Nurse staffing, medical staffing and mortality in Intensive Care: An observational study.

West, E; Barron, DN; Harrison, D; Rafferty, AM; Rowan, K;

Sander-son, C (2014) Nurse staffing, medical staffing and mortality in

Inten-sive Care: An observational study. International journal of nursing

studies. ISSN 0020-7489 DOI: https://doi.org/10.1016/j.ijnurstu.2014.02.007

Downloaded from:

http://researchonline.lshtm.ac.uk/1620740/

DOI:

10.1016/j.ijnurstu.2014.02.007

Usage Guidelines

Please refer to usage guidelines at

http://researchonline.lshtm.ac.uk/policies.html

or

alterna-tively contact

researchonline@lshtm.ac.uk

.

Nurse

staffing,

medical

staffing

and

mortality

in

Intensive

Care:

An

observational

study

Elizabeth

West

a,

*

,

David

N.

Barron

b

,

David

Harrison

c

,

Anne

Marie

Rafferty

d

,

Kathy

Rowan

c

,

Colin

Sanderson

e

a

FacultyofEducationandHealth,UniversityofGreenwich,UnitedKingdom b_{Saı¨dBusinessSchoolandJesusCollege,UniversityofOxford,UnitedKingdom} c_{IntensiveCareNationalAuditandResearchCentre,UnitedKingdom} d

KingsCollege,London,UnitedKingdom e

LondonSchoolofHygieneandTropicalMedicine,UnitedKingdom

ARTICLE INFO Articlehistory: Received10July2013

Receivedinrevisedform9February2014 Accepted11February2014

Keywords: Intensivecareunits Nursestaffing Medicalstaffing Mortality Multilevelmodelling Observationalstudies ABSTRACT

Objectives:Toinvestigatewhetherthesizeoftheworkforce(nurses,doctorsandsupport staff)hasanimpactonthesurvivalchancesofcriticallyillpatientsbothintheintensive careunit(ICU)andinthehospital.

Background:Investigationsofintensivecareoutcomessuggestthatsomeofthevariation inpatientsurvivalratesmightberelatedtostaffinglevelsandworkload,buttheevidence isstillequivocal.

Data:Informationaboutpatients,includingtheoutcomeofcare(whetherthepatient livedordied)camefromtheIntensiveCareNationalAudit&ResearchCentre(ICNARC) CaseMix Programme.An AuditCommission surveyof ICUsconducted in1998 gave informationaboutstaffinglevels.Themergeddatasethadinformationon65ICUsand 38,168patients.Thisiscurrentlythebestavailabledatasetfortestingtherelationship betweenstaffingandoutcomesinUKICUs.

Design:Across-sectional,retrospective,riskadjustedobservationalstudy. Methods:Multivariable,multilevellogisticregression.

OutcomeMeasures: ICUandin-hospitalmortality.

Results:Aftercontrollingforpatientcharacteristicsandworkloadwefoundthathigher numbersofnursesperbed(oddsratio:0.90,95%confidenceinterval:[0.83,0.97])and highernumbersofconsultants(0.85,[0.76,0.95])wereassociatedwithhighersurvival rates.Furtherexplorationrevealedthatthenumberofnurseshadthegreatestimpact onpatients athigh riskof death(0.98,[0.96,0.99]) whereas theeffect of medical staffingwas unchanged across the range of patient acuity (1.00,[0.97, 1.03]). No relationshipbetweenpatientoutcomesandthenumberofsupportstaff (administra-tive,clerical,technicalandscientificstaff)wasfound.Distinguishingbetweendirect careandsupernumerarynursesandrestrictingtheanalysistopatientswhohadbeen intheunitformorethan8hmadelittledifferencetotheresults.Separateanalysisof in-unitandin-hospitalsurvivalshowedthattheclinicalworkforceinintensivecare hadagreaterimpactonICUmortalitythanonhospitalmortalitywhichgivesthestudy additionalcredibility.

Conclusion:Thisstudysupportsclaimsthattheavailabilityofmedicalandnursingstaffis associatedwiththesurvivalofcriticallyillpatientsandsuggeststhatfuturestudiesshould

* Correspondingauthor.Tel.:+4407533457701. E-mailaddress:e.west@gre.ac.uk(E.West).

ContentslistsavailableatScienceDirect

International

Journal

of

Nursing

Studies

journalhomepage:www.elsevier.com/ijns

http://dx.doi.org/10.1016/j.ijnurstu.2014.02.007

Whatisalreadyknownaboutthetopic?

Thereisagrowingconsensus,supportedbyseveralhigh

quality systematicreviews, thatthenumber ofnurses

availableforpatientcareimprovespatientoutcomesin

acute medical and surgical wards, but there is less

agreementthatthisrelationshipholdsinIntensiveCare

Units.

Someevidence,mainlyfromtheUnitedStates,suggests

thattheorganisationofmedicalstaffinIntensiveCare

Unitsisrelatedtopatientoutcomes.

Anumberofothervariablesaffectpatientoutcomesin

ICUs, including, most importantly, the patient’s own

conditionandtheworkloadoftheunit.Thesevariables

needtobeincludedinthestatisticalanalysesascontrol

variables.

Whatthispaperadds

This studyshows a statistically significantassociation

betweenthenumberofnursesanddoctorsavailablein

IntensiveCareUnitsandpatients’chancesofsurviving

theirstayinICUandforupto30daysafteradmissionto

hospital.

ThesizeofthenursingworkforceinICUshasthegreatest

effect on the most severely ill patients, whereas the

number of doctors seemsto be important across the

rangeofpatientacuity,i.e.thereisnointeractioneffect

betweenthesizeofthemedicalworkforceandpatient

acuity.

The workload of the unit had an impact on patient

mortalityinadditiontothenumberofclinicalstaffon

theunitestablishment.

1. Introduction

Intensive care units (ICUs) were introduced in the

1950s based on the idea that the lives of severely ill

patientscouldbesavediftheyweretreatedinsmaller,well

staffed units with access to the most technologically

sophisticated equipment. Key features of this new

organisationalformincludedtriage(patientsshouldonly

beadmittedtoICUiftheirfutureisuncertain),surveillance

(closeandcontinuousobservationsbyhighlyskilledstaff)

and organ support, made possible by innovative new

technologies.Thismodelofcarediffusedrapidly

through-out thehealthcare systems of higher income countries.

However,ICUswere,andare,veryexpensivetorun;staff

salaries are themost expensive item of expenditure in

mosthealthcarebudgetsandICUsrequireamuchhigher

staff/patientratiothangeneralmedicalandsurgicalunits.

Theaimofthisstudyistoinvestigatewhetherthereisa

relationshipbetweenthenumberofstaff(nurses,doctors

and support workers) that are available in ICUs and

patients’chancesofsurvival.Totestthisrelationshipwe

use the best information that is currently available,

provided by two national datasets collected in England

around March1998.Thesedatasets allowus toinclude

importantcontrolvariablesinouranalyses,includingthe

patient’sownconditionandtheworkloadoftheunit.

1.1. Background

Bythelate1990s,theNationalHealthService(NHS)of

theUnitedKingdom(UK)wasspendingalargeproportion

ofitsbudgetonintensivecarebut whilethecostswere

rising,theperceivedneedforintensivecarewasnotbeing

met.Atragicincidentin1995whenaboydiedwhilebeing

transferredinsearchofanintensivecarebed,followedbya

flu epidemic in 1999 drew further attention to the

inadequacies in provision leading to sustained media

attention, questions in Parliament and vigorous debate

amongprofessionalgroups(Crocker,2007).

In1998,theAuditCommission,abodyestablishedby

theUKgovernmenttoconductvalue-for-moneystudies

acrossallpublicservices,publishedareportonICUstitled

‘‘Critical toSuccess: The place of efficientand effective

critical care services within the acute hospital’’ (Audit

Commission, 1999). This investigationshowed that the

outcomesofcarevariedwidelyacrossICUsinwaysthat

that werenoteasilyexplainedbystaffinglevelsorskill

mix.Inunitswithsimilarworkloads,thenumberofnurses

variedby50percentandconsultantcostsbyafactorof

three.Nursingcostsdifferedbyathirdbetween thetop

and bottom quartiles. Most importantly, mortality was

over50percentinsomeunits.Whileunitsvariedgreatly

instaffingcostsandinpatientoutcomes,thereappearedto

beverylittlerelationshipbetweenthetwo.Inotherwords,

higher spendingon staffdidnotalwaysresultin better

chancesofsurvivalforpatients.Theonlystaffingvariable

that theAudit Commissionteamfoundtoberelatedto

patient mortality was the pattern of consultant cover.

Lowerthanexpectedpatientsurvivalwasfoundinunits

whereeachconsultantworkedasetnumberofdaysper

weekcomparedtounitswhereconsultantsworkedashift

patternofoneweekon,twoweeksoff.Noneofthenursing

variables werefoundtoberelated topatient outcomes.

However,theanalysisconductedbytheAuditCommission

isnotdescribedindetailinthepublishedreportandthe

authorsmaynothavehadaccesstosomeoftheresources

and techniques that are available to analysts today,

includingbettermethodsofriskadjustment,andstatistical

methodsthatallowforthesimultaneousinclusionofdata

frommorethanonelevelofanalysis.Giventhatthecosts

ofcriticalcarecontinuetoconsumeexpensiveresources

andthattheevidenceforlinkingstaffinginputstopatient

outcomes in ICUs remains contentious, there are good

groundsforreanalysingthesedata.

focusontheresourcesofthehealthcareteam.Theresultsemphasisetheurgentneedfora prospectivestudyofstaffinglevelsandtheorganisationofcareinICUs.

1.2. Previousliterature

1.2.1. Nursestaffing

There is a large and growing literature on the

relationshipbetweennursestaffingandpatientoutcomes

inacutemedicalandsurgicalunits.Seminalpapersby,for

example,Aikenet al.(2002),Needleman etal.(2002),

Marketal.(2004)andTourangeauetal.(2007),conducted

mainlyinNorth America,havehada dramaticeffecton

empirical researchon workforceissues and have had a

demonstrableinfluenceonpolicyinmanycountries.There

is internationalconcern abouttherelationship between

nurse staffingandpatient outcomesand similarstudies

have beenconductedin several countries,including,for

example,Korea(Choetal.,2008);Belgium(VandenHeede

et al.,2009)andTaiwan(Lianget al.,2012).IntheUK,

where fewer quantitativestudies have beenconducted,

Rafferty et al. (2007) showed that patients in NHS

hospitals withthe mostfavourable staffing levelswere

morelikelytosurvive.

Astheevidencefromsinglestudiesoftherelationship

between nurse staffing and patient outcomes began to

accumulate, a number of systematic reviews have

attempted to consolidate their findings. For example,

based on ananalysis of 28 studies, Kane et al. (2007)

concludedthathigherstaffinglevelswereassociatedwith

lowerpatientmortalityinICUsandinsurgicalandmedical

patients. Theydescribed anemerging consensusthat in

acutemedicalandsurgicalsettings,thereisevidenceofa

‘‘...statistically and clinically significantassociation

be-tween RN staffing and adjusted odds ratio of

hospital-related mortality, failure to rescue, and other patient

outcomes’’(Kaneet al.,2007).

Anumberofsystematicreviewshavefocused

specifi-callyonnursestaffinginICUs.Numataetal.(2006)located

nine observational studies of the association between

nursestaffingand mortalityinICUs,fiveofwhichwere

includedina meta-analysis(Personetal.,2004;Dimick

etal.,2001;Pronovostetal.,2001;Amaravadietal.,2000; Tarnow-Mordi et al.,2000). Thefirstfourstudieslisted

wereconductedintheUSandthefifthinScotland.Numata

et al.(2006)concludedthatthereiscurrentlyinsufficient

evidencetosupporttheindependentassociationofnurse

staffinglevelsandthemortalityofcriticallyill patients.

However, they highlighted methodological problems in

manyofthepublishedstudies,perhapsmostimportantly,

in their frequent failure to control adequately for the

patient’sowncondition.

Asecondsystematicreviewofstudiesofnursestaffing

andpatientoutcomesinICUs(Westet al.,2009)located

15 studies, of which three reported a statistically

significant relationship between nursing resourcesand

mortality(Giraudetal.,1993;Robertetal.,2000;

Tarnow-Mordietal.,2000).Eachofthesestudieswasconductedin

a single unit. However, in seven studies there was

insufficientevidencetorejectthenullhypothesisofno

relationshipbetweenstaffinglevelsandmortality(Audit

Commission,1999;Bastosetal.,1996;Dimicketal.,2001; Pronovostetal.,1999, 2001;Reis-Mirandaet al.,1998; Shortellet al.,1994).Interestingly,thesewereall

multi-unitstudies.

Athird,morerecentreviewfocusedonnursestaffing

andpatientoutcomesinICUspublishedbetween1998and

2008 (Penoyer, 2010). Of the studies of mortality, six

reported an association with staffing (Cho et al., 2008;

Stoneetal.,2007;Tourangeauetal.,2007;Personetal., 2004;Tucker,2007;Tarnow-Mordiet al.,2000)whilefive

reportednoassociation(Kiekkasetal.,2008;Metnitzetal.,

2004; Dimick et al., 2001; Pronovost et al., 2001;

Amaravadiet al.,2000). Theconclusions ofthis review

are that the association between nurse staffing and

outcomesinICUsissimilartothatreportedinstudiesof

the same relationship in general medical and surgical

units.

Insum,threerelativelyrecentsystematicreviewsand

meta-analysesdo not provideconsistentevidenceas to

howtheliteratureonnursestaffingandpatientmortality

inICUshouldbeinterpreted.Eachreviewwasbasedona

relativelysmallnumberofpapersandfindingswillhave

beeninfluenced by theinclusion and exclusion criteria.

New empirical evidence is also contradictory.

Further-more, although the studies were conducted in many

differentcountriesacrossEuropeincluding,forexample,

Greeceand Austria,as wellasBrazil,in additiontothe

moreprolificliteraturefromNorth America,onlyoneof

thestudiesincludedinthesereviewswasconductedinthe

UK(Tarnow-Mordiet al.,2000).Thepaucityofstudieson

therelationshipbetweennursestaffingandmortalityin

adultICUsintheNHSisoneofthegapsintheevidence

basefornursingthatthispaperseekstoaddress.

1.2.2. Previousliteratureonmedicalstaffing

Studiesofthemedicalcontributiontopatientoutcomes

inintensivecarehavefocusedontheroleoftheintensivist

(aphysicianspecialisinginthecareofcriticallyillpatients

usuallyworkinginanICU)andontherelativeeffectiveness

of ‘‘open’’ and ‘‘closed’’ ICUs. In open units, which

predominateinthe USA,patients are caredfor bytheir

primaryphysician.Closedunits,wherean intensivistor

teamofintensivists providescare,aremorecommonin

Europe(BurchardiandMoerer,2001).Asystematicreview

oftheevidenceconcludedthattheclosedmodeltendsto

achieve betteroutcomes for patients (Pronovost et al.,

2002).Theseauthorsarguethat‘‘...physicianswhohave

the skills to treat critically ill patients and who are

immediately available to detectproblems and institute

therapieswillpreventorattenuatemorbidityand

mortal-ity.’’ More recently, Kahn et al. (2007) showed that

patientsreceivingprolongedmechanicalventilation,and

whose care was primarily the responsibility of an

intensivist,weremorelikelytoreceivearangeoftherapies

recommended by theInstitute for Healthcare

Improve-ment,suchasdeepveinthrombosis(DVT)andstressulcer

prophylaxis,suggesting thatthelinkbetween structural

variables,suchasmedicalstaffingandpatientoutcomes,

mightbemediatedthroughprocessvariablesrelatingto

thequalityofcarethatisdelivered.

Compellingevidenceofthebenefitsofintensivistcare

reviewedbyPronovostet al.(2002)hasledintheUStoa

call for staffing by intensivists over the entire 24h.

However, Kahn and Hall (2010), caution against this

that one reason for the lack of consistent results in

empiricalstudiesoftheeffectof24hintensiviststaffingis

ourcollectivelackofknowledgeaboutwhyintensivistsare

good. In other words, we have little knowledge of the

mechanisms linking physician staffing to patient

out-comes. There are a number of possibilities: care by

intensivists might increase the use of evidence based

practice,theymight facilitatemulti-disciplinarycare,or

theymightprovideurgenttreatmentatthebed-side.There

isacleargapinempiricalevidenceaboutwhatintensivists

doinICUsthatlinkstheirpresencetoimprovedclinical

outcomes.

Apartfromthesestudiesoftheorganisationofmedical

work, we have found only one study that investigated

whethertherewasanassociationbetweenthe

intensivist-to-bedratioandpatientmortality.DaraandAfessa(2005)

studied 2492 patients treated under different staffing

ratios(1:7.5,1:9.5.1:12.5and1:15)inoneunitintheUS

and controlled for demographic variables and patients’

APACHEIIIscores.However,theydidnotcontrolforthe

numberofnursesorbedoccupancy,bothofwhichmight

mediate the relationship between medical staff and

patient outcomes. They found no association between

medicalstaffingratiosandeitherICUorhospitalmortality

butatthehighestratio(1:15),lengthofstayintheICUwas

significantlyhigher.

Morerecently,aUKstudyof‘failuretorescue’(death

after a treatable complication) in surgical patients in

England(1997–2009)byGriffithset al.(2013)foundthat

lowerratesoffailuretorescuewereassociatedbothwith

greater numbers of clinically qualified staff per bed

(doctorsandnurses)andwithhighernumbersofdoctors

relativetonurses.Althoughfailuretorescuehas

previous-lybeenassociatedwiththenumberofnurses,thisstudy

showsthatthenumberofmedicalstaffisalsoimportant.

Insum,mostoftheworkonthemedicalcontributionto

ICUoutcomeshasfocusedontheorganisationofwork,and

hasbeenconductedmainlyintheUS.Littleattentionhas

yetbeenpaid internationallytothesize ofthemedical

workforceandtotestingwhetherthenumberofdoctors

whoareavailabletoprovidepatientcarehasanimpacton

patients’survivalchanceswiththeexceptionofonerecent

studyofsurgicalpatientswhere‘‘failuretorescue’’wasthe

outcomeofinterest.

1.2.3. Previousliteratureonworkload

One of the criticisms levelled against research on

staffing and patient outcomes is the lack of sufficient

controlsforpossibleconfoundingvariables,oneofwhich

might be workload. A review of the literature on

organisational factors in intensive care identified 13

studiesundertheheading‘‘volumeandpressureofwork’’

sotheimpactofworkloadisoflongstandinginternational

interest(CarmelandRowan,2001).

In a study of one UKhospital, Tarnow-Mordi et al.

(2000) found that patients who were treated when

workloadwashighwereabouttwice aslikely todieas

thosewhowereadmittedduringrelativelyquietperiods.

Threemeasuresofworkloadwereparticularlyimportant:

peakoccupancy,averagenursingrequirement(asdefined

bytheUKIntensiveCare Society)peroccupied bed per

shift, and theratio ofoccupied toappropriatelystaffed

beds.Althoughthiswasastudyofonlyoneunit,itwas

conductedover4yearsandmeetsmanyofthecriteriaofa

highqualitystudy(Westet al.,2009).

Both Unruh and Fottler (2006) and Evans and Kim

(2006)intheUShavearguedthatpatientturnover,defined

as number of admissions, transfers and discharges,

increases the demands on nurses and so affect patient

outcomes.This ideawastested in a recent USstudy of

nursestaffingandinpatientmortality,usingmodelsthat

includedmeasuresofday-to-day,shift-to-shiftvariations

innursestaffingandpatientturnover(Needlemanet al.,

2011). This large study includednearly 200,000

admis-sionsin43unitsinoneinstitutionthathadagoodstaffing

record and low mortalityrates. Using Cox proportional

hazardsestimation,theyfoundthattherewasasignificant

associationbetweenmortalityandexposuretoshiftswhen

therewerefewernursesthanthetargetnumber(setbya

wellcalibratedcommercialsystemfordeterminingnurse

staffing levels)and highturnover (admissions,transfers

and discharges). They found that ‘‘...the risk of death

increasedby2percentforeachbelow-targetshiftand4

percentforeachhighturnovershifttowhichapatientwas

exposed.’’Theauthorsarguethatstaffingdecisionsmight

best betaken on a shift-by-shift basis asworkload can

fluctuaterapidlyandisapotentialthreattopatientsafety.

In summary, interest in the effect of workload on

patient outcomesis long-standing.Some goodevidence

exists that workload, variously defined and measured,

mighthaveadirecteffectonpatientmortalityaswellas

beinganimportantmediatingvariableinamodellinking

staffinglevelstopatientoutcomes.

1.3. Hypotheses

What are the mechanisms linking staffing levels to

patientmortality?Inthenursingliterature,theconceptof

‘‘surveillance’’, thatisthecloseandcontinuous

observa-tionprovidedbyskillednursingstaff,playsakeyrole(e.g.,

Clarke and Donaldson, 2008; Kutney-Lee et al., 2009).

Becausenursesarewithpatientsforlongerperiodsoftime

thananyothermemberoftheteam,theyaremorelikelyto

see early warning signs, such as increasing pallor,

breathlessness or a change in vital signs that indicate

deteriorationinapatient’scondition.Morenursesonthe

unit should mean that patients will be more closely

observed;nursescanrespondmorerapidlywhenpatients

needlifesavinginterventionsandnursescanmobilisethe

resourcesofthehospitaltomeettheirneeds.Thisleadsto

thefollowinghypothesis:

H1. HighernumbersofnursesontheICUestablishment

willbeassociatedwithlowerratesofpatientmortality.

Historians claim that in the 1950s and 1960s, the

demandsofintensivecareintheUSmeantthatnursesand

doctors had to evolve a new kind of relationship, one

whichgavenursesmoreautonomyandindependenceand

forgedlinksamongalltheteammembers(Sandelowski,

2000). An ethnographic study conducted in a UK ICU

describesnursesanddoctorsasperformingverysimilar

undoubtedlythedominantoccupation, therelationship

between medicine and nursing in intensive care is

characterised more by convergence and incorporation

than competition...’’. This author describes observing

working relationships characterisedby informalityand

mutualrespectsuggestingthatICUsmightcomecloserto

theideal ofa multi-disciplinaryteamthan manyother

healthcaresettings.Thismakesitallthemoresurprising

that so many studies have adopted a uni-professional

focus.Itseemsreasonabletosuggestthatifnursesand

doctorsareworkingcloselytogether,andindeedifthereis

someoverlapintheirfunctions,weneedtoincludesome

measure of medical staffing, leading to the following

hypothesis:

H2. Higher numbers of consultants in an ICU will be

associatedwithlowermortalityrates.

Few studies have tested whether the numbers of

support staff, that is staff who provide administrative,

clericaland technicalsupporttotheUnit,arerelatedto

patient outcomes. Although not involved directly in

patientcare,thepresenceofsupportstaffshould,through

relievingthemofadministrativedutiesaswellastechnical

and scientific tasks, increase the amount of time that

nursesanddoctorsareabletospendwithpatients.The

hypothesisthenisthat:

H3. Higher numbers of supportstaff in an ICU willbe

associatedwithlowermortalityrates.

Theframeworkdevelopedaboveisbasedontheidea

thatadditionalresourcesleadtobetteroutcomes.

Howev-er,thereisanimportantmediatingvariable:theamountof

work that has to be done. Units vary greatly in key

workloadindicatorsincludingadmissions,bedoccupancy

andturnoverratessothatthesamenumberofnursesand

doctors,evenwhenstandardisedbythenumberofbeds,

maymeanverydifferentthingsinbusyratherthanquiet

ICUs.

H4. Thehighertheworkloadof theunit,theless likely

individualpatientsaretosurvive.

In summary, we have developed a simple input,

throughput, outputmodel linkinghumanresourcesand

patientoutcomesinICUs.Whentherearemoremembers

oftheclinicalstaffavailabletoprovidesurveillanceandto

intervene when required and holding constant the

workloadoftheunit,morepatientsarelikelytosurvive.

Fromthis conceptualframeworkand basedon previous

empirical research four hypotheses have been derived

basedontheassumptionthatthefactorsunderlyingthese

fourhypotheses haveindependent effects, detectablein

multivariableanalysis.

2. Methods

2.1. Studydesign

Thisisanobservationalstudywherestatisticalcontrols

are used to assess the relationship between the key

independent variables of interest and the dependent

variable.Thisstudyseekstounderstandtherelationship

between thesize of the clinical workforce (nurses and

doctors),aswellasthenumberofsupportstaff

(adminis-trative, clerical, technical and scientific), and patients’

chancesofsurvival.Althoughwehavehypothesisedthat

therewillbea relationshipbetween thesevariables, we

expectthatthepatient’sowncondition—age,numberof

diagnoses,severityoftheillnessandco-morbidities—will

contributemorethananyotherfactortotheoutcome.This

meansthatweneedtotaketheirriskofdyingduetotheir

ownconditionintoaccount,aswellasanyothervariables

thatcontributetotheoutcome,inordertogetunbiased

estimatesofthekeyvariablesofinterest.

2.2. Datasources/measurement

TheIntensiveCare NationalAudit &ResearchCentre

(ICNARC) Case Mix Programme is the national clinical

auditofpatientoutcomesfromICUsinEngland,Walesand

NorthernIreland.Participationisvoluntary;howeverthe

participatingunitsarerepresentativeofthepopulationin

terms of geographical spread, unit size and hospital

teaching status (university vs. non-university). Datafor

thesixmonthsbeforeandafterMarch1998,whichhad

beencollectedprospectively,weremergedonto

organisa-tionaldataon65ICUssurveyedbytheAuditCommission.

ThematcheddatasetcontainedinformationonlyonICUs

inEngland.Weusetwodichotomousdependentvariables:

whetherornotthepatientsurvivedtheirstayinICUand

whetherornotthepatientsurvivedtheirstayintheacute

hospital. Most previous studies have used the latter

measureof mortalityand this isan attractiveoptionin

our study for a number of reasons. First, it facilitates

comparability with prior research. Second, the risk

adjustment variable we use is based on this outcome.

ThirditispossiblethatpeoplearetransferredoutofICU

when death is thought to be inevitable, or when the

workloadoftheunitistoohigh.Ontheotherhand,itcould

bearguedthatstaffinglevelsinICUshouldnotbeexpected

to affect the chances of mortality outside the unit,

especiallywhenthekeymechanismlinkingstaffinglevels

andmortalityisarguedtobe‘‘surveillance’’.Therefore,we

analysebothoutcomestoexploretheideathatstaffingin

anICUwillhavethegreatestimpactonICUmortalityand,

althoughtheeffectwillcarryoverintothehospital,itwill

beattenuated.

2.3. Variables

2.3.1. Riskadjustment

Thereareanumberofriskpredictionmodelsforuse

withcriticalcarepatients.Recently,ICNARCdevelopeda

newmodel(theICNARCmodel)usingdatafromtheCase

Mix Programme, which we adopt. The purpose of this

variable is to control for the great influence that the

patients’ownhealthstatuswillhaveontheoutcomeof

their hospital stay. The ICNARC model was developed

specifically to underpin comparative studies of

risk-adjusted outcomes for adult critical care in the UK. To

thisend,itspecificallyavoidsfactorsrelatedtotreatment

only.DetailsofthemodelareprovidedbyHarrisonet al. (2006,2007),butbriefly,theICNARCscoreisbasedona

physiology model, includingblood pressure, respiratory

rate,oxygenation,andacidbasedisturbance,alongwitha

range of other factors known to be associated with

mortality,includingage,pastmedicalhistory,andsource

ofadmissiontoanICU.Themodelhasbeenvalidatedand

shown to perform better than other risk adjustment

models,suchasAPACHEIIandAPACHEIII.

2.3.2. Humanresourcevariables

Dataonseveralindependentvariablescamefromthe

Audit Commission’s survey of all ICUs in England and

Wales,onwhichthereportCriticaltoSuccess(1999)was

based.Keyexplanatoryvariablesderivedfromthisdataset

aredescribedbelow.

Number of nurses per bed: This variable counts the

numberoffull-timeequivalentnursesonthepermanent

staffoftheICUononespecificdate(thedateoftheAudit

Commissionsurvey).The questiononthesurveyasked

forseparateinformationonregisterednursesandhealth

careassistants.Thevariableusedintheseanalysesisa

count of theregistered nurses at differentgrades who

wereinpostonthecensusdate.Itisimportanttonote

thatthisisnotthenumberofnursingstaffavailablefor

dutywhenanyparticularpatientisadmitted.Wewere

able to break the number of nurses per bed into two

separatevariables:thenumberofdirectcarenursesand

the number of supernumerary nurses. This was made

possiblebythefactthatsurveyrespondentswereasked

tospecifyhowmanyofthenursesinpostonthecensus

dateweredesignatedassupernumerary.Weanalysethe

effectof the number of direct care nurses as they are

potentially more relevant to the concept of

‘‘surveil-lance’’thanthesupernumerarynursewhocontributeto

theunitin otherways.

Number of consultant Notional Half Days (NHDs) per

bed: Unitsreported to theAudit Commission thetotal

number of weekly fixed notional half daysfor critical

care clinical sessions. To those that were reported as

‘‘sharedwithotherunits’’wegaveaweightingofhalfas

thedatadidnotspecifyhowmuchtimewasallocatedto

eachunit.

Intensivist:TheACsurveyaskedwhetheroneormore

consultantsworked allthetimein theICU(andrelated

critical careunits)withno otherclinicalcommitments.

Thisis a dummyvariable, coded1 for unitsthat had a

dedicatedconsultant(intensivist).

Support Staff: This variable was the summation of

severaldifferentcategoriesofstaffincluding

administra-tiveandclericalstaff (e.g.,businessorservicemanager,

secretary,wardclerk,auditassistant)and technicaland

scientific staff (ICU technician, ECG technician). This

variabledoesnotincludeprofessionsalliedtomedicine,

such as pharmacists, dieticians, occupational therapists

andphysiotherapists.

Workload variables: We used several variables to

measure thepressure of workexperienced by thestaff

of theICU. Thefirst three aredrawn from the ICNARC

dataset, whereas the second two are derived from the

AuditCommissiondataset.

1.ProportionofbedsintheICUthatwereoccupiedatthe

timeofeachpatient’sadmission.

2.Mean ICNARC model predicted log odds of acute

hospitalmortalityofotherpatients intheunitatthe

timeofadmission(ameasureofhowseriouslyillthe

otherpatientswere).

3.AveragelengthofstayofpatientsintheICU,measured

inhours.

4.Admissionstotheunit,perbedperday.

5.Thenumberof transfersinfromanotherTrusttothe

unit,perbed,perweek.

Thefirststageofthisanalysiswashypothesistesting

but we also conducted further exploratory analysis

which included interactions between the number of

nursesandthenumberofconsultantsandthepredicted

log odds of acute hospital mortality from the ICNARC

model to see whether the impact of staffing differs

depending on the degree of severity of the patient’s

illness.

2.4. Statisticalmethods

Multilevel logistic regression was used to perform

alltheanalyses(GuoandZhao,2000;GelmanandHill,

2006).Allouranalyseswerecarriedoutusingthelmer

function thatis partof the lme4 package in Rversion

3.0 (R Development Core Team, 2013; Bates et al.,

2013).Multilevel regression is used forthese analyses

becauseitwouldbeinappropriate toconsiderpatients

treatedin thesame ICUasbeing independent

observa-tions.

Thebasicmodelthatweusedcanberepresentedbythe

followingequation: log pij 1pij ! ¼b0þ X k bkxijkþuj;

where i indexes patients, j indexes ICUs, there are k

explanatory variables in themodel, pij=Pr(yij=1) isthe

probability of death, and the following conditions are

assumed: E½uj¼0; varðujÞ¼

s

2u;

andcovðuj;uj0Þ¼0 forallj6¼j0:

Itwillbenotedthatthisisessentiallyastandardlogistic

regressionmodelwiththeadditionofaseparaterandom

effect,uj,foreachhospitaltrust.Thiscanbeconsideredto

controlforunmeasuredfactorsthatvaryacrossunitsbut

are constant over time. The assumptions regarding the

variance and covariances of the random effects are

conventional for this type of analysis. The estimates of

theeffectsoftheexplanatoryvariables(bk)areinterpreted

inthesamewayasstandardlogisticregressionestimates,

and confidence intervals for these estimates are also

interpretedintheusualway.Whenwereportestimated

log odds ratios weuse themean values of explanatory

3. Results

3.1. Descriptivestatistics

TheAuditCommissiongathereddatafrom69ICUsin

total.Wewereabletomergedatafromthetwosources

(Audit CommissionandICNARC) on 65 ICUs.Fourunits

weredroppedfromtheanalysesbecausethenumberof

bedstheyreporteddifferedbymorethanthreeinthetwo

datasets, which cast doubt on the reliability of the

information aboutthem. Themergeddatasethad

infor-mationon38,168admissions.Thenumberofpatientswas

reducedinsomeofthemultivariableanalysesbymissing

data;numbersofcasesineachanalysisareshowninthe

relevanttables.

Ofthe38,168patientsinthisstudy,6413(16.8percent)

diedintheICUwheretheywerebeingtreated.Afurther

4397 (11.5 per cent) died in hospital after leaving the

initialICU,with579(1.5percent)patientsbeinglostto

followup(seeFigs. 1and2).Theobservedcrudemortality

ratesintheunitsvariedbetween8.1percentand33.9per

cent,whilethehospitalmortalityrateswerebetween16.9

percent and 47.9percent. Thisvariation couldnot be

entirelyexplainedbyvariationinpatientriskfactors.The

correlationbetweentheICUandhospitalmortalityrates

was0.90.

The distribution of nurses per bed (whole time

equivalents) across different units is shown in Fig. 3.

OnlyasmallminorityofunitsintheUKhadsevenormore

nursesperbedon their permanentpayroll, thenumber

generallyaccepted asbeing requiredtomaintain a one

patientpernurseratiooverthreeshifts,withallowancefor

sickness and holiday leave. The number of consultant

notional half days per bed is also shown in Fig. 3.

Descriptivestatisticsandcorrelationsamongthevariables

used in the analysis are shown in Tables 1 and 2,

respectively.

Table1

Descriptivestatisticsoftheexplanatoryvariables.

Mean Standard deviation

IMlogodds 1.41 1.92

Proportionbedsoccupied 0.83 0.20 Admissionsperbedperday 0.18 0.061 Transfersinperbedperweek 0.090 0.073 Totalsupportstaffperbed 0.25 0.23 Directcarenursesperbed 4.97 1.29 Clinicalnotionalhalfdaysperbed 1.04 0.75 Unitlengthofstayinhours 100.0 27.0

3.2. Multivariableanalyses

Theresultsofourmultivariableanalysesareshownin

Table 3(ICUmortality)andTable 4(hospitalmortality).

Thefirstcolumnineach tableshowstheresultsforthe

testsofHypotheses1to4.

ConsideringfirsttheimpactonICUmortality,shown

in Table 3, we can see that the risk of mortality is

significantly increased by the severity of the patient’s

own condition (ICNARC model predicted log odds of

acute hospital mortality) and by our measures of

workload(hypothesis4).Theriskofmortalityincreases

when a higher proportion of beds in the unit are

occupiedandwhentherearealargenumberoftransfers

intotheunit.Thenumberofnursesandthenumberof

consultantsbothsignificantlyreducetheriskof

mortal-ity (Hypotheses 1 and 2). The hypothesis that the

numberofsupportstaffwouldaffectpatientmortalityis

not supported.

Further,aswebelievethattheeffectofstaffingmight

dependontheseverityofapatient’sillness;inthesecond

column we add an interaction between the number of

nurses and predicted log odds of mortality, while in

column3weaddaninteractionbetweenthenumberof

consultantsandpredictedlogoddsofmortality.Herewe

are exploring the relationship between staffing and

Fig.2.Proportionofpatientswhodiedinthehospitalinwhichtheywerebeingtreated.Eachpointonthegraphrepresentsahospital.

Table2

Correlationmatrixoftheexplanatoryvariables.

1 2 3 4 5 6 7 8 9

ICNARCmodellogoddsforeachpatient 1 1 Numberofdirectcarenursesperbed 2 .069 1 Proportionbedsoccupiedatthetimeofadmission 3 .044 .14 1

Admissionsperbedperday 4 .038 .36 .033 1

Transfersinperbedperweek 5 .029 .34 .006 .10 1

Supportstaffperbed 6 .043 .067 .13 .12 .07 1

NumberofconsultantNHDsperbed 7 .04 .014 .12 .21 .018 .54 1 MeanIMlogodds(otherpatientsinunit) 8 .083 .11 .040 .064 .043 .012 .029 1

Fig.3.Barplotshowingthenumbersofdirectcarenurses(WTEs)andconsultantNHDsperbedineachICU.

Table3

MortalityintheICU.Oddsratiosfrommultilevellogisticregressionmodels,with95percentconfidenceintervals.

1 2 3

ICNARCmodellogoddsofpatientmortality 2.62[2.56,2.69] 2.97[2.70,3.27] 2.97[2.68,3.29] MeanIMlogoddsofpatientmortalityd

0.95[0.91,0.99] 0.95[0.91,0.99] 0.95[0.91,0.99] AverageICUlengthofstay(h)d

1.54[1.11,2.13] 1.53[1.11,2.12] 1.53[1.11,2.12] Proportionbedsoccupiedd

1.24[1.03,1.48] 1.23[1.03,1.48] 1.23[1.03,1.48] Admissionsperbedperdayd

1.66[0.35,7.96] 1.65[0.35,7.89] 1.66[0.35,7.90] Transfersinfromanothertrustperweekperbedd _{5.06[1.47,17.4] 5.17[1.50,17.8] 5.17[1.50,17.8]} Supportstaffperbedc _{1.16[0.81,1.68] 1.15[0.79,1.66] 1.15[0.79,1.66]}

Intensivist 0.97[0.79,1.19] 0.98[0.80,1.21] 0.98[0.80,1.21]

Numberofdirectcarenursesperbeda

0.90[0.84,0.97] 0.90[0.83,0.97] 0.90[0.83,0.97] NumberofconsultantNHDsperbedb

0.85[0.76,0.95] 0.85[0.76,0.95] 0.85[0.76,0.95]

IMlo*Numberofnursesperbed 0.98[0.96,0.99] 0.98[0.96,0.99]

IMlo*NumberofconsultantNHDsperbed 1.00[0.97,1.03]

Intercept 1.38 1.36 1.36

Standarddeviationofrandomeffect 0.28 0.28 0.28

Deviance 23,293 23,286 23,286 Numberofadmissions 38,168 38,168 38,168 Numberofunits 65 65 65 a Hypothesis1. b Hypothesis2. c Hypothesis3. d Hypothesis4.

Fig.4.EstimatedeffectontheoddsofmortalityintheICUofthenumberofdirectcarenursesintheunitforpatientsatdifferentlevelsofriskasmeasuredby theICNARCmodel,witharugplotshowingthedistributionofobservedvalues.OtherexplanatoryvariablesaresetatthemeanvaluesshowninTable1, withunitsassumedtohaveanintensivist.IMlomeansICNARCModellogoddsofmortality.

Table4

Mortalityinacutehospital.Oddsratiosfrommultilevellogisticregressionmodels,with95%confidenceintervals.

1 2 3

IMlogoddsofpatientmortality 2.61[2.56,2.67] 2.72[2.50,2.96] 2.64[2.42,2.89] MeanIMlogoddsofpatientmortalityd

0.96[0.93,0.99] 0.96[0.93,0.99] 0.96[0.93,0.99] AverageICUlengthofstay(h)d

1.26[.95,1.68] 1.26[0.95,1.68] 1.26[0.95,1.68] Proportionbedsoccupiedd

1.22[1.05,1.43] 1.22[1.04,1.43] 1.22[1.04,1.43] Admissionsperbedperdayd

1.04[0.27,4.04] 1.04[0.27,4.02] 1.05[0.27,4.06] Transfersinfromanothertrustperbedperweekd _{11.7[3.93,34.7] 11.7[3.93,34.6] 11.6[3.91,34.5]} Supportstaffperbedc _{1.08[0.78,1.49] 1.08[0.78,1.49] 1.08[0.78,1.49]}

Intensivist 0.99[0.83,1.19] 0.99[0.83,1.19] 0.99[0.83,1.19]

Numberofdirectcarenursesperbeda

0.92[0.87,0.98] 0.92[0.86,0.98] 0.92[0.86,0.98] NumberofconsultantNHDsperbedb

0.89[0.81,0.99] 0.90[0.81,0.99] 0.91[0.82,1.00]

IMlo*Numberofnursesperbed 0.99[0.98,1.01] 0.99[0.98,1.01]

IMlo*NumberofconsultantNHDsperbed 1.03[1.00,1.06]

Intercept 0.22 0.19 0.21

Standarddeviationofrandomeffect 0.24 0.24 0.24

Deviance 30,335 30,334 30,330 Numberofadmissions 37,590 37,590 37,590 Numberofunits 65 65 65 a Hypothesis1. b Hypothesis2. c Hypothesis3. d Hypothesis4.

mortalityas thereis noprevious literatureon whichto

baseahypothesis.

Column two shows that the interaction variable is

negative and significant,implying that thereduction in

mortality risk from having more nurses is larger for

patientswhoarethemostseverelyill.Inthethirdcolumn,

however,wecanseethattheequivalentinteractionisnot

significant whenit involves thenumber of consultants.

Fig. 4isaneffectplot,basedontheresultsincolumn2of

Table 3,toillustratehowtheinfluenceofnursestaffing

varieswithconditionseverity.Bywayofexample,wecan

calculatethattheeffectofincreasingthenumberofnurses

per bedfromfourtosix is toreducetheprobability of

mortality for patients with a predicted log odds of

mortality of -3 from .017 to .016. This represents

approximatelyone extradeath for every1000 patients.

However,therealimportanceoftheseresultsisthatthis

effectissomuchlargerforthosepatientswhohaveahigh

riskofdeath.Ifwelookatpatientswithapredicted log

oddsofmortalityof2,thenthereductionintheprobability

ofmortalityingoingfromfourtosixnursesisfrom.72to

.65,oraboutsevenextradeathsper100patients.About5

percentofICUadmissionsinthisdatasethaveapredicted

logoddsofmortalityatleastthishigh,sothisrepresentsa

significantnumberofpatients.Theseresultssuggestthat

themostseverelyillpatientsareasub-groupthatismost

vulnerabletolownursestaffinglevels.

InTable 4weseeabroadlysimilarpatternofresultsfor

theeffectsonhospitalmortality.Again,thenumberofnurses

and the numberof consultants perbed havea negative

association with the risk of mortality. The estimated

magnitudeoftheseeffectsissimilartothecorresponding

estimatesoftheireffectsontheriskofmortalityintheICU.

However, although both main effects are statistically

significant,thisisnottrueofeitherofthe twoestimated

interactioneffects.ThemaineffectofICUstaffingisonthe

outcomeofthepatients’stayinICU.Althoughwecanshow

aneffect onhospitalmortality,theinteractioneffect between

nursestaffingandpatientacuityisnotsignificant.

4. Discussion

4.1. Keyresults

Thisstudyinvestigatedwhetherthesizeoftheclinical

workforce—the number of nurses and doctors—was

associated with the survival chances of critically ill

patientsbothintheintensivecareunitandinthehospital.

The most significant findings are that, controlling for

patientcharacteristicsandtheworkloadoftheunit,higher

numbersofnursesperbedontheunit’sestablishmentand

higher numbers of consultants per bed were both

associatedwithhighersurvivalrates.Nursesanddoctors

doseemtomakeadifferencetopatientoutcomesinICU.

We foundno evidencethatthenumberofsupportstaff

workingontheunitimprovespatients’survivalchances

even thoughtheir presenceon theunit might havethe

effectofreleasingclinicalstafftocareforpatients.Inthis

study,clearsupportemergedfortheeffectofworkload.

Highworkload,measuredinanumberofdifferentways,

wasassociatedwithhighermortality.

There was also a statistically significant interaction

between the number of nurses and patient’s risk of

mortality, suggesting that nursing staff availability has

thegreatestimpactonthoseatgreatestriskofdeath.This

isconsistentwiththeclaimthatnursingsurveillanceisone

ofthekeymechanismslinkingnursingnumberstopatient

outcomes.Thefactthattheinteractionsbetweenthesize

oftheclinicalworkforceandthepatient’sriskofmortality

werestatisticallysignificantinestimatesofICUmortality

butnotinthecaseofhospitalmortalityalsolendssupport

to the suggestion that the effects are due to the key

surveillanceroleofnursesinICUs.

4.2. Interpretation

Takentogetherthesefindingssuggesttheneedtostudy

thewholeteaminICU.There isalarge literatureonthe

nursingcontributiontopatientoutcomeswhichfocuseson

thesizeandlevelofqualificationsofstaff.Thereisalsoa

large literature on the medical contribution to ICU

outcomes.Wearguethatfuturestudiesneedtoconsider

thesetwogroupsatthesametime.Nursesanddoctors,in

particular,mayinsomeclinicalsettingsbesubstitutesfor

each other,so that units that are short of doctors, for

example,maycompensatebyhiringmorenurses.Thisis

supported by ethnographic evidence of how closely

clinicians fromdifferentprofessionalbackgrounds work

togetherinthissetting(Carmel,2006).Furtherempirical

supportfortheimportanceofteamworkingisprovidedby

arecentUSstudy(Kimet al.,2010)whichshowedalink

between daily rounds by a multi-disciplinary team

(physicians,nurses, respiratory therapists, clinical

phar-macists,social workersandothers)and lowermortality

among medical ICU units, especially when this was

combinedwithhighintensityphysicianstaffing

(manda-torycarebyanintensivistormandatoryconsultation).This

studydidnotcontrolforthesizeoftheclinicalworkforce

butdoessuggestthatfurtherworkonthecontributionof

theteamasawholewillbejustified.

Futurestudiesalsoneedtoneedtocontroladequately

for workload. The nurse-to-patient or physician to bed

ratiodoesnotmakesensewithoutincludingcontrolsfor

thethroughputoftheunitaswellasamountofcarethat

eachpatientneeds.Furtherresearchontheworkforceis

urgentlyrequired toguidedecisions about safestaffing

levelsinavarietyofhealthcaresettingstoensurepatient

safetyaswellasequityandcost-effectivenessacrossthe

wholesystem.

This study, which tests a simple model of human

resources, demonstrates a relationship between those

inputs and perhaps the most important measure of a

hospitalsperformance—whetheror notpatients survive.

Thebaseline modelwehave producedwillbeusefulin

testingmorecomplexmodelsof,forexample,theimpactof

teammemberslevelofeducationandexperience,quality

ofteamworkorprocessesofcareinICUs.

4.3. Strengthsandweaknessesofthestudy

Theuseoftwohighqualitynationaldatasetsisoneof

NationalAuditandResearchCentre(ICNARC)enabledusto

estimate models of ICU and hospital mortality using

sophisticatedmethodsofriskadjustment,andtocontrol

forimportantsourcesofworkload;theproportionofbeds

thatwereoccupiedatthetimethepatientwasadmitted

andthelevelofacuityoftheotherpatientsthatwereinthe

unit.Dataon characteristicsoftheunit collectedbythe

Audit Commission in 1998 and a merged dataset with

information on 65 units and 38,168 patients was

constructedand multilevel logistic regression was used

to analyse the outcomes of their admission. The large

numberofunitsandpatientsanalysedinthisstudy,aswell

astheuseofappropriatemethodsforthestructureofthe

dataareadditionalstrengthsofthisstudy.Thestudymay

alsobe importanttheoretically, because it is consistent

withtheoperationof mechanisms,suchas surveillance

andtheadoptionofevidence-basedpractice,proposedas

key mechanisms linking staffing inputs and patient

outcomesinICU.Theseresultswillhopefullybedeveloped

infuturequalitativeandquantitativeresearch.

Butthisstudyalsohasweaknesses.Thedataare

cross-sectionalwhichlimitstheextenttowhichcausalclaims

canbemade.Iflongitudinaldatawereavailable,recording

theeffectsofchangesinstaffinglevelsandpatientacuity

over time, we would be able to produce more robust

evidenceaboutthe impactof staffinglevelsonpatient

outcomes. The data are also several years old. This is

acceptable to us because we set out to investigate a

relationship between human resources and

organisa-tionalperformancewhichisnottemporallyor

geographi-cally bounded. If there were more recent data of this

quality,thenitwouldbedifficulttojustifythis

investiga-tion, but such data do not yet exist. Changes in the

distribution of staff across units since the data were

collected will mean that the results are probably less

usefulforpracticalpurposessuchasworkforceplanning,

butstrongly suggest thatmanagersand policy makers

needtoconsiderhumanresourcesaskeytoqualityand

safetyinUKICUs.Thereareotherweaknessesinthedata.

KeyvariablesderivedfromtheAuditCommissionSurvey

onstaffingandworkloadaremeasuredfortheICUasa

wholenotatthepatientlevelandthereisnodirectpatient

levelmeasureofworkload.Ideally,wewouldliketobe

abletomeasurethenumberofnurses, doctorsandthe

activityoftheICUforeachpatient,shift-by-shift;some

studiesintheUSarenowbasedondataofthisquality.The

results of this study then underscore the need for a

prospectivestudyinthearea,giventhecrucialimportance

ofstaffcostsintheNHS.

Since2000therehasbeenaprocessofmodernisationin

criticalcare,includingtheformationof29geographically

based networks (several hospitals working together to

common protocols and standards); the integration of

criticalcareintotherangeofadultservicesinhospitals

(outreachandresponseteams),andaplannedapproachto

workforcedevelopment.TheDepartmentofHealth(2000)

alsoproposedthatcriticalcareshouldbebasedonseverity

ofillness,ratherthanthelocationofthepatientwithina

designatedunitandthatpatientdependency,ratherthan

bednumbers,shouldbethebasisofstaffingallocations.So

theboundariesaroundcritical careasa unitof analysis

may be less clear. Increased funding was also made

availabletodevelopservicesandtherewasariseinthe

numberof,mainlyhighdependency,beds,whichmayhave

ledtoadiminutionoftheaveragelevelofacuityofpatients

who are now receiving critical care. The modernisation

agendapursuedsince2000doesseemtohaveproduced

improvements overall in the effectiveness and

cost-effectiveness of critical care, although it is difficult to

attribute thesetooneormoreofthechangesthathave

takenplace(Hutchingset al.,2009).Itispossiblethatthe

modernisation agenda has decreased the amount of

variationinstaffinglevelsandpatientmortalityfromunit

tounit,whichwasasourceofgreatconcernwhenAudit

Commission data were published in 1998, or that the

organisational changesto theservice have changedthe

relationship between staffingnumbersand patient

out-comes.However,thisremainsanempiricalquestionwhich

willrequiretheformulationofnewtheoreticalmodelsand

thecollectionofnewdatathatwillenableustotestthe

relationshipbetweenstaffinglevelsandpatientmortality

inthecurrentorganisationalcontext.

5. Conclusions

Thisisoneofthefewstudiesthatshowasignificant

relationshipbetweenthesizeoftheclinicalworkforceand

patient’schancesofsurvivalinICU.Thisstudytherefore

makesacontributiontotheinternationalliterature.Atthe

sametime,thisstudyisparticularlyimportantbecauseso

fewstudiesinthisresearchtraditionhavebeenconducted

intheUK.Whiletheinternationalevidenceaboutnursing

numbershasbeenextremelyinfluentialinshapingdebates

aboutresources,thisstudyshouldhaveadditionalimpact

on policy and practice because it shows that the

relationshipbetweennursingnumbersandpatient

mor-talityalsoholdsinthecontextoftheNHS.

Thisstudyhasalsoproducedsomenewknowledge.Itis

thefirststudytoincludenurses,doctorsandsupportstaff

inoneanalysisandtocontrolfortheimpactofworkloadon

patients’ survivalchances. Although there is a growing

internationalliteratureonnursestaffing,fewstudieshave

investigatedtheimportanceofotherprofessionalgroups.

It is also the first analysis to test whether there is an

interaction effect between the number of nurses and

number ofdoctorsandtheseverityof thepatients own

illness. Itseemsreasonabletoarguethat skillednurses,

whohavethetimetoobservepatientsclosely,tointervene

ormobilisetheteamiftheybegintodeteriorate,wouldbe

mostimportanttopatients whoareatthegreatestrisk.

Thisstudyisthefirsttoproduceevidencethatthisisthe

case.

Theresultssuggesttheneedtostudytheeffectsofthe

wholeteamthatisavailabletocareforpatients,ormore

radically,thatweshiftourfocusawayfromprofessional

groupsand ontotheskills thatarerequiredinICU.In a

recent paper,Needleman et al. (2011) call for further

researchtotryto understand‘‘...thecomplexinterplay

among nurse staffing, patient preferences, and other

factors,includingstaffinglevelsforphysiciansandother

non-nursing personnel,technology, work processes and

programmeofworkthatwillinformpolicyandpracticeto

improvepatientcare.

Conflictofinterest:Nonedeclared.

Funding:HealthFoundationFellowshiptofirstauthor.HFhad

noroleintheconductoftheresearch.

Ethical approval: London School of Hygiene and Tropical

Medicineethicscommittee.

AppendixA. Supplementarydata

Supplementary material related to this article can be

found, in the online version, at http://dx.doi.org/10.1016/

j.ijnurstu.2014.02.007. References

Aiken, L.H., Clarke, S.P.,Sloane, D.M., Sochalski,J., Silber, J.H.,2002. Hospitalnursestaffingandpatientmortality,nurseburnout,and jobdissatisfaction.TheJournaloftheAmericanMedicalAssociation 288(16) 1987–1993.

Amaravadi,R.K.,Dimick,J.B.,Pronovost,P.J.,Lipsett,P.A.,2000.ICU nurse-to-patientratioisassociatedwithcomplicationsandresourceuse afteresophagectomy.IntensiveCareMedicine26(12) 1857–1862. AuditCommission,1999.CriticaltoSuccess:ThePlaceofEfficientand Effective Critical Care ServicesWithin the AcuteHospital. Audit Commission,London.

Bates,D.,Maechler,M.,Bolker,B. lme4:Linearmixed-effectsmodels usingS4classes.2013.

Bastos,P.G.,Knaus, W.A.,Zimmerman,J.E.,MagalhaesJr.,A.,Sun, X., Wagner,D.P.,1996July.TheImportanceofTechnologyforAchieving SuperiorOutcomes fromIntensive Care,BrazilAPACHEIII Study Group.IntensiveCareMedicine22(7) 664–669.

Burchardi,H.,Moerer,O.,2001.Twenty-fourhourpresenceofphysicians intheICU.CriticalCare5,131–137.

Carmel,S.,2006.Boundariesobscuredandboundariesreinforced: incor-porationasastrategyofoccupationalenhancementforintensive care.SociologyofHealthandIllness(March(2)) 154–177. Carmel,S.,Rowan,K.,2001Aug.Variationinintensivecareunit

out-comes:asearchfortheevidenceonorganizationalfactors.Current OpinioninCriticalCare7(4) 284–296.

Cho,S.H.,Hwang,J.H.,Kim,J.,2008.Nursestaffingandpatientmortalityin intensivecareunits.NursingResearch57(September-October(5)) 322–330.

Clarke,S.P.,Donaldson,N.,2008.PatientSafetyandQuality:An Evidence-basedHand-bookforNurses(AHRQPublicationNo.08-0043)Agency forHealthcareResearchandQuality,Rockville,MD.Nursestaffing andpatientcarequalityandsafety.Availablefrom http://www.ahrq.-gov/qual/nurseshdbk/

Crocker,C.,2007.ThedevelopmentofcriticalcareinEngland.Intensive andCriticalCareNursing23,323–330.

Dara,S.I.,Afessa,B.,2005Aug.Intensivist-to-bedratio:associationwith outcomesinthemedicalICU.Chest128(2) 567–572.

Dimick,J.B.,Pronovost,P.J.,Heitmiller,R.F.,Lipsett,P.A.,2001.Intensive careunitphysicianstaffingisassociatedwithdecreasedlengthof stay, hospitalcost,andcomplications afteresophageal resection. CriticalCareMedicine29(April(4)) 753–758.

Evans,W.N.,Kim,B.,2006.Patientoutcomeswhenhospitalsexperiencea surgeinadmissions.JournalofHealthEconomics25,365–368. Gelman,A.,Hill,J.,2006.DataAnalysisUsingRegressionandMultilevel/

HierarchicalModels.CambridgeUniversityPress,Cambridge. Giraud,T.,Dhainaut,J.F.,Vaxelaire,J.F.,Joseph,T.,Journois,D.,Bleichner,

G.,etal.,1993.Iatrogeniccomplicationsinadultintensivecareunits: aprospectivetwocentrestudy.CriticalCareMedicine21(1) 40–51. Griffiths,P.,Jones,S.,Bottle,A.,2013Feb.Is‘‘failuretorescue’’derived fromadministrativedatainEnglandanursesensitivepatientsafety indicatorforsurgicalcare?Observationalstudy.InternationalJournal ofNursingStudies50(2) 292–300.

Guo,G.,Zhao,H.,2000. Multilevelmodeling forbinarydata.Annual ReviewofSociology26,441–462.

Harrison,D.A.,Brady,A.R.,Parry,G.J.,Carpenter,J.R.,Rowan,K.,2006Mar 21.Recalibrationofriskpredictionmodelsinalargemulticenter

cohortofadmissionstoadult,generalcriticalcareunitsintheUnited Kingdom.CriticalCareMedicine.

Harrison,D.A.,Parry,G.J.,Carpenter,J.R.,Short,A.,Rowan,K.,2007.Anewrisk predictionmodelforcriticalcare:theIntensiveCareNationalAuditand ResearchCentre(ICNARC)model.CriticalCareMedicine35,1091–1098. Hutchings,A.,Durand,M.A.,Grieve,R.,Harrison,D.,Green,J.,Cairns,J., Rowan,K.,Black,N.,2009.Evaluationofthemodernizationofadult critical care services in England. Research Report. NIHR Service DeliveryandOrganisationProgramme.

Kahn,J.M.,Brake,H.,Steinberg,K.P.,2007.Intensivistphysicianstaffing andtheprocessofcareinacademicmedicalcentres.Qualityand SafetyinHealthCare16,329–333.

Kahn,J.M.,Hall,J.B.,2010.MoredoctorstotherescueintheIntensiveCare Unit:Acautionarynote.AmericanJournalofRespiratoryandCritical CareMedicine181(11) 1160–1161.

Kane,R.L.,Shamliyan,T.A.,Mueller,C.,Duval,S.,Wilt,T.J.,2007.Nursing staffingandqualityofpatientcare.MedicalCare45(12)1195–1204. Kiekkas,P.,Sakellaropoulos,G.C.,Brokalaki,H.,Manolis,E.,Samios,A., Skartsani,C.,Baltopoulos, G.I.,2008.Association betweennursing workloadandmortalityofICUpatients.JournalofNursingScholarship 40(4) 385–390.

Kim,M.M.,Bernato,A.E.,Angus,D.C.,Fleisher,L.F.,Kahn,J.M.,2010.The effectofmultidisciplinaryteamsonIntensiveCareUnitmortality. ArchivesofInternalMedicine170(4) 369–376.

Kutney-Lee,A.,Lake,E.T.,Aiken,L.H.,2009.Developmentofthehospital nursesurveillancecapacityprofile.ResearchinNursingandHealth 32,217–228.

Liang,Y.W.,Tsay,S.F.,Chen,W.Y.,2012.Effectsofnursestaffingratioson patientmortalityinTaiwanacutecarehospitals:alongitudinalstudy. JournalofNursingResearch20(March(1)) 1–7.

Mark,B.A.,Harless,D.W.,McCue,M.,Xu,Y.,2004Oct;.Alongitudinal examinationofhospitalregisterednursestaffingandqualityofcare. HealthServicesResearch39(5) 1629.

Metnitz,P.G.,Reiter,A.,Jordan,B.,Lang,T.,2004.Moreinterventionsdo notnecessarilyimproveoutcomeincriticallyillpatients.Intensive CareMedicine30(August(18)) 1586–1593.

Needleman,J.,Buerhaus,P.,Mattke,S.,Stewart,M.,Zelevinsky,K.,2002. Nurse-staffinglevelsandthequalityofcareinhospitals.TheNew EnglandJournalofMedicine346(22) 1715–1722.

Needleman,J.,Buerhaus,P.,Pankratz,S.,Leibson,C.L.,Stevens,S.R.,Harris, M.,2011.Nursestaffingandinpatienthospitalmortality.New Eng-landJournalofMedicine364,11.

Numata,Y.,Schulzer,M.,vanderWal,R.,Globerman,J.,Semeniuk,P., Balka,E.,Fitzgerald,J.M.,2006.Nursestaffinglevelsandhospital mortalityincriticalcaresettings:literatureandmeta-analysis. Jour-nalofAdvancedNursing.55(4) 435–448.

Penoyer,D.A.,2010.Nursestaffingandpatientoutcomesincriticalcare:a concisereview.CriticalCareMedicine38(7) 1521–1529. Person,S.D.,Allison,J.J.,Kiefe,C.I.,Weaver,M.T.,Williams,O.D.,Centor,

R.M.,Weissman,N.W.,2004.Nursestaffingandmortalityfor Medi-care patientswith acute myocardial infarction.Medical Care 42 (January(1)) 4–12.

Pronovost,P.J.,Jenckes,M.W.,Dorman,T.,Garret,E.,Breslow,M.J., Rosen-feld,B.A.,Lipset,P.A.,Bass,E.,1999.Organisationalcharacteristicsof intensivecareunitsrelatedtooutcomesofabdominalaorticsurgery. JournaloftheAmericanMedicalAssociation281(14) 1310–1459. Pronovost,P.J.,Dang,D.,Dorman,T.,Lipsett,P.A.,Garrett,E.,Jenckes,

M.W.,etal.,2001.Intensivecareunitnursestaffingandtheriskfor complicationsafterabdominalaorticsurgery.EffectiveClinical Prac-tice4(5) 199–206.

Pronovost,P.J.,Angus,D.C.,Dorman,T.,Robinson,K.A.,Dremsizov,T.T., Young,T.L.,2002.Physicianstaffingpatternsandclinicaloutcomesin criticallyillpatients:asystematicreview.JournaloftheAmerican MedicalAssociation288(November(17)) 2151–2162.

Rafferty,A.M.,Clarke,S.P.,Coles,J.,Ball,J.,James,P.,McKee,M.,Aiken,L.H., 2007.OutcomesofvariationinhospitalnursestaffinginEnglish hospitals: cross-sectional analysis of survey data and discharge records.InternationalJournalofNursingStudies44(2) 175–182. RDevelopmentCoreTeam,2013.R:ALanguageandEnvironmentfor

Statistical Computing. R Foundation for Statistical Computing, Vienna,Austria.,In:http://www.R-project.org.

Reis-Miranda,D.,Ryan,D.,Schaufeli,W.(Eds.),1998. Organisationand ManagementofIntensiveCare:AProspectiveStudyin12European Countries.Springer,Berlin.

Robert,J.,Fridkin,S.K.,Blumberg,H.M.,Anderson,B.,White,N.,Ray,S.M., etal.,2000.Theinfluenceofthecompositionofthenursingstaffon primarybloodstreaminfectionratesinasurgicalintensivecareunit. InfectionControlandHospitalEpidemiology21(1) 12–17. Sandelowski,M.,2000.DevicesandDesires:Gender,Technologyand

Shortell,S.M.,Zimmerman,J.E.,Rousseau,D.M.,Gillies,R.R.,Wagner,D.P., Draper,E.A., etal.,1994.Theperformanceofintensivecareunits: doesgoodmanagementmakeadifference? MedicalCare32(May (5)) 502–525.

Stone,P.W.,Mooney-Kane,C.,Laarson,E.L.,Horan,T.,Glance,L.G., Zwan-ziger,J.,Dick,A.W.,2007.Nurseworkingconditionsandpatientsafety outcomes.MedicalCare45(June(6)) 571–578.

Tarnow-Mordi,W.O.,Hau,C.,Warden,A.,Shearer,A.J.,2000.Hospital mortalityinrelationtostaffworkload:afour-yearstudyinanadult intensivecareunit.Lancet356(July(9225)) 185–189.

Tourangeau,A.E.,Doran,D.M.,McGillisHall,L.,O’BrienPallas,L.,Pringle, D.,Tu,J.V.,Cranley,L.A.,2007.Impactofhospitalnursingcareon30 daymortalityforacutemedicalpatients.JournalofAdvancedNursing 57(January(1)) 32–44.

Tucker,J.,2007Jan12;.Patientvolume,staffingandworkloadinrelation to risk-adjusted outcomesin a random stratified sample ofUK neonatalintensivecareunits:aprospectiveevaluation.Lancet359 (9301) 99–107.

Unruh,L.Y.,Fottler,M.D.,2006.Patientturnoverandnursingstaff ade-quacy.HealthServicesResearch41(April(2)) 599–612.

VandenHeede,K.,Sermeus,W.,Diya,L.,Clarke,S.P.,Lesaffre,E.,Vleugels, A.,Aiken, L.H.,2009Jul;.Nursestaffingandpatientoutcomesin Belgianacutehospitals:cross-sectional analysisofadministrative data.InternationalJournalofNursingStudies46(7) 928–939. West,E.,Sanderson,C.,Mays,N.,Rafferty,A.M.,Rowan,K.,2009.Nursing

resourcesandpatientoutcomesinintensivecare:asystematicreview oftheliterature.InternationalJournalofNursingStudies46(7)993– 1011.