Effects of health-information-based diabetes shared care ...

Available from https://visualizinghealthdata.idv.tw/?route=article/faq&faq_id=10. Accessed 10 Sep 2019. National Health Insurance Administration ... Skiptomaincontent Advertisement SearchallBMCarticles Search Effectsofhealth-information-baseddiabetessharedcareprogramparticipationonpreventablehospitalizationsinTaiwan DownloadPDF DownloadPDF Researcharticle OpenAccess Published:27November2019 Effectsofhealth-information-baseddiabetessharedcareprogramparticipationonpreventablehospitalizationsinTaiwan Yia-WunLiang1,Hsiao-FengChang2&Yu-HsiuLin  ORCID:orcid.org/0000-0002-0049-267X3  BMCHealthServicesResearch volume 19,Article number: 890(2019) Citethisarticle 1590Accesses 7Citations 269Altmetric Metricsdetails AbstractBackgroundTaiwan’sDiabetesSharedCareProgramhasbeenimplementedsince2012,andthehealthinformationsystemplaysavitalroleinsupportingmostservicesofthisprogram.However,littleisknownregardingtheeffectivenessofthisinformation-basedprogram.Therefore,thisstudyinvestigatedtheeffectsoftheparticipationoftheDiabetesSharedCareProgramonpreventablehospitalizations.MethodsThislongitudinalstudyexaminedthedataofhealth-careclaimsfrom2011to2014obtainedfromthediabetesmellitushealthdatabase.Patientswithdiabetesaged≥18 yearswereincluded.PreventablehospitalizationswereidentifiedonthebasisofpreventionqualityindicatorsdevelopedforadministrativedatabytheUSAgencyforHealthcareResearchandQuality.AmultilevellogisticregressionwasperformedtoexaminetheeffectsoftheparticipationoftheDiabetesSharedCareProgramonpreventablehospitalizationsafteradjustmentforothervariables.Analyseswereconductedinlate2018.ResultsAmediumlevelofparticipation(p = 0.05),agebetween40and64 years(p  9.0%indicatefavorableandpoorglycemiccontrol,respectively),andlowdensitylipoproteinlevels( 130 mg/dLindicatefavorableandpoorcontrol,respectively).TNHIalsopaysextrabonusesbasedonthequalityindicatorsforhospitals/clinics,orontherateofnewcaseandthequalityindicatorsforphysicians[7].Moreover,ifapatientisenrolledintheDSCPinhospitalA,thenthatpatientcannotenrollintheDSCPinhospitalBbecausetheNHIISSconnectsthehospitalinformationsystemtoconstituteasingleregistryforpatientsinhospitals.Ambulatorycare-sensitiveconditions(ACSCs)offeravaluableperspectiveonsystemperformanceandthuscanbeusedtoevaluateprimarycarephysicians’access,availability,andeffectiveness[10,11,12].TheUSAgencyforHealthcareResearchandQuality(AHRQ)developedpreventionqualityindicators,whicharedefinedbytheICD-9-CM,toidentifyACSCs[13].Since1993,theInstituteofMedicinehasrecommendedusingACSCstomonitoraccessofcare[12].HospitalizationscausedbyACSCsareconsideredpreventablehospitalizations(PHs).Todate,nostudyhasevaluatedtheeffectoftheinformation-basedDSCPontheincidenceofPH.Therefore,thepresentstudyinvestigatedtheeffectoftheparticipationoftheDSCPontheincidenceofPH.MethodsStudysampleTheDSCPhasbeenapay-for-performanceprogramintheTNHIsince2012,andwefocusedontheeffectsobservedinthefirstthreeyears.Analyseswereconductedinlate2018.Thislongitudinalstudyincludedanalysesatindividualandcountylevels.Datafortheindividual-levelanalysiswereobtainedfromthe2011–2014DiabetesMellitusHealthDatabase(DMHD),whichisbasedonthenationwideTaiwanNationalHealthInsuranceResearchDatabase.Alldataaredeidentifiedandencryptedtoprotectparticipants’privacy.AllpatientsintheDMHDwerediagnosedwithTypeIandTypeIIdiabetes(ICD-9-CMcode250).Theexclusioncriteriaincludeage ≤ 18,gestationaldiabetesmellitus,andmissingdata.Inthisstudy,weusedfoursubsetsofthedatabase:registryforbeneficiaries(DM_ENROL),ambulatorycareexpendituresbyvisits(DM_OPDTE),detailsofambulatorycareorder(DM_OPDTO),andinpatientexpendituresbyadmissions(DM_IPDTE).County-leveldatawereretrievedfromthe2014TaiwanHospitalandClinicStatistics.ThetimeflowofthisstudyisdisplayedinFig. 1. Fig.1TimeflowofthestudyFullsizeimageFirstly,merging2011–2014DM_ENROL,DM_OPDTE,DM_OPDTO,andDM_IPDTE,atotal404,418peopleincludedinthemergeddataset(noted:noteveryonecontains4-yearclaimdata).Secondly,inaccordancewiththestudypurpose,onlythedataofadultsaged≥18 years,diagnosedastypeIorIIdiabetes,andcontained2011–2014claimswerecollected,andremained61,032peopleintheanalyticdataset.Finally,afterdisregardingmissingvaluesandoutliersinthestudyvariablesandmergingcounty-leveldata,atotalof60,962patientsfrom22countieswereincludedinthisstudy.ApprovalfortheanalysisofthedatabasewasobtainedfromtheInstitutionalReviewBoardofChungKangBranch,ChengChingHospital,Taiwan(IRBNo.HP180005).MeasuresDependentvariablesWeuseddefinitionsfromtheAHRQfordiagnosingACSCs.ACSCsforadultsincludedspecificdiagnosesofasthma,angina,congestivehealthfailure,bacterialpneumonia,chronicobstructivepulmonarydisease,dehydration,long-termandshort-termdiabetescomplications,hypertension,lower-extremityamputationforpatientswithdiabetes,perforatedappendix,uncontrolleddiabetes,andurinarytractinfection[13].HospitalizationforanyofthesediagnoseswasconsideredhospitalizationforanACSC,alsoknownasaPH.KeyexplanatoryvariablesParticipationoftheDSCPTheparticipationoftheDSCPisdividedintofourlevels:none,low,medium,andhigh.NoneindicatesthatthepatientdidnotenrolledintheDSCP;lowparticipationindicatesthatthepatientonlyenrolledintheDSCPorfollowedup,butwithoutannualexam(procedurecode:P1407CandP1408C);mediumparticipationindicatesthatthepatientcompletedthefirststageoftheDSCP(threefollow-upsandanannualreview)(procedurecode:P1409C);andhighparticipationindicatesthatthepatientbeganthesecondstageoftheDSCPfollow-uporcompletedthesecondstageannualreview(procedurecode:P1410CandP1411C).CovariatesCovariatespotentiallyassociatedwithPHsincludedcharacteristicsatindividualandcountylevels.Individual-levelcharacteristicsincludedpatients’sociodemographicandhealth-relatedvariables.Sociodemographicvariablesincludedsex,incomelevel,andregionofresidence.Regionofresidencewasdefinedaccordingtopatients’healthinsuranceadministrationdivision,namelyTaipei,Northern,Central,Southern,Kaoping,andEasterndivisions.Health-relatedvariablesincludedcomorbiditiesandcatastrophicillnesses.ACharlsoncomorbidityscorewascalculatedforeachpatienttomeasurethecomorbidities[14].CatastrophicillnessesareapprovedbytheBureauofNationalHealthInsurance,including29catastrophicillnesses,suchasmalignantneoplasm,systemiclupuserythematosus,etc.,andotherrarediseases.Patientswithcatastrophicillnessescertificatescanapplycatastrophicillnessregistrationcards,andeligibleforexemptionfrominsurancepremiumsandcopayments.County-levelcharacteristicswererepresentedbyhealth-careresourcesandhealth-carepersonneldensity.Health-careresourcesincludedthenumberofgeneralhospitalsandclinics.Thenumberofphysiciansrepresentedhealthpersonneldensity.County-levelcharacteristicswereadjustedbypopulationsizeandwerecalculatedasthenumberofeachvariabledividedbytotalpopulationintheareamultipliedby100,000.County-levelcharacteristicsweredividedintotwogroupsbasedonthemeanscore:low(deprived)andhigh(affluent).StatisticalanalysisDescriptivedataanalysiswasperformedforindividual-levelcharacteristics.Chi-squareandindependentttestswereusedtoexaminebivariatecorrelationsbetweeneachindividual-levelcharacteristicandPHs.Thedatahadahierarchicalstructure,inwhichindividualdata(level1)werenestedwithincounties(level2).AmultilevelanalysiswasperformedtocontrolforthecountyeffectonPH.ArandominterceptmultilevelmodelwaspreferredoverotherstatisticalapproachesbecauseittestedwhethertheparticipationoftheDSCPwasassociatedwithPHsamongpatientsacrosscounties.RegressioncoefficientsandvariancecomponentsatcountyandindividuallevelswereestimatedforPHs.Threemodelswerefitted.Thefirstmodelintheoutputwasanemptymodel;thatis,amodelwithnopredictors.TheemptymodelwasusedtodeterminewhethertheoveralldifferencebetweencountiesandindividualsintermsofPHswouldbesignificant.Thesecondmodelincludedonlyindividual-levelvariables,andthethirdmodelincludedbothindividual-levelandcounty-levelvariables.Amultilevellogisticregressionwasperformedtoestimateadjustedoddsratios(ORs)with95%confidenceintervals(CIs)andpvalues.Dependentvariablesusedinmultilevellogisticregressionmodelsweredichotomous;patientswithPHwerecodedas1,andthosewithoutPHwerecodedas0.Theequationforthemultilevellogisticmodelisasfollows: $$\mathrm{logit}\left({\pi}_{ij}\right)=\alpha+{u}_j+{\beta}^{\tau}{X}_{ij}$$where\({u}_j\sim\mathrm{N}\left(0,{\sigma}_u^2\right)\),ujistherandomeffectandjrepresentscounty-levelcharacteristics.αandβarefixedeffects,αrepresentstheintercept,andirepresentsindividual-levelcharacteristics.Intraclasscorrelationcoefficients(ICCs)werecalculatedtodeterminethecontributionofvarianceatthecountyleveltothetotalvariance.Formultilevellinearmodels,theICCwascalculatedusingthefollowingformula: $$\mathrm{ICC}=\frac{\sigma_n^2}{\sigma_i^2+{\sigma}_n^2}$$where\({\sigma}_n^2\)=county-levelvarianceand\({\sigma}_i^2\)=individual-levelvariance.Becausethevarianceofalogisticdistributionwithascalefactorof1isπ2/3(approximately3.29)inahierarchicallogisticregressionmodel,thisformulacanbereformulatedasfollows[15]: $$\mathrm{ICC}=\frac{\sigma_n^2}{\sigma_n^2+\left(\frac{\uppi^2}{3}\right)}$$AllstatisticalanalyseswereperformedusingSASversion9.4(SASInstitute,Inc.,Cary,NC,USA).Statisticalsignificancewasdeterminedfordifferenceswithatwo-sidedpvalueof