Are Patient Falls and Pressure Ulcers Sensitive to Nurse Staffing? - Western Journal of Nursing Research

This article evaluates the state of the science linking nurse staffing to falls and pressure ulcers. Studies that employed multivariate analysis to discern the effect of nurse staffing on patient falls and pressure ulcers in hospitals were evaluated. The evidence of an effect of nursing hours or skill mix on patient falls and pressure ulcers is equivocal.

654
Are Patient Falls and
Pressure Ulcers Sensitive
to Nurse Staffing?
Eileen T. Lake
Robyn B. Cheung
University of Pennsylvania
Research has demonstrated an association between more nurses and more
qualified nursing staff in hospitals and better patient outcomes. Patient falls
and pressure ulcers have been advanced as nursing-sensitive outcomes. This
article evaluates the state of the science linking nurse staffing to falls and
pressure ulcers. Studies that employed multivariate analysis to discern the
effect of nurse staffing on patient falls and pressure ulcers in hospitals were
evaluated. Eleven studies that met inclusion criteria were contrasted on their
data sources and measures, data analysis, risk adjustment, and results. The
evidence of an effect of nursing hours or skill mix on patient falls and pres-
sure ulcers is equivocal. Substantial differences in research methods across
studies may account for the mixed findings. Two study types were identified
based on the level at which nurse staffing was measured, hospital or nurs-
ing unit, which exhibited systematic differences in measures and methods.
Improvements in measurement and methods are suggested.
Keywords: nurse staffing; pressure ulcers; patient falls; multilevel research
methods
The quantity and quality of nursing care are expected to influence the
occurrence of patient falls and pressure ulcers. Some of these events are
preventable given sufficient nursing resources to assess risk and implement
Western Journal of
Nursing Research
Volume 28 Number 6
October 2006 654-677
© 2006 Sage Publications
10.1177/0193945906290323
http://wjn.sagepub.com
hosted at
http://online.sagepub.com
Authors’ Note: This article was funded in part by grants from the University of Pennsylvania
Research Foundation, the Xi Chapter, Sigma Theta Tau International, and the National Institute
of Nursing Research, National Institutes of Health (Grant NR009068) to Dr. Lake and by an
institutional postdoctoral fellowship (National Institute of Nursing Research, National Institutes
of Health, Grant T32-NR07104) to Dr. Cheung. Dr. Cheung compiled and summarized the
empirical literature. Dr. Lake prepared the critical review of the empirical literature, developed
the tables, and developed the contributions to the science on staffing, falls, and pressure ulcers.
We appreciate the thoughtful review and helpful comments of the anonymous reviewers.
at University of Ulster Library on January 4, 2015wjn.sagepub.comDownloaded from

Lake, Cheung / Nurse Staffing and Adverse Events 655
prevention strategies. Since 1996, the American Nurses Association (1999)
has considered falls and pressure ulcers to be “nursing-sensitive quality indi-
cators.” The National Quality Forum (2004) recently endorsed both falls and
pressure ulcers as core measures of nursing care performance in hospitals.
The Institute of Medicine’s (IOM) landmark volume (Wunderlich, Sloan, &
Davis, 1996) revealed the dearth of empirical evidence on the relationships
between nurse staffing and quality of care. The theoretical basis for a link
among staffing, falls, and pressure ulcers, coupled with the IOM’s research
recommendations prompted scientific investigation of this link. A decade of
research can now be assessed to determine whether there is clear evidence
to guide practice and research. More broadly, the state of the science on this
question can be considered an exemplar of theoretical issues and method-
ological challenges in the field of nursing outcomes research.
The purposes of this article are to present and evaluate the state of the
science linking nurse staffing to patient falls and pressure ulcers and to
identify research methods to advance the science. To prepare the reader for
an evaluation of this science, the Background section provides an overview
of theoretical frameworks in nursing outcomes research, an initial descrip-
tion of the adverse events and measurement issues, and an overview of risk
adjustment in outcomes research.
Background
The Theoretical Link
Several theoretical or conceptual frameworks for the relationships
between nursing organizational factors and outcomes have been proposed
(Aiken, Sochalski, & Lake, 1997; Irvine, Sidani, & Hall, 1998; Mark,
Sayler, & Smith, 1996; Mitchell, Ferketich, Jennings, & American Academy
of Nursing Expert Panel on Quality Health Care, 1998). Mark et al. (1996)
theorized that both the hospital organizational context (i.e., environment and
technology) and the nursing unit structure (e.g., nurse–physician collabora-
tion) influence outcomes. They considered nursing skill mix and education
mix as context variables but did not mention staffing ratios. Aiken etal.
(1997) proposed that organizational models,such as magnet hospitals or
dedicated AIDS units, achieve better patient and nurse outcomes via
enhanced nurse autonomy, nurses’ control over the work environment, and
nurses’ relations with physicians. A subsequent framework incorporated
nurse staffing ratios and skill mix as hospital organizational features that
at University of Ulster Library on January 4, 2015wjn.sagepub.comDownloaded from

affect outcomes (Aiken,Clarke, & Sloane, 2002). Mitchell et al. (1998)
developed the quality health outcomes model, positing reciprocal relations
among system characteristics, interventions, client characteristics, and out-
comes; they identified nursing skill mix as a system characteristic. Irvine
et al. (1998) developed a nursing role effectiveness model that specified
nursing staff mix and workload as structural variables in a causal chain influ-
encing outcomes. Collectively, these models describe a theoretical link
between nurse staffing and patient outcomes and provide the basis for empir-
ical investigations of nurse staffing, patient falls, and pressure ulcers.
The nursing role, encompassing both surveillance and care, makes nurses
uniquely suited to prevent falls and pressure ulcers. The patients at greatest
risk of these events and the best prevention protocols have been well docu-
mented (Ayello & Braden, 2001; Olson et al., 1996; Rutledge, Donaldson,
& Pravikoff, 2003). Several basic elements of prevention, such as systematic
falls risk assessment once a shift or more and repositioning every 2 hours
for pressure ulcer prevention,depend directly on staff availability. Thus,
patients in better staffed hospitals may be expected to experience fewer falls
or pressure ulcers. Some preventive efforts may be provided by staff that are
not registered nurses (RNs). The key staffing component may be hours of
nursing care by all licensed staff (i.e., RNs and licensed practical nurses,
LPNs) or by total nursing staff rather than RN hours.
Occurrence of Falls and Pressure Ulcers
Falls and pressure ulcers occur rarely. Their frequency can be calculated
as rates of incidence (number of new cases occurring over a given period;
e.g., per 1,000 patient days) or rates of prevalence (a count or proportion of
a population observed at one time point that has the condition of interest;
Frantz, 1997). Falls incidence among hospital inpatients has been reported
as ranging from 2.3 to 7.0 falls per 1,000 patient days (Hitcho et al., 2004).
A large observational study revealed that 7% of medical–surgical and inten-
sive care patients developed pressure ulcers (Stages I through necrotic)
after hospital admission (Whittington, Patrick, & Roberts, 2000).
Risk Adjustment in Outcomes Research
Differences across patients influence the likelihood of an adverse event
(Iezzoni, 2003). In outcomes research, if differences in patients across orga-
nizational settings (hospitals or nursing units) are not controlled for, spuri-
ous relationships might be identified. Better staffed units may appear to
656 Western Journal of Nursing Research
at University of Ulster Library on January 4, 2015wjn.sagepub.comDownloaded from

have poorer outcomes, when, in fact, the better staffing is because of a
sicker, higher risk patient group. The matter of differential risk is important
because most of the variation in adverse outcomes derives from patient
characteristics not hospital characteristics such as resources available for
treatment (Silber, Rosenbaum, & Ross, 1995). “It is vastly better to be a
little healthier than to go to a well-equipped hospital with a superior staff”
(Silber et al., 1995, p. 11).
In a multivariate model, there are several ways to account for the
patient’s contribution to the likelihood of the adverse event. The best way
is to include a measure of the patient’s risk for the specific outcome. Several
risk assessment scales for these adverse events are commonly used in acute
care: Morse (1996) and Schmid (1990) for falls risk and Braden (Braden &
Maklebust, 2005) and Norton (1996) for pressure ulcers. The patient’s risk
status (at risk or not) and risk score would be included in a patient-level
model predicting whether the patient did or did not have the adverse event.
Unfortunately, such risk data are not often available.
The second best way to account for differences across patients is to
include a measure of the patient’s severity of illness. A variety of severity
of illness measures have been developed based on medical record review or
discharge abstracts (Kane, 2006). These measures may focus on the physi-
ological severity of the principal diagnosis (the principal basis for the hos-
pital stay) and/or the number and severity of comorbid conditions.
A third way to measure differences across patients is by their differential
nursing care needs, or nursing acuity. The expectation is that patients with
greater nursing care needs are more likely to be at risk for adverse events.
Patients are classified daily according to a variety of factors that may include
physical activity, hygiene, feeding, medications, vital signs, and treatment.
Researchers may use aggregate measures to adjust for patient mix. An
example is the Medicare case mix index (CMI), which reflects the average
resources needed to care for the diagnostic mix of elderly patients in a hos-
pital or nursing unit. Although the CMI may be a proxy for patient mix in
an institution, it may not be suitable or sufficient for case mix adjustment
of younger adults or children. Sometimes researchers account for case mix
differences across institutions by conducting separate analyses for different
clinical subgroups of patients identified by nursing unit types or diagnoses.
This method does not account for severity of illness differences within clin-
ical subgroups.
In summary, risk adjustment approaches cover a range from the most
preferable but least available data (patient-specific risk data) to the least
preferable but most available (aggregated case mix or clinical data). The
Lake, Cheung / Nurse Staffing and Adverse Events 657
at University of Ulster Library on January 4, 2015wjn.sagepub.comDownloaded from

choice of risk adjustment approach is generally dictated by data availa-
bility. Researchers may use multiple approaches simultaneously to take
advantage of all available data.
Method
The study design was critical analysis of empirical studies of the rela-
tionships among nurse staffing, falls, and pressure ulcers in hospitals. To
identify articles, multiple databases (CINAHL, PubMed, OVID) and terms
(nurse staffing, work environment, falls, pressure ulcers, patient outcomes)
were searched from their earliest dates through October 2005. The most
efficient search was obtained via PubMed with the terms nurse staffing and
pressure ulcer, or nurse staffing and falls. This database and strategy
yielded 37 articles. The inclusion criteria were publication in a peer-
reviewed journal, research about an acute care setting,falls or pressure
ulcers as outcomes of interest, and multivariate analysis of staffing and one
or both outcomes. Eleven studies met the criteria.
The analysis began with a description of the research design and findings.
Then, studies were compared on the clinical composition of the patient sam-
ple, the adverse event rates, the data sources and measures of nurse staffing
and adverse events, the data analyses, and the risk adjustment approaches.
These features were assessed to determine if certain samples,measures,
methods, or events consistently yielded significant or null findings.
Findings
In one study, nurse staffing was measured at the nurse level (Krauss etal.,
2005). In this innovative and intensive, prospective, case-control study, 98
patients who fell were matched to 318 controls randomly selected from all
inpatients and matched on length of stay at the point the fall occurred. The
other studies could be differentiated by the organizational level for nurse
staffing: nursing unit or hospital. The organizational level was the key design
feature associated with systematic differences in adverse event measurement
and risk adjustment methods across studies. Studies were classified into two
groups by level (Table 1). In five studies, hospital-level staffing was linked to
hospital- or patient-level adverse events (Table 1, Panel A). These studies had
from 200 to 800 hospitals and from 100,000 to 6 million patients. In five stud-
ies, both staffing and adverse events were measured at the nursing unit level
658 Western Journal of Nursing Research
at University of Ulster Library on January 4, 2015wjn.sagepub.comDownloaded from

(Panel B). These studies had from 39 to 1,751 nursing units in 1 to 282 hos-
pitals. The Krauss et al. (2005) study is displayed in Panel B of Table 1.
Effect of Staffing on Falls and Pressure Ulcers
For each study, for the respective adverse events, Table 1 shows whether
the results across one or more analyses linking staffing measures to the
event were all significant, all not significant, or mixed (some significant and
others not). Falls were an outcome in eight studies. The results were sig-
nificant in two, mixed in three, and not significant in three. Pressure ulcers
were analyzed in seven studies. The results were significant in two, mixed
in three, and not significant in two. This initial picture suggests that the
evidence linking nurse staffing to these two events is inconclusive.
The inconclusive results may indicate a weak effect of staffing on these
events. An alternative explanation is that methodological limitations in
the literature prevent significant effects from being detected. An informed
judgment about the evidence requires a careful assessment of the methods.
Adverse Event Measurement and Frequency
Falls. For hospital-level studies, adverse events were identified by diag-
nosis or event codes in patient discharge abstracts in administrative data sets.
Cho, Ketefian, Barkauskas, and Smith (2003) used diagnosis codes indicat-
ing falls “with injury” among patients hospitalized for common surgeries in
California in 1997. Unruh (2003) used event codes indicating a low-level fall
or an accidental fall from a bed or chair for all medical and surgical patients
in Pennsylvania hospitals in 1991 to 1997. The fall rates from these studies
were trivial—less than 1% (Table 1). The rate detected by Unruh was more
than double the rate detected by Cho et al. This difference may reflect the
fact that most falls do not result in injury and that medical patients have
higher fall rates than do surgical patients (Hitcho et al., 2004). Rate com-
parisons across these different data sources may be inappropriate.
For nursing unit studies, falls were identified from incident reports. The
incidence ranged from 2 to 4 falls per 1,000 patient days. Blegen, Goode,
and Reed (1998) studied all 42 acute units in one hospital. Blegen and Vaughn
(1998) studied 39 medical/surgical, intensive care, obstetrics, and skilled
units in 11 hospitals. The fall rate was highest for skilled (4.0) and medical/
surgical units (2.7) and lowest for intensive care (1.4) and obstetrics (0.4).
Dunton, Gajewski, Taunton, and Moore (2004) studied data from 2,351
medical, surgical, step-down, and critical care units in 282 hospitals in
Lake, Cheung / Nurse Staffing and Adverse Events 659
at University of Ulster Library on January 4, 2015wjn.sagepub.comDownloaded from

660
Table 1
Sample Sizes, Rates of Adverse Events, Statistical Analysis, and Results of Studies
Linking Nurse Staffing to Patient Falls and Pressure Ulcers
Nursing
Rates of Adverse Events
Study Hospitals (n) Units (n) Fall % Pressure Ulcer % Analysis Results
Panel A—Hospital level
Cho, Ketefian, Barkauskas, 232 0.21a 0.26 Multilevel logistic ns; MIX
and Smith (2003) regression
Lichtig, Knauf, and Milholland (1999) 126-131 (NY) Not reported Regression MIX
352-295 (CA)
Mark, Harless, McCue, and Xu (2004) 244-256b 0.48-0.74c Generalized method ns
of moments
Needleman, Buerhaus, Mattke, Stewart, 799 5.8 S; 7.2 (M) Negative binomial ns
and Zelevinsky (2002) regression
Unruh (2003) 211d 0.55e 0.69 Random effects SIG; SIG
poisson regression
at University of Ulster Library on January 4, 2015wjn.sagepub.comDownloaded from

661
Panel B—Nursing unit level
Blegen, Goode, and Reed (1998) 1 42 2.69 1.73 Linear regression ns; SIG
Blegen and Vaughn (1998) 11 39 2.20 Generalized estimating MIX
equations
Dunton, Gajewski, Taunton, 282 2,351 3.73 Generalized linear MIX
and Moore (2004) mixed model
Krauss et al. (2005) 1f 3.29g Logistic regression SIG
McGillis Hall, Doran, and Pink (2004) 19 77 Not reported Regression ns
Sovie and Jawad (2001) 29 29 (M) 3.97-4.11 (M) 2.23-2.61 (M) Regression MIX; MIX
29 (S) 2.42-2.69 (S) 1.88-2.68 (S)
Note: SIG =significant; MIX =mixed (some results significant and others not); NY =New York; CA =California; S =surgical; M =medical.
a. Falls with injury.
b. Mark et al. (2004) studied a cohort of 422 hospitals during 6 years. The displayed values are the range in the sample size during the 6-year
period. The number of observations in their analyses of pressure ulcers was 945.
c. Ratio of risk-adjusted observed/expected decubitus ulcers.
d. Unruh (2003) studied a cohort of 211 hospitals during 7 years. The number of observations in her analyses was 1,477.
e. Unruh (2003) reported this value as “mean incidence/1000 patients.” This measure was converted to a proportion for comparison with the study
by Cho et al. (2003).
f. Sample comprised 98 cases matched with 318 controls in one hospital.
g. Hospital-wide fall rate.
at University of Ulster Library on January 4, 2015wjn.sagepub.comDownloaded from

45 states from 2002. They noted that falls were most common on medical
units and least common on critical care units. McGillis Hall, Doran, and
Pink (2004) studied 77 medical, surgical, and obstetric units in 19 teaching
hospitals but did not report rates. Sovie and Jawad (2001) studied one med-
ical and one surgical unit in 29 hospitals. They found a significantly higher
fall rate for medical than surgical units. Krauss et al. (2005) identified falls
through an online adverse event reporting system.
In summary, falls were studied more often from incident reports (five stud-
ies) than from administrative data (two studies). Extensive incident report data
from the National Database of Nursing Quality Indicators (Dunton etal.,
2004) yielded a rate of 3.73 falls per 1,000 patient days for the most common
nursing unit types. The fall rate was highest on medical and skilled units.
Pressure ulcers.All five hospital-level studies measured pressure ulcers,
identified by secondary diagnoses. Cho etal. (2003) included cases only when
the diagnosis was not present on admission, based on an uncommon variable in
their data. The two studies that analyzed both adverse events found equally low
rates of pressure ulcers as falls (i.e., 1% or less; Cho etal., 2003; Unruh, 2003).
Lichtig, Knauf, and Milholland (1999) studied pressure ulcers among all dis-
charges from all California and New York hospitals in 1992 and 1994. They did
not report rates. Needleman,Buerhaus, Mattke, Stewart, and Zelevinsky’s
(2002) study of all medical and surgical discharges from 799 hospitals in 11
states in 1997 identified much higher rates (6%-7%) than other studies, possi-
bly because of their selection of broad diagnosis codes 682 (other cellulitis and
abscess) and 707.0 (decubitus ulcer). The other studies did not report the codes.
Mark, Harless, McCue, and Xu (2004) reported risk-adjusted ratios of the
observed to the expected number of pressure ulcers for the 6 years of data they
studied in a cohort of 422 hospitals in 11 states.
Two nursing unit studies measured hospital-acquired pressure ulcers. Both
excluded Stage I ulcers (persistent redness of a defined area of skin). Blegen
et al. (1998) measured pressure ulcers by chart review. Sovie and Jawad
(2001) measured pressure ulcer prevalence through observation. On one day
each month, all patients were examined to identify nosocomial ulcers.
In summary, pressure ulcers were studied more often using administra-
tive data (five studies) than observational or chart data (two). For this
adverse event, administrative data are much more practical to obtain than
are primary data. The rates of pressure ulcers calculated from administra-
tive data appear to be highly influenced by the choice of diagnosis codes.
This cannot be stated conclusively because the two studies that detected
low pressure ulcer rates did not report the diagnosis codes.
662 Western Journal of Nursing Research
at University of Ulster Library on January 4, 2015wjn.sagepub.comDownloaded from

Nurse Staffing Measurement
These studies used variants of two staffing measures: hours of care per
patient day and skill mix (Tables 2 and 3). The most common measures,
found in five studies, were total nursing hours per patient day (RN, LPN,
and nursing assistant, NA) and RN skill mix (RN proportion of nursing
hours). The next most common measure was RN hours per patient day (four
studies). Unruh (2003) studied licensed hours. The skill mix measures
of McGillis Hall et al. (2004) and Unruh combined licensed staff in the
numerator and total nursing staff in the denominator (in daily hours and full
time equivalents—FTEs—respectively). Needleman et al.’s (2002) skill
mix measure was the RN proportion of licensed staff. The staffing measures
used by Needleman et al. and Lichtig et al. (1999) were adjusted using
nursing intensity weights (NIWs). Mark, Harless, et al. (2004) measured
RN FTEs per 1,000 patient days. Krauss et al. (2005) studied the patient-
to-nurse ratio as the number of patients assigned to the nurses caring for the
cases and controls at the time of the fall. The last two measures were clas-
sified for comparative purposes in the analysis below as variants of RN
hours per patient day.
In each study, the effect of each staffing measure on each adverse event
was estimated in a multivariate model. Each study had from one to three
staffing measures and one or both adverse events. Two studies had analyses
for different nursing unit types (Dunton et al., 2004; Sovie & Jawad, 2001).
Lichtig et al. (1999) had separate analyses for New York and California data
in each of 2 years. Across the studies, from 1 to 16 separate analyses were
conducted. In all cases but one (Sovie & Jawad, 2001), the multiple nursing
unit types, states, or years that were the basis for multiple analyses were suf-
ficiently large to be considered multiple samples within the same study. The
Sovie and Jawad (2001) study was excluded from the subsequent analysis of
the patterns of effects of different staffing measures because of the dispro-
portionate number of analyses that Sovie and Jawad conducted (16) com-
pared to the other studies (1-6), given their modest sample sizes (n =29 for
each analysis) and the concomitant uncertainty about power. Only 3 of their
16 analyses yielded significant results. Tables 2 and 3 display the results of
all analyses by study. In contrast to Table 1, which summarizes results across
analyses within each study, Tables 2 and 3 note the statistical significance of
every analysis and the direction of the relationship for significant results.
The findings of Sovie and Jawad were excluded from the assessment.
The findings were analyzed across three measure sets for each adverse
event: total hours per patient day, skilled (RN or licensed) hours per patient
Lake, Cheung / Nurse Staffing and Adverse Events 663
at University of Ulster Library on January 4, 2015wjn.sagepub.comDownloaded from

664
Table 2
Staffing Measures and Findings Reported in Studies Linking Nurse Staffing to Pressure Ulcers
Hours of Care Skill Mix
Study Staffing Measure Effect of Staffing Staffing Measure Effect of Staffing
Blegen, Goode, and Reed (1998) Alla POS RN/Alla NEG
Cho, Ketefian, Barkauskas, and Smith (2003) All POS RN/All ns
RN ns
Lichtig, Knauf, and Milholland (1999) Alla,b ns (CA 92, NY 94) RN/Alla NEG (CA 92, 94)
NEG (CA 94, NY 92) (NY 92, 94)
Mark, Harless, McCue, and Xu (2004) RN ns
Needleman, Buerhaus, Mattke, RN ns (M, S) RN/LIC ns (M, S)
Stewart, and Zelevinsky (2002)
Sovie and Jawad (2001) All NEG (S 98) ns (S 97, M 97, 98)
RN ns (S 97, 98, M 97, 98)
Unruh (2003) LIC NEG LIC/All NEG
Note: All =RN + LPN + unlicensed nursing staff hours per patient day; RN =RN hours per patient day; LIC =RN + LPN hours per patient day;
NEG =a negative coefficient; POS =a positive coefficient; M =medical; S =surgical; CA =California; NY =New York; 92, 94, 97, 98 =the year
of the data set.
Sovie and Jawad (2001) were excluded from text discussion.
a. Indicates results for both measures are from a combined model.
b. This measure was calculated as total nursing hours per nursing intensity weight.
at University of Ulster Library on January 4, 2015wjn.sagepub.comDownloaded from

665
Table 3
Staffing Measures and Findings Reported in Studies Linking Nurse Staffing to Patient Falls
Hours of Care Skill Mix
Study Staffing Measure Effect of Staffing Staffing Measure Effect of Staffing
Blegen, Goode, and Reed (1998) Alla ns RN/Alla ns
Blegen and Vaughn (1998) Alla ns RN/Alla NEG
Cho, Ketefian, Barkauskas, All ns RN/All ns
and Smith (2003)
RN ns
Dunton, Gajewski, Taunton, Alla NEG (M, M-S, S-D); ns (S) RN/Alla NEG (M, S-D);
and Moore (2004) ns (S, M-S)
Krauss et al. (2005) Patient-to-nurse ratio POS
McGillis Hall, Doran, LIC/All ns
and Pink (2004)
Sovie and Jawad (2001) All NEG (S 97); ns (M 97, 98; S 98)
RN NEG (M 98); ns (S 97, 98; M 97)
Unruh (2003) LIC NEG LIC/All Pos
Note: All =RN + LPN + unlicensed nursing staff hours per patient day; RN =RN hours per patient day; LIC =RN + LPN hours per patient day;
NEG =a negative coefficient; POS =a positive coefficient; M =medical; S =surgical; M-S =medical–surgical; S-D =step-down; 97, 98 =the year
of the data set.
Sovie and Jawad (2001) were excluded from text discussion.
a. Indicates results for both coefficients are from a combined model.
at University of Ulster Library on January 4, 2015wjn.sagepub.comDownloaded from

day, and skill mix. The proportion of all analyses across studies that yielded
significant effects was calculated. First, the findings for pressure ulcers are
reviewed (Table 2). The measure with the highest proportion of significant
results was RN skill mix of total nursing staff. Five out of six analyses
across three studies had significant negative findings. Two variants of the
skill mix measure yielded disparate findings. Needleman et al. (2002) had
nonsignificant results for the RN proportion of licensed staff. Unruh (2003)
found that the licensed (RN and LPN) proportion of total staff was nega-
tively associated with pressure ulcers.
Four out of six analyses linking total nursing hours to pressure ulcers
had significant findings, although two of the results were positive and two
were negative. The positive results from two different studies may reflect
that high total staffing levels are the result of a more complex patient mix,
which may not be adequately controlled in the analyses. The risk adjust-
ment for all studies will be discussed in a subsequent section. For skilled
hours per day, none of the three studies (Sovie and Jawad, 2001, excluded)
that analyzed RN hours had significant findings. By contrast, Unruh’s
(2003) licensed hours measure had significant findings.
The evidence linking staffing to falls exhibited about the same proportion
of significant results as that of pressure ulcers. About half (11 of 20) of the
pressure ulcer analyses had significant findings. Nearly half (9 of 19) of the
falls analyses had significant findings (Table 3). RN skill mix was significant
in three out of seven analyses. Licensed proportion of total staff was signifi-
cant in one study but not the other. Total nursing hours were associated with
falls in three out of seven analyses. Skilled hours per patient day was a sig-
nificant predictor of falls in two out of three analyses: one of patient-to-nurse
ratio (Krauss et al., 2005), the other of licensed hours (Unruh, 2003).
The only measure with consistent evidence of a relationship with both
adverse events was licensed hours per patient day (Unruh, 2003). By con-
trast, RN hours was not significant for either outcome in Cho etal.’s (2003)
study and not significant in Needleman et al.’s (2002) or Mark, Harless,
et al.’s (2004) analyses of pressure ulcers. The RN and licensed skill mix
measures exhibited significant negative results in 56% (9 of 16) of the
analyses, particularly for pressure ulcers, and a positive result for falls in
Unruh’s (2003) study. Needleman et al.’s nonsignificant results for the RN
proportion of licensed staff in large medical and surgical samples may indi-
cate that this aspect of skill mix is less important than the RN complement
alone or the licensed complement of total staff.
In the eight studies with more than one staffing measure, the coefficients
were estimated in a combined model in four studies. The estimation of
666 Western Journal of Nursing Research
at University of Ulster Library on January 4, 2015wjn.sagepub.comDownloaded from

coefficients in separate or combined models may influence their significance.
Staffing measures are often correlated, whether because of shared numerators
or consistent management decisions. Testing correlated measures in com-
bined models may yield lower effects than if the measures are tested sepa-
rately. Both separate and combined models should be analyzed and reported.
Separate models permit comparison with measures in other studies. A com-
bined model reveals the effects controlling for other staffing factors.
Data Analyses
Table 1 notes the statistical analyses for the studies. Two studies pre-
dicted the adverse event itself at the patient level from nurse-level staffing
(Krauss et al., 2005) or hospital-level staffing (Cho et al., 2003) in logis-
tic regression models. In the other nine studies, both the dependent and the
independent variables were measured at the same level (hospital or nursing
unit). The dependent variables in these studies were adverse event rates,
counts, or indexes. Lichtig et al. (1999) constructed a hospital-level pres-
sure ulcer index as the ratio of actual to expected outcome rates, adjusted
by the hospital’s diagnosis-related group (DRG) mix. The state-specific
outcome rate for each DRG was considered the expected rate. Needleman
et al. (2002) and Unruh (2003) used count data for their dependent vari-
ables. They regressed the hospitals’ annual number of adverse events on the
staffing measures while controlling for the number of patients at risk for the
adverse events. Needleman et al. conducted negative binomial regression
to account for the maldistribution of adverse events and their relatively low
occurrence. Unruh used random effects poisson regression. Her study was
the only one to identify significant effects for both staffing measures
(licensed hours; licensed skill mix) for both outcomes. Mark, Harless,
et al. (2004) conducted the only longitudinal analysis:a dynamic panel
study. Hospital-level pressure ulcer ratios were analyzed using generalized
method of moments. Mark,Harless, et al. found that results on another
outcome (mortality) differed across statistical models (ordinary least
squares—OLS, fixed-effects, or dynamic panel). Staffing effects that were
significant in an OLS model were not significant in a dynamic panel model.
They did not present this comparison for the pressure ulcer ratio, for which
staffing was nonsignificant in a dynamic panel model.
In three nursing unit level studies, the dependent variable was the nurs-
ing unit’s adverse event rate (Blegen etal., 1998; McGillis Hall et al.,
2004; Sovie & Jawad, 2001). Sovie and Jawad (2001) first transformed the
pressure ulcer rate to a logarithmic form because of the skewed distribution.
Lake, Cheung / Nurse Staffing and Adverse Events 667
at University of Ulster Library on January 4, 2015wjn.sagepub.comDownloaded from

Blegen et al. (1998) tested for curvilinear relationships between skill mix
and outcomes by including a dummy variable for RN skill mix greater than
.875. It appears that Blegen et al. and McGillis Hall et al. (2004) did not
account in statistical models for the nesting of their nursing units within
hospitals. Conventional analysis considers all nursing units independent of
each other. The effect may be the underestimation of standard errors and
concomitant inflation of Type I error: finding significant effects when there
are none. Blegen et al. found significant effects for pressure ulcers but not
falls. McGillis Hall et al.’s results were not significant. Blegen and Vaughn
(1998) used generalized estimating equations to analyze repeated staff-
ing and adverse event measures within the same nursing units over time.
Dunton et al. (2004) used a generalized linear mixed model with a nega-
tive binomial distribution and a nested design to predict falls.
Three studies used multilevel models (Blegen & Vaughn,1998; Cho
et al., 2003; Dunton et al., 2004). Typically, these models account for nest-
ing of patients or nursing units within the same hospital in samples that
contain multiple hospitals. This typical application of multilevel models
was used by Cho et al. (2003) and Dunton et al. (2004). By contrast,
Blegen and Vaughn (1998) used a multilevel model for repeated measures
over time within nursing units. All nursing units in their study were in the
same hospital. There was no nesting of organizational units.
Five of the nine studies with cross-sectional designs and aggregate
staffing measures used statistical models tailored to the infrequent nature of
the adverse events or the nested data structure. In this subset, findings were
significant (Unruh, 2003), mixed (Cho et al., 2003), and not significant
(Needleman et al., 2002) for pressure ulcers and significant (Unruh, 2003),
mostly significant (Dunton et al., 2004), mixed (Blegen & Vaughn, 1998),
and not significant (Cho et al., 2003) for falls. The evidence from this sub-
set is no clearer than that from the full set of studies.
Risk Adjustments for Patient Characteristics
It is important to adjust for differing risks of adverse events across patients.
Patient characteristics are the principal determinant of patient adverse events.
Krauss et al. (2005) matched cases (patients who fell) with randomly sampled
controls to identify differences between those who fell and those who did not.
The researchers included patient-related factors,such as confusion and gait
imbalance, in their multivariate model linking staffing to falls.
Four hospital-level studies used demographic, insurer, and clinical data
to adjust for the patient distribution across hospitals. One used clinical data
668 Western Journal of Nursing Research
at University of Ulster Library on January 4, 2015wjn.sagepub.comDownloaded from

from the admission record (Unruh, 2003). Unruh (2003) adjusted for each
hospital’s average patient severity with MediQual severity scores. MediQual
scores are calculated from patient age, sex, race, ethnicity, and 23 key clin-
ical findings from the admission record. Admission data are considered
superior to discharge data for calculating severity of illness because the
severity measure reflects the patient’s baseline status and is not confounded
with diagnoses and procedures that may have developed during the hospi-
tal stay and potentially be related to the quality of care. The other three
studies used discharge abstract data (Cho et al., 2003; Mark, Harless,
et al., 2004; Needleman et al., 2002). Cho et al. (2003) controlled for
patient age, sex, race, primary insurer, DRG, number of diagnoses at admis-
sion, and whether the admission was scheduled or not. Needleman et al.
(2002) controlled for the rate of the adverse event in the patient’s DRG
and the patient’s age, sex, state of residence, primary insurer, whether the
admission was emergent or not, and the presence or absence of 13 chronic
diseases. Needleman et al. used these characteristics in a model to predict
each patient’s probability of the adverse event. The probabilities were
summed to yield an expected number of adverse events for each hospital
and entered as a control variable in their model. Mark,Harless, et al.
(2004) used the Medstat complications-of-care software to estimate each
patient’s risk for developing a pressure ulcer. This software uses diagnosis
and procedure codes and other patient characteristics. Their outcome vari-
able was a complication ratio that was the observed number of complica-
tions divided by the expected number of complications based on the
Medstat probabilities for that hospital’s discharges. In addition, their model
controlled for the hospital’s Medicare CMI and the proportion of discharges
that were Medicare or Medicaid, to account for payer mix. The fifth hospital-
level study, by Lichtig et al. (1999), adjusted for patient mix by dividing
the nurse staffing hours by each hospital’s average NIW. The average NIW
was calculated for each year from the patient data. The NIW reflects the
average nursing resources needed for the patients in a DRG.
The approaches to adjust for patient mix across hospitals have their
respective merits. Mark, Harless, et al.’s (2004) approach focused on the
risk for the particular adverse event,pressure ulcer. This is an excellent
approach, although primary data on risk status,which are not routinely
available, would be even better. Unruh (2003) had the best data for generic
severity of illness adjustment: admission clinical findings. Needleman
et al. (2002) and Cho et al. (2003) had roughly equivalent sets of charac-
teristics that capitalize on administrative data sources although Needleman
et al. included 13 chronic diseases, whereas Cho et al. had the number of
Lake, Cheung / Nurse Staffing and Adverse Events 669
at University of Ulster Library on January 4, 2015wjn.sagepub.comDownloaded from

diagnoses on admission. The study that did not incorporate patient-specific
characteristics (Lichtig et al., 1999) focused instead on adjusting for nurs-
ing resources based on the DRG mix. This approach lacks the comprehen-
siveness of the other approaches. It makes the importance about patient
differences the nursing resources needed rather than clinical and demo-
graphic factors that may influence patients’ risks for adverse events.
Of the five nursing unit level studies,two used patient-specific data
to adjust for patient mix differences. Blegen etal. (1998) adjusted for
patients’ average daily nursing acuity, measured from monthly reports.
McGillis Hall et al. (2004) used a patient complexity score, which is
assigned to all inpatients discharged from Canadian hospitals. The com-
plexity scores are 1 (no complexity) through 4 (highly complex) and 9
(complexity not related to the case mix group). The researchers did not
describe how they aggregated the complexity score at the nursing unit level.
Their table with regression results contained “average resource intensity
weights,” not a complexity score, as an independent variable. The intensity
weights were not described in the text. The other three nursing unit level
studies adjusted for patient mix differences by including the nursing unit
type as a control variable (Blegen & Vaughn, 1998) or by doing separate
analyses for each nursing unit type (Dunton et al., 2004; Sovie & Jawad,
2001). In addition, Blegen and Vaughn (1998) adjusted for each hospital’s
Medicare case mix. Of all approaches across the 11 studies, adjusting for
patient mix differences by the nursing unit type is the weakest approach
because it does not incorporate differing distributions of risk or severity of
illness across nursing units of the same type.
Discussion
Collectively, these studies have not identified the contributions of nurse
staffing to patient falls and pressure ulcers. A recent systematic review of
nurse staffing and a variety of outcomes (Lang, Hodge, Olson, Romano, &
Kravitz, 2004) reached the same conclusion: “The evidence does not sup-
port relationships between nurse staffing and the incidence of pressure
ulcers, patient falls, [etc.]” (p. 335). Only 3 of the 11 studies noted a theo-
retical model (Cho et al., 2003; Dunton et al., 2004; Sovie & Jawad,
2001), and none was a framework that was presented initially. Cho et al.
(2003) referenced Cho’s (2001) nurse staffing and patient outcomes model,
which focuses on adverse outcomes rather than outcomes more generally.
The latter 2 studies referenced Donabedian’s (1992) health quality model,
670 Western Journal of Nursing Research
at University of Ulster Library on January 4, 2015wjn.sagepub.comDownloaded from

which was the basis for Mitchell et al.’s (1998) model. Differences in the-
oretical approaches or the lack of theoretical bases for the studies may
account in part for the divergent findings. A better articulation of theory and
empirics would accelerate the science.
Two features of the literature stand out: inconclusive evidence and
methodological variety. The variety begins with differing focal levels of the
independent variables (hospital and nursing unit), continues with multiple
measures of the adverse events, principally because of administrative ver-
sus incident or observational data sources and multiple variations of the
core staffing measures, and concludes with diverse analytic approaches and
risk adjustment strategies.
The key question is whether the methodological variety accounts for the
inconclusive evidence. A logical answer would derive from emphasis on the
evidence from the studies with the best methods. The methodological vari-
ety, however, is not marked by a mix of inferior and superior methods per
se but rather by the best methods available to researchers given common
research constraints or by different ideas about which staffing variant to
study. The diversity of methods and measures revealed in this area of
inquiry provides a catalog of methods in nursing outcomes research more
generally. The methodological variety could be considered a strength of
these empirical areas. Multiple data sources and methods yield richer
science. Unfortunately, despite the methodologic richness, the evidence is
weak. As detailed above, five cross-sectional studies with more sophisti-
cated analytic approaches yielded equivocal results. Cross-sectional research
has a limited capacity to establish causality. The one longitudinal study of
pressure ulcers resulted in nonsignificant findings. More longitudinal
research is needed to provide additional evidence based on this stronger
design.
This literature has two distinct subsets, reflecting nursing outcomes
research more generally. One set linked staffing to adverse events at the
nursing unit level. Adverse events were measured from primary data on
pressure ulcers and incident reports of falls. Few of these studies adjusted
for differences in patient mix across nursing units or accounted for cluster-
ing of nursing units in hospitals. The second set linked nurse staffing to
adverse events at the hospital level. Pressure ulcers and falls were measured
from diagnosis or event codes in administrative databases. These studies
incorporated extensive patient-specific risk adjustment.
Future research on falls should be based on online adverse event report-
ing systems, on incident reports, or, in administrative data, on diagnosis
codes. A drawback to incident report data is the overwhelming problem of
Lake, Cheung / Nurse Staffing and Adverse Events 671
at University of Ulster Library on January 4, 2015wjn.sagepub.comDownloaded from

the lack of systematic incident reporting. The “fall with injury” event code
in administrative data has theoretical and practical weaknesses. All falls are
potentially injurious and may reflect poorer quality of care. Injuries may
not be identified in all instances.
Pressure ulcers should be measured by observational prevalence studies,
where possible, because of the accuracy of the rate because of the compre-
hensive assessment of all patients and the certainty about whether an ulcer
was acquired in the hospital. Researchers should analyze all pressure
ulcers, including Stage I. The majority (56%) of nosocomially acquired
pressure ulcers are Stage I (Whittington & Roberts, 2000). Not only does
the exclusion of Stage I ulcers underestimate the incidence rates, but Stage
I ulcers are at a stage of development where interventions can reverse pro-
gression to a Stage II ulcer. Stage I ulcers may be the most sensitive to the
adequacy of nurse staffing. Although observational prevalence studies
obtain the highest quality pressure ulcer data, the data collection burden is
great. For this reason, pressure ulcers measured from administrative data
are still an acceptable alternative. Data that indicate whether a pressure ulcer
developed during the hospital stay are the most important. Researchers
should report the specific diagnosis codes used.
The multiple staffing hours and skill mix measures used in these studies
make the findings difficult to compare. The variety reflects a debate about
the most valid measures of nurse staffing and the lack of a conceptual def-
inition of nurse staffing that could guide the measure selection (Mark,
Hughes, & Jones, 2004). It also reflects data availability. The American
Hospital Association Annual Survey Database had a decade-long gap (1994
to 2002) in the reporting of NA data. Since 2003, all three categories of staff
(RN, LPN, and NA) are reported. Other data sets have neither LPNs nor
NAs. Although it will result in a proliferation of models, at this point the
investigation of multiple staffing measures is necessary. The evidence sup-
ports a focus on skill mix and skilled hours of care. Two variants of both
measures should be evaluated and reported separately: RNs alone and RNs
and LPNs. No studies contrasted an RN measure to an RN and LPN mea-
sure to differentiate their effects. The principal effects were noted for RN
skill mix and licensed hours of care but not RN or total hours of care. LPNs
may be an important complement to RNs, but NAs less so. A measure of
total hours of care deserves additional testing given the multiple potential
bases for the mixed evidence. For consistent interpretation,the specific
operational definitions for nursing care hours per patient day and skill mix
endorsed by the National Quality Forum (2004) should be used. To best
reveal the relative contributions of hours versus mix, future research should
672 Western Journal of Nursing Research
at University of Ulster Library on January 4, 2015wjn.sagepub.comDownloaded from

test measures in separate and combined models. As Lang etal. (2004)
pointed out, a weakness in the literature is that researchers merely report
whether results are significant but do not interpret the clinical significance
of effect sizes. Also, researchers must report baseline rates to assess whether
staffing changes could reduce the rates.
The existence of studies oriented at two different organizational levels
raises theoretical and empirical questions. The effect of nurse staffing may be
stronger at one level of organization. A recent evidence report prepared for
the Agency for Healthcare Research and Quality (Hickam etal., 2003) noted,
Studies of nurse staffing and patient safety with data aggregated at the unit
level were judged to have better quality than studies that aggregated data at
the hospital level, due to the elimination of data pooling across divergent
types of nursing units. (p. 25)
Among the 10 organization-level studies described herein, neither level was
more informative. The findings were mixed within each subset.
Research to date has necessarily focused on one organizational level or
the other, rather than both, because of data constraints. All patients at risk for
falls or pressure ulcers are cared for in particular nursing units in particular
hospitals. Research at the hospital level necessarily ignores that patients are
clustered in particular nursing units. When hospital-level data are available,
the data set usually has a large number of hospitals but no unit-level data.
Research at the unit level often ignores that nursing units are clustered in
hospitals. When unit-level data are available, they are usually from a small
set of hospitals. In this instance, hospital-level data are not very informative
because variation at the hospital level is modest given the small N. Future
studies should employ statistical techniques that address the multilevel
nature of staffing and adverse event data and that are tailored to rare events.
Dunton et al. (2004), who had an exceptional database, accounted analyti-
cally for the clustering of nursing units in hospitals and used models appro-
priate to rare events. A large data set with both hospital and unit data is
required to support a multilevel purpose that has not been attempted: com-
parison of hospital-level-only results to unit-level-only results. This com-
parison could reveal whether, how, and for which measures and outcomes
the nursing effect predominates at the nursing unit or hospital level.
The study by Krauss et al. (2005) is exceptional in this literature for its
absence of organizational context. The focus was on a specific nurse–patient
dyad. The staffing effect in a multivariate model was remarkable. Compared
to patients whose nurse had three or fewer patients, the likelihood of falling
Lake, Cheung / Nurse Staffing and Adverse Events 673
at University of Ulster Library on January 4, 2015wjn.sagepub.comDownloaded from

was 3 times higher for patients whose nurse had four to six patients and was
7 times higher when the nurse had seven or more patients. This study’s sig-
nificance cannot be overstated. In an equivocal literature,it demonstrates
unequivocally that falls are sensitive to nurse staffing. It establishes the theo-
retical and empirical importance of both the particular nurse–patient dyad and
the patient-to-nurse ratio. This study suggests that designs that focus on par-
ticular situations, in contrast to aggregate staffing and outcome indicators, can
be informative in outcomes research. This study focused on an event (falling)
with a nature that is distinct from most outcomes (e.g., pressure ulcers, satis-
faction, or death). An outcome such as falling, which is temporally discrete,
directly reflects the patient care situation. An analogous event would be a med-
ication error. Other outcomes reflect a cumulative episode of nursing care.
This distinction yields a theoretical advance in the understanding of how
nurses and nursing affect patient outcomes. The effect may be observed and
understood through several perspectives: particular situations and episodes.
The methods used in these studies to adjust for patient differences across
settings (nursing units or hospitals) ranged from minimal to extensive. Future
research must incorporate patient-specific risk or health characteristics.
Adjustments based on nursing acuity may not account fully for consequential
patient differences. Adjustments limited to nursing unit type are insufficient.
Hospital-level studies have the advantage over nursing unit studies of rich
patient-level data in administrative databases for risk adjustment.
It is unclear whether findings may differ in hospital-level studies
depending on whether patient-specific characteristics are incorporated into
organization-level probabilities rather than retained as independent vari-
ables in logistic models of the adverse event at the patient level. Four of the
hospital-level studies took the former analytic approach; the fifth took the
latter approach. Neither approach yielded consistent findings. Similarly, all
nursing unit level studies predicted nursing unit level adverse event rates,
rather than patient-level events. Models of patient-level events require data
on all patients at risk for the events. These data cannot be obtained from
incident reports of falls, which include only the patients who fell, not all
patients. Observational data on pressure ulcers have the potential to support
patient-level analytic models as data on all patients are collected.
For nurse-sensitive quality indicators to guide the assessment of where pro-
blems reside in nursing care,comprehensive frameworks for nursing care
should be tested. Broader coverage of the structural elements that support the
delivery of high-quality care might prove more illuminating. The fact that the
research to date is equivocal (at times there is a staffing effect and at other
times not) underscores the point that having more nurses, rather than more of
674 Western Journal of Nursing Research
at University of Ulster Library on January 4, 2015wjn.sagepub.comDownloaded from

the right ones and in the right environment,does not necessarily achieve
better outcomes. The likely influences of differences among RNs and prac-
tice environments on these outcomes are nearly absent from the literature.
Key areas for growth in the science are to explore how differences among
RNs and nurses’ practice environments are associated with adverse events.
Recent evidence revealed that a higher proportion of RNs with bachelor’s
degrees on a nursing staff was associated with fewer patient deaths follow-
ing common surgeries (Aiken,Clarke, Cheung, Sloane, & Silber, 2003).
More evidence about how levels of nursing education are associated with
outcomes is critical as policy responses are developed to address the nurse
shortage. Blegen, Vaughn, and Goode (2001) advocated including nursing
experience and the number of agency or float staff as staffing measures.
Dunton et al. (2004) included the percentage of hours provided by contract
nursing staff of all skill levels. This variable was not significant in models
that included total nursing hours and skill mix. Measures of clinical exper-
tise could be introduced to establish the theoretical basis for staffing effects
associated with differences among RNs and to inform managers how to cul-
tivate and optimize nursing expertise. Differences in the nursing practice
environment may have a direct effect on adverse events or a moderating
effect on the relationship between staffing and adverse events.
This thorough appraisal of the design, measurement, and analysis fea-
tures of 11 multivariate studies has revealed that the title question,“Are
patient falls and pressure ulcers sensitive to nurse staffing?” cannot be
answered definitively at this stage of the science. Discerning the precise
measures and approaches used in each study posed a considerable chal-
lenge. Interpreting the influence of myriad factors on the disparate results
was formidable. To advance the science and build the evidence needed to
guide practice and policy, multivariate research on staffing and adverse
events should continue with attention to the selection of superior design,
measurement, and analytic and reporting approaches in studies that test
questions derived from comprehensive theoretical frameworks.
References
Aiken, L. H., Clarke, S. P., Cheung, R. B., Sloane, D. M., & Silber, J. H. (2003). Educational
levels of hospital nurses and surgical patient mortality. Journal of the American Medical
Association, 290, 1617-1623.
Aiken, L. H., Clarke, S. P., & Sloane, D. M. (2002). Hospital staffing, organization, and qual-
ity of care: Cross-national findings. International Journal for Quality in Health Care,
14(1), 5-13.
Lake, Cheung / Nurse Staffing and Adverse Events 675
at University of Ulster Library on January 4, 2015wjn.sagepub.comDownloaded from

Aiken, L. H., Sochalski, J., & Lake, E. T. (1997). Studying outcomes of organizational change
in health services. Medical Care, 35(Suppl. 11), NS6-NS18.
American Nurses Association. (1999). Nursing-sensitive quality indicators for acute care set-
tings and ANA’s Safety and Quality Initiative (PR-28). Washington, DC: Author.
Ayello, E. A., & Braden, B. (2001). Why is pressure ulcer risk assessment so important?
Nursing 2001, 31(11), 75-79.
Blegen, M. A., Goode, C. J., & Reed, L. (1998). Nurse staffing and patient outcomes. Nursing
Research, 47(1), 43-50.
Blegen, M. A., & Vaughn, T. (1998). A multisite study of nurse staffing and patient occur-
rences. Nursing Economic$, 16(4), 196-203.
Blegen, M. A., Vaughn, T. E., & Goode, C. J. (2001). Nurse experience and education: Effect
on quality of care. Journal of Nursing Administration, 31(1), 33-39.
Braden, B. J., & Maklebust, J. (2005). Preventing pressure ulcers with the Braden Scale.
American Journal of Nursing, 105(6), 72.
Cho, S.-H. (2001). Nurse staffing and adverse patient outcomes: A systems approach. Nursing
Outlook, 49(2), 78-85.
Cho, S.-H., Ketefian, S., Barkauskas, V. H., & Smith, D. G. (2003). The effects of nurse staffing
on adverse events, morbidity, mortality and medical costs. Nursing Research, 52(2), 71-79.
Donabedian, A. (1992). The role of outcomes in quality assessment and assurance. Quality
Review Bulletin, 11, 356-360.
Dunton, N., Gajewski, B., Taunton, R. L., & Moore, J. (2004). Nurse staffing and patient falls
on acute care hospital units. Nursing Outlook, 52(1), 53-59.
Frantz, R. A. (1997). Measuring prevalence and incidence of pressure ulcers. Advances in
Wound Care, 10(1), 21–24.
Hickam, D. H., Severance, S., Feldstein, A., Ray, L., Gorman, P., Schuldheis, S., etal. (2003).
The effect of health care working conditions on patient safety (Evidence Report/Technology
Assessment 74; Oregon Health & Science University Contract 290-97-0018; AHRQ
03-E031). Rockville, MD: Agency for Healthcare Research and Quality.
Hitcho, E. B., Krauss, M. J., Birge, S., Claiborne Dunagan, W., Fischer, I., Johnson, S., et al.
(2004). Characteristics and circumstances of falls in a hospital setting:A prospective
analysis. Journal of General Internal Medicine, 19, 732-739.
Iezzoni, L. I. (Ed.). (2003). Risk adjustment for measuring health care outcomes (3rd ed.).
Chicago: Health Administration Press.
Irvine, D., Sidani, S., & Hall, L. M. (1998). Linking outcomes to nurses’ roles in health care.
Nursing Economic$, 16(2), 58-64, 87.
Kane, R. L. (Ed.). (2006). Understanding health care outcomes research (2nd ed.). Sudbury,
MA: Jones and Bartlett.
Krauss, M. J., Evanoff, B., Hitcho, E., Ngugi, K. E., Dunagan, W. C., Fischer, I., etal. (2005).
A case-control study of patient, medication, and care-related risk factors for inpatient falls.
Journal of General Internal Medicine, 20(2), 116-122.
Lang, T. A., Hodge, M., Olson, V., Romano, P. S., & Kravitz, R. L. (2004). Nurse-patient
ratios. Journal of Nursing Administration, 34(7/8), 326-337.
Lichtig, L. K., Knauf, R. A., & Milholland, D. K. (1999). Some impacts of nursing on acute
care hospital outcomes. Journal of Nursing Administration, 29(2), 25-33.
Mark, B. A., Harless, D. W., McCue, M., & Xu, Y. (2004). A longitudinal examination of hos-
pital registered nurse staffing and quality of care. Health Services Research, 39, 279-300.
Mark, B. A., Hughes, L. C., & Jones, C. B. (2004). The role of theory in improving patient
safety and quality health care. Nursing Outlook, 52(1), 11-16.
676 Western Journal of Nursing Research
at University of Ulster Library on January 4, 2015wjn.sagepub.comDownloaded from

Lake, Cheung / Nurse Staffing and Adverse Events 677
Mark, B. A., Sayler, J., & Smith, C. S. (1996). A theoretical model for nursing systems out-
comes research. Nursing Administration Quarterly, 20(4), 12-27.
McGillis Hall, L., Doran, D., & Pink, G. H. (2004). Nurse staffing models, nursing hours, and
patient safety outcomes. Journal of Nursing Administration, 34(1), 41-45.
Mitchell, P. H., Ferketich, S., Jennings, B. M., & American Academy of Nursing Expert Panel
on Quality Health Care. (1998). Quality health outcomes model. Image:Journal of
Nursing Scholarship, 30(1), 43-46.
Morse, J. M. (1996). Preventing patient falls. London: Sage.
National Quality Forum. (2004). National voluntary consensus standards for nursing-sensitive
care: An initial performance measure set—A consensus report (NQFCR-08-04). Washington,
DC: Author.
Needleman, J., Buerhaus, P., Mattke, S., Stewart, M., & Zelevinsky, K. (2002). Nurse-staffing lev-
els and the quality of care in hospitals. New England Journal of Medicine, 346, 1715-1722.
Norton, D. (1996). Calculating the risk: Reflections on the Norton Scale. Advances in Wound
Care, 9(6), 38-43.
Olson, B., Langemo, D., Burd, C., Hanson, D., Hunter, S., & Cathcart Silberberg, T. (1996).
Pressure ulcer incidence in an acute care setting. Journal of Wound Ostomy and
Continence Nursing, 23(1), 15-22.
Rutledge, D. N., Donaldson, N. E., & Pravikoff, D. S. (2003). Update 2003: Fall risk assessment
and prevention in hospitalized patients. Online Journal of Clinical Innovations, 6(5), 1-55.
Schmid, N. A. (1990). Reducing patient falls: A research-based comprehensive fall prevention
program. Military Medicine, 155, 202-207.
Silber, J. H., Rosenbaum, P. R., & Ross, R. N. (1995). Comparing the contributions of groups
of predictors: Which outcomes vary with hospital rather than patient characteristics?
Journal of the American Statistical Association, 90(429), 7-18.
Sovie, M. D., & Jawad, A. F. (2001). Hospital restructuring and its impact on outcomes.
Journal of Nursing Administration, 31(12), 588-600.
Unruh, L. (2003). Licensed nurse staffing and adverse events in hospitals. Medical Care,
41(1), 142-152.
Whittington, K., Patrick, M., & Roberts, J. L. (2000). A national study of pressure ulcer preva-
lence and incidence in acute care hospitals. Journal of Wound, Ostomy and Continence
Nursing, 27(4), 209–215.
Wunderlich, G. S., Sloan, F. A., & Davis, C. K. (Eds.). (1996). Nursing staff in hospitals and
nursing homes: Is it adequate? Washington, DC: National Academy Press.
Eileen T. Lake, PhD, RN, is an assistant professor at the School of Nursing,Secondary
Faculty, Department of Sociology, Center for Health Outcomes and Policy Research,
University of Pennsylvania.
Robyn B. Cheung, PhD, RN, is a research associate at the Center for Health Outcomes and
Policy Research, University of Pennsylvania.
at University of Ulster Library on January 4, 2015wjn.sagepub.comDownloaded from