 were selected, the county would appear
to have 120 RNs, which might prevent it
from qualifying as a shortage county.
These kinds of sampling errors would be
random and not systematic, so less populous
counties should not be consistently advantaged
or disadvantaged by the method.
It is important that any method used
by HRSA be easily updated using existing
sources of data. Updating the decennial
U.S. Census data can only be done every
ten years, which creates estimation problems
that grow over time, especially for counties
that are rapidly growing or shrinking.
Starting in 2008, another option will
become available when the Census Bureau’s
American Community Survey (ACS) begins
to provide estimates for smaller areas
using three-year moving averages. Although
the ACS sample will be smaller than the
Census long-form data, it will be larger
than any other interim data set, which
increases the potential for sampling errors
in small counties. Each person sampled
in the ACS in one year will represent
more than 100 people.
C. Adjusting for Commuting
Estimates of where RNs live were inadequate
measures of supply because in some areas
commuting inflows or outflows were very
substantial. For example, only 16% of
workers in New York County in 2000 actually
resided in New York County. Using numbers
of RNs living in New York County would
thus substantially overestimate the degree
of shortage in that county.
The U.S. Census Bureau provides data
collected in the decennial census on commuting
flows between every pair of counties in
the U.S. From these data, commuting inflow
was estimated based on the percentage
of persons employed in county who lived
in a different county, and commuting outflow
was calculated based on the percentage
of employed residents of the county who
worked in a different county. These rates
of county inflow and outflow were applied
to RNs on the assumption that RN commuting
patterns were not different from commuting
patterns overall. (Preliminary analyses
did not indicate that RNs were any more
or less likely to work outside of their
county of residence.)
D.
Geography-Based Methods
There are a number of ways to conceptualize
and measure RN supply at the county level,
ranging from simple to sophisticated.
All of the methods below were calculated
using RN supply data adjusted for commuting
patterns.
1.
RNs to Population Ratio Method
This method is based upon the assumption
that RNs should be evenly distributed
across the U.S. in direct proportion to
population (e.g., that 70 people in Los
Angeles County, California require the
same number of RNs as the 70 people who
make up the entire population of Loving
County, Texas). The estimated number of
RNs required in a county is calculated
based on population need rather than demand
for RNs created by the existing health
care infrastructure, and assumes that
people receive nursing services where
they live.
This ratio is very simple to compute
(#RNs/#Population) and the data needs
are also relatively clear. On the other
hand, this ratio is also very crude, ignoring
actual use of services (i.e., where people
actually receive care), and demographic
variations in health care needs (e.g.,
the greater needs of the older adults).
2.
RNs to Adjusted Population Method
The project team explored two methods of
adjusting the population. The first was
based on rates of primary care utilization
by gender and age (with weights based
on the new primary care HPSA methodology)
and the second was based on rates of utilization
of multiples types of services estimated
on age alone (with weights based on age-specific
utilization rates for different types
of services, gleaned from a variety of
sources, most commonly Health, United
States, 2005).
Because it accounts for population demographics,
this method, which assumes that age-specific
patterns do not vary across counties,
should more accurately reflect population
need than a simple RN to population ratio.
However, this method, like the first,
is based on estimated need for RNs rather
than estimated demand for RNs.
3.
RN to Physician Ratio
Both previous methods fail to account for
the location of health care infrastructure.
Regardless of the needs of the population,
if an area has no health care employers
to hire RNs there is no labor market demand
for RNs and therefore no shortage. Places
with more health care employers should,
however, have more physicians, so physician
supply can be used as a crude proxy for
RN employer demand.
On the other hand, the net effect of
this method is that areas that are short
on both physicians and RNs appear comparable
to areas that have surpluses of both physicians
and RNs if the ratio is similar. This
is particularly concerning because physician
shortage areas may have the greatest need
for RNs to provide basic primary care.
This raises the RN shortage standard for
exactly those counties—they must be short
of RNs relative to the number of physicians
when they are already short of physicians.
4.
County Cluster Adjustments
All of the previous methods discussed ignore
the flow of patients between adjacent
counties to receive health care. An attempt
was made to adjust for this by recalculating
the previous ratios based on county clusters
(RN, population, and/or physician counts
summed for each county and its contiguous
counties). The effect of this adjustment
was higher shortage scores for nurse-poor
counties surrounded by other nurse-poor
counties, compared to nurse-poor counties
surrounded by nurse-rich counties. This
was theoretically appropriate in that
it accounts for the unavailability of
RNs in neighboring counties as well as
in the county of residence.
This method showed some promise, but
it still did not address some of the fundamental
problems of the previous ratio methods.
Furthermore, it did not account for the
effects of multiple counties drawing on
each others’ resources. For example, it
is tempting to say that County A’s shortage
isn’t really so bad because it is bordered
on the west by County B, which has a surplus
of RNs. The situation of both County A
and County B would be accounted for in
County A’s county cluster, but what would
not be accounted for is the possibility
that County B is bordered on the west
by County C, which is also short of RNs
and draws on County B’s resources. County
B’s surplus may be sufficient to share
between its own population and County
A’s population, but not between its own
population, County B’s population, and
County C’s population.
5.
Cross-County Patient Flow Adjustments
Another attempt to adjust for the flow
of patients between counties involved
adjusting population figures based upon
commuting flows. This assumed that the
flows for seeking health care services
were similar to those for commuting in
general, and that areas that drew more
commuters had more health care infrastructure
and would also draw more health care consumers.
Unfortunately, it was not clear that this
is a reasonable assumption. It seemed
likely to be true for many counties, but
may not be true for some (particularly
counties with large outflows of “extreme
commuters” who travel more than sixty
minutes to their jobs).
After reviewing the various versions
of these ratio models, it was unclear
whether county clusters or adjustments
for cross-county patient flows were consistently
an improvement on base ratios. It was
concluded that an ideal method should
use actual measures of health care utilization
rather than attempting to estimate patient
flows.
6.
Factor Analysis of Nursing Shortage Indicators
A more sophisticated attempt to create
a typology of counties based on the RN
labor market involved factor analysis,
an advanced statistical technique used
to collapse a large set of characteristics
of objects (counties in this case) into
a smaller set of “factors” that represent
different aspects of the objects. In this
case, different characteristics of counties
related to the supply of and demand for
nurses (e.g., #RNs per capita, per capita
income) load into different factors that
represent different aspects of the supply
and demand for nurses (e.g., a factor
related to the economic conditions in
the county).
This technique identified three broad
factors relevant to nursing shortages
at the county level: RNs relative to infrastructure
(demand); RNs relative to population (need);
and economic conditions. These factors
are shown in Table 12.
Table
12. Standardized Factor Analysis Coefficients
Related to Nursing Shortages in Counties
in the U.S.
Metropolitan
dummy variable |
-0.025 |
-0.044 |
0.188 |
RNs
per 1,000 individuals |
0.003 |
0.256 |
0.012 |
RNs
per 1,000 individuals < 5 years |
-0.007 |
0.259 |
0.002 |
RNs
per 1,000 individuals >=65 years |
0.005 |
0.109 |
0.122 |
RNs
per hospital bed |
0.213 |
-0.052 |
0.048 |
RNs
per MD |
0.136 |
0.096 |
-0.132 |
RNs
per 1,000 civilian labor force |
0.020 |
0.274 |
-0.045 |
RNs
per 1,000 inpatient days |
0.272 |
-0.059 |
-0.058 |
RNs
per 1,000 outpatient visits |
0.158 |
0.007 |
-0.016 |
RNs
per 1,000 emergency room visits |
0.134 |
0.066 |
0.016 |
Infant
mortality rate |
0.028 |
0.019 |
-0.140 |
RNs
per 100 Medicare inpatient days |
0.278 |
-0.053 |
-0.038 |
RNs
per 100 Medicaid inpatient days |
0.220 |
-0.018 |
-0.069 |
Median
household income ($10,000) |
-0.027 |
-0.091 |
0.310 |
Percent
persons in poverty |
0.037 |
0.052 |
-0.297 |
Unemployment
rate |
0.064 |
-0.037 |
-0.151 |
Percentage
of manufacturing workers |
0.057 |
-0.102 |
0.036 |
Percentage
of health service workers |
-0.041 |
0.232 |
-0.168 |
Percentage
of Blacks and Hispanics |
0.010 |
-0.053 |
-0.098 |
Percentage
of AIAN |
0.020 |
0.061 |
-0.119 |
Note: The three factors can explain 50.3
percent of total variation of all variables
Using factor analysis, a typology of
eight categories was created based upon
their scores on the three dimensions (Table
13). Using this approach, the counties
with the greatest shortages were low on
all three factors (i.e., category 111),
indicating high levels of unmet need,
unmet demand, and socioeconomic disadvantage.
The counties with the least shortages
were high on all three factors (i.e.,
category 222).
Table
13. Numbers and Percentages of Counties
in Factor Analysis Categories, by Census
Division
Category(a) |
East
North Central |
70 |
14 |
40 |
23 |
50 |
20 |
111 |
34 |
75 |
437 |
16.0% |
3.2% |
9.2% |
5.3% |
11.4% |
4.6% |
25.4% |
7.8% |
17.2% |
100% |
East
South Central |
80 |
38 |
22 |
32 |
20 |
71 |
35 |
46 |
20 |
364 |
22.0% |
10.4% |
6.0% |
8.8% |
5.5% |
19.5% |
9.6% |
12.6% |
5.5% |
100% |
Middle
Atlantic |
13 |
3 |
13 |
26 |
48 |
2 |
8 |
9 |
28 |
150 |
8.7% |
2.0% |
8.7% |
17.3% |
32.0% |
1.3% |
5.3% |
6.0% |
18.7% |
100% |
Mountain |
67 |
49 |
55 |
29 |
18 |
13 |
22 |
22 |
5 |
280 |
23.9% |
17.5% |
19.6% |
10.4% |
6.4% |
4.6% |
7.9% |
7.9% |
1.8% |
100% |
New
England |
4 |
1 |
2 |
2 |
25 |
1 |
2 |
4 |
26 |
67 |
6.0% |
1.5% |
3.0% |
3.0% |
37.3% |
1.5% |
3.0% |
6.0% |
38.8% |
100% |
Pacific |
25 |
20 |
30 |
10 |
12 |
18 |
29 |
15 |
5 |
164 |
15.2% |
12.2% |
18.3% |
6.1% |
7.3% |
11.0% |
17.7% |
9.2% |
3.0% |
100% |
South
Atlantic |
167 |
66 |
58 |
55 |
41 |
66 |
56 |
44 |
36 |
589 |
28.4% |
11.2% |
9.8% |
9.3% |
7.0% |
11.2% |
9.5% |
7.5% |
6.1% |
100% |
West
North Central |
147 |
34 |
50 |
62 |
89 |
21 |
35 |
97 |
83 |
618 |
23.8% |
5.5% |
8.1% |
10.0% |
14.4% |
3.4% |
5.7% |
15.7% |
13.4% |
100% |
West
South Central |
103 |
102 |
22 |
52 |
18 |
72 |
30 |
58 |
12 |
469 |
22.0% |
21.8% |
4.7% |
11.1% |
3.8% |
15.4% |
6.4% |
12.4% |
2.6% |
100% |
Total |
676 |
327 |
292 |
291 |
321 |
284 |
328 |
329 |
290 |
3,138 |
21.5% |
10.4% |
9.3% |
9.3% |
10.2% |
9.0% |
10.4% |
10.5% |
9.2% |
100% |
Note: (a) An example
of how to interpret the category: 121
means F1<median, F2>median, F3<median
This analysis showed promise in theory,
but was based on primary care utilization,
with no basis for examining long-term
care, home health care, or public agency
services, and no way of reflecting variations
in staffing intensity across types of
care. While acute care hospitals are the
primary driver of RN demand, the focus
on hospital care does not make this method
applicable to counties without hospitals.
|