Human Capital and Productivity in a Team Environment: Evidence from the Healthcare Sector Ann P. Bartel Nancy Beaulieu Ciaran S. Phibbs Patricia W. Stone* July 2013 Abstract Using panel data from a large hospital system, this paper presents estimates of the productivity effects of human capital in a team production environment. Proxying nurses’ general human capital by education and their unit-specific human capital by experience on the nursing unit, we find that greater amounts of both types of human capital significantly improve patient outcomes. Detailed data on team composition enables us to model productivity effects of team disruptions caused by the departure of experienced nurses, the absorption of new hires, and the inclusion of temporary contract nurses. These disruptions to team functioning are associated with significant decreases in productivity beyond those attributable to changes in nurses’ skill and experience. ___________________________ *Ann Bartel: Columbia University, NBER and Graduate School of Business, 623 Uris Hall, 30122 Broadway, New York, NY 10027, apb2 @ columbia.edu. Nancy Beaulieu: NBER, [email protected]. Ciaran S. Phibbs: VA Health Economics Resource Center and Stanford University School of Medicine, Health Economics Resource Center, Veterans Affairs Medical Center,795 Willow Road, Menlo Park, California 94025, [email protected]. Patricia W. Stone: Columbia University School of Nursing, 617 W. 168th St, Room 228, New York, NY 10032, [email protected]. The authors gratefully acknowledge the generous support of a grant from The Robert Wood Johnson Foundation, institutional support from the VA HSR&D program, outstanding research assistance from Lakshmi Ananth, Bruno Giovannetti, Cherisse Harden, Cecilia Machado, Raymond Lim, Susan Schmitt, Andrea Shane, and Anukriti Sharma and helpful comments from the referees of this journal and seminar participants at the NBER Summer Institute, the American Society of Health Economists Annual Meetings, MIT Sloan, Columbia University Mailman School of Public Health, and CUNY Graduate Center. The views expressed in this manuscript are those of the authors and do not necessarily reflect the position or policy of the Department of Veterans Affairs or the United States government.
48
Embed
Human Capital and Productivity in a Team Environment ... · PDF fileHuman Capital and Productivity in a Team Environment: Evidence from the Healthcare ... capital by experience on
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Human Capital and Productivity in a Team Environment:
Evidence from the Healthcare Sector
Ann P. Bartel
Nancy Beaulieu
Ciaran S. Phibbs
Patricia W. Stone*
July 2013
Abstract
Using panel data from a large hospital system, this paper presents estimates of the productivity
effects of human capital in a team production environment. Proxying nurses’ general human
capital by education and their unit-specific human capital by experience on the nursing unit, we
find that greater amounts of both types of human capital significantly improve patient outcomes.
Detailed data on team composition enables us to model productivity effects of team disruptions
caused by the departure of experienced nurses, the absorption of new hires, and the inclusion of
temporary contract nurses. These disruptions to team functioning are associated with significant
decreases in productivity beyond those attributable to changes in nurses’ skill and experience.
___________________________
*Ann Bartel: Columbia University, NBER and Graduate School of Business, 623 Uris Hall, 30122
Broadway, New York, NY 10027, apb2 @ columbia.edu. Nancy Beaulieu: NBER,
[email protected]. Ciaran S. Phibbs: VA Health Economics Resource Center and Stanford
University School of Medicine, Health Economics Resource Center, Veterans Affairs Medical Center,795
Willow Road, Menlo Park, California 94025, [email protected]. Patricia W. Stone: Columbia
University School of Nursing, 617 W. 168th St, Room 228, New York, NY 10032,
[email protected]. The authors gratefully acknowledge the generous support of a grant from The
Robert Wood Johnson Foundation, institutional support from the VA HSR&D program, outstanding
research assistance from Lakshmi Ananth, Bruno Giovannetti, Cherisse Harden, Cecilia Machado,
Raymond Lim, Susan Schmitt, Andrea Shane, and Anukriti Sharma and helpful comments from the
referees of this journal and seminar participants at the NBER Summer Institute, the American Society of
Health Economists Annual Meetings, MIT Sloan, Columbia University Mailman School of Public Health,
and CUNY Graduate Center. The views expressed in this manuscript are those of the authors and do not
necessarily reflect the position or policy of the Department of Veterans Affairs or the United States
government.
1
We provide new insights into an important, but under-studied, factor that
shapes the cost and quality of healthcare in the United States – the structure and
composition of nursing teams on acute care hospital units. The number of micro-
econometric studies of the productivity of health care delivery is small relative to
the large and expanding role that the health care sector plays in the American
economy. Nursing care is a frequently overlooked but critical factor of health care
productivity.1 While doctors make the majority of decisions about when and how
to treat patients, nurses fill a pivotal role in implementing treatment plans.
Moreover, nurses monitor the progress of their patients, facilitate the frequent
adjustments that customize treatments to the evolving needs of individual
patients, and coordinate care delivery. These actions, in turn, speed recovery,
economize on resources, and enhance patient satisfaction. Importantly, nurses
work closely with patients and family caregivers to encourage them, and to help
patients understand their treatment so they may play an active role in their care.
Using monthly data from the Veterans Administration (VA) hospital
system, we study how changes in the human capital attributes of the nursing team
impact patient outcomes. We are able to identify when new nurses join the team
and when experienced nurses depart, and also observe whether nurses on the unit
are regular staff members or agency nurses (who are not part of the regular
nursing team) contracted to cover for absences by regular staff nurses. Unlike
other contexts in which teams are endogenously formed (for example, Bandiera,
Barankay, and Rasul, forthcoming), hospital nurses are assigned to existing units
as vacancies become available or staff expands. They are compensated based on
their seniority and credentials, and do not receive individual or group incentive
pay. Month-to-month variations in characteristics of the nursing staff (education,
1 However, see the study of British hospitals by Propper and Van Reenen (2010) who found that
higher outside wages for nurses significantly worsened hospital quality because it was harder to
attract and retain skilled nurses.
2
hospital experience, unit experience, contract status) result from absences
(vacations, sick days, personal leaves), separations (turnover and retirement) and
new hires. We focus on how these changes in the composition of the nursing
team impact productivity.
We base our productivity measure on the length of time patients stay in
the hospital. Length of stay (LOS) is a relatively inclusive proxy for the cost and
quality of a hospital episode of care. To control for variations in patient severity
of illness, we compute each patient’s residual length of stay as the difference
between his actual length of stay and his expected length of stay; the latter
measure is based on the patient’s admitting diagnoses and other characteristics.
The VA data are unique in that they link each patient to the nursing units
in which he was actually treated. 2
This feature of the data enables us to relate
changes in the composition of the nursing team within individual hospital units to
changes in residual length of stay for patients on those same units. By estimating
this relationship in a fixed effects framework (with hospital unit fixed effects), we
base our identification on within-hospital unit changes over time. Annual unit
fixed effects control for any characteristics of the nursing unit that might
influence patient outcomes and which are unlikely to vary within a year.
In this econometric framework, one concern might be that the nurse
staffing changes are endogenous (e.g. that nurses change their labor supply in
response to the quality of care on the unit). We show that monthly mobility
between units, and separations from the VA, are not correlated with patient
outcomes on the unit. Another possible concern is that management may adjust
nursing staff based on unit performance, reallocating staff from well-performing
2 Another advantage of the VA data is that all the hospitals belong to the same umbrella
organization with data collection standardized across member hospitals, ensuring that variable
definitions and data coding algorithms are identical across the nursing units in our study. The
major difference between patients in VA hospitals and patients in non-VA hospitals is that the
former do not include children and are less than 10% female.
3
units to poorly performing units. To address this concern we show that the rate of
transfers between like units is less than 1 percent. Furthermore, restricting our
analysis to units that are the only one of their type in the hospital (thereby
lessening the likelihood of internal transfers), leaves our regression results
unchanged. In sum, our unique monthly data enable us to provide convincing
estimates of the impact of various dimensions of nurse human capital on patient
outcomes.
We find that higher levels of general human capital and specific human
capital among nurses on the unit are associated with shorter patient stays in the
hospital. The degree of specificity of the registered nursing staff’s human capital
is shown to be an important determinant of patient outcomes; while unit-level
tenure is significant, the effect of a nurse’s hospital tenure outside of the unit is
insignificant. Further evidence of the importance of specific human capital is that
staffing by contract nurses does not improve patient outcomes; while the presence
of a contract nurse increases staffing intensity, these additional resources are not
productive in improving patient outcomes.
A unique feature of our study is that we are able to model human capital in
ways that are different from previous studies.3 The essence of team production is
that it involves interaction among team members, typically of the sort involving
communication, knowledge sharing, and coordination. When experienced teams
are disrupted by the absence of a key member, the presence of an outsider, or the
addition of a new member, these activities that manage interdependencies are
likely to be impaired. We find evidence of negative productivity effects when
3 Unlike organizational level studies that relate aggregate human capital measures of the
workforce to firm-level outcomes (Fox and Smeets 2011; Black and Lynch 2001), our study is
more closely related to studies of peer effects. See Mas and Moretti (2009) on the impact of
monitoring by more productive peers and Chan, Li, and Pierce (2011) on the role of informal
teaching done by an experienced salesperson who is co-located with an inexperienced salesperson
.
4
nursing teams are disrupted by the departure of experienced nurses or the
absorption of new hires.
The paper is organized as follows. Section I describes the hospital setting,
the relevance of various dimensions of nurse human capital, and the VA dataset.
Section II describes our empirical strategy and addresses potential challenges to
the exogeneity of our measures of monthly changes in nurse staffing. Regression
results, including a number of robustness checks, are presented in Section III.
Section IV concludes.
I. The Context and Data
A. Nurse Staffing in Hospital Units
Hospital patients are assigned to nursing units based on the type of care
they require (e.g. acute care units such as medical, surgical, neurology, oncology,
cardiac care, and intensive care units). In our sample of acute care units, there is
an average of sixteen patients who are cared for by a team of three registered
nurses (RNs) on a given shift (or eight to nine RNs on a given day). Registered
nurses are assisted by licensed practical nurses (LPNs) and unlicensed assistive
personnel (UAPs) (commonly referred to as nursing aides) who have less
extensive educational requirements and clinical training. LPNs are not allowed to
conduct patient assessments or care planning or administer intravenous
medications. UAPs are restricted to very basic patient tasks. Units are managed by
nurse managers.
While specific RNs are assigned primary responsibility for a patient, some
tasks, such as checking certain medications, wound care, or administering blood,
require two RNs, and, if a patient’s primary nurse is busy with another patient or
off the unit, other RNs provide help. In addition, nursing care is provided by
multiple shifts per day requiring nurses on one shift to share information
regarding a patient’s condition and treatment with nurses on other shifts. Hence,
5
the work on a nursing unit is best described as a group production process that
utilizes knowledge workers, i.e. individuals who apply their knowledge to solving
specific problems and communicating solutions to co-workers (Garicano and
Hubbard 2007).
B. Human Capital
General human capital is higher in units that have a greater proportion of
RNs compared to LPNs and UAPs, or a greater proportion of RNs with more
prior nursing experience. Since hospitals often use their own systems, policies,
procedures and protocols, RNs acquire knowledge and skills that may be specific
to the hospital in which they work.4 Within a hospital, human capital can be
specific to the unit in which the RN works, because the nature of care that patients
require differs across units and because unit managers are free to establish their
own norms and work processes. Survey data on RNs changing jobs (Blythe,
Baumann, and Giovannetti 2001) suggests that they do indeed acquire significant
amounts of hospital-specific and unit-specific human capital.
Although the licensing requirements are the same for VA and non-VA
RNs, the VA RN workforce is older, slightly more educated, more ethnically
diverse and has a larger proportion of males than the non-VA RN workforce
(National Commission on VA Nursing, 2004). The VA pegs the wages of its RNs
to the wages of RNs in non-VA facilities in the local labor market (Staiger, Spetz,
and Phibbs 2010).
In order to provide adequate nurse staffing at all times, hospitals use
overtime as well as temporary agency contracts (50% of the units in our sample
employ contract nurses at some point during the study period). Under an agency
contract, the RN is employed by another firm (an agency) but provides nursing
services on site at the contracting hospital for a fixed period of time, ranging from
4 For evidence of hospital-specific human capital for cardiac surgeons, see Huckman and Pisano
(2006). .
6
one day to 13 weeks. Contract nurses receive little or no orientation training and
are typically brought into the unit on very short notice; they are likely to be
unfamiliar with the procedures, practices and equipment in the unit as well as with
their nursing colleagues, and are therefore expected to have less specific human
capital than regular staff RNs. 5
Individual knowledge and skills specific to a production process in a
particular location is one commonly studied dimension of specific human capital.
Another dimension of specific human capital, particularly important in team
settings, relates to relationships among co-workers. First, relationships with co-
workers that facilitate communication and coordination are likely to generate
positive externalities when work is interdependent (Gittell, Seidner, and Wimbush
2010). In our context, this would occur when the productivity of one nurse spills
over to positively impact the productivity of a team member. Second, mentoring
of less experienced nurses by more experienced nurses has the potential for
improving performance of the team while also building human capital to improve
future performance. When the absence of an experienced regular staff nurse is
covered by either an inexperienced nurse or a contract nurse, these mentoring
activities are less likely to occur: inexperienced nurses lack the knowledge and
skills to draw on in mentoring, and contract nurses lack the incentives to mentor
and the relationships with regular staff nurses that would facilitate mentoring.
C. Data
We use data from the Veterans Administration Healthcare System which
is one of the largest healthcare systems in the U.S. with over 7.2 million veterans
enrolled for health services (National Commission of VA Nursing, 2004).
5 Gruber and Kleiner (2012) found that in-hospital mortality was higher for patients admitted
during nurse strikes when hospitals often use contract nurses to replace staff nurses. For evidence
from other sectors of the economy, see Rebitzer (1995) and Guadalupe (2003) who find that the
use of contract workers is associated with an increased incidence of work accidents and Herrmann
and Rockoff (2012) who find that replacing absent teachers with temporary substitutes negatively
impacts students’ test scores.
7
Measuring the impact of human capital on productivity in hospitals requires a
dataset that links patients to the actual nursing teams that provided their care. The
VA hospital data systems are uniquely qualified for this task. Unlike the system
used by most hospitals, the VA’s integrated accounting system (DSS) creates a
separate record for each nursing unit stay for each patient so that it is possible to
identify the nursing units in which the patient was treated during his hospital
stay.6 This dataset provides monthly data on the number of nursing hours actually
worked on each unit for each type of nursing labor tracked by the VA (RN, LPN,
UAP), the number of overtime hours by staff RNs as well as the number of
contract nursing hours charged to each unit.
The VA’s Personnel and Accounting Integrated Data (PAID) includes
employee qualifications and employment history data for all nursing staff. It is an
individual-level dataset with information on each nurse’s age, education, prior
experience, VA hire date, start date at the VA hospital where he/she is currently
working, and when the employee started at his/her current nursing unit. This
dataset enables us to link each nurse to the unit in which he/she worked during
each two week pay period and provides information on the actual number of
hours worked on the unit for each nurse in each pay period. We are also able to
identify if a nurse transfers to a different unit from one pay period to the next (i.e.
an internal transfer) or if a nurse who is new to the hospital joined the unit at the
start of a pay period (external hire).
The Patient Treatment File (PTF) is a patient-level data set that includes
the dates of admission and discharge for each bed section as well as the admission
6 Each bed-section in the hospital corresponds to a type of care, not a specific unit. There is a 1-to-
1 correspondence between unit and bed-section for 89% of the acute-care bed-section stays. An
additional 6% of the patients were assigned to a specific unit based on the fact that the patient
spent less than one day on the second unit; the remaining 5% of the patients were dropped. An
examination of the excluded patient records showed no systematic differences in the
characteristics of the excluded patients.
8
and discharge dates for the overall hospitalization. It also includes International
Classification of Diseases, 9th
version Clinically Modified (ICD) diagnoses, the
Medicare Diagnosis Related Group (DRG) , the Elixhauser index which measures
co-morbidities (Elixhauser et al. 1998), and the patient’s age.
D. Sample
During our study period (fiscal years 2003 through 2006, i.e. October 1,
2002 through September 30, 2006) the VA operated 143 hospitals with acute
inpatient care units located across the United States. Many VA hospitals are
located in rural, non-metropolitan areas; these hospitals are quite small and, in
particular, have very small in-patient facilities, often focusing on outpatient
services. After deleting nursing units that had fewer than 100 patient days (i.e.
about 3 patients per day) or fewer than two RNs per shift, our final dataset
includes 907,993 patients who were admitted to 151 acute care units (excluding
intensive care units) in 76 hospitals. The hospitals that were deleted as a result of
these exclusion rules are all in rural areas or very small metropolitan areas and the
final sample accounts for 90 percent of all acute care stays in the VA system in
fiscal years 2003-2006. Figure 1 shows the geographic location of the hospitals
included in our final sample.
II. Empirical Strategy and Specification
Our empirical objective is to identify the effects of nurse staffing on
patient outcomes as measured by patient length of stay in the hospital. We
estimate this relationship using fixed effects regression analyses. Our basic
p<.10; ** p<.05; ***p<.01 a Each observation is a nurse-month. In columns (1) and (2), the dependent variable equals one if the nurse is working in a different unit in month t+1 compared to the unit he/she worked in month t, while in columns (3) and (4), the dependent variable equals one if the nurse left the VA in month t+1 Unit characteristics are measured at month t. All regressions include time dummies for each month, and an annual unit fixed effect. Robust standard errors in parentheses. Variables definitions are provided in the glossary in Appendix. b Log of average residual length of stay for patients admitted to this unit in month t. c Average complication rate for patients admitted to this unit in month t.
41
Table 3: Nurse Human Capital and Patient’s Residual Length of Staya
*p<.10; ** p<.05; ***p<.01 a Dependent variable is log(patient’s residual length of stay in hospital). N = 907.993. All regressions include patient age, Elixhauser co-morbidity index, number of patient admissions, time dummies for each month, and unit fixed effects that vary by year. Robust standard errors, reported in parentheses, are clustered by nursing unit. Variable definitions are provided in the Glossary in Appendix.
*p<.10; ** p<.05; ***p<.01 a Dependent variable is log(patient’s residual length of stay in hospital). N = 907.993. All regressions include patient
age, Elixhauser co-morbidity index, number of patient admissions, time dummies for each month, and unit fixed
effects that vary by year. Robust standard errors, reported in parentheses, are clustered by nursing unit. Variable
definitions are provided in the glossary in Appendix. b In columns (1), (3) and (4),, departures are restricted to RNs who had at least one year of unit tenure. In column (2),
departures include RNs with less than one year of unit tenure.
*p<.10; ** p<.05; ***p<.01 a Dependent variable is log(patient’s residual length of stay in hospital). These regressions use the specification in column (1) of
Table 4. All regressions include patient age, Elixhauser co-morbidity index, number of patient admissions, time dummies for each
month, and unit fixed effects that vary by year. Robust standard errors, reported in parentheses, are clustered by nursing unit. Variable
definitions provided in glossary in Appendix.
45
Table 6
Cost-Benefit Estimates a
(1) (2) (3)
Cost Days Saved Benefit of Days Saved e
A. RN Unit Tenure Increases by 4.3 years b
$18,196c 10.46 $26,487
B. Change daily 8-hr shift from Aide to RN
$5,280 1.224 $3,098
C. Change 420 hours from Contract to RN Overtime d
$10,920 5.35 $13,387
a Monthly estimates
b Difference between the 90th and 10th percentiles of average RN Unit tenure
c Includes additional wages and fringe benefits
d This is the average number of contract hours for unit-months with nonzero contract
hours
e Based on VA's estimate of $2500 cost per patient day
46
Appendix: Glossary of Variables
Variable Name Definition
Dependent Variables
Residual LOS Patient's actual length of stay on the unit minus the DRG-specific
Medicare expected length of stay
Inter-unit mobility Dummy variable=1 if RN is working on a different unit in month t+1
compared to month t
Separation Dummy variable=1 if RN was working in VA in month t but had left by
month t+1
Independent Variables
Tenure on unit RN's tenure on the unit
Bachelors Degree Dummy variable =1 if RN has bachelor's degree
RN Age Age of RN
RN regular hours Total RN regular work hours on the unit in month t, divided by number of
patient days in month t
RN overtime hours Total RN overtime hours on the unit in month t, divided by number of
patient days in month t
RN hours Sum of RN regular hours + RN Overtime hours
LPN hours Total LPN hours on the unit in month t, divided by number of patient days
in month t
UAP hours Total UAP hours on the unit in month t, divided by number of patient days
in month t
Contract hours Total contract hours on the unit in month t, divided by number of patient
days in month t
Avg RN unit tenure Average unit tenure of RNs working on the unit in month t
Avg RN net facility tenure Average of net faciilty tenure (faciity tenure minus unit tenure) for RNs
working on the unit in month t
Avg RN experience Average of total nursing experience of RNs working on the unit in month t
% RN hours with 1-2 yrs unit tenure Percentage of RN hours on the unit in month t accounted for by RNs with
1-2 years unit tenure
% RN hours with 2-3 yrs unit tenure Percentage of RN hours on the unit in month t accounted for by RNs with
2-3 years unit tenure
% RN hours with 3-4 yrs unit tenure Percentage of RN hours on the unit in month t accounted for by RNs with
3-4 years unit tenure
% RN hours with 4-5 yrs unit tenure Percentage of RN hours on the unit in month t accounted for by RNs with
4-5 years unit tenure
% RN hours with 5-6 yrs unit tenure Percentage of RN hours on the unit in month t accounted for by RNs with
5-6 years unit tenure
% RN hours with 6-7 yrs unit tenure Percentage of RN hours on the unit in month t accounted for by RNs with
6-7 years unit tenure
% RN hours with 7-8 yrs unit tenure Percentage of RN hours on the unit in month t accounted for by RNs with
7-8 years unit tenure
% RN hours with 8-9 yrs unit tenure Percentage of RN hours on the unit in month t accounted for by RNs with
8-9 years unit tenure
% RN hours with 9-10 yrs unit tenure Percentage of RN hours on the unit in month t accounted for by RNs with
9-10 years unit tenure
% RN hours with > 10 yrs unit tenure Percentage of RN hours on the unit in month t accounted for by RNs with
>10 years unit tenure
47
Experienced departure and no hire Experienced RN departed during the month and there was no new hire
Hire and no experienced departure RN joined the unit during the month and there was no experienced
departure
Experienced departure and hire Experienced RN departed and new hire joined the unit
Internal hire and no experienced
departure
RN transferred from another unit and there was no experienced departure
External hire and no experienced
departure
RN joined the unit from outside the hospital and there was no experienced
departure
Internal hire and experienced
departure
RN transferred from another unit and experienced departure also occurred
in the month
External hire and experienced
departure
RN joined the unit from outside the hospital and experienced departure
also occurred in the month
Internal hire RN transferred from another unit in the hospital
External hire RN joined the unit from outside the hospital
Any departure and no hire Any RN departed the unit during the month and there was no hire
Hire and no departure RN joined the unit during the month and there was no departure
Any departure and hire RN joined the unit during the month and a departure also occurred
Avg patient age Average age of patients treated on the unit during the month
Avg Elixhauser Average of Elixhauser index of patients treated on the unit during the
month
Complication Rate Average rate of reported complications experienced by patients treated on
the unit during the month
Admissions Number of patients admitted to the unit during the month