Nursing Resources and Patient Outcomes in Intensive Care: A Systematic Review

This systematic review evaluates the empirical evidence linking nursing resources to patient outcomes in intensive care settings as a framework for future research in this area. The review found that there is a relationship between nursing resources and adverse events in intensive care, but more research is needed to establish the mechanisms and design large multi-centre prospective studies.

Review
Nursing resources and patient outcomes in intensive car
A systematic review of the literature
Elizabeth Westa,*, Nicholas Maysb, Anne Marie Raffertyc,
Kathy Rowand, Colin Sandersonb
aSchool of Health and Social Care,University of Greenwich,Southwood Site,Avery Hill Road,Eltham, London SE9 2UG,UK
bLondon School of Hygiene and Tropical Medicine, Keppel Street,London WC1E 7HT,UK
c Florence Nightingale School of Nursing and Midwifery,King’s College London,James Clerk Maxwell Building,
57 Waterloo Road, London SE1 8WA,UK
dIntensive Care National Audit and Research Centre,Tavistock House,Tavistock Square,London WC1H 9HR,USA
Received 7 November 2006; received in revised form 25 May 2007; accepted 3 July 2007
Abstract
Objectives:To evaluate the empirical evidence linking nursing resources to patient outcomes in intensive care se
framework for future research in this area.
Background:Concerns about patient safety and the quality of care are driving research on the clinical and cost-e
health care interventions, including the deployment of human resources. This is particularly important in intens
large proportion of the health care budget is consumed and where nursing staff is the main item of expenditure
tions about staffing levels have been made but may not be evidence based and may not always be achieved in
Methods:We searched systematically for studies of the impact of nursing resources (e.g. nurse–patient ratios, n
education,training and experience) on patientoutcomes,including mortality and adverse events,in adultintensive care.
Abstracts of articles were reviewed and retrieved if they investigated the relationship between nursing resourc
outcomes.Characteristics of the studies were tabulated and the quality of the studies assessed.
Results:Of the 15 studies included in this review, two reported a statistical relationship between nursing resourc
mortality and adverse events, one reported an association to mortality only, seven studies reported that they c
null hypothesis of no relationship to mortality and 10 studies (out of 10 that tested the hypothesis) reported a r
adverse events. The main explanatory mechanisms were the lack of time for nurses to perform preventative m
patientsurveillance.The nurses’ role in pain controlwas noted by one author.Studies were mainly observationaland
retrospective and varied in scope from 1 to 52 units. Recommendations for future research include developing
linking nursing resources to patient outcomes, and designing large multi-centre prospective studies that link pa
nursing care on a shift-by-shift basis over time.
# 2007 Elsevier Ltd.All rights reserved.
Keywords:Nursing; Outcomes assessment; Hospital mortality; Complications; Intensive care; Health services research
What is already known about the topic?
A previous systematic review concluded thatthere is
currently insufficientevidence to rejectthe hypothesis
www.elsevier.com/ijns
Available online at www.sciencedirect.com
International Journal of Nursing Studies 46 (2009) 993–1011
* Corresponding author.Tel.: [phone].
E-mail address: [e-mail] (E.West).
0020-7489/$ – see front matter # 2007 Elsevier Ltd.All rights reserved.
doi:10.1016/j.ijnurstu.2007.07.011

of no association between nursing resources and mortality
in intensive care.
Several non-systematic reviews suggest that there may be
a link between nurse staffing and the developmentof
adverse events among patients in intensive care.
There are many methodologicaldifficulties to be over-
come in conducting research in this area.
What this paper adds?
This paper describes and critiques studies of the impact of
nursing resources on mortality and adverse events in one
systematic review.
Focuses attention on the methods used and devises a way
of assessing the scientific rigour of observational studies
which are notoriously difficult to evaluate.
Finds that studies of adverse events in ICU were likely to
report significant relationships between staffing and out-
comes butthatthe number of positive associations was
small relative to the number of hypothesised relationships
tested.
Links study findings to features of research design.The
three studies thatfound a relationship between nursing
and mortality were small prospective studies whereas the
seven studies that found no association were large multi-
unit studies based on administrative data.
Shows that several studies that failed to reject the hypoth-
esis ofno association between staffing levels and out-
comes had little variation in staffing levels.
1. Introduction
Historians of criticalcare nursing trace its origins to
the increasing demand for health care in the 1950s and to
the invention of the ‘iron lung,’ a precursor of the modern
ventilator (Reis Miranda etal., 1998;Bennettand Bion,
1999;Fairman and Lynaugh,1998).Intensive care units
(ICUs) were notdesigned simply to care forthe most
seriously ill, but for those for whom survival was possible,
but not certain. Patients admitted to intensive care were to
be closely observed by skilled nurses capable ofinter-
vening clinically and of mobilising the resources of the
hospitalon their behalf.Although intensive care is often
associated with high technology,this can obscurethe
importance ofthe two cardinalorganisationalfeatures
of intensive care:triage and surveillance (Fairman and
Lynaugh,1998;Sandelowski,2000).Advances in tech-
nology are tools to support staff in monitoring and treating
patients who are critically ill,rather than a substitute for
skilled health care staff (Sandelowski,2000;Westet al.,
2004).
For many, intensive care is central to the activities of the
hospital because its function is so clearly aligned with the
main goalof saving lives.Clinicians and managers have
often maintained staffing levels in ICU by,for example,
closing beds in other parts of the hospital, transferring staff
from other parts of the hospital or employing agency nur
(DoH, 2000a).Recommendations aboutstaffing in ICUs
have been in place since the late 1960s in the UK and the
‘gold standard’ of one nurse to each ICU patient is widely
accepted (British Medical Association, 1967; Royal Colleg
of Nursing, 2000, 2003). However, there is now a great d
of uncertainty and debate aboutthe levels of staffing and
skill mix that are required for patient safety.
Criticalto Success,a reportby the AuditCommission
(1999), found wide variations in the numbers and grades
staff employed in ICUs in the UK and in the number of
nurses who were supernumerary (i.e. not engaged in dire
patient care). They also found that mortality was unexpec
edly high in some UK ICUs,but they were unable to
establish whether or not this was linked to different staffi
levels. The report recommended a more flexible approach
nurse staffing but this is difficult to implement in the abse
of good measuresof staffresources,patientneedsand
outcomes.There isclearly a need forsound empirical
evidence to guide decisions about the deploymentof staff
to ensure patientsafety and improve both the quality and
cost-effectiveness of care.
2. Purpose of the study
This paper reviews empiricalevidence aboutthe link
between nursing resources and patient outcomes in inten
sive care,assessesits strengthsand weaknesses,and
identifies where furtherresearch is required.The focus
is on whetherand to whatextentcharacteristics ofthe
nursing workforce,such asthe numberof nursesper
patient and the skill mix of the nursing staff,affect rates
of mortality and adverse events,such as post-operative
complications and hospital-acquired infections.Although
there is a burgeoning literature on organisational behavio
in intensive care, including topics such as, communicatio
and collaboration within teams,we focuson variables
associated with the conceptof human capital(Becker,
1964),in this case,the number of nurses and their levels
of education,training and experience.Limiting the scope
of the review in this way allows closerscrutiny ofthe
quality of papers included. This is important because man
of the studies in this area are observationalratherthan
experimental,and as such arenotoriously difficultto
evaluate (Downs and Black,1998).
Much is already known about the importance of medic
staff in intensive care.Pronovost et al. (2002) reviewed 26
observational studies of the impact of ICU physician staffi
patterns on patient outcomes and concluded that high in
sity physician staffing (a closed ICU or one where consult
tion with an intensivist is mandatory) was associated with
reduced hospital and ICU mortality and length of stay. Th
research question addressed here is whetherthere is any
evidence that patterns of nurse staffing are similarly imp
cated in patient outcomes.
E. West et al. / International Journal of Nursing Studies 46 (2009) 993–1011994

3. Methods
3.1.Search strategy
Previous authors have drawn attention to the difficulties
of defining standard search terms in this area.Pronovost
et al. (2002), for example, reported that their comprehensive
search for empirical evidence on physician staffing patterns
did not uncover some of their own work. Carmel and Rowan
(2001) note that their review of the literature on organisa-
tional factors related to patient mortality was hampered by
the fact that electronic databases tend to adopt a biomedical-
intervention perspective. We therefore used many different
terms and strategies to try to ensure thatthe search for
relevant articles was as complete as possible:
Unit labels intensive care units; intensive care; surgical
intensive care; critical care; high dependency;
open,closed,critically ill
Outcomes outcome assessment, treatment outcomes,
mortality,morbidity,adverse events,infections,
length of stay,complications,error/s,
readmission/s,admission/s
Nursing nursing staff,hospital staffing
Workforce labour force,health care workforce,manpower,
workforce policy,training,education,nurse/s,
number,size,staffing levels,ratio/s, skill mix,
substitution, specialisation,training,education,
grade/s, staff development, human resources, HR
management, personnel staffing and scheduling
Workload volume of activity,workload
Methods multi-level modelling, case mix adjustment, risk
adjustment, APACHE, SAPS, case-control study,
retrospective study
The ‘related articles’feature in Pubmed was used in
locating studies, as was Google Scholar. The bibliographies
of articles retrieved atthe beginning ofthe search were
scanned for new references.
3.2.Inclusion and exclusion criteria
Studies were included if they were conducted exclusively
in intensive or critical care settings and allowed data on one or
more of the nursing workforce variables to be related to data
on mortality or adverse events.Studies conducted in acute
medical and surgical units or in neonatal or paediatric inten-
sive care were excluded because these settings are so different
that they warrant separate reviews. Single-unit studies were
included. Only studies published in English in refereed jour-
nals between 1990 and 2006 were included. The only quality
criterion that was used to selected studies was that they had to
have employed some method of risk adjustment. In summary,
studies were included if they met the following criteria:
Conducted exclusively in one or more adult ICUs
Dependent variable was mortality or adverse events
Human capitalcharacteristics of the nursing workforce
formed at least one of the independent variables
Published in English between 1990 and 2006
Used some form of risk adjustment
The first author read all the abstracts and retrieved all t
articles that in her judgement met these criteria.
3.3.Previous reviews
Several reviews have already been conducted in this ar
Carmel and Rowan (2001) found in 63 publications about 5
different studies of the organisation of intensive care, whic
they grouped into eightcategories:staffing,teamwork,
volume and pressure of work, protocols, admission to inte
sive care, technology, structure and error. Articles on staffi
were the most common and fell into a number of different
strands of research on management and personnel, intens
vist-led units,medicaland nursing intensity and nursing
autonomy.Five studies focussed on the impactof nursing
intensity (mainly nurse–patientratios,butone study also
examined levelof nurse qualifications)on mortality,but
none were able to rejectthe hypothesis of no association.
This was a usefulsource ofinformation aboutstudies
published prior to 2000.
Numata etal. (2006) reviewed the literature on nurse–
patientratiosand mortality in ICUsand identified nine
observationalstudies,five of which were included in a
meta-analysis.The unadjusted risk ratios of nurse staffing
(high versus low) on hospitalmortality were combined to
give a pooled estimate of 0.65 (95% CI 0.47–0.91).How-
ever,after adjusting for covariates the association between
staffing and mortality became non-significant in all but on
study (Tarnow-Mordiet al., 2000).Numata etal. (2006)
acknowledged that their meta-analysis was based on a sm
number of studies, four of which were conducted in the sa
region of the United States. They concluded that there wa
insufficient evidence to support an independent associatio
between nurses staffing levels and the mortality of critical
ill patients.They identified arangeof methodological
challenges: lack of an agreed operational definition of nurs
staffing, lack of variation in staffing levels in some studies
staffing levels which varied from shift-to-shift measured at
one pointin time,confounding factorsnot included or
controlled in any way,failure to use statisticalmethods
appropriate for the research design,crude methods of risk
adjustment due to the use of administrative databases and
tendency to analyse hospital mortality without considering
the care that the patient received outside the unit.
Both of the above reviews focussed on mortality as the
key variable to be explained, but there is a growing literat
on the impact of nursing on a much broader range of patie
outcomes,including safety,adverse events and complica-
tions, as well as patients’ and relatives’ satisfaction with c
and their subjective experiences of it. Williams et al. (2003
conducted a rapid review of different aspects of workload
E. West et al. / International Journal of Nursing Studies 46 (2009) 993–1011 995

ICUs, including the policy context and secular trends. They
identified 12 studies of nurse staffing,and presented evi-
dence for a positive effectof increased nurse staffing on
patient outcomes. More recently, Carayon and Gurses (2005)
conducted a wide ranging review ofthe literature.Their
main goalwas to devise a ‘human factorsengineering’
framework to guide modification ofthe work setting to
improve patient safety and to improve the quality of nurses’
work lives in ICUs. They cite several studies that have shown
thatmedicalerrorsare common in ICUsand may be
attributable to problems in communication between nurses
and physicians,to impaired access to information,and to
high intensity nursing workloads. They identified 22 studies
in all,but made no attempt to evaluate them or aggregate
their findings. They concluded that there were at least four
differentkinds of workload associated with the unit,job,
patient and situation. Most studies to date have focussed on
workload associated with the first two: the unit or the patient.
Each of the above reviewsmakesa contribution to
knowledge aboutworkforce,workload and outcomesin
intensive care.The emerging consensus seems to be that
while there is some evidence that the nursing workforce has
an impacton patients’risksof adverse events,there is
insufficient evidence of a link with mortality. This presents
a puzzle because some of the adverse events studied have
known links to mortality.Why would nursing workforce
characteristics affect the more proximate events but not the
final outcome? The process of dying is likely to be one of an
accumulation of adverse events as the patientdeteriorates
and clinicians take increasing risks in an attempt to save their
life. It seems important then to begin to integrate these two
separate strands of work into one coherent analysis.
This study uses systematic review methods to identify
and appraise empiricalevidence aboutthe impactof any
human capital characteristic of the nursing workforce (such
as skill mix, education or employments status, in addition to
nurse–patient ratios) on any patient outcomes,particularly
adverse events such as post-operative infections and mor-
tality.It adds to the literature a deeper description of the
methodsused and attemptsto evaluate the quality and
scientific rigour of the studies reviewed.
4. Results of the literature search
Fifteen studies were identified. Table 1 in the appendix to
this paper shows the key features of each study,including
location,research design,sample,sources of data,nursing
variables,outcome variables,factors adjusted for,findings
and comments.
4.1.Study quality
Research designs employed in this field divide into the
quasi-experimentalsuch as case controlor cohortstudies
with an emphasison directcomparison ofresultsfor
matched groups,and more ‘natural’ observationalstudies
which rely on multivariate analysis to adjust for the effect
known confounders. Quasi-experimental studies (case co
trol and cohort designs) try to come as close to an experi
as possible with random assignmentto ‘treatment’and
‘control’groupswith differentlevels ofexposure to an
intervention,e.g.patients exposed to high levels ofstaff
compared with patients exposed to low staffing levels. No
experimental research designs are more familiar in the so
sciences,where ithasbeen argued thatif a significant
relationship persists after partialing out the effects of con
founding variables (robust dependence), then a causal re
tionshipcan be said to exist (Goldthorpe,2000).In
observationalstudiesof this type there are no ‘control’
group or ‘treatment’ groups and the effects of confoundin
variables are controlled statistically. Although quasi-expe
mental studies do also use multi-variate statistics they te
do so to compensate for failures in matching case and co
groups rather than their main strategy.For the purposes of
this paperthereforewe makea very cleardistinction
between these two types of studies because the differenc
are relevantto the ways in which the studies should be
evaluated.
A great number of tools exist for evaluating randomise
trials (Moher etal.,1995).Downs and Black (1998) also
devised a quality checklistfor non-randomised as wellas
randomised studies buttheir work only extends as far as
quasi-experimental designs. To date, we are not aware of
systematic attempt to develop measures of quality for stu
of this type and so we designed a rudimentary measure
focussing on the following areasas a preliminary step
towards more formal attempts to establish criteria for obs
vational studies.
1. Geographical scope (4): The numbers of patient and un
in the study. The concern here is with how far the resu
can be generalised to other populations, and the geogr
phical coverage of the study is salient. High scores wou
be given to a census or random sample of units across
nation ora large geographicalarea.Medium scores
would be given to a study of a smaller number of units
and a low score would be given to a single-unit study.
2. Quality ofdata (4):How and why were the data were
collected.High scores would be assigned to data that
were collected prospectively to answer the specific que
tions in the study. Low scores would be given to studie
that use data collected for other purposes,e.g. adminis-
trative data.
3. Validity of key independent variables (4): How sensitive
and specific are the key workforce measures as indicat
of the number and quality of nurses available to indivi-
dualpatients? Nurse–patientratios are often averaged
overlong periods oftime and may be simply dichot-
omised into ‘high’ or ‘low.’ High scores would be given
the information on nurses was collected over time and
linked to individualpatients.Medium scores would be
E. West et al. / International Journal of Nursing Studies 46 (2009) 993–1011996

Table 1
Studies of nursing resources and patient outcomes: Main characteristics and results
Reference N of units Nursing variables Risk adjustment,
other controls
Relationship to
mortality,size of
effect,significance
Relationshi
adverse ev
size of effe
significance
Location N of patients
Period studied Data analysis
Design
Giraud et al.
(1993)
France
1989
Prospective
observational
study
2 ICUs
382 patients (400
consecutive
admissions)
Monitored daily
by physicians
for iatrogenic
complications
(defined as
adverse events
that was
independent of the
patients disease)
Non-parametric
Kruskal–Wallis test
Multi-variate
logistic regression
Cox survival analysis
Nursing workload
subjectively
assessed by each
nurse on each
shift,with score
for each patient
in their care.Patient’s
total nursing workload
was sum of all shift
scores during their stay.
OMEGA system
measures nursing
workload based
on 47 diagnostic
and therapeutic items.
Age
Organ System
Failure Score
SAPS
Prognosis
For patients who
stayed >24 h,
mortality was
2 higher for
patients who
developed
complications after
adjusting for Organ
System Failure Score
and prognosis
RR=1.92 (1.28–2.56)
Total num
iatrogen
complica
316 whic
in 31% o
in sampl
complica
defined
with 3 le
death.
Increased
of major
when nu
workload
using Om
and sub
assessm
or exces
The tota
workload
patients
go on to
develop
was 16.1
of 52.85
associat
patients
moderat
and pati
develop
complica
score on
workload
11515 w
significa

Table 1 (Continued )
Reference N of units Nursing variables Risk adjustment,
other controls
Relationship to
mortality,size of
effect,significance
Relationshi
adverse ev
size of effe
significance
Location N of patients
Period studied Data analysis
Design
p<.0001
Kruskal–
test.
Shortell et al.
(1994)
USA
1988–1990
Prospective,
multi-centre
observational
study
42 ICUs
26 units in stratified
random sample;
14 volunteer units
‘largely representative
of national population.’
17,440 patients
used to calculate
unit SMRs using
logistic regression
Organisational
assessment
questionnaire
Patient data
on mainly
consecutive admissions
average study
period=10 m/unit
OLS used to test
hypotheses with unit
SMR as dependent
variable
Nurse/patient ratio
(data collected
on each shift during
the study period)
Apache III
Primary disease
category
Duration of
hospitalisation
Location prior to
admission
Elective or
emergency surgery
No association
between nurse patient
ratio and unit SMR.
(OLS estimated
coefficient .12 with
SE=.137,standardised
coefficient Beta=.14
in standard deviations
of both the dependent
and predictor variables.
Not tested

Thorens et al.
(1995)
Switzerland
Descriptive and
case control
study
1 ICU
87 COPD patients:
15 in study year+
72 from earlier years
1 year prospective
data collection
compared to 5 years
of data already
collected.
Each patient followed
individually by lead
author
Head nurse kept data
on nurses
Correlations
1-way ANOVA
Index comparing
number of nurses
and their qualifications
with ideal for
dependency-adjusted
number of
patients
Patient dependency
in calculation of
the nursing index
Risk of developing
ventilator-assisted
pneumonia increases
linearly by 1%/day
of mechanical
ventilation with a
mortality rate of
>50% implies a
link to nurse staffing,
but not directly
tested in the study.
Duration o
off mech
ventilati
with COP
Negative
with nur
coefficie
given,Sp
rank cor
p=.025
In the firs
duration
ventilati
from 7.3
to 38.22
In the 6th
number
increase
duration
to 9.913
(p<.001y
vs year 6
Bastos et al.
(1996)
Brazil
1990–1991
Multi-centre,
prospective,
observational
study
10 ICUs
1734 consecutive
adult admissions
Questionnaire about
hospital
Unit data given to
ICU Director.
Multi-variate
regression
Nurse staffing ratios
across all shifts.
Apache III
May have controlled
for hospital factors
but not reported.
No association
with
SMR (b=0.32,
p=0.12)
Not tested

Table 1 (Continued )
Reference N of units Nursing variables Risk adjustment,
other controls
Relationship to
mortality,size of
effect,significance
Relationshi
adverse ev
size of effe
significance
Location N of patients
Period studied Data analysis
Design
Fridkin et al
(1996)
VA, Tucson
USA
1992–3
Cohort study
1 centre
All SICU patients
in period
Correlations
Logistic
regression.
Average monthly
nurse–patient ratio
(registered nurses
only)
Age
Mortality
Length of SICU
hospital stay.
N of patients
with >14 days
assisted ventilation
or on total parenteral
nutrition
Period of
hospitalisation
(outbreak vs
pre-outbreak)
Not tested Patient to
nurse ra
increase
in the ou
period c
with the
period fr
to 1.40 (
Correlatio
CVC-BSI
nurse to
ratio; Sp
Rank Co
Coefficie
Logistic re
showed
occurren
one CVC
strongly
with a h
to patien
Reis-Miranda
et al.(1998)
12 European
countries
4 months
Prospective
observational
89 units
(non-random
selection)
Daily data on
patients and nursing
workload (TISS)
Questionnaires
and site visits.
Fixed effects—DV
measure of
(perceived) ICU
performance
(measure developed
by Shortell).
Multiple regression
Nurses/bed
N of nurses included
in a composite
independent variable
‘wealth of the unit’
NB: these are
not the same—the
first refers only to
nurses and the
second includes
a number of
other variables
in one composite
score.
SAPSII
Country
ICU
Technology
Centralisation
Organisational
factors
Nurses’ morale/stress
Process vs results
culture
Budgetary factors
Mortality decreased by
Organisational
commitment
Results culture
Not significant
in final models.
Not tested

Random effects—dv
patient outcome.
Logistic regression
EOF
Country
Audit Commission
(1999)
England & Wales
Prospective
survey conducted
in 1998 and linked
to a benchmarking
database (ICNARC)
Observational study
227 (100%) Trusts
in England and Wales
and 243 (85%)
adult ICUs returned
surveys
ICNARC supplied data
on 15,805 patients
in 79 ICUs.52 units
agreed to allow the
use of their survival
data.
Report does not state
exactly which statistical
tests were used.
Nurse staffing ratios
and skill mix
Apache II
Not clear whether
any other controls
used.
No association Not tested
Vicca (1999)
England
19 months
Retrospective
(descriptive)
cohort study
1 ICU
50 patients with
MRSA
Information on nurse
staffing for each 8 h
shift
Correlations
N of trained ITU
nurses per shift/
number of patients,
for each 8 h shift.
Trained+extra
staff/number of
patients
Staffing level:
total number nurses—
total dependency score
Total number of
nurses/shift dependency
score.
Patient
dependency
included in
calculations of
workload.
Not tested Weak but s
(t-test) ve
correlation
MRSA and
staff/patien
r=0.150 (0
to 0.229);
p<0.001.Pe
staff/patien
r=0.145 (0
to 0.224);
p<0.001.M
nurse/patie

Table 1 (Continued )
Reference N of units Nursing variables Risk adjustment,
other controls
Relationship to
mortality,size of
effect,significance
Relationshi
adverse ev
size of effe
significance
Location N of patients
Period studied Data analysis
Design
Sum of daily staffing
levels on each
of three shifts
Peak, trough and
mean scores in a day
r=0.145 (0
0.225); p<
staff/patien
r=0.137 (0
to 0.216); p
Number of
cases and
surplus or d
peak staffin
r=0.147 (0
to 0.226); p
Trough sta
r=0.171 (0
-0.249); p<
Daily total
r=0.166 (0
0.245); p<
Pronovost et al.
(1999)
Maryland,USA
1994–1996
Retrospective
and prospective
observational
study
39 ICUs (85%) units
invited to participate
returned their
questionnaires)
2606 patients with
abdominal aortic
surgery
Patient data linked
to prospective
survey of ICU directors
Multi-level multi-
variate regression
Nurses/patient during
day & evening Fewer
(less than or equal
to 1:2) vs More (>1:2)
Socio-demographic
variables
Romano–Charlson
illness severity
Hospital volume
Surgeon volume
No association
between nurse/patient
ratio and in-hospital
mortality in
multivariate
regression
No associ
between
ratios an
complica
a list sel
experts
given in
Associatio
hospital
nurses in
associat
mean in
(7–33%)
Associatio
ICU LOS
nurses in
associat
increase
(17–91%

Robert et al.(2000)
France? or US?
Large inner city
public hospital
1994–1995
Nested case-control
study
1 20-bed SICU
28 ‘cases’ with BSI
compared to 99
randomly selected
controls who stayed
in SICU at least
3 days
Multi-variate logistic
regression
Regular staff,pool
nurses and nurse-
patient ratio
Periods 1 (8 months)
vs 2 (5 months)
Regular nurses
hours per
patient=10.6 vs
9.1 and pool nurses
hours per patient=2.2
vs 4.4
Total nurse hours
per patient=12.8
vs 13.5
Apache score
BSI and control
patients were
similar on many
dimensions but
did differ on a
number of
characteristics that
might be associated
with BSI (e.g.
longer stays in
SICU)
BSI patients were
significantly more
likely to die
BSI cases
cantly lo
nurse to
and high
to patien
the 3 da
Increased
in period
pool nur
multivar
OR 3.8 (
Overall N–
not sign
Tarnow-Mordi
et al.(2000)
Scotland
1992–1995
Prospectively
collected data,
observational
cohort study
1 unit
1050 patients
(all admitted
that met Apache
II criteria)
Data collected
each shift
Multiple logistic
regression
Occupancy per shift
Total ICU nursing
requirement as
defined by UK
ICS for each shift
Ratio of occupied
to appropriately
staffed beds per shift
Patients grouped on
composite measure of
ICU workload based on
Average nursing
requirement/occupied
bed/shift
Peak occupancy in
any shift during stay
Apache II
Patients divided
into 4 categories
of admissions.
Medical establishment
constant over the
period.
Patients exposed
to high ICU
workload were more
likely to die than
those exposed to
lower workloads.
Measures of workload
most associated with
mortality were peak
occupancy,average
nursing
requirement/occupied
bed/shift and the ratio
of occupied to appro-
priately staffed beds.
Adjusted odds ratio for
mortality with moderate
workload asthe refer-
ence
category:
Low workload OR
2.0 (1.2–3.3)
Intermediate OR 1.9
(1.2–3.1)
High OR 3.1 (1.9–
5.0)
Not tested

Table 1 (Continued )
Reference N of units Nursing variables Risk adjustment,
other controls
Relationship to
mortality,size of
effect,significance
Relationshi
adverse ev
size of effe
significance
Location N of patients
Period studied Data analysis
Design
Pronovost
et al.(2001)
Maryland,USA
1994–1996
Retrospective and
prospective
observational
study
38 ICUs
See above for
sample size and
data characteristics
Bi-variate
analysis to assess
relationship with
each complication,
significant ones put
into multi-variate
multi-level models
Nurses/patient during
day
Fewer (1:3 or 4) vs
More (1:1 or 2)
Socio-demographic
variables
Romano–Charlson
illness severity
Hospital factors
Surgeon factors
No association 14 compli
identifie
physicia
Fewer nu
associat
Reint
1.6 (1.1–
Comp
RR 1.7 (
Any m
complica
2.1 (1.5–
Pulm
insufficie
after pro
RR 4.5 (
Amaravadi
et al.(2000)
Maryland,USA
1994–1998
Multi-centre,
cross-sectional
retrospective
observational
with elements
of a cohort
design
35 ICUs
366 patients with
oesophageal resection
Hospital discharge
data
Staffing survey data
for 1996,multi-level
logistic regression
Average night-time
nurse/patient ratios
Low <1:2 vs
High >1:2
Age, sex,race
Type of operation
Type of admission.
Hospital
Surgeon volume
No association NNPR<1:2
with:
Reintubatio
2.6 (1.4–4.
Pneumonia
(1.2–4.7); p
Septicaemi
(1.1–12.5);

Dimick et al.
(2001)
Maryland,USA
1996–8
Multi-centre
retrospective
cohort study
51 ICUs
556 patients with
hepatic surgery
Hospital discharge
data
Survey of
organisational factors
Bi-variate analysis
for variable selection
Logistic regression
for mortality
Linear regression
for adverse events
Average night-time
nurse/patient ratios
Fewer (51:3) vs
More (41:2)
Demographic
characteristics
Romano–Charlson
co-morbidity index
Ruptured vs
non-ruptured aorta
Type of admission
Type of operation
Hospital volume
Surgeon volume
No association Fewer vs
nurses r
OR 2.9 (
p=0.04
No signifi
relations
aspiratio
insufficie
pneumo
septicae
post-ope
infection
cardiac c
cardiac a
myocard
and acu
Dang et al.
(2002)
Maryland,USA
1994–6
Multi-centre
retrospective
observational
study
38 units
2606 with
abdominal aortic
surgery
Hospital discharge
data on patients
Survey of
organisational factors
Multiple logistic
regression for
14 complications
Nurse/patient
ratio:
Low=1:3 day and
night
Medium=1:3 on
day or night
High 41:2 day and
night
Socio-demographic
variables
Romano–Charlson
Ruptured vs non-
ruptured aorta
Type of admission
N of cases of
AA surgery
at hospital
each year
N of ICU beds
Not tested Respirato
complica
(low vs h
2.33 (1.5
Cardiac c
(medium
OR 1.78
Other com
(medium
OR 1.74

given if the nursing variables were measured at one po
in time.Low scores would be given if there was little
variation in the variable because itcannotproduce a
satisfactory test of the hypothesis.
4. Controls (3):Are there adjustments for‘supply-side’
confounderssuch as variationsin medicalstaffing,
equipment,use of protocols,etc.? High scores would
be given when the authors include a range of variables
that mightaffectthe dependentvariableover and
abovethe nursing variables.Low scoresare given
when there is little discussion of possible rival hypoth-
eses and no attempt to include control variables in the
models.
5. Risk Adjustment(1):Are variations in the case mix of
patients adjusted for using an appropriate setof prog-
nostic variables or a validated prospective scoring syst
such as APACHE or SAPS? Studies that did not use any
form of risk adjustmentwere excluded from the study.
High scores would be given to studies that use standar
well-known formsof risk adjustmentand low scores
would be given where the procedure by which risk
adjustmentis achieved isby a processthatis non-
standard.
6. Statisticalanalysis (3):Were the methods used appro-
priate to the data? For example, did the analysis explo
the multi-level nature of the data if present? In time-se
data, was autocorrelation taken into account? Multi-lev
modelling would obtain a high score, time-series analy
a medium score and comparison of means would achie
a low score.
7. Reporting (2): Was the description of the study clear an
complete? Was enough information given in the paper
replicate the study? High scores are given for clearly
written descriptions of the work and low scores follow
the conclusion that it would be difficult to replicate this
study.
8. Interpretation (3): Was the interpretation of results obj
tive and balanced? Where the conclusions supported b
the data? Is there any discussion of the possibility that
significant results could have occurred by chance? Hig
scores would be given to a study where the conclusion
are clearly supported by the analysisand there isa
balanced discussion of the strengths and weaknesses o
the study. Medium scores are given when the interpret
tion is not closely linked to the analysis and which fails
mention some of the weaknesses of the study. Low sco
are given to studies where the interpretation seems bia
or subjective.
This scoring system was applied to eight studies1 and the
results are shown in Table 2.
E. West et al. / International Journal of Nursing Studies 46 (2009) 993–10111006
Table 1 (Continued )
Reference N of units Nursing variables Risk adjustment,
other controls
Relationship to
mortality,size of
effect,significance
Relationship to
adverse events,
size of effect;
significance
Comments
Location N of patients
Period studied Data analysis
Design
and 3 complication
groups
Multi-level
modelling
Use of critical
paths for this
population.
and complications is
through their
responsibilities for
monitoring patients and
care co-ordination
and in the case of
respiratory symptoms
for pulmonary hygiene.
1 Although the Audit Commission (1999) study met all the criter
for inclusion in thistable the published reportdid notcontain
enough information about how the study was conducted to pursu
the issue of quality.

4.2.Studies of the impact of nursing on adverse events
Ten studies focussed on whether nursing resources affect
the risk of an adverse event. Five were large-scale observa-
tional studies, and five were smaller case-control and cohort
studies.
Giraud et al. (1993) followed 382 patients in two ICUs in
France. Nursing workload was subjectively assessed by each
nurse on each shift using carefully defined and operationa-
lised variables.Complications occurred in 31% of admis-
sions, with 13% being describedas major (severe
hypotension, respiratory distress, pneumothorax and cardiac
arrest). There was an increased risk of major complications
when nursing workload was described as high or excessive.
The authors called for preventative measures to be targeted
at the elderly and the mostseverely illpatients as most
vulnerable to adverse events.
Robertet al. (2000)focussed on the influence ofthe
composition ofthe nursing staffon primary bloodstream
infection (BSI) rates in a surgical ICU. This was described as
a nested case-controlstudy covering two periods thatdif-
fered both in staff/patient ratio and in the proportion of pool
nurses (i.e.nurses who were members of the hospital pool
service or agency nurses,as opposed to nurses who were
permanently assigned to the SICU) on duty.Patients who
contracted BSI were more likely to have been in hospital
during the 5-month period when the unit was more depen-
dent on pool nurses than during an 8 month reference period.
Also ‘case’ patients had significantly lower regular nurse-to-
patient ratios and higher pool nurse-to-patient ratios for the 3
daysbeforecontracting BSI.In multi-variateanalyses,
admission to hospitalwhen the ratio ofpoolnurses was
high, total parenteral nutrition and CVC days were shown to
be significant independent risk factors for BSI. This study is
importantbecauseit focuseson differencesin nursing
resources above and beyond simple nurse–patient ratios that
might affect patients,but the study design is vulnerable to
concurrentconfounding;other possible factors thatmight
have changed between the two periods were not controlled
for.
Thorens et al. (1995) investigated whether nursing vari-
ables were related to the duration of weaning from mechan-
ical ventilation in 87 patientswith chronicobstructive
pulmonary disease (COPD) on one ICU unit over 6 years.
A composite ‘index of nursing’ was constructed to compar
the effective workforce of the nurses (number and qualific
tions) with the ideal workforce required by the number and
condition of the patients in the unit at the time. They foun
thatbelow a threshold in the available workforce of ICU
nurses, the weaning time for patients with COPD increased
dramatically. As with the study by Robert et al. (2000), the
main threat to validity in this study comes from comparing
event rates for different periods in time; it seems quite like
that there were confounders not accounted for in the mod
For example,at the same time as the numberof nurses
increased, the number of doctors did too, and there is no w
of knowing whether the change in the duration of weaning
due to the increase in nurses, increase in doctors, or to so
other differences between the two periods not included in
model.
Another single centre study was conducted by Fridkin
et al (1996) who gathered data on all patients who develo
a central venous catheter (CVC) associated BSI during an
outbreak in 1992–1993. Controlling for a number of factor
that pre-dispose patients to contracting CVC BSI,such as
total parenteral nutrition, assisted ventilation and duration
hospitalisation,they found that the nurse/patient ratio also
had a significantimpacton the probability ofinfection.
Vicca (1999) found that the acquisition of methicillin resis-
tant staphylococcus aureus (MRSA) in the ICU of a tertiary
referral centre correlated with peaks of nursing staff work-
load and times of reduced nurse/patientratios within the
unit.This study used correlation analysiswhich is less
powerfulthan the statistical analyses in some of the other
studies.
A team atJohns Hopkins has published 5 large scale
studiesin this area.Pronovostet al. (1999),in a study
focussed on ICU physicians, showed a relationship betwee
nurse staffing and length ofstay.Using a questionnaire
developed by Shortell et al. (1994), they gathered informa
tion from medicaldirectorsaboutthe organisation and
staffing of 46 ICUs in Maryland and linked this to outcome
data on patients undergoing abdominal aortic surgery. The
found thathaving fewernurseson duty during the day
increased hospital length of stay and number of days spen
in the ICU.
E. West et al. / International Journal of Nursing Studies 46 (2009) 993–1011 1007
Table 2
Quality of non-experimental observational studies
Study Scope (4) Data (4) IV (4) Control (3) RA (1) Analysis (3) Clarity (2) Interp.(3). Total (24)
Shortell et al.(1994) 4 4 0 2 1 2 2 2 17
Bastos et al.(1996) 3 4 0 1 1 2 1 2 14
Reis-Miranda (1998) 4 4 1 3 1 3 1 2 19
Pronovost et al.(1999) 3 2 2 1 0 3 2 2 15
Tarnow-Mordi et al.(2000) 1 4 4 2 1 2 2 3 19
Dimick et al.(2001) 3 2 2 2 0 3 2 2 16
Pronovost et al.(2001) 3 2 2 1 0 3 2 3 16
Dang et al.(2002) 3 2 2 2 0 3 2 3 17

A later study using the same data investigated whether
the link between nursing resources and length of stay was
due to patients developing medical and surgical complica-
tions after abdominal aortic surgery (Pronovost et al., 2001).
Control variables in the models included hospital character-
istics,surgical volume and daily rounds by an ICU physi-
cian. The Romano–Charleson co-morbidity index was used
to adjust for case mix, and nurse-to-patient ratios on the day
shifts were dichotomised into ‘more ICU nurses’ (1:1 and
1:2) and ‘fewerICU nurses’(1:3 and 1:4).Patientsin
hospitals with fewer ICU nurses were more likely to have
post-operative complications, particularly pulmonary insuf-
ficiency and reintubation.This was found in models that
included daily rounds by an ICU physician, suggesting that
medicine and nursing inputs had independenteffects on
patient outcomes.
Pronovost et al. (1999, 2001) investigated the impact of
nurse to patient ratios on the day shift but two later studies
explored the theory thatnursing takeson an increased
importance atnightwhen fewerphysicians and ancillary
staff are present.Amaravadiet al. (2000)investigated
whether the night time nurse to patient ratio, dichotomised
into one nurse caring for one or two patients (>1:2) versus
one nurse caring for three or more patients (<1:2) in the
ICU, affected patients undergoing oesophagealresection.
Multi-variate analysis was used to adjust for case mix, and
for hospital and surgeon volume. A ratio of nurses to patients
greater than 1:2 was associated with an increased the prob-
ability of complications such as pneumonia,reintubation,
and septicaemia,increased length ofstay and increased
costs.
Dimick etal. (2001)asked whethernurse to patient
ratios at night had an effect on patients’ experiences after
anotherhigh-risksurgicalprocedure,hepaticsurgery.
They found a significant increase in post-operative pulmon-
ary complicationsand use of resourcesfor patients
receiving post-operative care in ICUs in which one nurse
provided care forthree ormore ICU patients atnight.
However,only reintubation was significantin the multi-
variate analysis,
Building on these studies that focussed on nursing work-
force characteristics on either day or night shifts, Dang et al.
(2002)investigated the impactof nurses on allshifts on
cardiac, respiratory and other complication in patients who
had abdominalaorticsurgery in 38 unitsin Maryland
between 1994 and 1996.They argued thatnurse staffing
has an impact on outcomes because nurses are responsible
for monitoring patients,co-ordinating care and more spe-
cifically forpost-operative pulmonary hygiene.Multiple
logistic regression and multi-levelhierarchicalmodelling
showed that there was a statistically significant increase in
the likelihood of respiratory complications in patients cared
for in low nursing intensity versus high nursing intensity
conditions and thatthere was an increased likelihood of
cardiac and othercomplicationsin medium versushigh
intensity nurse staffing.
Taken together, the five studies from Johns Hopkins ar
very similar and it is difficult to give them as much analyt
weight as independent studies. Their strengths are that t
cover many sites, use high quality administrative and sur
data,articulate the mechanisms by which they expect nur
sing to be implicated in patient outcomes and consult exp
about the most appropriate complications to be used as t
dependent variables. The fact that they focus on one surg
procedure in one state of the US is a limitation, and the c
measures of nurse staffing and risk adjustment procedure
well as the cross-sectional retrospective research designs
which they are based are also weaknesses. They tend no
discussthe factthatsometimesonly a minority ofthe
hypotheses they testare supported in any one study and
thatsome of the statistically significantrelationships may
have occurred by chance.
4.3.Studies of the impact of nursing resources on
mortality
There were 10 studies of the effect of nursing resource
on mortality. These included studies of nurse staffing rati
(Bastos et al.,1996; Dimick et al.,2001; Pronovost et al.,
1999, 2001; Reis Miranda et al., 1998; Shortell et al., 199
and skill mix (Audit Commission, 1999). Only three studie
detected a significant relationship.
Giraud et al. (1993), as well as finding an increased ris
of major complications when nursing workload was high,
found that the patients who developed complications wer
twice as likely to die. This study is distinguished by the us
of survival analysis,and adjustment for age,organ system
failure,SAPS and diseaseprognosis.However,as the
authors note at the end of the paper, it is possible that w
death seems imminentstaffmay become more interven-
tionistand take more risky clinicaldecisions.This would
expose the patientto harm as wellas benefit,and so the
complications mightarise from an increase risk of dying,
rather than death being the result of the complications. T
concluded: ‘‘Major iatrogenic complications were frequen
associated with increased morbidity and mortality rates,
related to high orexcessive nursing workload and were
often secondary to human errors.’’
Robertet al. (2000)found significantrelationships
between staffing characteristics,BSI and mortality.This
study was described in detail above and in Table 1.
One UK study also found a relationship with mortality.
Tarnow-Mordiet al. (2000) investigated whether hospital
mortality was independently related to nursing requireme
and othermeasures ofworkload in one Scottish hospital
between January 1992 and December 1995. They control
for patient characteristics using APACHE II scores. Using a
formula to calculate the number of ‘appropriately staffed
beds,’ they found thatpatients who were treated in times
when the ICU workload was high were more likely to die
than those who were there during periods of low workloa
In factmortality was more than twice as high in patients
E. West et al. / International Journal of Nursing Studies 46 (2009) 993–10111008

treated when the unit workload was high. Three measures of
workload wereparticularly important:peak occupancy,
average nursing requirementper occupied bed pershift
and the ratio ofoccupied to appropriately staffed beds.
The main limitation of this study is thatit was based on
data from only one unit. However the study is unique in that
levels of nurse staffing relate to individual patients, whereas
most studies use a measure of nurse staffing for a unit at one
point in time. The authors draw attention to the fact that they
did not exclude each patient’s own scores in calculating the
measure of unit level dependency which may introduce bias.
Two studies of technology in ICU included nurse staffing
as control variables; Bastos et al. (1996) studied 10 units in
Brazil and Shortell et al. (1994) studied 42 ICUs across the
USA. Although neither found a relationship between nursing
characteristics and mortality,both point out that there was
very little variation in nurse staffing across the units that they
studied, so neither can be considered a satisfactory test of the
hypothesis.
Lack of variation in staffing was nota problem in the
study conducted by the Audit Commission (1999) in the UK,
butthey too failed to find an association between staffing
characteristics(nurse–patientratiosand skill mix) and
patient mortality. This study is difficult to evaluate because
very few details of the statistical analysis are included in the
published report. The investigation was designed primarily
to assess the extentto which units across the UK were
performing in termsof ‘value formoney’so the focus
was on comparing expenditure and resource use and only
secondarily on linking these to patient outcomes.
In spite of finding relationshipsbetweennursing
resources and other adverse outcomes neitherAmaravadi
et al. (2000) nor Dimick et al. (2001) were able to detect a
significant effect on mortality.
5. Discussion
A systematic search of the literature identified 15 studies
of the link between nursing resources and patient outcomes
in ICUs. This review builds on previous work in the area by
bringing togetherstudiesof both mortality and adverse
events into one systematic review. It also devotes attention,
not just to the findings,but to the methods by which they
were obtained.We devised a rudimentary system for eval-
uating aspects of observationalstudies.This showed that
studies vary in quality and suggest that this may be relevant
to the interpretation of their findings.
All the included studiesthatexamined relationships
between nursing resources and adverse events found a link
with at least one outcome. However, studies that investigated
the impactof workforce variables on a large numberof
adverse events sometimes reported positive associations for
only a few of the relationships tested, without discussing the
possibility thatsome ofthese mighthave occurred by
chance. Five of the studies of adverse events emerged from
the same research team (Amaravadi et al., 2000; Dang et
2002; Dimick et al., 2001; Pronovost et al., 1999, 2001) an
use differentparts ofthe same data,and so may notbe
regarded as independentsources of evidence.Five of the
other six (Fridkin et al.,1996; Giraud et al.,1993; Robert
et al. (2000);Thorenset al., 1995;Vicca,1999)were
conducted in single units.Taken together these considera-
tions suggest that while there may be more evidence of a
between ICU nursing resources and complications than the
is between ICU nursing and mortality, the evidence is not
convincing.
In only three of ten tests of the link between nursing
resources and mortality was the nullhypothesis rejected.
The three studies concerned were based in one or two uni
whereas those that found little or no evidence for an assoc
tion were large multi-centre studies. The small-scale studie
were based on detailed descriptive information aboutthe
links between nursing resources and patient outcomes, wi
careful articulation of the mechanisms involved, attention
the operationalisation ofthe variables,and prospectively
collected data. They tended to use quasi-experimental cas
controland cohortdesigns.Howeverthese smallstudies
have limited generalisability.Some have design flaws and
some rely on descriptive statistics, such as correlations rat
than more powerful inferential statistics such as regression
analysis.
It seems premature to conclude that there is no associa
tion between nursing resources and mortality. If the link is
weak one itmay notbe detectable in large studies using
crude indicators and poor adjustment for confounding, wh
smallstudies with good data may detectit unreliably,i.e.
with large confidence intervals.
Table 3 summarises the findings.
5.1.Future research
Mortality is the most important dependent variable in th
intensive care setting because about30% of all patients
admitted to ICUs die,and mortality varies across units in
ways thatare currently difficultto explain.However this
review suggeststhat the relationshipbetweennursing
resources and mortality may be quite weak,at least above
some threshold level,and large studiesusing carefully
developed and prospectively collected measurements may
be needed to estimate the strength of this relationship wit
precision.An alternative would be meta-analyses of small
studies,butthis requires standardisation ofthe measure-
ments and classifications used,which may not be possible
retrospectively.
Studies of adverse events are also important because th
may tellus more aboutthe processes by which patients
deteriorate towards death.Adverse events can be linked to
mortality through the concept of ‘failure-to-rescue,’ define
as death after an adverse occurrence that could have bee
amenable to medical intervention (Silber et al.,1992).
E. West et al. / International Journal of Nursing Studies 46 (2009) 993–1011 1009

The hypotheses in the studies reviewed above are derived
from the theory thatoverwork and staffshortageswill
interferewith task performance,including surveillance,
monitoring, early detection of adverse events and preventa-
tive measures (e.g., hand washing, pulmonary hygiene, early
ambulation).One study also linksnursing to patients’
experience of pain which may affectthe developmentof
complications and progress through the unit. Experience or
at least familiarity with the unit and its routines and practices
was cited in one study,as was the nurses’role in co-
ordinating care. Several authors argued that nurses are more
important at night when there are fewer doctors and ancillary
staff around. The conceptual frameworks for studies of this
type could be enhanced by greater attention to what nurses
actually do in intensive care and by interviewing experts
about those characteristics of the nursing staff that would be
most relevant to patient outcomes.
In most of the studies reviewed above there appears to be
an assumption thatmore nurseswill alwaysbe better.
However,thatmay notalways be the case.Theoretically
at least, one might imagine a situation where there might be
too many nurses, or perhaps more realistically where adding
more nurses brings little or no additional benefit to patients
in ICUs, at the expense ofpatients in otherparts ofthe
hospital.Future research needs not only to further develop
the mechanisms linking nursing and outcomes, but to specify
morecomplex and interesting functionalformsof that
relationship,including threshold effects.
The majority of large studies identified in the course of
this review were observational, which means that the impact
of confounding is controlled by including in the model all
variables that are thought to affect the dependent variable.
This is difficult to achieve using existing databases. Future
studies using this design need to devote much more atte
to specifying the fullmodeland justifying the control
variables included in the model.
The majority of studies in this review were cross-sec-
tional and there are well known problems in inferring cau
relationships from studies at one point in time. If studies
this tradition could also incorporate temporalinformation
into their analysis and use time series for multiple units,
survivalanalysis or eventhistory methods they would be
able to makestrongercausalclaims.It is particularly
importantthatin future studiesan attemptis made to
measure each patient’s exposure to nursing as a variable
that changes from shift to shift, rather than simply measu
the number of nurses in a unit at one point in time.
To date,studies have focussed on providing empirical
evidence of a link between nursing resources and patient
outcomes. An important further step will be to work out t
clinical and cost implications of the results. In their study
volume ofpatients,Iapichino etal. (2004)showed that
relative mortality decreased by 3.4% and 17.0% for every
five extra patients treated per bed per year in overall vol
and high-risk volume,respectively.They concluded that
while totalvolume was statistically significant,only high
risk volume wasclinically significant.Pronovostet al.
(2001) also tried to work outthe clinicalimplications of
their findings and more recently report results of financia
modelling of the Leapfrog Intensive Care Physician (ICP)
staffing standards (Pronovost et al.,2004).
6. Conclusions
ICUs consume a large amount of the health care budge
nurses are the biggest single expense. However, the evid
base for current staffing decisions is not well developed a
the subject of considerable debate (Adomat and Hicks, 20
This review confirms that the implications for outcomes a
patientsafety of possible changes in ICU staffing are still
uncertain. The relationship between nursing workforce ch
acteristics and patientoutcomes should continue to be the
focus of audit and research at unit and national level.
Acknowledgements
The authors would like to thank D.N.Barron forhis
critical reading of previous drafts and colleagues at LSHT
for their support of the larger project of which this study i
part. Funding for the project was provided by a post-doct
fellowship to the first author from the Health Foundation.
References
Adomat,R.R.,Hicks,C.C., 2003.Measuring nursing workload in
intensive care; an observational study using closed circuit vide
cameras.Journal of Advanced Nursing 42 (4),402–412.
E. West et al. / International Journal of Nursing Studies 46 (2009) 993–10111010
Table 3
Main findings of 15 studies
First author Adverse
events
Mortality
Amaravadi et al.(2000) U
Audit Commission (1999) X
Bastos et al.(1996) X
Dang et al.,2002 U
Dimick et al.(2001) U X
Fridkin et al (1996) U
Giraud et al.(1993) U U
Pronovost et al.(1999) U X
Pronovost et al.(2001) U X
Reis Miranda et al.(1998) X
Robert et al.(2000) U U
Shortell et al.(1994) X
Tarnow-Mordi et al.(2000) U
Thorens et al.(1995) U
Vicca (1999) U
Proportion positive 10/10 3/10
A tick means that the study found an association; a cross means that
they were unable to reject the null of no association and a blank cell
means that the hypothesis was not tested.

Amaravadi, R.K., Dimick, J.B., Pronovost, P.J., Lipsett, P.A., 2000.
ICU nurse-to-patient ratio is associated with complications and
resource use after esophagectomy. Intensive Care Medicine 26
(12),1857–1862.
Audit Commission, 1999. Critical to Success: The Place of Efficient
and Effective Critical Care Services Within the Acute Hospital.
Audit Commission Publications,Oxford.
Bastos,P.G.,Knaus,W.A., Zimmerman,J.E., Magalhaes Jr.,A.,
Sun, X., Wagner, D.P., 1996. The importance of technology for
achievingsuperioroutcomesfrom intensivecare. Brazil
APACHE III Study Group.Intensive Care Medicine 22 (7),
664–669.
Becker,G.S.,1964.Human Capital:A Theoreticaland Empirical
Analysis with SpecialReference to Education.University of
Chicago Press,Chicago.
Bennett, D., Bion, J., 1999. ABC of intensive care: organisation of
intensive care.British Medical Journal 318,1468–1470.
British Medical Association,1967.Intensive care.BMA Planning
Unit Report No.1, British Medical Association,London.
Carayon,P., Gurses,A.P., 2005.A human factorsengineering
conceptualframework of nursing workload and patientsafety
in intensive care.Intensive and CriticalCare Nursing 21 (5),
284–301 (epub.2005 Feb.24).
Carmel,S., Rowan,K., 2001.Variation in intensive care unit out-
comes:a search forthe evidence on organizationalfactors.
Current Opinion in Critical Care 7 (4),284–296.
Dang,D., Johantgen,M.E., Pronovost,P.J., Jenckes,M.W., Bass,
E.B., 2002. Post-operative complications: does intensive care unit
staff nursing make a difference?Heart & Lung (3), 219–228.
Department of Health (DoH), 2000a. Comprehensive Critical Care:
A Review of AdultCriticalCare Services.Stationary Office,
London.
Dimick,J.B., Swoboda,S.M.,Pronovost,P.J., Lipsett,P.A.,2001.
Effectof nurse-to-patientratio in the intensive care uniton
pulmonary complications and resource use after hepatectomy.
American Journal of Critical Care 10 (6),376–382.
Downs, S., Black, N., 1998. The feasibility of creating a checklist
for the assessmentof methodologicalquality both of rando-
mized and non-randomised studies ofhealth care interven-
tions.Journalof Epidemiology and Community Health 52,
377–384.
Fairman,J., Lynaugh,J., 1998.CriticalCare Nursing: A History.
University of Pennsylvania Press,Philadelphia.
Fridkin, S.K., Pear,S.M., Williamson, T.H., Galgiani, J.N., Jarvis,
W.R., 1996. The role of understaffing in central venous catheter-
associatedbloodstream infections.Infection Control and
HospitalEpidemiology 17 (3),150–158 (comment,pp.147–
149).
Giraud, T.,Dhainaut, J.F.,Vaxelaire, J.F.,Joseph, T.,Journois, D.,
Bleichner,G., Sollet,J.P., Chevret,S., Monsallier,J.F., 1993.
Iatrogenic complications in adultintensive care units:a pro-
spective two centre study. Critical Care Medicine 21 (1), 40–51.
Goldthorpe, J.H., 2000. Numbers, Narratives and the Integration of
Research and Theory.Oxford University Press,Oxford.
Iapichino, G., Gattinoni, L., Radrizzani, D., Simini, B., Bertolini, G.,
Ferla, L., Mastraletti, G., Porta, F., Miranda, D.R., 2004. Volume
of activity and occupancy rate in intensive care units. Associa-
tion with mortality.Intensive Care Medicine 30 (2),290–297
(epub.2003 Dec.19).
Moher, D., Jadad, A.R., Nichol, G., Penman, M., Tugwell, P., Walsh,
S., 1995. Assessing the quality of randomised controlled trials:
an annotated bibliography ofchecklists.Controlled Clinical
Trials 16,62–73.
Numata, Y., Schulzer, M., van der Wal, R., Globerman, J., Semeniuk
P.,Balka,E., Fitzgerald,J.M., 2006.Nurse staffing levels and
hospitalmortality in criticalcare settings:literaturereview
and meta-analysis. Journal of Advanced Nursing 55 (4), 435–448
Pronovost,P.J., Jenckes,M.W.,Dorman,T., Garrett,E., Breslow,
M.J., Rosenfeld, B.A.,Lipsett,P.A.,Bass, E.,1999. Organiza-
tional characteristics of intensive care units related to outcome
of abdominal aortic surgery.281(14),1310–1317.
Pronovost, P.J., Waters, H., Dorman, T., 2001. Impact of critical car
physician workforce for intensive care unit physician staffing.
Current Opinion in Critical Care 7 (6),456–459.
Pronovost, P.J., Angus, D.C., Dorman, T., Robinson, K.A., Dremsi-
zov,T.T., Young,T.L., 2002.Physician staffing patterns and
clinical outcomes in critically ill patients: a systematic review.
Journal of the American Medical Association 288 (17),2151–
2162.
Pronovost,P.J., Needham,D.M., Waters,H., Birkmeyer,C.M.,
Calinawan,J.R., Birkemyer,J.D., Dorman,T., 2004.Intensive
care unitphysician staffing:financialmodeling ofLeapfrog
Standard. Critical Care Medicine 32 (6),1406–1408.
Reis Miranda, D., Ryan, D., Schaufeli, W. (Eds.), 1998.Organisa-
tion and Management of Intensive Care: A Prospective Study in
12 European Countries. Springer,Berlin.
Robert, J., Fridkin, S.K., Blumberg, H.M., Anderson, B., White, N.,
Ray,S.M., Chan,J., Jarvis,W.R.,2000.The influence of the
composition of the nursing staff on primary bloodstream infec-
tion rates in a surgical intensive care unit. Infection Control and
Hospital Epidemiology 21 (1),12–17.
Royal College of Nursing, (2000, 2003). Guidance for nurse staffing
in critical care.Accessed on the RCN website.
Sandelowski, M., 2000. Devices and Desires: Gender, Technology
and American Nursing. The University of North Carolina Press,
Chapel Hill and London.
Shortell,S.M., Zimmerman,J.E., Rousseau,D.M., Gillies,R.R.,
Wagner, D.P., Draper, E.A., Knaus, W.A., Duffy, J., 1994. The
performance ofintensive care units:does good management
make a difference?Medical Care 32 (5),508–525.
Silber,J.H., Williams,S.V.,Krakauer,H., Schwartz,J.S., 1992.
Hospital and patient characteristics associated with death after
surgery.A study of adverse occurrence and failure to rescue.
Medical Care 30 (7),615–629.
Tarnow-Mordi,W.O.,Hau,C., Warden,A., Shearer,A.J., 2000.
Hospital mortality in relation to staff workload: a 4-year study in
an adult intensive-care unit. Lancet 356 (9225),185–189.
Thorens,J.B., Kaelin, R.M., Jolliet, P., Chevrolet,J.C., 1995.
Influence of the quality of nursing on the duration of weaning
from mechanical ventilation in patients with chronic obstructive
pulmonary disease.Critical Care Medicine 23 (11),1807–
1815.
Vicca,A.F., 1999.Nursing staffworkload asa determinantof
methicillin-resistantstaphylococcus aureus spread in an adult
intensive therapy unit. Journal of Hospital Infection 43 (2), 109–
113.
West, E., Rafferty, A.-M., Rowan, K., 2004. Preserving the humanity
of care in high-tech environments. Paper Given at the In Sick-
ness and in Health Conference Iceland,June 2004.
Williams,S.,Coombs,M., Lattimer,V., 2003.Workforce planning
for critical care: a rapid review. Accessed at hhttp://www.neevia
comi.
E. West et al. / International Journal of Nursing Studies 46 (2009) 993–1011 1011