October 2016
Does ACO Adoption Change
the Health Workforce
Configuration in U.S.
Hospitals?
AUTHORS:
Avi Dor, PhD
Patricia Pittman, PhD
Clese Erikson, MPAff
Roberto Delhy, MA
Xinxin Han, MS
Mullan Institute Health
Workforce White Paper No. 2
Prepared By
The George Washington University
Fitzhugh Mullan Institute for Health Workforce Equity
Questions
For questions regarding this report, please contact
Patricia Pittman at ppittman@gwu.edu.
Suggested Citation
Dor A, Pittman P, Erikson C, Delhy R, Han X. Does ACO Adoption Change the Health
Workforce Configuration in U.S. Hospitals?. Washington, DC: Fitzhugh Mullan Institute for
Health Workforce Equity, George Washington University; 2016.
Funding
This white paper was supported by the Bureau of Health Workforce (BHW), National
Center for Health Workforce Analysis (NCHWA), Health Resources and Services
Administration (HRSA) of the U.S. Department of Health and Human Services (HHS) as
part of an award totaling $450,000, with zero percent financed with non-governmental
sources. The contents are those of the author[s] and do not necessarily represent the
official views of, nor an endorsement by HRSA, HHS, or the U.S. Government.
1
October 1, 2016
f
Does ACO Adoption Change the Health Workforce
Configuration in U.S. Hospitals?
Table of Contents
INTRODUCTION ............................................................................................................................................. 2
JOBS IN PREMIER DATA ................................................................................................................................. 3
JOBS IN AHA DATA ........................................................................................................................................ 4
CONCLUSIONS ............................................................................................................................................... 6
REFERENCES .................................................................................................................................................. 9
List of Tables
Table 1: Hospital staffing by ACO status, 2014 ........................................................................................... 11
Table 2: Distribution of Shared Savings Programs in 2013 and 2014 ......................................................... 12
Table 3: Changes in Hospital staffing by ACO status, 2014......................................................................... 13
2
October 1, 2016
f
INTRODUCTION
An Accountable Care Organization (ACO) refers to a group of physician and other healthcare
providers and suppliers of services, notably hospitals that form a collaborative network in order to reduce
costs while improving quality of inpatient care, and thereby meet contractual requirements and targets
set by third-party payers. Although the number of ACOs nationwide has been growing rapidly since they
were first recognized in Medicare regulations in 2011, little is known about the way delivery systems
adjusted to the change, and specifically about how those changes affect the health care workforce in
hospitals).1 The purpose of this report is to provide a descriptive analysis of workforce differences
between hospitals that participate in an ACO and those that do not.
The most well-known type of ACO is the Medicare Shared Risk Program (MSPP). However, recent
surveys suggest that commercial ACO contracts are rising rapidly. It is estimated that by the end of 2015
the number of lives covered by commercial ACO contracts was more than double the number of lives in
Medicare ACOs – 17.2 and 8.3 million, respectively.2
Nearly all Medicare ACOs opted for one-sided risk contracts offered by CMSi, whereby they would
share savings with CMS if costs of the patient pool are below some threshold payment level, up to 50% of
the spending difference. In addition, in order to be eligible to share in any savings generated, an ACO must
meet the established quality performance standard that corresponds to its performance year.3
One idea put forth by experts was that to ensure better outcomes ACO hospitals would likely
increase primary care clinicians, including nursing care staff.4 The premises was the belief that an increase
in staffing would help ACOs to adjust to the new regulations and standards arising from the
implementation of the ACA, and would help them gain legitimacy and credibility among payers and
patients.4
Extending beyond the immediate transition of care, hospital engagement in care management of
complex patients is associated with reduced readmissions.5 Given that a relatively small portion of the
population accounts for the majority of health care expenditures, identifying high risk patients, including
patients with social service or behavioral health needs, and focusing care on that population is an
important cost containment strategy.6,7 This can take the form of developing disease registries to track
patients with one or more chronic diseases8 to complex algorithms to prospectively identify patients using
claims data and other patient information.9 The personnel handling these activities may be listed as data
analysts, or they may have other titles such as medical assistants.1
While all of these activities would suggest an increase in nurse staff, as well as, possibly, care
coordinators, data analysists, pharmacists, and others, the manner in which payments are determined
may actually provide a disincentive for hospitals to spend on labor once they begin the program. Douven
and colleagues point out that benchmark calculations of payment include the last three years of providers’
spending, but that it is the most recent year that carries the greatest weight.10 Thus, the incentives to
increase spending are strongest in the last year prior to ACO adoption. They argue that providers that
incur the greatest costs during the year before entering or renewing an ACO arrangement are actually
rewarded, while providers with the lowest cost during the last year are in fact penalized.22 Under these
conditions, it is possible that joining an ACO would have the effect of reducing previously inflated staff
levels in high-cost and high-benchmark ACOs.
Given the conflicting effects that ACO adoption may have on nursing and other staff hiring,
coupled with the fact that incentives in commercial ACOs are not known, we believe that an exploratory
analysis of staffing in ACO programs is warranted.
i Only 2% of ACOs opted for the two-sided risk model, where they split both losses and gains with CMS
3
October 1, 2016
f
To conduct this preliminary analysis, we used two different data sources as follows:
•
First, we used an operations database maintained by Premier to examine a set of jobs that
might be related to care coordination to see if ACO hospitals differed from non ACO hospitals
in their use of each type of personnel. This analysis was limited to a cross-sectional comparison
for 2014, the only year for which Premier had tagged ACOs.
•
Second, we used the American Hospital Association’s (AHA) Survey of Care Systems and
Payment to identify ACOs, and the AHA’s regular annual survey to examine nurse staffing
ratios in the two types of hospitals in 2013 and in 2014, the two years for which data is now
available.
The methods and findings for each of these analyses is further detailed below in each section. We
conclude with a section that lays out the next steps in this program of research and describes what we
have learned about the relative strengths of our data sources for workforce research.
JOBS IN PREMIER DATA
Methods & Data
To examine the extent to which becoming an ACO, and the increased use of care coordination as
documented above, might lead to new jobs or changes in the configuration of staff in hospitals, we
compared hospitals in 2014 that were ACOs to those that were not. We used an operational database
maintained by Premier Inc. that tracks labor hours, hospital units, and facility characteristics.
A total of 317 unique hospitals were included in 2014 data. The ACO flag variable was linked by
Premier from American Hospital Association (AHA) 2014 Annual Survey. The sample included 135
hospitals that were ACOs, and 182 hospitals were non-ACOs.
Measures
Based on a review of job titles in the Premier data, we identified 16 jobs that might be affected
by ACO status. These jobs included advanced practical registered nurse (APRN), physician assistant (PA),
registered nurse (RN), licensed practical nurse (LPN), unlicensed assistive personnel (UAP), nurse assistive
personnel (NAP), case manager and case management assistant, managed care coordinator, risk
management, patient educator, social worker, medical social worker, clinical social worker, pharmacists,
and pharmacy technician.
Hospital staffing was measured by annual average number of hours worked (including only regular
and overtime hours) by each type of workforce examined. The labor hours were also adjusted by case-
mix index adjusted total patient days, a similar measure used in previous nursing studies. Compared to
full- time equivalent workers, this measure allowed us to capture the impacts of absences from work, as
well as overtime hours.
We then conducted a cross-sectional analysis of ACOs staffing and used t-tests to compare staffing
levels between ACOs and non-ACOs in 2014.
Results
As presented in Table 1, in 2014, ACOs and non ACO hospitals used similar levels of PA, UAP, NAP
and APRN staffing (0.073 vs. 0.053, p=0.216; 1.053 vs. 1.142, p=0.242; 0.989 vs. 1.105, p=0.117; 0.167 vs.
0.169, p=0.938). However, ACOs used significantly lower level of RN but higher level of LPN staffing than
non-ACOs in 2014 (3.891 vs. 4.772, p=0.000; 0.290 vs. 0.216, p=0.029).
In 2014, ACOs and non ACOs used similar levels of case manager and case management assistant
staffing (0.122 vs 0.143, p=0.088). Likewise, ACOs used similar levels of managed care coordinator, risk
management, and patient educator staffing as non-ACOs (0.011 vs. 0.016, p=0.364; 0.013 vs. 0.016,
p=0.467; 0.018 vs. 0.019, p=0.692; 0.018 vs. 0.019, p=0.692). However, ACOs used lower levels of social
4
October 1, 2016
f
workers, pharmacists, and pharmacy technician staffing than non-ACOs (0.076 vs. 0.097, p=0.028; 0.138
vs. 0.188, p=0.000; 0.169 vs. 0.198, p=0.005).
Discussion
We are cautious in our interpretation of these results, given that the Premier data and the AHA
ACO flag are both convenience sample. However, it is notable that there are no significant increases in
hours for care coordinators, case managers, patient educators, or risk managers, as might have been
expected. It is also notable that RN and UAP hours are lower in ACO hospitals than in non ACO hospitals,
as are pharmacy and social worker hours.
These differences are the opposite of what we might have expected, given the evidence that
higher RN staffing levels improve outcomes11,12,13, and the increased interest in pharmacists for
medication management14 and social workers to help manage patients with challenges in the realm of
social determinants15. They may suggest that ACO hospitals are engaged in cost containment strategies
that include constraining staff growth.
Further analyses that includes additional years and multivariate regressions that can control for
facility and regional characteristics are needed and will be conducted during the fall of 2016.
JOBS IN AHA DATA
The purpose of this second section of the report is to continue to explore workforce differences
in ACO hospitals using the AHA Annual Survey Database, and the AHA Survey of Care Systems for the years
2013 and 2014. Henceforth we will refer to these as the 2013 and 2014 AHA and ACO surveys respectively.
From the AHA and ACO surveys, our 2013 data included information for 1,795 hospitals. Out of
these 1,795 hospitals, we excluded 358 government hospitals, leaving us with 1,250 non- governmental,
not-for-profit hospitals and 187 investor-owned for-profit hospitals. We excluded government hospitals
to date the literature does not address the role that federal/local hospitals in ACO. In later analysis we
aim to include/compare in our analyses federal/local hospitals. Out of the resulting 1,437 private
hospitals, therefore, 286 had “established or were part of an ACO”, while 853 of them did not have any
type of ACO arrangementii.
For the year 2014, both the AHA and ACO surveys included information for 1,697 hospitals. Out
of these 1,697 hospitals, we excluded 317 government hospitals, leaving us with 1,239 non-governmental,
not- for-profit hospitals, and 141 investor-owned for-profit hospitals. Out of the resulting 1,380 private
hospitals, 279 had “established or were part of an ACO”, and 526 did not have any type of ACO
arrangement.
After merging our 2013 and 2014 information of private hospitals with ACO information for both
years, our final dataset was composed by 403 hospitalsiii. Out of those, 117 hospitals were part of an ACO
in both years 2013 and 2014 (group #1 – always ACO), 268 hospitals were not part of an ACO in either
2013 or 2014 (group #2 – never ACO), 63 hospitals were not part of an ACO in 2013 but joined an ACO in
2014 (group #3) and finally 25 hospitals were part of an ACO in 2013, but reported having left their ACO
arrangements by 2014 (group #4).
With regards to the rural/urban distribution of the hospitals described above, using US Census
defined Core Based Statistical Areas, we determined that the number of rural hospitals was very small in
all groups except the never-ACO group. It included 8 rural hospitals in group #1, 104 rural hospitals in
group #2, 8 rural hospitals in group #3 and 6 rural hospitals in group #4.
ii For 298 of these hospitals, ACO status information was missing
iii Out of 1,380 non-governmental hospitals present in both years, 575 of them had missing ACO status information for either year.
5
October 1, 2016
f
In regard to the variety of shared savings programs pursued by ACO hospitals, the three most
common shared savings arrangements in 2013 and 2014 were the following: commercial payer
partnerships, Medicare shared savings programs, and joint Medicare shared savings programs together
with commercial payer partnerships (see Table 2).
As observed in Table 2, the distribution of ACO participating hospitals in different shared savings
programs is complex and dynamic. A significant number of hospitals entered and exited the existing
shared savings programs just within the two-year observed period. Additionally, hospitals may have
multiple shared savings programs which can also change from year to year.
For the workforce analysis, we consider only hospitals in groups #1 (always ACO) and #2 (never
ACO) as described above. We excluded the two groups of hospitals were ACO membership status changed
between 2013 and 2014, because of small sample sizes which would prevent us from obtaining
meaningful statistical comparisons.
We focus on the number of staff hours per adjusted patient day. Total patient days are adjusted
by the hospital level proportion of inpatient and outpatient revenue, per AHA’s methodology, and we
then also adjusted for CMS’ yearly case mixed indexes (CMI).
We use a modified measure of full-time equivalent employment developed by Spetz and
colleagues.16 They assume that productive hours per year are fewer than 2,080 and instead use 1,768
hours per year; this is equivalent to an 85% productive level over 52 weeks per year at 40 hours per week.
Based on staffing data available from the AHA, we focused this analysis on various levels of nursing
staff: registered nurses (RNs), licensed practical nurses (LPNs), nursing assistive personnel (NAP) (the term
used in the AHA survey), and advanced practice nurses (APNs). As discussed in the introduction, there is
extensive research demonstrating that higher RN staffing levels result in better outcomes. LPN and APN
staffing levels may affect RN workload and therefore may also affect outcomes. APNs do various types of
tasks in hospitals, and we know of no research suggesting APN staffing levels bear a relationship with
outcomes. We include them simply as an additional point of interest that could be explored in future
research.
Thus, for each type of nurse we use the following formula in order to calculate the number of
hours per CMI adjusted patient day:
After calculating the average number of nursing hours for RNs, LPNs, NAPs and APNs, we then
proceed to compare the absolute difference, as well as the percentage change across occupations among
ACO and non-ACO hospitals between 2013 and 2014.
Results
As presented in Table 2, when comparing the baseline (2013) staffing levels for ACOs and non
ACOs, we find that non ACOs have higher levels of LPNs and APNs. Given the higher proportion of rural
hospitals in the non ACO group, this is not surprising. We know that nationwide employment of LPNs in
hospitals is falling, although it is most dramatic in urban areas. We also know that employment of APNs is
rising across the country, but especially in rural areas.17
In comparing changes in the two groups from 2013 to 2014, we find that the average number of
RN and APN hours has fallen in the always-ACO group and that the changes are statistically significant.
This finding was surprising, as discussed below, given the idea that ACOs may be more advanced in the
organization of care than non ACOs.
6
October 1, 2016
f
We also find that LPN and NAP hours fell in both groups, while APN hours increased significantly
in both groups, especially in the always-ACO group.
Discussion
We are again cautious not to interpret our results as a definitive study of labor patterns among
ACO/non-ACO hospitals, given the potential for sample selection bias and the fact that this is simply a
descriptive analysis. Nevertheless, findings relating to the small drop in RN and the large increase in APN
hours per patient day in ACOs echo our findings using Premier data for just one year (2014).
CONCLUSIONS
We were surprised that not only did there not appear to be major ACO related differences in the
workforce either cross sectionally or when comparing changes across two years in the “always ACO” and
“never ACO” groups. Indeed the early signs of directionality of change among those with ACO status are
the opposite of what we would have anticipated. These two descriptive analyses suggests the need for
further research with longer time series (more data), as well as, adjustments that take into account other
important characteristics of hospitals.
ACO’s are required to report quality outcomes and their payment is linked to results. Indeed, we
know that ACOs that participated in the Medicare Shared Savings program in both 2013 and 2014
improved on 27 of 33 measures.18
Given, increased interest in pharmacists for medication management19 and social workers to help
manage patients with challenges in the realm of social determinants20 makes the lower staffing levels of
these groups of professionals in ACO hospitals unforeseen.
Even more surprising was the reduction in RN and APN hours among ACO hospitals, as compared
to non ACO hospitals, given that there is such a strong body of research suggesting higher nurse staffing
results in better outcomes. Moreover, nurse organizations, especially labor unions, have been advocating
for various types of mandatory and voluntary nurse staffing laws. Indeed, California and Massachusetts
now have mandatory laws, and 14 other states have either public reporting or staffing committee
requirements that aim to push hospitals towards higher nurse staffing levels.21 At the same time, recent
research conducted by this team (forthcoming in HSR) using Premier data also shows a nationwide decline
in RN and nurse support staff hours, suggesting that something important may be occurring.
With regard to reduced or constrained staffing, possible explanations to be explored in further
research include the following:
•
A suggested by Douven and colleagues, the existing ACO payment formula may lead hospitals
to reduce spending just after joining ACOs. This may be due simply to the savings incentives,
or it may be linked to the payment calculations which give the greatest weight to the year
prior to ACO adoption.
•
Hospitals in ACOs are expanding their market power and, therefore, may be less concerned
about public reputation regarding nurse staffing, and perhaps even their nurse related quality
outcomes;
•
ACO hospitals are identifying ways to reduce labor costs without affecting outcomes;
•
The reduction in RN staff could be related to the retirement of older nurses, and hospitals in
certain regions may be having difficulty replacing them.
•
It is possible that ACO hospitals are redeploying some RNs and support staff to other settings
with partner organizations and they are therefore not captured in either data source.
The next steps in our research will be to use a longer longitudinal data set and multivariate
regression analyses to control for a variety of facility and regional characteristics to ensure that our
preliminary findings are robust. Among the control variables that should clearly be included in our next
7
October 1, 2016
f
analysis are rurality, local nurse supply, hospital market share, types of shared savings programs, and
percent of patients covered under shared savings.
Reflections on the Strengths and Weaknesses of our Data Sources
One of the important objectives of this phase of our research is to fully comprehend these new
data sources for workforce research. In the paragraphs that follow, we reflect on the relative strengths
and weaknesses of Premier and AHA data sources that we have identified to date.
Premier Data
The Premier operational database captures information on more than 500 Premier membership
hospitals that cover healthcare systems in all 50 states and the District of Columbia. It includes basic
facility characteristics, department codes and descriptions, job titles and descriptions, and staffing
information such as labor hours, expenses, and skill-mix category. GW has purchased 2010-2014 and 2015
will be provided to us soon. Premier collects clinical, financial, pharmacy, supply chain, and operational
data from its member hospitals on a daily, biweekly, monthly, or quarterly basis. The data provide a
unique opportunity to track hospital-based workforce, and in this instance it allows us to identify a variety
of job titles across different hospital departments and specialty areas.
Hospital staffing was measured by calculating the annual number of total worked hours (including
regular and overtime hours) for each of the selected jobs, adjusted by case-mix index adjusted total
patient days. Compared to full-time equivalent workers, this labor hour measure allows us to capture the
impacts of absences from work, and thus may reflect the actual hours that workers spend on assisting
clinical tasks. Premier was able to link the ACO flag from AHA for us (we are not provided access to
provider identifying numbers), allowing us to compare the differences in care coordination related
workforce in hospitals.
The weakness of the Premier data is the small sample size, which may affect the statistical power
to detect statistical differences. That is to say, while some of our results did not have significant
differences, the magnitude of the estimates still provide practical implications. In addition, Premier’s
member hospitals are essentially a convenience sample of all U.S. hospitals and thus may not necessarily
be representative of all U.S. hospitals. However, Premier’s hospitals characteristics are still similar to the
characteristics of U.S. community hospitals as reported by other national hospital databasesiv suggesting
that our findings are likely to reflect hospital staffing trends nationwide. Lastly, the ACO flag linked from
AHA indicates ACOs, being part of ACOs, however, it also indicates hospitals that are not ACOs but are
actively thinking to become an ACO as the ACO flag. This may affect the accuracy of our estimates. Future
study may need to exclude these hospitals out.
American Hospital Association (AHA) data
The AHA Annual Survey Database (ASDB) covers all U.S. community health hospitals, and the
response rates are high in each year. GW has purchased 2009-2014. This annual survey contains over
6,300 hospitals and almost 1,000 fields of information in the following categories: organization structure,
facility characteristics, inpatient and outpatient utilization, staffing, and geographic indicators. Due to its
reliability across time, the data are used for a variety of purposes. They are seldom used, however, to
examine the hospital workforce. Two previous works we found are using AHA data as regression
controls.22,23 The AHA annual survey provides data on number of FTE for a smaller number of jobs than
iv The proportion of teaching hospitals and average occupancy rate in our dataset is comparable to the national average, while our sample
consists of a larger portion of not-for-profit, urban, and system-affiliated hospitals and hospitals with more staffed beds and admissions as
compared to the national sample from the American Hospital Association Annual Survey and Healthcare Cost and Utilization Project data.
8
October 1, 2016
f
Premier. These include physician, dentists, medical residents, nurses, nurse assistive personnel,
technicians, pharmacists, therapists, other clinical professionals, and support personnel.
The AHA Survey of Care Systems and Payment was established in 2013 to understand payment
arrangements. GW purchased 2013-2015. The most important component of this survey is its focus on
Accountable Care Organization (ACO) participation and structure. The survey identifies which hospitals
are participating in ACOs, or are actively thinking to become an ACO. Using this information, we are able
to link the AHA annual data workforce variable to the ACO flag created from the ACO survey and examine
staffing variation by hospital ACO status.
In 2013, the survey was sent to all registered community hospitals (4,999) and received 1,517
responses from the field. Of these, 309 hospitals indicated they are part of an ACO. The respondent profile
for the overall survey was broadly representative of the universe of U.S. hospitals, as indicated by the
AHA.24
Both of the AHA surveys have limitations. First, the AHA ASDB does not distinguish between
outpatient and inpatient registered nurses (RNs). Second, surveyed hospitals may use different definitions
to calculate the number of FTE workers. Thirdly, when we convert the number of FTE nurses to annual
nursing hours, we use a formula previously used by Spetz and colleagues in which they assume that one
FTE nurse would work for 1,768 hours per year. Under this formula, potential work hours equal to 52
weeks per year at 40 hours per week and actual productive hours equal to 85% of potential hours.71 This
method yields similar results to Premier data.
Lastly, the ACO surveys’ response rates are relatively low, resulting in the possibility of selection
bias. In other words, hospitals that chose to answer the survey may differ from hospitals that did not
answer it in ways that we do not understand. Future research will need to account for this issue.