Nursing Research  January/February 2011  Vol 60, No 1, 1–8 Nurses’ Work Schedule Characteristics, Nurse
Staffing, and Patient Mortality
Alison M. Trinkoff 4 Meg Johantgen 4 Carla L. Storr b Background: Although nurse staffing has been found to be
related to patient mortality, there has been limited study of
the independent effect of work schedules on patient care
outcomes.
b Objective: To determine if, in hospitals where nurses report
more adverse work schedules, there would be increased
patient mortality, controlling for staffing.
b Methods: A cross-sectional design was used, with multilevel
data from a 2004 survey of 633 nurses working in 71 acute
nonfederal hospitals in North Carolina and Illinois. Mortality
measures were the risk-adjusted Agency for Healthcare
Research and Quality Inpatient Quality Indicators, and
staffing data were from the American Hospital Association
Annual Survey of hospitals. Principal components analysis
was conducted on the 12 work schedule items to create eight
independent components. Generalized estimating equations
were used to examine the study hypothesis.
b Results: Work schedule was related significantly to mortality
when staffing levels and hospital characteristics were
controlled. Pneumonia deaths were significantly more likely
in hospitals where nurses reported schedules with long work
hours (odds ratio [OR] = 1.42, 95% confidence interval [CI] =
1.17Y1.73, p G .01) and lack of time away from work (OR =
1.24, 95% CI = 1.03Y1.50, p G .05). Abdominal aortic
aneurysm was also associated significantly with the lack of
time away (OR = 1.39, 95% CI = 1.11Y1.73, p G .01). For
patients with congestive heart failure, mortality was associated with working while sick (OR = 1.39, 95% CI =
1.13Y1.72, p G .01), whereas acute myocardial infarction
was associated significantly with weekly burden (hours per
week; days in a row) for nurses (OR = 1.33, 95% CI =
1.09Y1.63, p G .01).
b Discussion: In addition to staffing, nurses’ work schedules
are associated with patient mortality. This suggests that
work schedule has an independent effect on patient
outcomes.
b Key Words: mortality & patient outcomes & working conditions &
work schedule D ata from various sources indicate that lower nurse
staffing levels contribute to poor patient outcomes.
Patient mortality has been one of the most frequently 4 Ayse P. Gurses 4 Yulan Liang 4 Kihye Han assessed outcomes because it is represented reliably in administrative data and is related conceptually to poor care.
In some studies using Medicare data, higher nursing skill
mix was associated with lower mortality (al-Haider & Wan,
1991; Hartz et al., 1989). Also, in studies using hospital
data, a link between staffing and mortality was reported
(Aiken, Clarke, Cheung, Sloane, & Silber, 2003; Person et al.,
2004). In a recent meta-analysis, it was concluded that, overall, an increase in nurse staffing is related to improved patient
outcomes (Kane, Shamliyan, Mueller, Duval, & Wilt, 2007),
yet the authors cautioned that there may be other factors
such as nurse scheduling that could be related independently to patient care.
In U.S. hospitals, most nurses work extended schedules,
that is, schedules that extend beyond the typical 9:00 A.M. to
5:00 P.M., Monday through Friday work day, because hospitals need to provide continuous nursing coverage.
Extended work schedules for nurses can cause fatigue and
performance deficits because of increased exposure to job
demands and insufficient recovery time (Geiger-Brown &
Trinkoff, 2010a). Nurses in one study reported being more
fatigued on 12-hour shifts compared with 8-hour shifts (IskraGolec, Folkard, Marek, & Noworel, 1996). In contrast, in a
study of nurses in 13 New York City hospitals, researchers
compared 8- and 12-hour work shifts (Stone et al., 2006) and
found that nurses on 12-hour shifts reported less emotional
exhaustion with no differences in patient outcomes. Unfortunately, only 4 of 13 hospitals offered both shifts, so findings
may reflect facility-level differences as opposed to withinhospital comparisons of nurses working 8- versus 12-hour
shifts. Although the effects of such schedules on nurses’ health
(Trinkoff, Le, Geiger-Brown, & Lipscomb, 2007), turnover
(Stordeur, D’Hoore, & the NEXT-Study Group, 2007), and
errors (Rogers, Hwang, Scott, Aiken, & Dinges, 2004) have
been documented, their impact on patient outcomes is largely
unknown.
Alison M. Trinkoff, ScD, RN, FAAN, is Professor; Meg Johantgen,
PhD, RN, is Associate Professor; and Carla L. Storr, ScD, is
Professor, School of Nursing, University of Maryland, Baltimore.
Ayse P. Gurses, PhD, is Assistant Professor, School of Medicine
and Bloomberg School of Public Health, Johns Hopkins University, Baltimore, Maryland.
Yulan Liang, PhD, is Associate Professor; and Kihye Han, MS,
RN, is Graduate Assistant, School of Nursing, University of
Maryland, Baltimore.
DOI: 10.1097/NNR.0b013e3181fff15d Nursing Research January/February 2011 Vol 60, No 1 Copyright @ 2010 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited. 1 2 Work Schedule and Mortality
Currently, most U.S. hospitals exclusively use 12-hour
shifts (Geiger-Brown & Trinkoff, 2010a). Although many
nurses like these schedules because of the compressed
nature of the workweek (e.g., three 12-hour shifts vs. five
8-hour shifts), this schedule as well as shift work in general
have been shown to lead to sleep deprivation (GeigerBrown & Trinkoff, 2010a). Nursing staff working nights
also face two major sleep challenges: getting insufficient
sleep because long work hours reduce sleep opportunity
and getting inadequate or poor quality sleep because of
circadian misalignment from shift work. Alertness not only
has a strong circadian element but also depends on having
an adequate duration of quality sleep (Akerstedt, Folkard,
& Portin, 2004; Arnedt, Owens, Crouch, Stahl, &
Carskadon, 2005; Shen, Barbera, & Shapiro, 2006). In
addition, nurses who work night shifts have notoriously
poor sleep including inadequate quantity and quality of
sleep and resultant fatigue and illness (Geiger-Brown &
Trinkoff, in press; Surani, Subramanian, Babbar, Murphy, &
Aguilar, 2008). Acute or chronic sleep deprivation is associated with deficits in neurobehavioral functioning such
as reduced or impaired vigilance, reaction time, memory,
psychomotor coordination, information processing, and
decision-making ability (Dinges et al., 1997; van Dongen,
Maislin, Mullington, & Dinges, 2003). Because the alertness
and vigilance required in nursing depends upon having
an adequate duration of quality sleep (Geiger-Brown &
Trinkoff, 2010a; Surani et al., 2008), long work hours can
impact nursing care and can increase the potential for error
(Hinshaw, 2006).
Similarly, among physicians, fatigue has been attributed to increased errors as the number of hours worked
increases (Gaba & Howard, 2002). To combat this situation, the medical profession has taken steps to limit the
hours a physician in training may work (Jagsi & Surender,
2004). However, there have been only voluntary recommendations for nurses that they limit their work hours to
no more than 60 per week or 16 in a 24-hour period (Institute of Medicine, 2004). Ironically, physician work hour
limits have led to task shifting; work hours among nurses
may be increasing to compensate for reduced physician
hours (Trinkoff, Geiger-Brown, Brady, Lipscomb, &
Muntaner, 2006).
The conceptual framework tying nurse-level working
condition factors with hospital-level patient outcomes is based
on balance theory. Balance theory is a human factors or systems engineering approach intended to measure organizationlevel conditions, incorporating data from individual employees
(Gurses & Carayon, 2007). According to this theory, job
performance is affected adversely by an imbalance of excessive demands with more positive aspects of the job (Carayon
& Smith, 2000). On the basis of balance theory, many
factors in nurses’ work environments may affect their performance and patient outcomes. However, except for staffing, other aspects of the nurses’ work environment that can
affect nursing care practices, such as work schedule characteristics, have not been examined substantially in relation
to patient outcomes and hence deserve further study. It was
hypothesized that, in hospitals where nurses report more
adverse work schedules, mortality rates will reflect poorer
quality care. Nursing Research January/February 2011 Vol 60, No 1 Methods
A cross-sectional design was used, incorporating data from
nurses and from the hospital where they worked. Patient outcome and staffing data from 71 acute care nonfederal hospitals in Illinois and North Carolina were merged with survey
data from 633 nurses working in these hospitals. The nurse
survey data came from the Nurses Worklife and Health Study,
Part 3 (Trinkoff, Geiger-Brown, et al., 2006). The Nurses
Worklife and Health Study was conducted originally as a
three-wave longitudinal study of nurse injury in relation to
work schedule and job demands. Out of 5,000 randomly
selected registered nurses (RNs) in the two states, 4,229 were
sent surveys and 2,624 returned usable questionnaires in
Wave 1. Follow-up responses in Waves 2 and 3 were received, from 85% and 86%, respectively, of Wave 1 nurses.
For this analysis, responses were included from nurses
working in hospitals with four or more RNs responding to
the survey in Wave 3, averaging nine nurses per facility. This
technique has been used similarly in surveys of hospital administrators and managers (e.g., Singer et al., 2003). Only
nurses who had been in their job for at least 1 year were
included in this analysis.
Work Schedule
Work schedule variables were derived from the Standard
Shiftwork Index, which has been used internationally to standardize self-report measures used in shift-work research
(Barton, Spelten, Totterdell, Smith, & Folkard, 1995; Folkard,
Spelten, Totterdell, Barton, & Smith, 1995). Three experts
from the National Institute for Occupational Safety and
Health with a career focus on work schedules examined the
survey for content validity (Trinkoff, Le, Geiger-Brown,
Lipscomb, & Lang, 2006). Work schedule data were derived
from the following variables: (a) hours worked per day, (b)
hours worked per week, (c) weekends worked per month, (d)
number of breaks lasting 10 minutes or more including meals
during a workday, and (e) shift rotation. To take into account
newer work schedule characteristics occurring among nurses,
seven additional items were included, for a total of 12 variables: how often nurses worked (a) 13 hours or more at a
stretch, (b) with less than 10 hours off between shifts, (c) on a
scheduled day off or vacation day, (d) while sick, (e) with
mandatory overtime, (f) required on call, and (g) the usual
number of days worked in a row (Trinkoff, Le, et al., 2006).
While completing the survey, nurses were asked to consider their typical work schedule for the past 6 months on
average. The use of a 6-month period minimized the chance
that participants would provide an unusual or atypical work
experience (Barton et al., 1995; Folkard et al., 1995). Nurses
were asked to report the hours they actually worked including overtime, as opposed to what they were scheduled to
work (Schernhammer et al., 2001; Trinkoff, Le, et al., 2006).
Hours worked per day were defined as the number of hours
worked at a stretch, so that it was possible for this item to
exceed 24 hours.
Staffing Data
Staffing data were obtained from the American Hospital
Association Annual Survey of Hospitals, including data on
full- and part-time RNs and licensed practical nurses (LPNs). Copyright @ 2010 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited. Work Schedule and Mortality 3 Nursing Research January/February 2011 Vol 60, No 1 The National Quality Forum nurse-sensitive indicators,
including staffing and skill mix (National Quality Forum,
2010), were used. Staffing was calculated as nursing (RN +
LPN) hours per patient day as devised by Kane et al. (2007),
which assumes there are 37.5 work hours per week and 48
weeks per year (excluding vacation, holidays, and sick time).
An adjustment formula of inpatient to outpatient gross revenues was applied to account for inpatient staffing (Mark,
Harless, McCue, & Xu, 2004). The RN proportion, used to
measure skill mix, was calculated by dividing RN hours by
the total hours for both RNs plus LPNs.
Hospital characteristic variables such as ownership and
teaching status were examined; only 2 of 71 hospitals were
for-profit hospitals; this did not vary sufficiently to add ownership to the analysis. Teaching status was constructed using
data on resident physician FTEs. Finally, state (Illinois or
North Carolina) and teaching status were included in the
analysis as control variables (Currie, Mehdi, & MacLeod,
2005; Jones, 2004).
Mortality
Mortality was measured from discharge data using the Agency
for Healthcare and Quality (AHRQ) In-patient Quality
Indicators (IQIs) because these indicators have been shown
to be related conceptually to nursing care (Davies et al.,
2001). The AHRQ IQIs include mortality rates for certain
medical conditions. To calculate hospital-level mortality, the
discharge data were applied to the AHRQ IQIs Windows
version 3.2, with risk adjustment using the All Payer-Refined
Diagnostic Review Groups. Outcomes were selected that
were relevant to nursing.
In addition, to provide valid estimates, Healthcare Cost
and Utilization Project selection rules were followed, and
some IQIs (those with extremely low rates of occurrence
[e.g., esophageal and pancreatic resection] as well as those
with denominators that were too small) were eliminated
because of rarity. The IQIs included in this analysis were
pneumonia, congestive heart failure (CHF), acute myocardial infarction (AMI) and stroke, and postsurgical procedures related to abdominal aortic aneurysm (AAA) repair
and craniotomy.
Analysis
All data analysis was performed using Predictive Analytics
Software (PASW) (Version 17.0; SPSS/IBM Inc., Somers,
NY). Univariate, descriptive statistical analyses were conducted; the mean and the standard deviation values of the key
variables were calculated and compared with state and
national data. Principal components analysis was conducted
on the 12 work schedule items to remove correlations across
the items and to create independent components. Construction of independent components for work schedule addressed
the multicollinearity issues in the follow up GEE modeling.
Sampling adequacy was good (KaiserYMeyerYOlkin = .67).
Components accounting for 82% of the variance for the
underlying dimension were retained in the analysis. Component scores were then derived for each of the eight
components (Table 1). Work schedule variables all contained less than 10% missing data, with the percentage of
missing data ranging from 1.7% to 7.9%. To investigate the relationship between nurses’ work
schedules and patient mortality and to describe it using the
best model, binomial logistic models with generalized estimating equation (GEE) methods were tested. Using GEE, it was
possible to account for within-hospital correlation arising
from the nested nature of the data (first level = nurse and
second level = hospital). Before using GEE, the original work
schedule items were examined for missing completely at
random using Little’s test. Findings supported missing completely at random, indicating that GEE using pairwise deletion
for missing data was appropriate (# 2 = 264.83, df = 236, p =
.096; Roth, 1994). Staffing was included along with skill mix
in all models because this was felt to be the most conservative
estimate of the independent effect of work schedule on mortality. The eight component scores extracted from work
schedule were used as explanatory variables for the given outcome variables. As outcome rates were not distributed normally, they were divided into quartiles, and study hospitals
with rates that met or exceeded the 75th percentile were defined as hospitals with higher than expected mortality within
the sample. This method has been used successfully in other
outcomes studies (Kane et al., 2007; Needleman, Buerhaus,
Mattke, Stewart, & Zelevinsky, 2001). Binomial logistic
models were therefore generated for each of the six outcomes
(mortality from pneumonia, AAA, CHF, AMI, stroke, and
craniotomy), with state and hospital teaching status included
in all models as potentially confounding variables. Institutional
review board approval from University of Maryland Baltimore
was obtained. Results
Study hospitals had higher levels of staffing and skill mix than
overall acute general hospitals in Illinois and North Carolina,
although differences were not significant: study hospitals
averaged 7.5 hours (SD = 2.3) of RNs and LPNs (licensed
hours) per patient day, approximately 1.3 times the overall
Illinois and North Carolina level, which was 5.7 hours (SD =
2.5). Skill mix, measured by RN proportion, was also higher
in study hospitals (94.7% RNs) than in Illinois and North
Carolina overall (88.9%). Of the 71 study hospitals, almost
half (47.9%) were teaching hospitals (i.e., had resident physicians). This was significantly higher than for Illinois and
North Carolina hospitals overall (31.2% teaching hospitals;
# 2 = 7.36, p G .01).
Study nurses averaged 43.9 years of age, compared with
43.4 years on average reported for U.S. hospital nurses (Health
Resources and Services Administration [HRSA], 2006). Study
nurses also were more racially diverse (15% non-White) than
national hospital nurses (7%). The proportion of nurses with
a Bachelor’s degree or higher was 60% compared with
approximately 50% in all hospital nurses in national data
(HRSA, 2006). For shifts worked, almost half of the nurses
reported working shifts other than during the day as part of
their typical schedule. Among the nurses, 13% reported mandatory overtime, yet more than 40% had required on call as
part of their jobs, indicating that the majority had some form
of required additional hours, above those which they were
already scheduled to work (Table 2).
For patient outcome data, risk-adjusted rates of selected IQIs
were summarized in Table 3. When comparing risk-adjusted Copyright @ 2010 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited. 4 Work Schedule and Mortality Nursing Research January/February 2011 Vol 60, No 1 q TABLE 1. Principal Components Analysis of Work Schedule Characteristics
Components Component 1: long work hours
Hours worked per workday
Work 13 hours or more
Component 2: off shift and weekends
Shift rotation
Weekends worked per month
Component 3: weekly burden
Usual number of days worked in a row
Hours worked per week
Component 4: lack of time away
Work on scheduled day off or vacation
G10 hours off between shifts
Component 5: mandatory overtime
Mandatory overtime
Component 6: working while sick
Work while sick
Component 7: required on call
On call required
Component 8: insufficient work breaks
No. breaks lasting 10 minutes or more 1 2 3 4 5 .83
.81 .30
.04 j.04
.02 .04
.34 j.07
.06 .20
.09 .82
.78 .05
j.17 j.03
.21 j.29
.30 j.04
j.06 .82
.79 .08
.34 .05
.16 .00 6 7 8 % Variance Explained .07
.00 j.04
.13 j.08
.05 21.49 j.11
.08 j.08
.14 .06
j.08 .03
j.01 13.25 .07
.03 j.02
.04 j.09
.11 .00
.05 .06
j.10 9.67 .00
.16 .89
.61 j.10
.16 .07
j.05 j.02
.14 .01
j.05 8.74 j.04 .02 j.01 .98 .00 .10 .00 8.06 .04 .04 .02 .04 .00 .98 .04 .03 7.76 .06 j.01 .04 .07 .10 .04 .98 j.04 6.91 j.03 .01 j.03 j.01 .00 .03 j.03 .99 6.53 Note. Bolded items represent loadings greater than 0.60; rotation method: varimax. rates in study hospitals (N = 71) with national benchmarks,
data were similar, with the exception of rates of AAA: In
study hospitals, the rate averaged 50.81 deaths per 1000
patients (SE = 7.87) versus 66.33 (SE = 1.15) nationally.
Nonetheless, the cut points for rates above the 75th percentile that were used for analysis well exceeded U.S. benchmarks for all IQIs.
Findings comparing hospitals with higher than expected
mortality (those above the 75th percentile for the IQIs vs. all
other hospitals) adjusted for staffing and skill mix, state, and
hospital teaching status are presented in Table 4. Consistent
with the study hypothesis, in which adverse work schedule
was proposed to increase the odds of mortality, outcomes
with odds ratios greater than 1.0 support the hypothesis.
Pneumonia deaths were significantly more likely to occur in
hospitals where nurses reported schedules that included long
work hours (odds ratio [OR] = 1.42, 95% confidence interval [CI] = 1.17Y1.73, p G .01) and lack of time away from
work (OR = 1.24, 95% CI = 1.03Y1.50, p G .05). Deaths
from AAA also were associated significantly with lack of
time away (OR = 1.39, 95% CI = 1.11Y1.73, p G .01). For
CHF, mortality was associated with working while sick
(OR = 1.39, 95% CI = 1.13Y1.72, p G .01), whereas AMI
was associated significantly with weekly burden for nurses
(OR = 1.33, 95% CI = 1.09Y1.63, p G .01). For staffing, the
analysis showed significantly lower licensed staffing in highmortality hospitals for pneumonia, CHF, and stroke, whereas lower skill mix was related significantly to AMI and craniotomy. Teaching status was not related to any of the outcome indicators. Discussion
In addition to staffing, nurses’ work schedules are associated
independently with patient mortality. The work schedule
component most frequently related to mortality was that of
lack of time away from the job. This has been found also to be
important for nurse injury and fatigue because nurses need
time off to rest and recuperate to protect their health. Similarly, the lack of recovery time may affect performance.
Geiger-Brown and Trinkoff (2010a), in the Nurses’ Sleep
Study, showed that nurses working long hours in successive
shifts averaged only 5.5 hours of sleep between shifts.
In previous studies, long hours were shown to be related
to nurse fatigue and health, suggesting that they also affect
performance or ability to practice. The continued vigilance
required of nurses can be affected by excessive work hours,
limiting their ability to detect adverse changes in patients in
time to address them and prevent consequences. This could
have profound consequences for patient safety and health.
The impact of working conditions on patient mortality
in the context of staffing was examined in this study. The
finding that work schedule can impact patient outcomes is
new and important and should lead to further work. Of the Copyright @ 2010 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited. Work Schedule and Mortality 5 Nursing Research January/February 2011 Vol 60, No 1 q q TABLE 2. Sample Distribution of Nurses’ Work
Schedule Characteristics (N = 633) TABLE 2. (continued)
Characteristics Characteristics
Hours worked per day (range = 0Y25)
Hours worked per week (range = 3.5Y96.0)
No. weekends worked per month
(range = 0Y4)
No. breaks lasting Q10 min including meals
during a workday (range = 0Y3)
Shift
Evening only
Day only
Day + evening
Day + nights
Night + evening or all three
Night only
Work Q13 hours at a stretch
Never/NA
Few times a year
Once a month
Every other week
Once a week
More than once a week
Work G10 hours off between shifts
Never/NA
Few times a year
Once a month
Every other week
Once a week
More than once a week
Work on a scheduled day off/vacation day
Never/NA
Few times a year
Once a month
Every other week
Once a week
More than once a week
Work while sick
Never/NA
Few times a year
Once a month
Every other week
Once a week
More than once a week
Mandatory overtime
Never/NA
Yes, with more than an 8-hour notice
Yes, with 2- to 8-hour notice
Yes, with less than a 2-hour notice 10.3 T 2.3
37.3 T 11.8
1.4 T 1.3
1.6 T 0.8 49
328
77
15
50
95 (8.0)
(53.4)
(12.5)
(2.4)
(8.1)
(15.5) 196 (32.7)
211 (35.2)
54 (9.0)
28 (4.7)
44 (7.3)
67 (11.2)
323
167
35
12
18
30 (55.2)
(28.5)
(6.0)
(2.1)
(3.1)
(5.1) 199
286
50
32
10
8 (34.0)
(48.9)
(8.5)
(5.5)
(1.7)
(1.4) 126
425
22
5
2
3 (21.6)
(72.9)
(3.8)
(0.9)
(0.3)
(0.5) 538
23
12
45 (87.1)
(3.7)
(1.9)
(7.3) Required on call/How often called into work
Never/NA
Yes, but never called in
Yes, few times a year
Yes, approximately once a month
Yes, approximately once a week
Yes, more than once a week
Usual number of days worked in a row
(range = 0Y50) 360
16
113
80
38
15
3.5 (57.9)
(2.6)
(18.2)
(12.9)
(6.1)
(2.4)
T 2.6 Note. Values are presented as mean T SD and n (%). significant work schedule components, work hours and a lack
of time off also have been shown to be related to nurse injuries
when examined prospectively (Trinkoff, Le, et al., 2006). In
addition, schedule may a...

Attachments:

• 1505217541-nursing research 1.pdf