ModelChain:DecentralizedPrivacy-PreservingHealthcarePredictiveModelingFrameworkonPrivate BlockchainNetworks Tsung-TingKuo,PhD,1Chun-NanHsu,PhD,1andLucilaOhno-Machado,MD,PhD1,2 1HealthSystemDepartmentofBiomedicalInformatics,UniversityofCaliforniaSanDiego,LaJolla,CA 2DivisionofHealthServicesResearch&Development,VASanDiegoHealthcareSystem Abstract Cross-institutionalhealthcarepredictivemodelingcanaccelerateresearchandfacilitatequalityimprovement initiatives,andthusisimportantfornationalhealthcaredeliverypriorities.Forexample,amodelthatpredicts riskofre-admissionforaparticularsetofpatientswillbemoregeneralizableifdevelopedwithdatafrom multipleinstitutions.Whileprivacy-protectingmethodstobuildpredictivemodelsexist,mostarebasedona centralizedarchitecture,whichpresentssecurityandrobustnessvulnerabilitiessuchassingle-point-of-failure (andsingle-point-of-breach)andaccidentalormaliciousmodificationofrecords.Inthisarticle,wedescribea newframework,ModelChain,toadaptBlockchaintechnologyforprivacy-preservingmachinelearning.Each participatingsitecontributestomodelparameterestimationwithoutrevealinganypatienthealthinformation (i.e.,onlymodeldata,noobservation-leveldata,areexchangedacrossinstitutions).Weintegrateprivacy- preserving online machine learning with a private block chain network, apply transaction metadata to disseminatepartialmodels,anddesignanewproof-of-informationalgorithmtodet