Sudden survival improvement in critical neurotrauma: An exploratory analysis using a stratified statistical process control technique

Sudden survival improvement in critical neurotrauma: An exploratory analysis using a stratified statistical process control technique

Signe Søvik

*, Nils Oddvar Skaga

, Rolf Hanoa

, Torsten Eken

aDepartmentofAnaesthesiaandIntensiveCare,DivisionofSurgery,AkershusUniversityHospital,Norway

bInstituteofClinicalMedicine,FacultyofMedicine,UniversityofOslo,Norway

cDepartmentofAnaesthesiology,DivisionofEmergenciesandCriticalCare,OsloUniversityHospital,Norway

dDepartmentofNeurosurgery,DivisionofSurgeryandClinicalNeuroscience,OsloUniversityHospital,Norway

Introduction

Outcomeaftertraumaisafunctionofpatientcharacteristics, qualityofcare,andrandomevents[1].Forhospitalbenchmarking itisnecessarytoadjustfortheriskprofileofthepatientstoremove sources of variation that are independent of quality of care.

Without risk stratification, trauma centres treating the most severelyinjuredpatientswillappeartohaveworseresultsthan others. This may influence on referral patterns and resource allocationandevendiscouragethetreatmentofhigh-riskpatients.

ARTICLE INFO

Articlehistory:

Accepted30May2014

Keywords:

Traumasystem Outcomemonitoring VariableLife-AdjustedDisplay CUSUMtechnique

Qualityofcare

ABSTRACT

Background:Outcome aftertraumadependsonpatientcharacteristics, qualityofcare, andrandom events.TheTRISSmodelpredictsprobabilityofsurvival(Ps)adjustedforInjurySeverityScore(ISS), RevisedTraumaScore (RTS),mechanismof injury,and age. Qualityof careis oftenevaluated by calculatingthenumberof‘‘excess’’survivors,yearbyyear.Incontrast,theVariableLife-AdjustedDisplay (VLAD)techniqueallowsrapiddetectionofalteredsurvival.VLADadjustseachdeathorsurvivalbythe patient’sriskstatusandgraphicallydisplaysaccumulatednumberofunexpectedsurvivorsovertime.

Weevaluatedoutcomechangesandtheirtimerelationtotraumaserviceimprovements.

Methods:Observational,retrospectivestudyofthetotal2001–2011traumapopulationfromaLevelI traumacentre.Outcomewas30-daysurvival.PswascalculatedwiththeTRISSmodel,2005coefficients.

VLADgraphswerecreatedfortheentirepopulationand forsubpopulationsstratifiedbyISS level, ISSbodyregion(Head/Neck,Face,Chest,Abdomen/Pelviccontents,Extremities/Pelvicgirdle,External), andmaximumAbbreviatedInjuryScale (maxAIS)scoreineachregion.Piecewiselinearregression identifiedVLADgraphbreakpoints.

Results:12,191consecutivetraumapatients(medianage35years,72%males,91%bluntinjury,41%

ISS!16) formed thedataset. TheirVLADgraph indicated performance equaltoTRISSpredicted survival until a sudden improvement in late 2004. From then survival remained improved butunchangedthrough 2011.Totalnumberofexcesssurvivorswas141.Inspectionofsubgroup VLADgraphsshowedthattheincreasedsurvivalmainlyoccurredinpatientshavingatleastone Head/NeckAIS5injury.Theeffectwaspresentinboth isolatedand multitraumatisedmaxAIS5 Head/Necktrauma.Theremainingtraumapopulationshowedunchangedsurvival,superiortoTRISS predicted,throughoutthestudyperiod.

Importantgeneralandneurotrauma-targetedimprovementsinourtraumaservicecouldunderlie ourfindings:Aformalisedtraumaservice,damagecontrolresuscitationprotocols,structuredtraining, increased helicopter transfer capacity, consultant-based neurosurgical assessment, a doubling of emergencyneurosurgicalprocedures,andimprovedneurointensivecare.

Conclusions: StratifiedVLADenablescontinuous,high-resolutionsystemanalysis.Weencouragetrauma centrestoexploretheirdataandtomonitorfuturesystemchanges.

!2014TheAuthors.PublishedbyElsevierLtd.ThisisanopenaccessarticleundertheCCBY-NC-ND license(http://creativecommons.org/licenses/by-nc-nd/3.0/).

* Corresponding author at: Department of Anaesthesia and Intensive Care, DivisionofSurgery,AkershusUniversityHospital,1478Lørenskog,Norway.

Tel.:+4791502900.

E-mailaddresses:signe.sovik@medisin.uio.no(S.Søvik), noskaga@online.no(N.O.Skaga),uxroha@ous-hf.no(R.Hanoa), torsten.eken@medisin.uio.no(T.Eken).

ContentslistsavailableatScienceDirect

Injury

j ou rna l h ome p a ge : w ww . e l se v i e r. co m/ l oc a te / i n j ury

http://dx.doi.org/10.1016/j.injury.2014.05.038

0020–1383/!2014TheAuthors.PublishedbyElsevierLtd.ThisisanopenaccessarticleundertheCCBY-NC-NDlicense(http://creativecommons.org/licenses/by-nc-nd/

3.0/).

Such consequences are particularly undesirable as high-risk patients probably gain most from referral to the highest level ofcare.

Severalriskstratificationmodelsintraumaexist,incorporating e.g.anatomicalinjury,physiologicalderangement,mechanismof injury, age, gender, and pre-injury comorbidity to predict probabilityofsurvival.Byapplyingsucha modeltoallpatients inacohort,theactualnumberofsurvivorscanbecomparedtothe predictednumberofsurvivorscalculatedasthesumofallsurvival probabilities.Thus,atraumacentrewillbeabletofindandfollow discrepanciesbetweenitsownperformanceandastandard,evenif itsreferral patternsand injuryprofiles aredifferentfromother hospitals and change over time. A norm in this kind of benchmarkingistheWstatistic,definedasthedifferencebetween observed and predicted survival rates, i.e. the number of unexpectedsurvivors,per100patients[2].

Traumacareperformanceisoftenevaluatedyearbyyear.This approacheliminatesapparentperformancefluctuationsduetoe.g.

seasonal changes in injury patterns. However, year-by-year evaluationalsoprevents detectionofshort-lasting variationsin performance that may be of particular interest. Examples are recurring performance deteriorations caused by vacations or rotation of personnel groups, and random performance drops caused by coincidental pile-ups of sick leaves or by hospital reorganisation.

Alternative approachesbased on statistical process control techniques allow rapiddetection of events that affect patient survival. One such technique is the Variable Life-Adjusted Display(VLAD), a refinement of the cumulative sum method.

VLADadjustseachdeathorsurvivalbythepatient’sriskstatus and provides a graphical display of accumulated number of unexpected survivors over time, with a high time-resolution.

Thenumberofexcesssavedlivesper100patients,readfromthe y-axisoftheVLADgraph,isequaltotheWstatistic.VLADcharts arealsoknownascumulativerisk-adjustedmortalitycharts.

TheVLADtechniquewasdevelopedincardiacsurgeryandis muchusedtomonitorintensivecareunit(ICU)performance[3–5].

In the trauma literature, few publications using the VLAD technique exist. In a study of a seven-year trauma population (2002–2008)fromtheOsloUniversityHospitalTraumaRegistry (TR–OUH)theVLADtechniquedemonstratedasuddenincreasein survival among the most injured patients, starting from the beginningof2005[6,7].Theperformancechangecoincidedintime withanorganisationalchangeinthehospitaltraumacare.Inthe currentstudywe utilisedtheVLADtechniquetosystematically exploreasupersetofthatpopulation(2001–2011),withtheaimof finding the patient subgroups in which the survival changes occurred.Patientswerestratifiedbythebodyregionsandinjury severity codes that represent the anatomical injuries in the underlyingTraumaScore– InjurySeverity Score(TRISS)model usedforriskadjustment[8].

A number of qualitative and quantitative changes were implementedinourtraumaserviceduringthestudyperiod,e.g.

inthepre-hospitalservices,inthesystematiccareofalltrauma admissions, including a strengthened focus on neurotrauma patients,inthetreatmentofmassivebleeding,andinstructured trainingoftraumateammembers.Weexploreanddiscussclinical and organisational changes that may underlie the increased survivalobservedintheidentifiedpatientsubgroups.

Methods Population

Thiswasanobservational,retrospectivestudyofelevenyearsof anonymisedtraumaregistrydatafromasingleNorwegianLevelI

traumacentre.Thestudywasapprovedandtheneedforwritten informed consent was waived by the institutional Privacy Ombudsman for Research, on behalf of the Norwegian Data Protection Authority and the Regional Committee for Medical Research Ethics.We aimedtoadhere totheSTROBEGuidelines (http://www.strobe-statement.org)inourreporting.

OsloUniversityHospital,Ulleva˚l(OUH–U)isaLevelItrauma referral hospital currentlycovering a population of 2.8 million inhabitants.Thestudyevaluatedtraumacareperformanceinall patientsincludedintheOsloUniversityHospitalTraumaRegistry (TR–OUH)intheperiod2001through2011.Inclusioncriteriafor TR–OUHwere(a)alltraumapatients admittedthroughtrauma teamactivation,irrespectiveofInjurySeverityScore(ISS)[9],(b) patientswithpenetratinginjuriesproximaltotheelboworknee, (c)patientswithheadinjuryofAbbreviatedInjuryScale[10](AIS) severitycode!3,and(d)patientswithISS!10admittedtoOUH–

U directly or via a local hospital <24h after injury. Patients transferred !24h after injury and patients with an isolated fracture ofa singleextremitywereincludedonlyifthetrauma team was activated upon their arrival at OUH–U. All eligible patients transported totheOUH–U emergencyroom(ER)were included;patientsclassifiedasdeadonarrivalwerenotexcluded.

Variables

Outcome was survival30 daysafter injury, determined by hospital records andinformationfrom theNorwegian Popula- tion Registry. Some foreign citizens were discharged alive to their homecountry lessthan 30daysafter injury;these were codedassurvivors.Predictive,systemcharacteristic,andprocess mappingvariablesused(Table 1)weredefinedaccordingtothe revisedUtsteintemplateforuniformreportingofdatafollowing major trauma[11].

Probability of survival(Ps)was calculatedaccording tothe TRISSmodelusingtheNationalTraumaDataBank(NTDB)2005 coefficients [12]. W statistics for selected subgroups were calculatedfromthePsvalues;standarderroroftheWstatistic wasdefinedasW/Z[2].

VariablesusedintheTRISSmodelweremechanismofinjury (blunt orpenetrating), patientage,RevisedTraumaScore (RTS) [13], and ISS. The ISS was calculated from anatomical injury descriptors according to the Abbreviated Injury Scale 1990 RevisionUpdate98(AIS98)[10],bysummingthesquareofthe highestAISseverityscoresforthethreemostseriouslyinjuredISS bodyregions.NewInjurySeverityScore(NISS)wasalsocalculated fromAIS98,bysummingthesquareofthepatient’sthreehighest AISseverityscoresindependentofinjuredISSbodyregion[14].

RTS category scores (0–4) for Glasgow Coma Scale (GCS) score,systolicbloodpressure,andrespiratoryrateuponarrival intheER[11]wereusedtocalculatetheweightedRTSscorein the TRISSmodel. Forpatients with missing physiologicaldata from the ER (e.g. missing respiratory rate due to artificial ventilation),TR–OUHassignsthelastdocumentedphysiological measurementbeforeadmission,eitherbythegroundambulance or HelicopterEmergencyMedicalSystem(HEMS) personnelor fromthephysicianintheERofthereferring hospital.Incases where noactual values are documented, the trauma registrar assignsanRTScategoryscorejudgingfromclinicaldescriptions inpatientrecords,ifpossible.AnormalRTScategoryscore(4)is assigned ifnodocumentationexists,toavoidbiasingoutcome data inthehospital’sfavour.

TriageRevisedTraumaScore(T-RTS)(0–12)wascalculatedfor the pre-hospital scene,at any transferring hospital, and in the OUH–UERbysummingtherelevantRTSclinicalcategoryscores (0–4)[13]forGCS,systolicbloodpressure,andrespiratoryrate.

Pre-injurycomorbiditywasclassifiedaccordingtotheAmerican

SocietyofAnesthesiologistsPhysicalStatusClassificationSystem (ASA-PS)[11,15].

VLADgraphanalysis

From each consecutive patient’s contribution, VLAD graphs werecalculatedwithJMP10statisticalsoftware(www.jmp.com, SASInstituteInc.,Cary,NC,USA)usingtheCUSUMplatform.Every patient was assigned a value corresponding to gained or lost

fractionallife,bysubtractingthatpatient’scalculatedprobability of survival(Ps) fromthe actual outcome,where 1 represented survivaland0death.Thus,everysurvivalcontributedarewardof 1"Psandeverydeathapenaltyof"Ps.Startingfromzero,each patient’scontributionwasaddedtothesummedcontributionofall previouspatients,andtheresultingnumberwasplottedvs.patient number.Thegraphofthecumulativesumofpenaltiesandrewards shows the difference over time between expected and actual cumulativesurvival.Thisrepresentsthenumberofexcesssaved Table1

Valuesofdescriptivevariablesinpatientsubgroups.

Variable Period No

Head/Neckinjury ofAIS5

Present Head/Neckinjury ofAIS5

Isolated Head/Neckinjury ofAIS5

Multitraumatised Head/Neckinjury ofAIS5 Numberofadmittedpatientsinperiod

(yearlyaverage)

1 3075(769) 363(91) 174(44) 189(47)

2 7865(1123) 888(127) 453(65) 432(62)

Age,years 1 33(15–68) 45(15–80) 56(15–83)^## 34(14–74)

2 34(13–68) 48(18–79) 56(18–81)^## 38(17–71)

Malegender,n(%) 1 2203(72) 260(72) 123(71) 137(72)

2 5659(72) 655(74) 331(73) 324(75)

Pre-injuryASA-PS!3,n(%) 1 215(7.0) 73(20) 48(28)^## 25(14)

2 753(9.6)** 221(25) 173(38)^## 48(11)

DominantinjuryBlunt,n(%) 1 2773(90) 343(94) 159(91)^# 184(97)

2 7091(90) 844(95) 433(95) 411(95)

Fall,n(%) 1 734(24) 158(44) 108(62)^## 50(27)

2 2154(27)** 439(49) 308(68)^## 131(30)

Transportinjury,n(%) 1 1668(54) 152(42) 32(18)^## 120(63)

2 3607(46)** 291(33)** 54(12)^## 237(55)

Violence,n(%) 1 314(10) 10(2.7) 9(5.2)^## 1(0.5)

2 931(12)* 61(6.9)** 34(7.5)^## 27(6.2)

Pre-hospitalGCSsum 1 15(7–15) 4(3–14) 5(3–15)^# 3(3–9)

2 15(8–15)* 6(3–15)** 8(3–15)^## 5(3–15)

Pre-hospitalT-RTSsum 1 12(10–12) 8(4–12) 10(6–12)^## 8(2–10)

2 12(10–12) 9(6–12)** 11(7–12)^## 9(5–12)

Physician-mannedprimarytransport (groundorair),n(%)

1 732(24) 121(33) 34(20)^## 87(46)

2 1735(22)* 288(32) 77(17)^## 211(49)

Pre-hospitalintubation,n(%) 1 289(9.4) 124(34) 32(18)^## 92(49)

2 511(6.5)** 244(27)* 55(12)^## 189(44)

Viatransferringhospital,n(%) 1 797(26) 159(44) 99(57)^## 60(32)

2 2299(29)** 452(51)* 292(64)^## 160(37)

TransferhospitalERGCSsum 1 15(7–15) 7(3–15) 7(3–14) 5(3–15)

2 15(10–15) 9(3–15)** 10(3–15) 9(3–15)

TransferhospitalERT-RTSsum 1 12(10–12) 10(8–12) 10(8–12) 10(8–12)

2 12(10–12) 11(8–12)** 11(8–12) 11(8–12)

Intubatedattransferringhospital,n(%)^! 1 165(5.4) 97(27) 57(59)^## 40(67)

2 338(4.3)** 250(28) 141(50)^## 109(76)

Physician-mannedsecondarytransport (groundorair),n(%)

1 229(29) 85(54) 44(45)^## 41(69)

2 650(29) 240(54) 125(43)^## 115(72)

IntubatedintraumacentreER,n(%)^! 1 368(12) 76(21) 35(40)^## 41(67)

2 789(10)** 168(19)* 90(35)^## 78(55)

TraumacentreERGCSsum 1 15(8–15) 6(3–15) 7(3–15)^## 4(3–14)

2 15(10–15)** 7(3–15)** 8(3–15)^## 6(3–15)

TraumacentreERT-RTSsum 1 12(10–12) 9(6–12) 10(7–12)^## 8(4–12)

2 12(10–12)** 10(7–12)** 10(8–12)^## 9(6–12)

ICUadmission,n(%) 1 2534(82) 338(93) 163(94) 175(93)

2 6657(85)** 839(94) 428(94) 411(95)

Ventilatortreatment,n(%) 1 690(22) 289(80) 123(76)^## 166(95)

2 1422(18)** 668(75)* 306(71)^## 362(88)

Ventilatordays 1 3(1–15) 4(1–16) 3(1–13)^# 5(1–18)

2 3(1–17) 6(1–25)** 5(1–19)^## 9(1–27)

Traumacentredays 1 4(1–13) 5(1–19) 5(2–16) 6(1–20)

2 3(1–13)** 8(2–28)** 6(2–22)^## 10(2–34)

Dischargedonaventilator,n(%oftotal) (%ofintubatedpts.)

1 299(9.7)(43) 196(54)(67) 65(37)^##(52)^## 131(69)(78)

2 472(6.0)**(33)** 358(40)**(53)** 166(36)^#(54) 192(44)(53)

Survivalat30days,n(%) 1 2936(96) 185(51) 96(55) 89(47)

2 7607(97)** 628(71)** 320(70) 308(71)

ISS 1 10(1–29) 30(25–50) 26(25–27)^## 38(30–57)

2 10(1–26)** 27(25–50) 26(25–27)^## 38(30–54)

NISS 1 13(1–34) 50(26–66) 43(25–66)^## 57(43–75)

2 12(1–34)** 50(30–66) 50(26–75)^## 57(43–75)

Wstatisticwith95%CI 1 1.30(0.73to1.88) "9.44("13.23to"5.64) "17.54("22.80to"12.27) "1.98("7.43to3.47) 2 1.35(1.00to1.70) 3.33(0.94to5.72) "7.03("10.27to"3.78) 14.20(10.68to17.72) Period1=2001–2004,Period2=2005–2011.Valuesaremedian(10–90percentile)ifnototherwisestated.^!:Percentofavailablepatients,i.e.thosenotarrivingintubated.Chi SquaredorMann–WhitneyUtest:StatisticallysignificantdifferencebetweenPeriod1andPeriod2:*p<0.05,**p<0.01.Statisticallysignificantdifferencebetweenisolated Head/NeckmaxAIS5patientsandmultitraumatisedHead/NeckmaxAIS5patientswithinsamePeriod:^#p<0.05,^##p<0.01.

livescomparedtothereferencemodelsincethefirstpatientwas admitted.

QualitativeinterpretationofVLADgraphs

When the cumulative sum is plotted against consecutive patientnumberthegraphwillbeunaffectedbychangesinnumber ofadmissionspertimeunit.Stableperformanceproducesalinear VLAD graph. A linear horizontal VLAD graph denotes stable performanceidenticaltothat ofthechosen referencemodel.A pointofdownwarddeflectionindicatesadeclineinperformance.

Anupward deflectionofthegraphsuggestsanimprovementof performanceatthistimepoint.IftheVLADgraphsubsequently riseslinearly,performanceisstablebutbetterthanthereference model. Continuous, ongoing improvement in performance will createanupward-curvinggraph.WequalitativelyevaluatedVLAD graphsfor different subpopulations withrespect tochanges in shape.

Quantitativeanalysis

QuantitativeanalysisoftheVLADgraphshapewasperformed using piecewise linear regression in the JMP 10 Nonlinear platform.This method could formallyidentifya breakpointin time,orrathertheconsecutivepatientnumberafterwhichthere wasanabruptchangeintheVLADgraphslope.Aformulawas specifiedthatrepresentedaninitiallinewithslopeb1andanother line with slope b2 added to it from a breakpoint C, i.e.

Y=b0+b1#X+b2#(X"C),withXbeingpatientnumberand (X"C)settozeroforallXvaluessmallerthanC.Thetwoslopes and the breakpoint C were calculated with 95% confidence intervals.

VLADgraphswerecreated fortheentire populationand for subpopulationsstratifiedby(a)injuryseverity(ISS1–15,ISS16–

24, ISS 25–75), (b) ISS body region (Head/Neck, Face, Chest, Abdomen/Pelviccontents,Extremities/Pelvicgirdle,andExternal), and(c)maximumAISscore(maxAIS;1–6)withineachISSbody region.

ThegroupofpatientshavingatleastoneAISscoreof5inthe Head/Neck region was studied more in depth. Separate VLAD graphswerecreatedforthosewithanisolatedcriticalheadorneck injury(ISS25–27,i.e.allowingAISscore1injuriesinuptotwo otherbodyregions)andformultitraumatisedpatientswithcritical headorneckinjury(ISS!28),sincethesetwosubgroupsdiffer substantiallywithregardtoepidemiological,injuryandtreatment factors.

Subgroupanalysis

Patientsubpopulations werecompared (Table 1)by demo- graphicdescriptors,injurymechanism,physiologicalresponseto injury, provided pre- and in-hospital services and treatment, anatomicinjury,30-daysurvival,andWstatistic.Distributionof descriptivevariablesaregivenasmedianswith10–90percentiles ifnototherwisestated.GroupswerecomparedwiththeWilcoxon RankSumtest forcontinuousdataandtheChiSquaredtestfor categorical data. Time trends in continuous variables were explored with linear regression or non-parametric Cochran–

Armitage test for trend as appropriate (JMP 10). A statistical significancelevelof0.05wasused.

ForpatientshavingatleastoneAISscoreof5intheISSHead/

Neckregion,survivalchangeswerestudiedforbroaddiagnostic groups,definedbycombiningAIScodesforanatomicallysimilar cerebral injuries. Rare diagnoses were omitted. Diagnostic categories were (a) brain swelling/oedema not including peri- lesionaloedema(AIS14066*),(b)contusion(AIS14060*,14061*,

14062* excluding 140628), (c) intracerebral haematoma (AIS 140638, 14064*, 140678), (d) subarachnoid haemorrhage (AIS 140684), (e) subdural haematoma (AIS 14065*), (f) epidural haematoma(AIS14063*excluding140638),and(g)diffuseaxonal injury (AIS 140628). These categories comprised AIS codes of differentseverities,andapatientwithseveralanatomicalinjuries would be represented in several categories. Change in yearly survivalratewithineachdiagnosticcategorywasanalysedwith theCochran–Armitagetestfortrend.

Results

Atotalof12,191patientsmettheinclusioncriteria;12,180had sufficient data tocompute Psand thus contributetotheVLAD graph.Medianagewas35years(14–70),72%weremales,10.4%

hadpre-injury ASA-PS!3,and91%sufferedbluntinjuries.Pre- hospital triage decided whether patients were transported to OUH–U, tolowerlevel traumahospitals, orto localemergency medical centres. Consequently, 41% of the study group had ISS!16. Median time from injury to arrival at OUH–U was 0.75h(0.3–1.75h)forpatientsarrivingdirectlyand4h(1.5–12h) for transferpatients. Over thestudyperiod therewas a steady increase in the number of admitted trauma patients per year (approx.68morepatients/year;linearregressionp<0.001).The fractionofpatientswithISS!16 decreasedslightlybyapproxi- mately0.5%peryear(p<0.01,Cochran–Armitagetestfortrend).

ThetotalnumberofAIScodeswas27,951.

VLADgraphanalysisbyISSlevel

The VLAD graph of the entire TR–OUH trauma population 2001–2011is showninFig. 1A. Duringthefirstfouryearsthe institutional performance was quite similar to what would be expected fromthe underlyingTRISS survivalprediction model, withneitheranaccumulationofexcesssurvivorsnordeaths.From late 2004theVLADgraph suddenlyrises and continuesalmost linearly through 2011. The total number of excess survivors amountedto141.

Fig. 1B shows thatalthough survival also appeared to be increasedin theotherISS groups,themajor changein perfor- manceseemedtooccuramongthemostseverelyinjuredpatients (ISS 25–75). This group’s VLAD graph can be described as consisting of twolinearsegments withdifferent slopes.Using piecewise linear regression, a breakpoint was identified at consecutive patient number 3371 (95% CI 3325–3417), corre- spondingtobeingadmittedNov28,2004(95%CINov10–Dec21).

ForpracticalreasonsJanuary1,2005waschosenasthecutofftime forbefore-and-afterperformanceanalyses.

VLADgraphanalysisbyISSbodyregion

VLADgraphswerecreatedforeachofthesixISSbodyregions, furtherstratifyingpatientsbytheirmaximumAISseverityscore (maxAIS)inthatregion(Fig. 2,PanelA–F).Apatientwithinjuries toseveralISSbodyregionswouldappearinVLADgraphpanelsfor all those regions, but only in a single maxAIS group within a specificbodyregion.

Mostcombinationsofbodyregionandinjuryseverity(Fig. 2) had almost linear VLAD graphs, indicating unchanged overall performance.TwoVLADgraphshowever,weredistinctlydifferent fromallothers.ThegraphforExternalregion,maxAIS1(Fig. 2F) showedaninflectionpointverysimilartothegraphforthetotal traumapopulation(Fig. 1A).As8947(73%)ofthepatientsinour populationwerecodedwithatleastoneminorinjurytothebody surface,theExternalregionmaxAIS1graphprobablyreflectedthe shapeoftheVLADgraphforthetotalpopulation,ratherthanbeing

causedbyimprovedperformanceinthetreatmentofminorskin injuries.

The graph forISSHead/Neckregion patients withmaxAIS 5 (1248patients)hadthemostdivergentshape(Fig. 2A).Thisgraph wasU-shaped,firstpointingdownwardindicating lowerperfor- mancethan expected, then takingan almost horizontal course indicatingperformancesimilartothereferencemodel,andlastly climbingupwardsindicatingbetterperformancethanexpected.

VLADgraphanalysisofsubpopulationsshowingchangesinsurvival

SincemultitraumatisedpatientswouldappearinVLADgraphs forseveralISSbodyregionpanelsinFig. 2,fluctuationsinsome graphsmightprimarilyreflectchangesinperformanceforinjuries inotherregions.Therefore,mutuallyexclusivepopulationswere definedandtheirVLADgraphsinvestigated.

Thecontributiontoinstitutionalperformanceinpatientswith maxAIS 5 in the Head/Neck region (shown in Fig. 2A) was comparedwiththat ofallothertraumapatients (Fig. 3A).The lattergroup(n=10,932;146excesssurvivors) showeda linear, upward-slopingVLADgraph,indicatingbetterperformancethan predictedbytheTRISSmodelbutunchangedthroughouttheentire study period (R²=0.997). It thus appears that the change in treatmentperformanceinlate2004,reflectedintheVLADgraph for the total trauma population (Fig. 1A), was caused by the considerablechangesinsurvivalamongpatientswithHead/Neck regionmaxAIS5(n=1248;"5excesssurvivors).

InFig. 3BseparateVLADgraphsaredisplayedfor(a)isolated maxAIS 5 Head/Neck injuries (ISS 25–27, n=627; "62 excess survivors)and(b)multitraumatisedmaxAIS5Head/Neckinjuries (ISS!28, n=621; 58 excesssurvivors). Improved survivalwas demonstratedinbothsubgroups.

Clinicaldescriptionofsubpopulations

ThesubpopulationswhoseVLADgraphsareshowninFig. 3A andBdifferedsubstantiallywithrespecttodemographicfactors, injurymechanism,physiologicresponsetoinjury,providedpre- and in-hospital treatment, injury severity, and 30-day survival (Table1).NoticeablechangesoccurredfromPeriod1(2001–2004), theyearsbeforetheinflectionpointintheVLADgraph,toPeriod2 (2005–2011)(Table 1).Whilethedistributionofageandgender wasunchanged,theproportionofpatients withsignificantpre- injury medical conditions increased, and more injuries were causedbyfallsorviolence.ISSandNISSdecreasedslightlyinthe populationwithoutcriticalheadorneckinjury,andGCSandT-RTS scoresweregenerallyhigher.

TimefromsustainedinjurytoarrivalatOUH–Uwasunchanged fromPeriod1 toPeriod2for patientsarrivingdirectly(median 44minvs.43min;p=0.92) butdeclinedmarkedly forpatients transferredfromotherhospitals.Transferpatientswithisolated maxAIS5Head/Neckinjuryseemedtohavethelargestreduction in mediantime frominjurytoarrival atOUH–U (reduction2h 11min;p<0.001).ThereductionformultitraumatisedmaxAIS5 Head/Neckinjury patients was1h 16min (p<0.02), for other headinjuredpatients1h5min(p<0.001),andfortheremaining population1h5min(p<0.01).Thisdevelopmentparalleledan almost linear increase in helicopter-based physician-manned transportsinourregion duringthestudyperiod(1067missions in2001,increasingbyapproximately137missions/year;R²=0.86, p<0.001).

AtOUH–U,traumateamassessmentofpatientsbecameswifter.

FromPeriod1toPeriod2,timefromadmissiontofirstCTscanwas reducedfrom39to29min(p<0.01forallsubgroups)andtimeto acutesurgerywasreducedfrom1h45minto1h25min(p<0.02 for all subgroups). For patients without maxAIS 5 Head/Neck injuries, the fractiontaken directly to surgery was unchanged (10.3% vs. 9.5%, p=0.16). For maxAIS 5 Head/Neck patients, however,therewasamarkedincreaseinimmediateneurosurgical proceduresfromyear2005,measuredasadirecttransferfromthe ER or CT lab to the OR (Fig. 4). Their duration of ventilator treatment and length of trauma centre stay also increased substantially(Table 1).

AmongpatientswithatleastonemaxAIS5injuryintheHead/

Neck ISS region, the mean number of Head/Neck AIS codes registeredperpatientincreasedslightlyfrom1.1to1.5duringthe studyperiod(p=0.02;R²=0.46).Theyearlyfractionofsurvivors increasedinthediagnosticgroupsbrainoedema,braincontusion, subarachnoid haemorrhage, subdural haematoma, and diffuse axonalinjury(allp<0.001;Cochran–Armitagetestfortrend).No significantchangesinsurvivalwereseenforpatientswithepidural (p=0.17)orintracerebral(p=0.13)haematomas.

Fig.1.VLADgraphsdepictingcumulativesumofexcesssurvivorsasafunctionof consecutivepatientnumberfortheyears2001–2011.Dashedverticallinesseparate calendaryears;increaseddistancebetweenlinesareduetoincreasednumberof admittedpatientsperyear.LinearVLADgraphsdenotestableperformance.Dashed horizontallinesrepresentperformanceaccordingtotheunderlyingTRISSsurvival predictionmodel,i.e.noexcesssurvivors.SolidverticallineinallVLADgraphs representsJanuary1,2005,separatingPeriod1fromPeriod2.(A)Totalpopulation, 12,180patients.(B) Totalpopulation stratified byInjurySeverity Score(ISS), confirmingthemajorperformanceimprovementintheISS25–75group.Solidlines superimposedon the ISS 25–75 group showthe results of piecewise linear regressionoftheVLADgraph,highlightingboththeincreaseinexcesssurvivorsand thetimepointatwhichthesuddenchangeoccurred.

WstatisticsincreasedfromPeriod1 toPeriod2both forthe totalgroupofpatientswithmaxAIS5Head/Neckinjuryandforthe two patient subgroups (isolated or multitraumatised injury) (Table 1). The remaining trauma population had identical W statisticsin thetwo time periods, indicating stable,unchanged performance.

Discussion Mainfindings

UsingtheVLADtechniqueonsurvivalprobabilitypredictedby theTRISSmodel,weconfirmedandextendedourpreviousfindings that 30-day survivalfor theentire TR–OUH traumapopulation startedtoincreasefromtheendof2004andremainedsteadily

improvedthroughoutthestudyperiod[6,7].Themainimprove- ment in survival was seen among the most severely injured patients(ISS!25),andpatientswithmaxAIS5injuriesintheISS Head/Neckregionseemedtorepresentthesubpopulationinwhich thechangesinsurvivaltookplace.ComparedtoTRISSpredicted survival, maxAIS 5 Head/Neck patients at OUH–U initially had pooreroutcome,from2005hadsimilaroutcome,andfrom2008 had betteroutcome than predicted. All otherISS injury region groupsshowedbettersurvivalthanpredictedbyTRISS,buttoan unchangeddegreethroughoutthestudyperiod.

Limitationsofthestudy

Thiswasanexploratory,retrospectivestudy,andourfindings mustthereforebeviewedashypothesisgenerating.Weusedthe Fig.2.VLADgraphsforallpatientswithoneormoreinjuriesineachspecifiedISSbodyregion(A–F),stratifiedbymaximumAIS(maxAIS)severityscoreinthatregion.Asingle patientmayappearinseveralISSbodyregionsduetomultipletrauma,butcanonlybeassignedtoonemaxAISgraphwithineachISSregion.ThegraphforHead/Neckregion patientswithmaxAIS5(1248patients)hadthemostdivergentshape(A).TheExternalregionmaxAIS1graph(F)representsconcomitantminorskininjuriesinthetotal traumapopulation(cf.Fig. 1A).

VLADtechniqueonPsvaluescalculated fromtheTRISS survival predictionmodel,stratifyingourtraumapopulationbyinjuredISS body region and further by maximum degree of injury in that region(AIS1–6).TheTRISSmodelisderivedfromalarge,general trauma population and is not validated for subpopulations.

Consequently, findings in stratified analyses should only be comparedwithinstratafromthesameinstitutionovertime.

Alimitationofourmaterialisthatweonlyhave30-daysurvival asoutcomemeasure.Theproportionoftraumadeathsoccurring laterthanthistimepointmayvarybetweenpatientgroups[16].It wouldbeofgreatinteresttorepeatouranalyseson6-and 12- monthsurvivaldataandtoseethemethodusedondatasetsfrom other trauma registries. Moreover, there is a strong need to investigateandmonitornon-fataloutcomesaftertrauma[17,18].

TheTRISSmodelcoefficientsareregularlyupdated,buttheset of explanatory variables in the model and their mathematical representations(e.g.categoricalratherthancontinuous,choiceof cutoffpoints)remainfixed.Thismayintroducebiasifthemodelis usedinpatientpopulationsthatdiffermarkedlyfromtheTRISS derivationpopulation.Institutionsmaythereforeshowdifferent survivalrates thanpredictedfromTRISS duetotheir qualityof

care,butalsoduetodifferencesincasemixifthesearenotwell controlledforbythemodel.Ourstratifiedanalysesthereforeonly focused on whether there were changes in slope of the VLAD graphs.Moreover, implicationsof possiblechanges in casemix duringthestudyperiodweremainfocuspointsinouranalysis.

Thesubpopulationdisplayingimprovedsurvivalinourstudy wasdefinedbyhavingatleastonecriticalheadorneckinjury,AIS severityscore 5. Still, compared to Period1, in Period 2 these patientshadhigherGCSscoresandT-RTSvaluesandwerelessoften artificiallyventilatedbothprehospitallyandatthetraumacentre.

Thus,wecannotruleoutthe possibilitythata shiftinadmitted injuriesdidoccur,andthattheAISsystemdoesnotfullyreflect injuryseverity.AllpatientswerecodedaccordingtothesameAIS catalogue and by the same certified coders, and though there seemedtobeslowshiftsintherelativefrequencyofheadandneck diagnosesused(datanotshown),survivalamongmaxAIS5Head/

Neckpatientsimprovedinasimilarmanneracrossbroaddiagnostic groups.Moreover,theGCSandRTSscoresareconstituentsofthe TRISSmodelandadjustedforwhenPsiscalculated.

Whyimprovedsurvivalinonetraumasubpopulation?

Thesuddenimprovementinsurvivalinourtraumapopulation appeared to result predominantly from improved survival in patients withverysevere headorneck injury,both as isolated injuries and in multitraumatisedpatients. Thesefindings could result from specific changes in the treatment of patients with critical neurotrauma. Additionally, patients withcritical neuro- trauma could constitute a particularly vulnerable group, an

‘‘indicatorpopulation’’thattothehighestdegreebenefittedfrom generalimprovementsinthetraumatreatmentchain,fromthesite ofinjury,duringtransport,intheER,OR,ICUandhospitalward.

Referralpatternandreferraltimes

Thefractionofadmittedtraumapatientsthathadatleastone headorneckAIS5injurywasunchangedthroughthestudyperiod.

Fig.3.VLAD graphsfor mutuallyexclusive subpopulations. (A)Patientswith maxAIS5intheHead/NeckISSbodyregionversustheremainingstudypopulation.

NotetheU-shapedgraphforpatientswithcriticalheadorneckinjuriesindicating largeperformance changes, in contrast to the linear VLAD graph indicating unchangedoverallperformancefortheremaininggroup.(B)PatientswithHead/

NeckmaxAIS5,splitintotwosubgroups:Isolatedinjurypatients(ISS25–27)versus multitraumatised patients (ISS!28). Regression lines with 95% prediction intervalsforeachsubpopulationinPeriod1areextrapolatedthroughPeriod2.

Bracketsdenoteincreaseinexcesssurvivorsfromtheextrapolatedregressionlines.

Fig.4.IncreaseinfractionoftraumapatientsarrivingdirectlytoOUH–Uwho underwentemergencysurgery,definedasbeingtakendirectlyfromtheERorCTlab totheOR.Circles:PatientswithisolatedmaxAIS5Head/Neckinjury.Squares:

MultitraumatisedpatientswithmaxAIS5Head/Neckinjury.Triangles:Remaining traumapopulation.Continuouslines/Filledsymbols:Neurosurgery.Dottedlines/

Opensymbols:Orthopaedicorgeneralsurgery.Alargeincreaseinemergency neurosurgicalproceduresinpatientswithcriticalneurotraumaoccurredin2005.

Nochangesoccurredinothersubgroups.

However,anoverallhigherproportionofpatientswastransferred fromlower-levelhospitals,andamongtransferpatients,timefrom sustainedinjurytoarrivalatOUH–Udeclinedmostmarkedlyfor those with isolated critical neurotrauma. This pattern of more frequentandswiftertransfersmayhaveresultedfromchangesin traumacareatlocalhospitals,intransferservices,andatOUH–U.

Lower-levelhospitalsprobablybenefittedfromtheincreased formalisationoftraumacaretakingplaceatOUH–U,regionally, andnationally.Improvedawarenessofandcompetencetowards trauma patients would be expected as the crew resource management course BEST (Better & Systematic Trauma Care, www.bestnet.no)wasintroducedatlower-levelhospitalsinour regionfromyear2000through2005.ATLS(AdvancedTraumaLife Support)courseswereintroducedin 2004and DSTC(Definitive Surgical Trauma Care) courses in 2006. Concurrently, regular regional trauma meetings were organised, aiming at closer collaboration between OUH–U, the pre-hospital services, and referringhospitals.

In2006,OUH–Uformalisedthatallcontactconcerningpossible referraloftraumapatientsfromlocalhospitalsshouldbewiththe trauma team leader, not the perceived ‘‘organ surgeon’’ (e.g.

neurosurgeon ororthopaedic surgeon). Thisprobably increased uniformityofassessmentofrelevantpatientfactorsandensured thatalsotransferpatientswerereceivedbytheOUH–U trauma teamuponarrival.

Informalandgradual referraland admittancepolicychanges couldhaveoccurredwithincreasedawarenessofthepotentialin interventionalradiology,surgery,andintensivecare.Sucheffects might have been most pronounced for elderly patients with significantcomorbidity.Among admittedpatients withisolated criticalheadorneckinjuryinPeriod2,agewasunchangedbutthe proportionwithcomorbiditieswasincreased.Interestingly,this shouldhavebeenexpectedtodecreasesurvival.TheTRISSmodel does not adjust for comorbidities, thus the observed upward changeintheVLADgraphsmightbeanunderestimationoftrue improvedperformance.ASA-PShaspreviouslybeenshowntobe anindependentnegativepredictorofsurvivalaftertrauma[15,19].

Pre-andinter-hospitaltransferservices

Theavailabilityofspecialisedphysician-mannedpre-andinter- hospitaltransferinourregionincreasedmarkedlyfrom2005when thenumberofhelicoptersinservicewasincreased.Theimproved capacity couldhave contributedto themarkeddecline in time frominjurytoarrivalatOUH–Uamongtransferpatients.Reduced delaytoneurosurgicalassessment,withmaintainedhighlevelof careduringtransport,couldhavecontributedtoimprovedsurvival in vulnerable patients. Even so, approximately 30% of multi- traumatised and 55% of isolated maxAIS 5 Head/Neck injured patients weretransferredfromlower-level hospitalstoOUH–U attendedbyparamedics.

Traumateamfactors

Internalauditsandformalresearchontraumateamactivation andsurgicalcompetence[20,21]ledtostepwiseupgradingofthe OUH–Utraumasystemfromaroundyear2000.Requirementsfor surgicaleducationandexperiencefortraumateamleaderswere refinedin2002[7],adedicatedtraumaservicewasformalisedin 2005,andcriteria for traumateamusewererevised. ATLSand DSTCcourseswererequiredforboththetraumateamleaderand theconsultantanaesthesiologist in thetraumateam. A trauma coursefornurseanaesthetists, ERnurses,ORnursesand radio- grapherswasintroduced.

Amarkedchangeinpracticeforfluidresuscitationandblood productusestartedaround2004withincreasingawarenessofthe

importance of preventing traumatic coagulopathy. Procedures wereimplementedforbetterpatienttemperaturecontrol,lessuse ofcrystalloids,balancedtransfusionwithredbloodcells,plasma and plateletsin massivebleeding, damagecontrolresuscitation targeted to normalise lactate and Base Excess, damage control surgery, and angio-embolisation. Hypertonic saline was intro- ducedasaresuscitationfluidinheadtraumapatientsandasrescue against high intracranial pressure. The practice changes may especiallyhavebenefittedpatientswithcriticalneurotrauma,who are exceedinglyvulnerabletopoor oxygen transportas wellas oedema. Effects on survival would be harder to detect in our remainingtraumapopulation,wheremortalitywasonly3.6%and deathsdirectlyduetoexsanguinationwereinfrequentgivenour settingwith91%blunttrauma.

Neurosurgicalpresenceandcompetence

Planningofimprovementsofneurotraumatologicalservicesat OUH–Uwasinitiatedin2001withanewheadoftheDepartment ofNeurosurgery.Theoverallresultwasanincreasedneurosurgical involvement in trauma patients. Active recruitment of skilled neurosurgeonsstarted in2003.From2005,rotaswerechanged fromatwo-leveltoathree-levelschedule,addinganexperienced specialtyregistraroryoungconsultantavailableatalltimestothe in-hospitaljuniordoctorandtheon-callconsultantneurosurgeon.

Thisensuredmuchmorecompetentneurosurgicalassessmentin theERandcontinuityofcompetenceintheICUduringthenight.A moreambitioustreatmentpracticeseemstohavefollowed.Fig. 4 showsthatamongdirectlyarriving(i.e.unselected)patientswith AIS 5 neurotrauma, the fraction who underwent emergency neurosurgicalinterventionsincludinginvasivemonitoringalmost doubledfrom2004to2005.

Upgraded neuroradiological services supported swifter and betterdiagnostics, and neuroradiologicaltechnology developed.

Interventional neuroradiologywasrestartedin 2002anddevel- opedtoafullyequippedinterventionalunit.MRIserviceimproved yearbyyear,achieving24/7accessibilityandcompetencelatein Period2.

Neurointensivecaresystems

From 2004 two senior neurosurgeons and from 2005 two seniorintensivistswerespecificallydedicatedtoneurointensive care, increasingconsistencyandcontinuity oftreatment.Mea- suresofearlyrehabilitationofneurotraumapatients startedin 2002andwereformalisedfortraumaticbraininjurypatientsin 2005. In2009 a multidisciplinarygroupof physiciansdecided upon a protocol onintensive care for serioustraumatic brain injury.

The treatment of subarachnoid haemorrhage (SAH) also developed through the years. Comprehensive guidelines for treatmentfromOUH–Uadmissionuntildischargewerereleased in 2006.Theseguidelineslistedindetail levelsofphysiological measurements that should trigger respiratory, circulatory, or surgicalinterventions,withtheirpathophysiologicalrationale.Itis reasonabletoassumethatimplementationofproceduresassigned to optimise brain oxygenation and perfusion in SAH patients throughout the neurocritical treatment chain may have had positivespillovereffectsalsoonneurotraumapatients,whowere treatedbythesamepersonnelinthesamehospitalareas.

Ventilatortreatmentwaslessfrequentinallpatientsubgroups in Period 2, but those who did receiveventilator treatment in Period2hadmoreventilatordays.Increaseduseofnon-invasive ventilatorysupportasanalternativetosedationandintubationisa possibleexplanation;unfortunatelywedonothavedataonthis.In contrasttoothertraumapatients,thosewithmaxAIS5Head/Neck

injuries had longer trauma centre stays in Period 2, possibly reflectinganoverallimprovedfocusonneurointensivecare.

Conclusions

InasurveyofallOUH–Utraumaregistrypatientsinaneleven- yearperiod,weusedtheVariable Life-AdjustedDisplay(VLAD) technique on Ps values calculated from the TRISS model, and confirmedasuddenimprovementin30-daysurvivalaroundthe endof2004[6,7].Stratifyingourmaterial,wefoundthatpatients withcritical neurotrauma (maxAIS 5 Head/Neck injury) repre- sented the patient group where survival improved most. Our remainingtraumapopulationshowedstablesurvival,superiorto TRISSpredicted.

Wediscussanumberofsystemchangestargetedtoimprove general and neurotrauma care that may underlie our findings.

Increased neurosurgical presence, competence, dedication and continuity from the ER through the ICU seems to be crucial.

Similarly important is a well-run general trauma system of surgeons,anaesthesiologists,intensivists,radiologists,nursesand paramedics.

Our study demonstrates a method for continuous system analysisthatcandetectsmallchangesinpatientoutcomewitha high time resolution, changes that may result from system interventions or incidental events. Availability of long-term survivaldatawouldincreasetheusefulnessofVLADfurther.We encourageothertraumacentrestoretrospectivelyexploretheir datawithstratifiedVLADanalysesandtoprospectivelymonitor futuresystemchanges.

Authors’contributions

NOSand TEdesigned and builttheOslo UniversityHospital TraumaRegistry.RHmonitoredanddocumentedchangesinthe traumaserviceingeneralandneurotraumaspecifically.SSandTE planned and designed the study, carried out the statistical analyses, drafted the manuscript and created the figures. All authorscriticallyevaluatedanddiscussedtheongoinganalyses, criticallyrevisedthemanuscript,andapprovedthefinalversion.

Conflictsofinterest

Theauthorsdeclarethattheyhavenoconflictsofinterest.

Acknowledgements

We thank trauma registrars Morten Hestnes and Hans Johansson for their continuing efforts at the Oslo University HospitalTraumaRegistry.Thestudywasfundedbytheauthors’

institutions and by the South-Eastern Norway Regional Health Authority.

References

[1]WuAW.Themeasureandmismeasureofhospitalquality:appropriaterisk- adjustment methods incomparing hospitals.AnnIntern Med 1995;122:

149–50.

[2]ChampionHR,CopesWS,SaccoWJ,LawnickMM,KeastSL,BainLW,etal.The MajorTraumaOutcomeStudy:establishingnationalnormsfortraumacare.J Trauma1990;30:1356–65.

[3]LovegroveJ,ValenciaO,TreasureT,Sherlaw-JohnsonC,GallivanS.Monitoring the results of cardiac surgery by variable life-adjusted display. Lancet 1997;350:1128–30.

[4]PagelC,UtleyM,CroweS,WitterT,AndersonD,SamsonR,etal.Realtime monitoringofrisk-adjustedpaediatriccardiacsurgeryoutcomesusingvari- able life-adjusted display: implementation in three UK centres. Heart 2013;99:1445–50.

[5]FlaattenH.Effectsofamajorstructuralchangetotheintensivecareunitonthe qualityandoutcomeafterintensivecare.QualSafHealthCare2005;14:270–2.

[6]EkenT,GrovenS,SkagaNO,NæssPA,GaarderC.Long-lastingperformance improvementafterinstitutionofadedicatedtraumaservice.In:American AssociationfortheSurgeryofTrauma–2010AnnualMeeting;2010.

[7]Groven S, EkenT, Skaga NO,Røise O,Næss PA, GaarderC. Long-lasting performanceimprovementafterformalizationofadedicatedtraumaservice.

JTrauma2011;70:569–74.

[8]BoydCR,TolsonMA,CopesWS.Evaluatingtraumacare:theTRISSmethod.

TraumaScoreandtheInjurySeverityScore.JTrauma1987;27:370–8.

[9]BakerSP,O’NeillB,HaddonW,LongWB.Theinjuryseverityscore:amethod fordescribingpatientswithmultipleinjuriesandevaluatingemergencycare.J Trauma1974;14:187–96.

[10]AssociationfortheAdvancementofAutomotiveMedicine(AAAM).TheAb- breviatedInjuryScale1990RevisionUpdate98;1998;p.1–79.

[11]RingdalKG,CoatsTJ,LeferingR,DiBartolomeoS,SteenPA,RøiseO,etal.The Utsteintemplateforuniformreportingofdatafollowingmajortrauma:ajoint revisionbySCANTEM,TARN,DGU-TRandRITG.ScandJTraumaResuscEmerg Med2008;16:7.

[12]AmericanCollegeofSurgeons.NationalTraumaDataBankreferencemanual:

background,caveatsandresources.JAmCollSurg2005;1–40.

[13]ChampionHR,SaccoWJ,CopesWS,GannDS,GennarelliTA,FlanaganME.A revisionoftheTraumaScore.JTrauma1989;29:623–9.

[14]OslerT,BakerSP,LongW.Amodificationoftheinjuryseverityscorethatboth improvesaccuracyandsimplifiesscoring.JTrauma1997;43:922–5.

[15]SkagaNO,EkenT,SøvikS,JonesJM,SteenPA.Pre-injuryASAphysicalstatus classificationisanindependentpredictorofmortalityaftertrauma.JTrauma 2007;63:972–8.

[16]MooreL,TurgeonAF,EmondM,LeSageN,LavoieA.Definitionofmortalityfor traumacenterperformanceevaluation:acomparativestudy.CritCareMed 2011;39:2246–52.

[17]ArdolinoA,SleatG,WillettK.Outcomemeasurementsinmajortrauma—

resultsofaconsensusmeeting.Injury2012;43:1662–6.

[18]MooreL, Stelfox HT, Boutin A,Turgeon AF.Trauma centerperformance indicatorsfornonfataloutcomes.JTraumaAcuteCareSurg2013;74:1331–43.

[19]JonesJM,SkagaNO,SøvikS,LossiusHM,EkenT.Norwegiansurvivalprediction modelintrauma:modellingeffectsofanatomicinjury,acutephysiology,age, andco-morbidity.ActaAnaesthesiolScand2014;58:303–15.

[20]LossiusHM,LanghelleA,Pillgram-LarsenJ,LossiusTA,SøreideE,LaakeP,etal.

EfficiencyofactivationofthetraumateaminaNorwegiantraumareferral centre.EurJSurg2000;166:760–4.

[21]GaarderC,SkagaNO,EkenT,Pillgram-LarsenJ,BuanesT,NæssPA.Theimpact ofpatientvolumeonsurgicaltraumatraininginaScandinaviantraumacentre.

Injury2005;36:1288–92.