Is the Hospital Anxiety and Depression Scale (HADS) a valid ...

The HADS (Hospital Anxiety and Depression Scale) aims to measure symptoms of anxiety (HADS Anxiety) and depression (HADS Depression). Skiptomaincontent Advertisement SearchallBMCarticles Search IstheHospitalAnxietyandDepressionScale(HADS)avalidmeasureinageneralpopulation65–80 yearsold?Apsychometricevaluationstudy DownloadPDF DownloadePub DownloadPDF DownloadePub Research OpenAccess Published:04October2017 IstheHospitalAnxietyandDepressionScale(HADS)avalidmeasureinageneralpopulation65–80 yearsold?Apsychometricevaluationstudy IngridDjukanovic  ORCID:orcid.org/0000-0001-6157-36441,JörgCarlsson1&KristoferÅrestedt1,2,3  HealthandQualityofLifeOutcomes volume 15,Article number: 193(2017) Citethisarticle 21kAccesses 111Citations 5Altmetric Metricsdetails AbstractBackgroundTheHADS(HospitalAnxietyandDepressionScale)aimstomeasuresymptomsofanxiety(HADSAnxiety)anddepression(HADSDepression).TheHADSiswidelyusedbuthasshownambiguousresultsbothregardingthefactorstructureandsexdifferencesintheprevalenceofdepressivesymptoms.ThereisalsoalackofpsychometricevaluationsoftheHADSinnon-clinicalsamplesofolderpeople.TheaimofthestudywastoevaluatethefactorstructureoftheHADSinageneralpopulation65–80 yearsoldandtoexampossiblepresenceofdifferentialitemfunctioning(DIF)withrespecttosex.MethodsThisstudywasbasedondatafromaSwedishsample,randomizedfromthetotalpopulationintheagegroup65–80 years(n = 6659).Confirmatoryfactoranalyses(CFA)wereperformedtoexaminethefactorstructure.OrdinalregressionanalyseswereconductedtodetectDIFforsex.ReliabilitywasexaminedbybothordinalaswellastraditionalCronbach’salpha.ResultsTheCFAshowedatwo-factormodelwithcross-loadingsfortwoitems(7and8)hadexcellentmodelfit.Internalconsistencywasgoodinbothsubscales,measuredwithordinalandtraditionalalpha.Flooreffectswerepresentedforallitems.NoindicationformeaningfulDIFregardingsexwasfoundforanyofthesubscales.ConclusionsHADSAnxietyandHADSDepressionareunidimensionalmeasureswithacceptableinternalconsistencyandareinvariantwithregardtosex.Despitepronouncedceilingeffectsandcross-loadingsforitem7and8,thehypothesizedtwo-factormodelofHADScanberecommendedtoassesspsychologicaldistressamongageneralpopulation65–80 yearsold. BackgroundPsychologicaldistressintermsofdepressionandanxietyisagrowingproblemamongolderpeople,withaprevalenceofdepressioninEuropeancountriesat12%forpeopleaged65 yearsorover[1].Foranxietythecorrespondingprevalencevariesfrom1%to14%inNorthAmericaandEurope[2].Late-lifedepressioncanhaveseriousconsequencessuchasincreasedcomorbiditywithphysicalillness,reducedfunctionandincreasedriskofsuicide[3].Alsoanxietyinthisagegroupcanleadtoconsiderabledistressandfunctionalimpairment[4].Asthereareaconsiderablenumberofolderpeoplewhosufferfromsymptomsofanxietyand/ordepression,thereisaneedforabriefandfeasibleinstrumenttoidentifypeopleatriskbutalsoforevaluationofinterventionsandforresearch.TheHospitalAnxietyandDepressionScale(HADS)isafrequentlyusedself-ratingscaledevelopedtoassesspsychologicaldistressinnon-psychiatricpatients.Itconsistsoftwosubscales,AnxietyandDepression[5].Overall,ithasdemonstratedsatisfactorypsychometricpropertiesindifferentgroups;inprimarycarepatients[6],cognitivelyintactnursinghomepatients[7],cancerinpatients[8]andingeneralpopulations[6,9].However,previousstudieshavesuggesteddifferentfactorstructuresoftheHADS.Thehypothesizedtwo-factorstructureismostoftenconfirmed[8,10,11]butstudieshavealsosuggestedonefactor[12],threefactors[13]andalsofourfactors[14].Inaddition,fewstudieshaveevaluatedthefactorstructureinolderpopulationsspecifically,andexistingstudiesshowdivergentfactorstructuresaswell.AstudybyHelviketal.[15]supportedatwo-factormodelinasampleofhospitalizedpatients65 yearsandolder.However,threeitemsintheirstudydidnotloadontheexpectedfactorsuggestedbytheconstructors[5].Onestudyinacommunity-dwellingpopulationaged60–80 yearsconfirmedthetwo-factorstructureasthemostplausibleandalsomoreclinicalrelevantincomparisonwithathree-factormodel[16].AlsoGaleetal.[10]foundatwo-factormodelasmoreappropriatecomparedtoathreefactormodelinnon-clinicalpopulationsofoldermenandwomen.Incontrast,athree-factormodelwassuggestedinastudyregardingolderveterans(>65 years)withlimbamputation[13].AsknowledgeaboutthelatentstructureoftheHADSinanolderpopulationislimitedandhasshowncontradictoryresults,thereisaneedtofurtherevaluatethefactorstructureinageneralpopulationofolderpeople.Previousresearchhascommonlyshownthatpsychologicaldistressismoreprevalentamongwomenthanmen[2,17].However,somestudieshaveshownnostatisticallysignificantdifferencesbetweenwomenandmenregardingprevalenceofdepression[15,18]whereasMartinetal.[19]showedahigherprevalenceformen.Apossibleexplanationtothesedivergingresultscouldbeheterogeneityregardingstudydesign,populationandmeasurements.AspreviousresearchalsohasshownambiguousresultswhenusingtheHADS[20,21],thereisaneedtoexamineifHADSisaninvariantmeasureforpsychologicaldistressforwomenandmen.Differentialitemfunctioning(DIF)isanoftenoverlookedaspectofvalidityandoccurswhendifferentsubgroupsresponddifferentlytospecificitemswithinascale,aftermatchingontheunderlyinglatentconstructthattheitemisintendedtomeasure[22].IfDIFispresenteditimpliesthatthescaleisnotmeasuringthesamethingforallrespondentsandthusmightleadtoincorrectconclusions.TheaimofthestudywastoevaluatethefactorstructureoftheHospitalAnxietyandDepressionscale(HADS)inageneralpopulation65–80 yearsoldandfurthertoexamthepossiblepresenceofdifferentialitemfunctioning(DIF)withrespecttosex.MethodThisvalidationstudywasbasedonacross-sectionalsurveyincludingarandomsampleofindividuals(n = 9968)selectedfromthetotalSwedishpopulationaged65–80 years(N = 1.276.307).Themainaimwastoinvestigatetheprevalenceof,andassociationbetweendepressivesymptomsandlonelinessinrelationtoageandsex[21].EthicalapprovalwasobtainedfromtheRegionalEthicReviewBoardinStockholm,Sweden(No.2010/823–314/4).SampleandproceduresParticipantswererandomlyselectedfromanationalregisterofthetotalpopulation,whichincludesallpersonsregisteredasresidentsinSweden.Theinclusioncriterionwasbeingintheagegroup65–80 years.Thestudywasbasedonpostalquestionnaires.StatisticsSwedenperformedtherandomizationanddistributedthequestionnairestogetherwithinformationaboutthestudyemphasizingvoluntarinesstoparticipateandanonymityinrelationtoresearchers.Questionnairesandfollow-uplettersweresenttonon-respondentsafterthreeweeksandresultedinaresponserateof67.0%forthetotalsample(66.6%forwomenand67.1%formen).Forthispsychometricevaluationstudy,37questionnaireshadmissingdatainallitemsregardingHADSandwerethereforeexcluded,leavingafinalsampleof6622participants.ThequestionnaireThequestionnairewasdividedintotwoparts;onepartwasspecificallyreflectinge.g.demographics,morbidity,andpharmacologicaltreatmentwhiletheotherpartconcernedpsychologicaldistress,symptomsofanxietyanddepression,measuredwithHADS.TheHADSaimstomeasuresymptomsofanxietyanddepressionandconsistsof14items,sevenitemsfortheanxietysubscale(HADSAnxiety)andsevenforthedepressionsubscale(HADSDepression).HADSAnxietyfocusmainlyonsymptomsofgeneralizedanxietydisorderandHADSDepressionisfocusedonanhedonia,themainsymptomofdepression[23].Eachitemisscoredonaresponse-scalewithfouralternativesrangingbetween0and3.Afteradjustingforsixitemsthatarereversedscored,allresponsesaresummedtoobtainthetwosubscales.Recommendedcut-offscoresaccordingtoZigmond&Snaith[5]are8–10fordoubtfulcasesand≥11fordefinitecases.Anoptimalbalancebetweensensitivityandspecificitywasfoundusingacut-offscoreof8oraboveforbothHADSAnxietyandHADSDepression[6].DataanalysisDemographiccharacteristicsarepresentedwithdescriptivestatistics(frequencies,means,standarddeviations,mediansandinterquartileranges)anddifferencesbetweensexeswereanalyzedwithindependentsamplet-testandchi-squaretest.Anitemanalysiswasconductedtoevaluatescoredistributions,floor/ceilingeffects,andmissingdatapatterns.Analysesofdistributionwerebasedondescriptivestatisticsforordinaldata.However,meanandstandarddeviationswerealsocalculatedforcomparisonswithpreviousstudies.TheD’Agostinotestwasconductedtoevaluateifitemandscalescoresdeviatedsignificantlyfromanormaldistribution.Floorandceilingeffects,whichrefertotheproportionsofparticipantswiththelowest(floor)andhighest(ceiling)possiblescores,wereevaluatedusingfrequencydistributions.Upto20%floor/ceilingeffectswereconsideredacceptableinthepresentstudy.Totestifthedatawerecompletelymissingatrandom(MCAR),Little’schi-squaredtestforMCARwasconductedforeachscaleseparately.Homogeneitywasevaluatedwithinter-itemcorrelationsbasedonpolychoriccorrelations.Aconfirmatoryfactoranalysis(CFA)wasconductedtoevaluatethehypothesizedtwo-factorstructureoftheHADS(modelI);7itemmeasuringanxietyand7itemmeasuringdepression,withoutanyothermodifications.Asthemodeldidnotperfectlyfitthedata,asecondmodelwasevaluated(modelII);two-factorswithcross-loadingsforitem7and8.Astheitemswerehighlyskeweddistributedwithpronouncedflooreffects,athirdmodelwasevaluatedtoidentifywhichimpactthisproblemhadonthefactorstructure(modelIII);atwo-factormodelwithcross-loadingsforitem7and8togetherwithcollapsedresponsecategories(category2and3forallitems).Theitemsweretreatedasorderedindicatorvariablesandconsequentlyadiagonallyweightedleastsquaremethod(WLSMV),basedonapolychoriccorrelationmatrix,wasusedtoestimatetheparametersofthemodels.Differentgoodness-of-fitstatisticswereusedtoevaluatetheCFAmodels.Anon-significantchi-squaretestindicatesaperfectmodelfitbetweenmodelanddata.However,sincethistestishighlysensitiveforlargesamplesizesitshouldbeinterpretedwithcaution.Thereforeweusedthefollowinggoodness-of-fitcriteria;rootmeansquareerrorofapproximation(RMSEA) ≤ 0.06,comparativefitindex(CFI) ≥ 0.95andTucker-Lewisindex(TLI) ≥ 0.95[24].Toevaluateinternalconsistencyreliability,anordinalvariantofCronbach’salphawascalculated[25].ThiscalculationisbasedonpolychoriccorrelationsratherthanPearsoncorrelations,butisinterpretedinthesamewayasthetraditionalCronbach’salpha.Thus,alphavaluesabove0.7indicatesufficientinternalconsistencyreliability[26]Forcomparisons,alsotraditionalCronbach’salphawascalculated.Examinationofdifferentialitemfunctioning(DIF)forsexwasconductedforeachitemusingordinalregressionanalyses.Thismethodenablestotestforbothuniform(effectsofgroupdifferences)andnon-uniformDIF(effectsofdifferencesingroupability)[22,27].Inthefirststep(BlockI),theitemresponsesweretreatedasoutcomevariablespredictedbytheconditionalvariable(i.e.totalscoreforHADSAnxietyandHADSDepressionrespectively).Inthesecondstep(BlockII),thegroupingvariable(i.e.sex)wasaddedascovariatetodetectuniformDIF.Inthethirdstep(BlockIII),theinteractiontermbetweentheconditionalvariableandgroupvariable(i.e.,sex×HADSAnxietyandsex×HADSDepression)wereaddedascovariatestotestfornon-uniformDIF[22].ThechangeinMcFaddenR2betweenthethreemodelswasusedtoevaluatetheeffectsizeofDIF.ForanitemtobeclassifiedasshowingDIF,thetwodegreeoffreedomchi-squaredtestinlogisticregressionmusthaveap-value<0.01andtheeffectsizemeasurehavetobeatleastR2 ≥ 0.13[22].TheanalyseswereconductedwiththeSPSSStatistics20.0(IBMCorp,Armonk,NY,USA),Mplus7.4(Muthén&Muthén,LosAngeles,CA,USA)andR3.3.0software(theRFoundationforStatisticalComputing,ViennaAustria).ResultsSamplecharacteristicsTheoverallmeanagewas71.2 years(SD = 4.5).Thesampleconsistedofalmostasmanymenaswomen,48.4%and51.6%respectively.Amajorityweremarried/cohabitating(70.5%),wasretired(80.2%)andreportedprimaryschoolasthehighesteducationlevel(49.1%).TheproportionofparticipantsscoringHADSAnxietywere10.7%fortheentiresample,significantlymorecommonamongwomenthanmen(14.1%vs.7.0%,p 0.5andrangedbetween0.74and0.85forHADSAnxietyand0.55and0.84forHADSDepression(Table 4,Fig. 1).Table4Goodness-of-fitindicesfortheconfirmatoryfactoranalysesmodels(n = 6622)Fullsizetable Fig.1Parameterestimates(i.e.,factorcorrelations,factorloadings,cross-loadingsandresidualvariances)frommodelI(outsidebrackets)andmodelII(insidebrackets)Fullsizeimage Astheitemscoreswerehighlyskeweddistributed,athirdmodelwasevaluatedtoaddressthisproblem,inwhichcategory2and3werecollapsed.ThisthirdmodeldemonstratedmodelfitatthesamelevelasmodelII(RMSEA = 0.05,CFI = 0.98,TLI = 0.98)(Table4).Inthismodel,factorloadingsvariedbetween0.33and0.83forHADSAnxietyand0.40and0.84forHADSDepression.AsinmodelII,onlyfactorloadingsandcross-loadingsforitem7and8were<0.5.InternalconsistencyreliabilityTheinternalconsistencyreliability,assessedwithordinalalpha,was0.92forHADSAnxietyand0.88forHADSDepression.ThecorrespondinginternalconsistencymeasuredwithtraditionalCronbach’salphawas0.87and0.81respectively.DifferentialitemfunctioningTheresultsfromtheordinalregressionsanalysisarepresentedinTables 5and6.Theconditionalvariable(i.e.HADSscalescores)wassignificantlyassociatedwithallitemresponsesforbothHADSAnxietyandHADSDepressioninBlockI.Thegroupvariable(i.e.sex)wasalsosignificantlyassociatedwithallitemsinbothHADSAnxietyandHADSDepression(p