Outstanding
Issues Related to the Preferred Method
There
were a number of outstanding issues related
to the Preferred Method that required
a closer look. While fully addressing
some of these issues was beyond the scope
and timeframe of the current study, limited
analyses were performed to investigate
some potential avenues for improving the
Preferred Method to address these shortcomings.
Two of the most important analyses are
summarized below.
A.
The Problem of Patient Acuity
Study staff did attempt to correct for
patient acuity in hospital settings because
hospital patient acuity will differ across
counties in ways that may systematically
disadvantage counties with major medical
and trauma centers. ARF data was used
to measure the number of surgeries and
to estimate percent of inpatient days
spent in the ICU. NSSRN data was used
to estimate the number of RNs working
in ICU units and operating rooms, and
the steps in the earlier model were followed
to obtain estimates of demand for operating
room RNs and ICU RNs calculated separately
from other hospital RNs. It was a problem,
however, that many counties lacked accurate
ICU bed data (especially large urban counties).
Perhaps, as a result, this adjustment
did not have the expected effects on RN
demand estimates. In fact, it often resulted
in lower rather than greater estimated
demand for RNs.
Figure
8. Estimated RN Shortage Percentages for
Counties in the U.S.
[D]
This additional adjustment for ICU and
surgical services is theoretically important
in that it is one of the few possible
adjustments for acuity. While the quality
of currently available data may prohibit
incorporating the adjustment into a national
methodology, hospitals that feel that
their higher patient acuity has disadvantaged
them in the standard process could potentially
submit ICU and surgical data through an
application and appeals process.
B.
RN Commuting Patterns
The original version of the Preferred Method
assumed that RN commuting patterns were
similar to those of the overall workforce.
This is generally true in the aggregate—RNs
are no more or less likely than other
workers to work outside the county where
they live. At the county level, however,
RN commuting patterns sometimes varied
dramatically from the patterns for all
workers. Additional analyses summarized
below revealed that RN commuting patterns
depended more on county characteristics
than on characteristics of RNs (e.g.,
gender, income level, etc.). These analyses
were based on data from New York, North
Carolina, and Mississippi, states for
which data were available on both county
of residence and county of employment.
1.
Models to Predict RN Commuting Patterns
To help identify the factors related
to RN commuting patterns, models based
on county characteristics were developed.
The commuting patterns of all workers
had been, on average, a good proxy for
the commuting patterns of RNs, so this
was retained as one independent variable.
This variable should reflect many of the
primary drivers of commuting behavior
(e.g., relative wages, cost of living,
etc.). Another independent variable was
assumed to be opportunities for RN employment
in a particular county. This was measured
by the extent to which the number of RNs
living in a county compared to the estimated
demand for RNs in that county (based on
infrastructure and service use). Counties
where resident RNs were in short supply
relative to service use were expected
to be net importers of RNs, while counties
where resident RNs were more than sufficient
for the county’s health care needs were
expected to be net exporters of RNs.
Other factors included in the analysis
were whether the county was a whole-county
HPSA, the county’s major industry, and
whether the county was a persistent poverty
county. The rural-urban characteristics
of the county (population size, proximity
to a metropolitan area) were also accounted
for, although these did not prove as crucial
as expected (probably because they did
not affect RN commuting any differently
than overall commuting, which was already
controlled for).
The intercept for the model was 0.495,
indicating that if all other variables
had a zero value, each resident RN would
be equal to 0.495 RNs working in the county.
The coefficient for overall commuting
was 0.601, indicating that for every one
percent increase in net incommuting, there
would be a 0.601 unit increase in RN incommuting.
The supply of resident RNs relative to
estimated demand was negatively related
to net incommuting (-0.148).
The percent of the population living
in an urban area within the county was
positively related to RN incommuting (0.001),
but this was not statistically significant
(p=0.059). For every increase of 10,000
population, RN incommuting increased by
0.0012. Whole-county HPSA status decreased
net RN incommuting (-0.157), as did persistent
poverty county status (-0.158) and dependence
on manufacturing (-0.09).
There was an interesting interaction
effect between population size and persistent
poverty status: being a persistent poverty
county had a greater depressant effect
on RN incommuting in small population
counties than in large population counties.
This model had an adjusted R2
of 0.702.
Table 15 shows the results of models
estimated separately for groups of counties
based on their relationship to a metropolitan
area (part of a metropolitan area, adjacent
to a metropolitan area, or not adjacent
to a metropolitan area). Although the
results show some potential to fine-tune
the RN incommuting estimates for different
groups of counties, the differences in
the model coefficients were not dramatic.
Differences in model fit were substantial,
however. The model for counties not adjacent
to a metropolitan area was the best fitting
model (adjusted R2 = 0.842).
The model for metropolitan counties also
fit well (R2 = 0.805). The
model for non-metropolitan counties adjacent
to metropolitan areas, however, explained
less variation (R2 = 0.509).
Table
15. Ordinary Least Squares Regression
Coefficients Predicting RN Incommuting,
By Type of County
0.495*** |
0.059 |
0.545*** |
0.089 |
0.357* |
0.167 |
0.536** |
0.165 |
0.601*** |
0.050 |
0.563*** |
0.057 |
0.559** |
0.187 |
0.664*** |
0.146 |
-0.148*** |
0.017 |
-0.221*** |
0.034 |
-0.094** |
0.025 |
-0.227*** |
0.043 |
0.001 |
0.001 |
0.0009 |
0.001 |
0.003 |
0.002 |
0.003 |
0.001 |
-0.157*** |
0.037 |
-0.117 |
0.059 |
-0.153* |
0.068 |
-0.151** |
0.046 |
-0.009** |
0.028 |
- |
- |
- |
- |
-0.134**
|
0.041
|
-0.158** |
0.053 |
-0.287* |
0.133 |
- |
- |
-0.100 |
0.058 |
0.001* |
0.000 |
0.003 |
0.000 |
- |
- |
- |
- |
- |
- |
0.050
|
0.065
|
- |
- |
0.100* |
0.046 |
- |
- |
- |
- |
0.361** |
0.118 |
- |
- |
- |
- |
- |
- |
-0.215 |
0.117 |
- |
- |
0.033** |
0.000 |
0.049* |
0.000 |
- |
|
- |
- |
- |
- |
-0.002
|
0.000
|
- |
|
- |
- |
- |
- |
- |
- |
- |
|
-0.003 |
0.002 |
0.702 |
0.805 |
0.509 |
0.842 |
* p < 0.05
** p < 0.01
*** p < 0.001
Interestingly, the most accurate method
for estimating commuting varied by county
type. In metro counties, the commuting
flow of all workers was the most accurate
estimate of the three models 39% of the
time; while in counties adjacent to metro
areas, the model for all counties was
the most accurate 47% of the time; and
in counties not adjacent to metro areas,
the best estimate was the RUCC-specific
estimate 51% of the time.
In many cases, however, the “best” estimate
was better than the “next best” estimate
by only a point or two. When the variable
used to evaluate was the percent of the
time that an estimate differed by more
than 10% from the actual RN commuting
value, the all-county estimate was accurate
more often for metro and adjacent-to-metro
counties, while non-adjacent-to-metro
counties did best when the RUCC-specific
estimate was used. It was never preferable
to use the overall commuting pattern.
Table
16. Percentage of Cases in Which Estimate
Differs From Actual by More Than 10%
Metro
County (93) |
Adjacent
to Metro County (86) |
Not
Adjacent to Metro County (65) |
Total |
(244) |
2.
Using Commuting Patterns to Estimate RN
Supply
Because the evaluation of the estimates
using data from the counties from which
they were derived was somewhat tautological,
it was decided to assess whether these
corrections brought estimates of RN employment
by county closer to actual employment
data in other states (for which real commuting
patterns were not available). The states
used in this preliminary validation process
were Tennessee, Texas, Pennsylvania, South
Dakota, and some counties in Iowa.
When compared to actual counts of RNs
working in particular counties, the revised
commuting adjustments did little to improve
the supply estimates. The estimated supply
was closer to the actual supply on average
when overall commuting was used as the
adjustment factor. There was some variation
by the Rural Urban Classification Code
(RUCC): the estimate of commuting based
on the all-county model produced somewhat
lower average differentials than other
estimates for counties adjacent to metro
areas, and somewhat lower absolute average
differentials for counties not adjacent
to metro areas.
It is important to remember that the
accuracy of the commuting estimates is
only one source of error in estimated
supply of RNs working in a county. Another
source of error is in the estimated numbers
of RNs living in the counties in which
they work. It may be possible in future
work to estimate confidence intervals
around commuting estimates and estimates
of supply of resident RNs, and for shortage
designations to be based on the lower
of the two estimates of RNs based on the
two confidence intervals. This would remove
some of the disadvantage potentially faced
by rural areas due to sampling errors,
although it would increase the likelihood
of designating some counties that should
not qualify.
|