corresponds4j4j.com

SOFTWAREOpenAccesswww.
common-metrics.
org:awebapplicationtoestimatescoresfromdifferentpatient-reportedoutcomemeasuresonacommonscaleH.
FelixFischer1,2*andMatthiasRose1,3AbstractBackground:Recently,agrowingnumberofItem-ResponseTheory(IRT)modelshasbeenpublished,whichallowestimationofacommonlatentvariablefromdataderivedbydifferentPatientReportedOutcomes(PROs).
WhenusingdatafromdifferentPROs,directestimationofthelatentvariablehassomeadvantagesovertheuseofsumscoreconversiontables.
ItrequiressubstantialproficiencyinthefieldofpsychometricstofitsuchmodelsusingcontemporaryIRTsoftware.
Wedevelopedawebapplication(http://www.
common-metrics.
org),whichallowsestimationoflatentvariablescoresmoreeasilyusingIRTmodelscalibratingdifferentmeasuresoninstrumentindependentscales.
Results:Currently,theapplicationallowsestimationusingsixdifferentIRTmodelsforDepression,Anxiety,andPhysicalFunction.
Basedonpublisheditemparameters,usersoftheapplicationcandirectlyestimatelatenttraitestimatesusingexpectedaposteriori(EAP)forsumscoresaswellasforspecificresponsepatterns,Bayesmodal(MAP),Weightedlikelihoodestimation(WLE)andMaximumlikelihood(ML)methodsandunderthreedifferentpriordistributions.
Theobtainedestimatescanbedownloadedandanalyzedusingstandardstatisticalsoftware.
Conclusions:ThisapplicationenhancestheusabilityofIRTmodelingforresearchersbyallowingcomparisonofthelatenttraitestimatesoverdifferentPROs,suchasthePatientHealthQuestionnaireDepression(PHQ-9)andAnxiety(GAD-7)scales,theCenterofEpidemiologicStudiesDepressionScale(CES-D),theBeckDepressionInventory(BDI),PROMISAnxietyandDepressionShortFormsandothers.
Advantagesofthisapproachincludecomparabilityofdataderivedwithdifferentmeasuresandtoleranceagainstmissingvalues.
Thevalidityoftheunderlyingmodelsneedstobeinvestigatedinthefuture.
Keywords:Item-ResponseTheory,Measurement,PatientReportedOutcomes,Depression,Anxiety,PhysicalfunctionBackgroundOneofthemajordevelopmentsintherecentyearsofPatient-ReportedOutcome(PRO)measurementhasbeentheadoptionofmethodsbasedonItem-ResponseTheory(IRT)[1].
Thosemethodshavebeenusedtodevelopshortermeasures[2],toapplycomputer-adaptivetests[3]ortoassesssystematicdifferencesinresponsebehaviorbetweengroups[4].
OneofthecoreadvantagesofIRTcomparedtoClassicalTestTheory(CTT)isthepossi-bilitytoestimatecommonmodelsfordifferentPROsmeasuringthesameconstructs,allowingcomparisonsofthemeasuredconstructoverdifferentmeasures[1].
WecallIRTmodelsthatcomprisetheitemparametersfromitemsofvariousmeasures,measuringacommonvariable,"commonmetrics".
Withsuchstatisticalmodels,onecanestimatethevariableofinterestbysubsetsofitems,e.
g.
whendifferentmeasuresareusedorwhendataismissing.
*Correspondence:felix.
fischer@charite.
de1DepartmentofPsychosomaticMedicine,ClinicforInternalMedicine,CharitéUniversittsmedizinBerlin,Berlin,Germany2InstituteforSocialMedicine,EpidemologyandHealthEconomics,CharitéUniversittsmedizinBerlin,Berlin,GermanyFulllistofauthorinformationisavailableattheendofthearticle2016TheAuthor(s).
OpenAccessThisarticleisdistributedunderthetermsoftheCreativeCommonsAttribution4.
0InternationalLicense(http://creativecommons.
org/licenses/by/4.
0/),whichpermitsunrestricteduse,distribution,andreproductioninanymedium,providedyougiveappropriatecredittotheoriginalauthor(s)andthesource,providealinktotheCreativeCommonslicense,andindicateifchangesweremade.
TheCreativeCommonsPublicDomainDedicationwaiver(http://creativecommons.
org/publicdomain/zero/1.
0/)appliestothedatamadeavailableinthisarticle,unlessotherwisestated.
FischerandRoseBMCMedicalResearchMethodology(2016)16:142DOI10.
1186/s12874-016-0241-0Intherecentyearssuchmodelshavebeendevelopedinvariousdomains:physicalfunctioning[5–7],pain[8,9],fatigue[10],headache[11],anxiety[12]anddepression[13–16].
Apromisingfieldofresearchisthelinkingofpediatricandadultmeasurestoallowmeaningfulcomparisonsoverthecourseoftime[17].
Dif-ferentmethodsyieldingcomparableresultshavebeenappliedtolinkmeasures,suchasfixed-parameterestima-tionorconcurrentestimationwithsubsequentlinking[12,13,18].
Sofar,thoseIRTmodelshavebeenfrequentlyusedtodevelopsumscoreconversiontablesbetweenmeasures[7,8,10,12,15]sinceitispossibletoderivelatenttraitestimatessolelyfromthesumscore[19].
Itisalsopossibletoestimatethelatenttraitdirectlyfromtheresponsepattern.
Thisapproachhassomeadvan-tagesovertheuseofsumscoreconversiontablessinceittakesintoaccountdifferencesintheresponsepat-tern,yieldingmoreaccurateresults[12,13]thancon-vertedsumscores.
Italsoisfavorableincaseofmissingitemresponse,sinceestimationofthelatentvariableisstillviableunderthatcondition[12,13].
EstimationofIRTscoresbasedoncommonmetricscancurrentlybedoneinanumberofdifferentstatisticalpackages,suchasIRTPRO,PARSCALE,RorSAS.
None-theless,itrequiressubstantialproficiencyinthefieldofpsychometricstofitthosemodels,hamperingaccessibilityofcommonmetricsforresearchersfromotherfields.
Wedevelopedawebapplication(http://www.
common-metrics.
org),whichallowsestimationoflatentvariablescoresmoreeasilyusingsuchcommonmetrics.
Ourgoalistoenableresearcherstocomparedataob-tainedwithdifferentmeasures,forexampleifinStudyAthePatientHealthQuestionnaire9(PHQ-9)hasbeenusedforthemeasurementofdepression,butinStudyBtheBeckDepressionInventory(BDI)wasthemeasureofchoice.
Inthispaper,wedescribethegen-eralorganizationoftheapplication,thetechnicalde-tailsoftheimplementedestimationaswellasaspectsofdatasafety.
Finally,advantagesandcaveatsoftheapplicationarediscussed.
ImplementationOverviewTheapplicationitselfconsistsofacontrolpaneland6tabs(seeFig.
1).
Metric:selectoneoftheavailablemetricsandreviewtheitemcodesforeachmeasure.
Currently,weimplementedcommonmetricsforthemeasurementofdepression[13,14],anxiety[12,20],andphysicalfunctioning[5,7]containingmeasuressuchasthePatientHealthQuestionnaireDepression(PHQ-9)andAnxiety(GAD-7)scales[21,22],theCenterofEpidemiologicStudiesDepressionScale(CES-D)[23],theBeckDepressionInventory(BDI)[24],PROMISAnxietyandDepressionShortForms[25–27]andothers.
Weprovidesomeinformationaboutthosemetrics,suchasestimationsamplesizeandincludeditems,butusersarereferredtotheactualpublications.
Additionalmetricscanbeaddedifrequested.
Data:selectexampledataoruploadyourowndataset.
Theidentificationofitemsinthedatasetiscase-sensitiveandcolumnnamesmustmatchtheitemcodesexactly.
Eachrowcorrespondstooneobservation.
Model:selectpriordistribution(N(0,1),N(0,10)andestimatedfromdata)andreviewitemparameters.
Estimates:selectestimationmethodEAP(expectedaposteriori),MAP(Bayesmodal),WLE(Weightedlikelihoodestimation),ML(Maximumlikelihood)orEAPSumScore)andreviewdescriptivestatistics(n,min,mean,median,maximum,standarddeviation,standarderrorofthemean,percentageofmissingvalues)includingahistogramofthedistributionoflatenttraitestimates.
Precision:reviewprecisionofestimates(standarderror)overlatentvariablecontinuum.
Ifestimationmethodismaximumlikelihood(ML),testprecisionoflegacyinstrumentscanbeshown.
Download:downloaddatasetwithscoreestimateandstandarderrorofmeasurement.
Thedefaultestimatorselection(EAPwithN(0,1)prior)canbeconsideredascurrentstandardandisappropriateforawiderangeofapplications.
However,weallowtheselectionofdifferentestimatorsandpriors,sincethosemightbemoreappropriateinagivensituation.
Forex-ample,comparisonoftheprecisionofasetofitemstolegacyinstrumentsisonlymeaningfulunderMLesti-mation.
Sincetheapplicationissolelyintendedtoallowresearcherstoestimatelatenttraitscoresonseveralpreviouslypublishedcommonmetrics,theapplicationdoesnotincludeanypossibilitytoreestimatetheunderlyingitemparameters.
TechnicaldetailsofthetaestimationTheapplicationsetsuptherespectiveIRTmodel(GradedResponseModelorGeneralizedPartialCreditModel)withallparametersfixedtotheitemparametersofthedesiredcommonmetric.
Priordistributioncanbeselectedbytheuser.
TheunderlyingRpackagemirt[28]usesamarginalmaximumlikelihoodmethodtoestimateitemparametersofIRTmodels,hence,estimationofpersonparameterscanbeconductedindependently.
Forpersonparameterestimationweincludedthesumscoreaswellasresponsepatternexpectedaposteriori(EAP),Bayesmodal(MAP),WeightedlikelihoodestimationFischerandRoseBMCMedicalResearchMethodology(2016)16:142Page2of5(WLE)andMaximumlikelihood(ML)methods.
Thetaes-timatesandstandarderrorsaretransformedtothet-metric(mean50,standarddeviationof10).
Forsomemet-rics,50issomemeaningfulanchorpointlikethegeneralpopulationmean[12–14].
Testspecificstandarderrorswerecalculatedformodelscomprisingallitemsfromonequestionnaire.
PleasenotethatthesestandarderrorsarevalidunderMLestimationonly.
ThewebsitewasbuildusingR3.
0.
2[29],Shiny[30]andggplot2[31].
IRTmodelsusedforthetaestimationwereestimatedusingtheR-packagemirt[28].
DatasafetyFromuploadeddata,allcolumnsaredisregardediftheirnamedoesnotmatchanyoftheitemcodesavailableintheselectedmetric.
AlthoughwedonotsaveuploadedFig.
1OverviewovertheapplicationworkflowFischerandRoseBMCMedicalResearchMethodology(2016)16:142Page3of5databeyondtheneedforprocessingwithintheactualsession,usersmustbeawarethatsensibledatasentthroughtheinternetisapotentialsecurityriskanddatamightbecomepublic.
Wehenceadviseusertouploadonlytherequiredamountofdata(inotherwords,onlytheitemresponses)andensurethatuploadeddatafulfillsdatasafetystandards.
Datashouldnotcontainanypersonalinformation,allowingtracingofsingleresponsestoindividuals.
TheapplicationwasapprovedinitscurrentversionbythedataprotectioncommissioneroftheCharitéUniver-sittsmedizinBerlin,Germany.
ResultsWepresentawebsitethatallowstheuseofcommonmetricstoestimatelatentvariableonacommonscaleindependentlyfromthemeasurebeingused.
ComparedtotraditionalIRTsoftwarethemajorstrengthofourapproachbyprovidingawebapplicationisthatthetaestimationfromdifferentPROsdoesnotrequiredetailedknowledgeonIRTmodelingnorestimationtechniques.
Weprovideasimpleinterfacetocheckbasicsummarydataanddatamaylaterbeusedinanyothersoftwaretheuserisfamiliarwith,suchasExcel,SPSS,SASorR.
Theapproachimplementedinwww.
common-metrics.
orgingeneralpromisesanumberofadvantagescomparedtotheuseofinstrumentdependentsumscores,suchas1.
comparabilityofdataderivedwithdifferentmeasures,e.
g.
whenassessingroutinedataorincaseofmeta-analysisonprimarydatalevel2.
moreprecisemeasurement(i.
e.
decreasedstandarderrorofindividualestimate)bytakingtheresponsepatternintoaccountaswellaswhenusingtwoormoremeasures3.
toleranceagainstmissingvalues4.
increasedvalidityofthescalecomparedtoinstrumentdependentscales.
However,usersshouldbeawareofthelimitationsofthisapproach.
Oneissueisthevalidityoftheunderlyingmodel.
Althoughfindingsliketheoverlapofdifferentcut-offvaluesfromstaticmeasuresonthecommonmetricmakeusconfidentinthevalidityofsomeofthemodels[12–14],agenerallackofexternalvalidationstudiesmustbeacknowledged.
However,providingatechnicalbasistousesuchmodelsinresearchmoreeasilymightbeacatalystforsuchvalidationstudies.
Furthermore,onemustbeawarethatmeasuresdifferintheircoverageoverthethetacontinuum.
WhileithasbeenshownthattheuseofIRTestimatesinsteadofsumscoresleadstosimilarresults[1,20],useofdifferentmeasuresinsteadofthesametoestimatethetashowedinonestudyanotableimpactontheeffectestimate[32].
Thiscanleadtoseverebiaswhencomparingscoresfromtestswithdifferingprecisionoverthecontinuum.
Sincemostinstrumentsweredevelopedinclinicalsam-plesthismightbeespeciallyproblematicinrelativelyhealthysamples,suchasthegeneralpopulation.
Apos-siblesolutionistotaketheuncertaintyaboutthethetaestimate–itsstandarderror–intoaccount,e.
g.
inaBayesianframeworkoradoptingtheplausiblevalueapproach[33–35].
Thisissuemustbeinvestigatedinthenearfuture.
Anotherthreadtovalidityisthepossibilityofdifferentialitemfunctioningbetweenthesampleswhichwereusedformodelcalibrationandthesamplesusedinapplication.
Forexample,itisunclearwhethercommonmetricdevel-opedfromGermansamples[14]canbeusedinEnglishspeakingsamplesaswell.
However,thisproblemisalsoapparentintheuseofsumscoreconversiontables.
ConclusionWefirmlybelievethatcommonmetricsincludingavar-ietyofmeasureshaveamuchstrongerchancetobe-comevalidandacceptedstandardsforaspecificdomainratherthanasinglequestionnaire.
Wehopethiswebsiteshowsthepotentialthatthedevelopmentofcommonmetricsholds,facilitatesstudiesinvestigatingthevalidityandclinicalusefulnessofsuchmetricsandcontributestothemovementtowardsinstrumentindependentscalesinmeasurementofPatient-ReportedOutcomes.
AvailabilityandrequirementOurwebapplicationisavailableathttp://www.
common-metrics.
orgwithinformationaboutthebackground,methods,andlimitationsofthisapproach.
Theapplica-tionmaybefreelyusedtoestimatethetascoresonacommonmetric.
AcknowledgementsWeacknowledgetheworkofallresearchersdevelopingcommonIRTmodelsforvariousoutcomes.
FundingNofundingwasreceivedforthepresentedwork.
AvailabilityofdataandmaterialsSourcecodeoftheapplicationcanberequestedfromFelixFischer.
Authors'contributionsFFandMRconceivedthedesignoftheapplication,FFprogrammedtheapplicationandwroteafirstdraftofthepublication.
Bothauthorsreadandapprovedthefinalmanuscript.
CompetinginterestsTheauthorsdeclarethattheyhavenocompetinginterests.
ConsentforpublicationNotapplicable.
EthicsapprovalandconsenttoparticipateNotapplicable.
FischerandRoseBMCMedicalResearchMethodology(2016)16:142Page4of5Authordetails1DepartmentofPsychosomaticMedicine,ClinicforInternalMedicine,CharitéUniversittsmedizinBerlin,Berlin,Germany.
2InstituteforSocialMedicine,EpidemologyandHealthEconomics,CharitéUniversittsmedizinBerlin,Berlin,Germany.
3DepartmentofQuantitativeHealthSciences,UniversityofMassachusettsMedicalSchool,Worcester,USA.
Received:23July2016Accepted:7October2016References1.
ReiseSP,WallerNG.
Itemresponsetheoryandclinicalmeasurement.
AnnuRevClinPsychol.
2009;5:27–48.
2.
TeresiJA,Ocepek-WeliksonK,KleinmanM,CookKF,CranePK,GibbonsLE,etal.
Evaluatingmeasurementequivalenceusingtheitemresponsetheorylog-likelihoodratio(IRTLR)methodtoassessdifferentialitemfunctioning(DIF):applications(withillustrations)tomeasuresofphysicalfunctioningabilityandgeneraldistress.
QualLifeRes.
2007;16Suppl1:43–68.
3.
ChoiSW,ReiseSP,PilkonisPA,HaysRD,CellaD.
Efficiencyofstaticandcomputeradaptiveshortformscomparedtofull-lengthmeasuresofdepressivesymptoms.
QualLifeRes.
2010;19:125–36.
4.
PazSH,SpritzerKL,MoralesLS,HaysRD.
EvaluationofthePatient-ReportedOutcomesInformationSystem(PROMIS())Spanish-languagephysicalfunctioningitems.
QualLifeRes.
2013;22:1819–30.
5.
McHorneyCA,CohenAS.
Equatinghealthstatusmeasureswithitemresponsetheory:illustrationswithfunctionalstatusitems.
MedCare.
2000;38:43–59.
6.
SchaletBD,RevickiDA,CookKF,KrishnanE,FriesJF,CellaD.
EstablishingaCommonMetricforPhysicalFunction:LinkingtheHAQ-DIandSF-36PFSubscaletoPROMISPhysicalFunction.
Med:J.
Gen.
Intern;2015.
7.
tenKloosterP,OudeVoshaarMAH,GandekB,RoseM,BjornerJB,TaalE,etal.
DevelopmentandevaluationofacrosswalkbetweentheSF-36physicalfunctioningscaleandHealthAssessmentQuestionnairedisabilityindexinrheumatoid.
HealthQualLifeOutcomes.
2013;11:199.
8.
ChenW-H,RevickiDA,LaiJ-S,CookKF,AmtmannD.
Linkingpainitemsfromtwostudiesontoacommonscaleusingitemresponsetheory.
JPainSymptomManageElsevierInc.
2009;38:615–28.
9.
CookKF,SchaletBD,KallenMa.
,RutsohnJP,CellaD.
Establishingacommonmetricforself-reportedpain:linkingBPIPainInterferenceandSF-36BodilyPainSubscalescorestothePROMISPainInterferencemetric.
QualLifeRes.
2015;24:2305–18.
10.
LaiJ-S,CellaD,YanezB,StoneA.
LinkingFatigueMeasuresonaCommonReportingMetric.
ElsevierLtd:J.
PainSymptomManage;2014.
11.
BjornerJB,KosinskiM,WareJE.
UsingitemresponsetheorytocalibratetheHeadacheImpactTest(HIT)tothemetricoftraditionalheadachescales.
QualLifeRes.
2003;12:981–1002.
12.
SchaletBD,CookKF,ChoiSW,CellaD.
Establishingacommonmetricforself-reportedanxiety:LinkingtheMASQ,PANAS,andGAD-7toPROMISAnxiety.
JAnxietyDisordElsevierLtd.
2014;28:88–96.
13.
ChoiSW,SchaletBD,CookKF,CellaD.
EstablishingaCommonMetricforDepressiveSymptoms:LinkingtheBDI-II,CES-D,andPHQ-9toPROMISDepression.
PsycholAssess.
2014;26:513–27.
14.
WahlI,LweB,BjornerJB,FischerHF,LangsG,VoderholzerU,etal.
Standardizationofdepressionmeasurement:acommonmetricwasdevelopedfor11self-reportdepressionmeasures.
JClinEpidemiol.
2014;67:73–86.
15.
FischerHF,TrittK,KlappBF,FliegeH.
Howtocomparescoresfromdifferentdepressionscales:equatingthePatientHealthQuestionnaire(PHQ)andtheICD-10-SymptomRating(ISR)usingItemResponse.
IntJMethodsPsychiatrRes.
2011;20:203–14.
16.
GibbonsLE,FeldmanBJ,CraneHM,MugaveroM,WilligJH,PatrickD,etal.
Migratingfromalegacyfixed-formatmeasuretoCATadministration:calibratingthePHQ-9tothePROMISdepressionmeasures.
QualLifeRes.
2011;20:1349–57.
17.
OlinoTM,YuL,McMakinDL,ForbesEE,SeeleyJR,LewinsohnPM,etal.
Comparisonsacrossdepressionassessmentinstrumentsinadolescenceandyoungadulthood:anitemresponsetheorystudyusingtwolinkingmethods.
JAbnormChildPsychol.
2013;41:1267–77.
18.
HaleySM,NiP,LaiJ-S,TianF,CosterWJ,JetteAM,etal.
Linkingtheactivitymeasureforpostacutecareandthequalityoflifeoutcomesinneurologicaldisorders.
ArchPhysMedRehabil.
2011;92:S37–43.
19.
ThissenD,PommerichM,BilleaudK,WilliamsVSL.
Itemresponsetheoryforscoresontestsincludingpolytomousitemswithorderedresponses.
ApplPsycholMeas.
1995;19:39–49.
20.
FischerHF,KlugC,RoeperK,BlozikE,EdelmannF,EiseleM,etal.
Screeningformentaldisordersinheartfailurepatientsusingcomputer-adaptivetests.
QualLifeRes.
2014;23:1609–18.
21.
SpitzerRL.
ValidationandUtilityofaSelf-reportVersionofPRIME-MD:ThePHQPrimaryCareStudy.
JAMA.
1999;282:1737–44.
22.
KroenkeK,SpitzerRL,WilliamsJBW,LweB.
ThePatientHealthQuestionnaireSomatic,Anxiety,andDepressiveSymptomScales:asystematicreview.
GenHospPsychiatryElsevierBV.
2010;32:345–59.
23.
RadloffLS.
TheCES-DScale:ASelf-ReportDepressionScaleforResearchintheGeneralPopulation.
ApplPsycholMeas.
1977;1:385–401.
24.
HautzingerM,BailerM,WorallH,KellerF.
BDIBeck-Depressions-InventarTesthandbuch.
2nded.
Bern:HansHuber;1995.
25.
PilkonisPA,ChoiSW,ReiseSP,StoverAM,RileyWT,CellaD.
ItembanksformeasuringemotionaldistressfromthePatient-ReportedOutcomesMeasurementInformationSystem(PROMIS):depression,anxiety,andanger.
Assessment.
2011;18:263–83.
26.
Patient-ReportedOutcomesMeasurementInformationSystem.
PROMISDepressionScoringManual[Internet].
2013[cited2016Mar18].
Availablefrom:https://www.
assessmentcenter.
net/documents/PROMIS%20Depression%20Scoring%20Manual.
pdf.
27.
Patient-ReportedOutcomesMeasurementInformationSystem.
PROMISAnxietyScoringManual[Internet].
2013[cited2016Mar19].
Availablefrom:https://www.
assessmentcenter.
net/documents/PROMIS%20Anxiety%20Scoring%20Manual.
pdf.
28.
ChalmersRP.
mirt:AMultidimensionalItemResponseTheoryPackagefortheREnvironment.
JStatSoftw.
2012;48:1–29.
29.
RDevelopmentCoreTeam.
R:Alanguageandenvironmentforstatisticalcomputing.
Vienna:RFoundationforStatisticalComputing;2008.
30.
RStudioInc.
shiny:WebApplicationFrameworkforR.
RpackageVersion0.
9.
1.
2014.
31.
WickhamH.
ggplot2.
NewYork:Springer;2009.
32.
FischerHF,WahlI,FliegeH,KlappBF,RoseM.
Impactofcross-calibrationmethodsontheinterpretationofatreatmentcomparisonstudyusing2depressionscales.
MedCare.
2012;50:320–6.
33.
GorterR,FoxJ-P,TwiskJ.
WhyItemResponseTheoryshouldbeusedforlongitudinalquestionnairedataanalysisinmedicalresearch.
BMCMedResMethodol.
2015;15:1–12.
34.
GorterR,FoxJ-P,ApeldoornA,TwiskJ.
Theinfluenceofmeasurementmodelchoiceforrandomizedcontrolledtrialresults.
ElsevierLtd:J.
Clin.
Epidemiol;2016.
35.
MarsmanM,MarisG,BechgerT,GlasC.
WhatcanwelearnfromPlausibleValuesPsychometrika.
SpringerUS;2016.
Weacceptpre-submissioninquiriesOurselectortoolhelpsyoutondthemostrelevantjournalWeprovideroundtheclockcustomersupportConvenientonlinesubmissionThoroughpeerreviewInclusioninPubMedandallmajorindexingservicesMaximumvisibilityforyourresearchSubmityourmanuscriptatwww.
biomedcentral.
com/submitSubmityournextmanuscripttoBioMedCentralandwewillhelpyouateverystep:FischerandRoseBMCMedicalResearchMethodology(2016)16:142Page5of5