Transcription
Session #22Data Operating System:What It Really Means and Why You Will Need ItImran QureshiSean StohlPete HessChief SoftwareDevelopment OfficerHealth CatalystSVP - Platform OperationsProduct DevelopmentHealth CatalystVP PlatformEngineering – DOSHealth Catalyst
Learning Objectives Describe the major problems health systems are facing today. Demonstrate why current analytics solutions aren’t good enough. Analyze why we need to evolve from data warehouses. Show how a data operating system addresses common use cases.
The State ofHealthcare Today
Hospital Margins Are at RiskHospital ProfitMargins6%20113.3%2025 (Projected)41%Hospitals withnegative margins2025 (Projected)Annual growth/costreduction needed tomaintain margins0.5%“Revenue growth will alsotemper amid decliningreimbursement from both privateand governmental payers.”“.uncertainty over federal healthcare policy also poses a strongheadwind (for hospitals' financialperformance).”Moody's: Preliminary FY 2016 US NFPCBO Publication 2016 – Projecting Hospitals’ Profit Margins
Clinicians Are Over-Worked and Over-MeasuredLloyd B. Minor, MD, Dean,Stanford School of MedicineSteven J. Stack, MD,President, AMAGary Botstein, MD,DeKalb Medical“Today’s searchengines are better athelping cliniciansdiagnose disease thanour EMRs.”50% Physicianspenalized 1% ofpayments byMeaningful Use“ the emphasis isreally on datacollection, but whatphysicians ought to bedoing is datasynthesis.”
Risk Management and Population Health RequireMore DataSource: Chilmark Research – 2017 Healthcare Analytics Market Trends Report
200 Current Data SourcesEMRs1. Affinity - ADT/Registration2. Allscripts - Ambulatory EMRClinicals3. AllscriptsEnterprise/Touchworks Ambulatory EMR4. Allscripts Sunrise - Acute EMRClinicals5. Aprima ERM6. Cerner - Acute EMR Clinicals7. Cerner - PowerWorksAmbulatory EMR8. Cerner HomeWorks - Other9. CPSI - Acute EMR Clinicals10. eClinicalWorks - AmbulatoryEMR Clinicals11. Epic - Acute EMR Clinicals12. Epic - Ambulatory EMRClinicals13. GE (IDX) Centricity Ambulatory EMR Clinicals14. McKesson Horizon - AcuteEMR Clinicals15. McKesson Horizon EnterpriseVisibility16. Meditech 5.66 EHR w/DR17. NextGen - Ambulatory PracticeManagement18. Quality Systems (Next Gen) Ambulatory EMR Clinicals19. Siemens Sorian Clinicals Inpatient EMRFinance/Costing1.2.3.4.5.6.7.Affinity - CostingAllscripts (EPSi) - BudgetAllscripts (EPSi) - CostingAllscripts (TSI) - CostingBOXI - GLCost Flex - CostingDigimax MaterialsManagement - InventoryManagement8. IOS ENVI - Costing9. Kaufman Hall Budget Advisor- Other10. Lawson - Accounts Payable11. Lawson - AccountsReceivable12. Lawson - GL13. Lawson - Supply Chain14. McKesson - AccountsPayable15. McKesson EnterpriseMaterials Management16. McKesson HPM - Costing17. McKesson HPM - GL18. McKesson PFM - AccountsPayable19. McKesson PFM - GL20. McKesson Series - AccountsReceivable21. Meditech - GL22. Microsoft Great Plains - GL23. Oracle (Hyperion) - Costing24. Oracle (PeopleSoft) - GL25. Oracle (PeopleSoft) - SupplyChain26. PARExpress27. PPM - Costing28. Smartstream - GL29. StrataJazz - CostingBilling1. Affinity - Hospital Billing2. CHMB 360 RCM - HospitalBilling3. CPSI - Hospital Billing4. Epic - Hospital Billing5. GE (IDX) Centricity - HospitalBilling6. GE (IDX) Centricity Professional Billing7. HealthQuest - PatientAccounting8. Keane - Hospital Billing9. McKesson Series - PatientBilling10. McKesson STAR - HospitalBilling11. MD Associates - ProfessionalBilling12. Siemens Sorian Financials Inpatient Registration andBillingPat. Sat1. Fazzi - Patient Satisfaction2. HealthStream - PatientSatisfaction3. NRC Picker - Patient Satisfaction4. PRC - Patient Satisfaction5. Press Ganey - PatientSatisfaction6. Sullivan Luallin - PatientSatisfactionHIE1. Adirondack ACO Clinical Datafrom HIXNY (HIE)2. ADT HIE Patient Programs3. Vermont HIEHR/ERP1. API Healthcare - Time andAttendance2. iCIMS3. Kronos - HR4. Kronos - Time andAttendance5. Lawson - HR6. Lawson - Payroll7. Lawson - Time andAttendance8. Maestro9. MD People10. Now Solutions Empath - HR11. Oracle (PeopleSoft) - HR12. PeopleStrategy/Genesys HR13. PeopleStrategy/Genesys Payroll14. Ultimate Software Ultipro HR15. WorkDayTerminology1. AHRQ Clinical ClassificationSoftware (CCS)2. Charlson Deyo and ElixhauserComorbidity3. Clinical Improvement Grouper(Care Process Hierarchy)4. CMS Hierarchical ConditionCategory5. CMS Place Of Service6. LOINC7. National Drug Codes (NDC)8. NPI Registry9. Provider Taxonomy10. Rx Norm11. CMS/NQF Value Set AuthorityCenterClaims1.2.3.4.5.6.7.8.835 – DenialsAdirondack ACO MedicareAetna - ClaimsAnthem - ClaimsAon Hewitt - ClaimsBCBS IllinoisBCBS VermontChildren's Community Health Plan(CCHP) - Payer9. Cigna - Claims10. CIT Custom - Claims11. Cone Health Employee Plan (UnitedMedicare) - Claims12. Discharge Abstract Data (DAD)13. Hawaii Medical Service Association(HMSA) - Claims14. HealthNet - Claims15. Healthscope16. Humana (PPO) - Claims17. Humana MA - Claims18. Kentucky Hospital Association (KHA) Claims19. Medicaid - Claims20. Medicaid - Claims - CCO21. Merit Cigna - Claims22. Merit SelectHealth - Claims23. MSSP (CMS) - Claims24. NextGen (CMS) - Claims25. Ohio Hospital Association (OHA) Claims26. ProHealth - Claims27. PWHP Custom - Claims28. QXNT - Claims29. UMR Claims Source30. Wisconsin Health InformationOrganization (WHIO) - ClaimsClinical Specialty1.2.3.4.5.6.Allscripts - Case ManagementApollo - Lumed X Surgical SystemAspire - Cardiovascular RegistryCarestream - OtherCerner - LaboratoryeClinicalWorks - Mountain Kidney DataExtracts7. GE (IDX) Centricity Muse - Cardiology8. HST Pathways - Other9. ImageTrend10. ImmTrac11. Lancet Trauma Registry12. MacLab (CathLab)13. MIDAS - Infection Surveillance14. MIDAS - Other15. MIDAS - Risk Management16. Navitus - Pharmacy17. NHSN18. NSQIPFlatFile19. OBIX - Perinatal20. OnCore CTMS21. Orchard Software Harvest - Pathology22. PACSHealth - Radiology23. Pharmacy Benefits Manager24. PICIS (OPTUM) Perioperative Suite25. Provation26. Quadramed Patient Acuity ClassificationSystem - Other27. QXNT/Vital - Member28. RLSolutions29. SafeTrace30. Siemens RIS - Radiology31. SIS Surgical Services32. StatusScope - Clinical Decisions33. Sunquest - Laboratory34. Sunrise Clinical Manager35. Surgical Information System36. TheraDoc37. TransChart - Other38. Varian Aria - Oncology39. Vigilanz - Infection Control
Machine Learning Has Potential, but It’s Not Scaling2xReadmissions predicted byUNC machine learningmodel vs LACEAdvisory Board 2017 - The Rise of Machine Learning.20%Decrease inCLABSI rateIndiana University System – Scottsdale Institute, Inside Edge (2016) Every machine learning project is a oneoff requiring a lot of work. Hospitals can’t find enough datascientists. Clinicians don’t trust “black box”machine learning.So, machine learning feelsout of reach
50 Health Catalyst Machine Learning Models
What Is the Solution?
First, Let Me Tell You a StoryA decade ago, best-of-breed applications ruled healthcare. Each fit the individual workflows well. But data remained in silos and apps did notinteroperate.Today, we have one monolithic application (EHR). Clinical data is in one place.-Although non-clinical data remains separate. But cited as:-Biggest cause of physician burnout.Forces one-size-fits-all workflows.Stifles innovation.
Single Monolithic App Model Isn’t WorkingAppAppAppApp (EHR)App(EHR)AppAppData Operating SystemDataDataDataDataDataSiloed apps don’t work.One app for everythingdoesn’t work.DataDataBest of breed appsworking together.
We Already Have a Model for ThisYour iPhone has best-of-breed appsthat all work together and share yourdata (e.g., your contacts).Imagine if you were forced to useyour email application to order Uber.
Precision eratingsystemCentralizedsecurityMachinelearningIn clinicalworkflow“EHRs would become commodity components in a larger platform that would include othertransactional systems and data warehouses running myriad apps, and apps could haveaccess to diverse sources of shared data beyond a single health system’s records.”“A 21st-Century Health IT System — Creating a Real-World Information Economy”, Kenneth D. Mandl, MD, MPH; Isaac S. Kohane, MD, MPH;NEJM, 18 May 2017.
Health Catalyst Workflow AppsCORUSMachine LearningResearch InformaticsUncompensatedCare SuiteComparative AnalyticsPatient SafetyPatient Relationship & CMPrecise Patient RegistriesPopulation Health & ACLeading WiselyFinancial ImprovementMACRA Measures &InsightsOperations & WorkflowImprovementMeasures BusinessLibraryClinical ImprovementHCC InsightsHealth Catalyst Data Operating System (DOS )Care ManagementCAFÉText Analytics
70 Health Catalyst Analytics AcceleratorsClinical Analytics & Decision SupportOperations & Performance ManagementAccountable CareAcute Coronary Syndrome (ACS)Blood Utilization DashboardBreast Milk FeedingCatheter Associated Urinary TractInfection (CAUTI) PreventionCentral Line Associated Blood StreamInfections (CLABSI) PreventionColorectal SurgeryEarly Mobility in the ICUGlycemic Control in the HospitalHeart FailureJoint Replacement - Hip & KneeKey Process Analysis (KPA)Labor and DeliveryPatient flight Path - DiabetesPatient Safety ExplorerPediatric AppendectomyPediatric AsthmaPediatric ExplorerPediatric SepsisPneumoniaPopulation ExplorerReadmission ExplorerSepsis PreventionSpine SurgeryStroke (Acute Ischemic & TIA)Surgical Site Infection PreventionDepartment Explorer: Emergency ServicesDepartment Explorer: Surgical ServicesLabor Management ExplorerLeading WiselyPatient Experience ExplorerPatient Flow ExplorerPractice Management: Patient AccessProvider ProductivitySupply Chain ExplorerMACRA Measures & InsightsCommunity CareACO ExplorerACO MSSP MeasuresAttribution ModelerBundled PaymentsHCC InsightsPMPM AnalyzerRisk Model AnalyzerFinancial Decision SupportCORUS (Clinical & Operational ResourceUtilization System)Advanced Billing MonitoringFinancial Management ExplorerGeneral Ledger ExplorerHIM Documentation Workflow AnalyzerRevenue Cycle Advisor: HospitalRevenue Cycle Advisor: ProfessionalRevenue Gap FinderCare Management & Patient RelationshipsPatient StratificationPatient IntakeCare CoordinationCare CompanionCare Team InsightsCatalyst 4 HealthResearch InformaticsCohort Definition FrameworkComparative AnalyticsCollective Analytics for Excellence (CAFÉ)
The Evolution fromData Warehouseto aData Operating System
Evolution from a Data Warehouse toa Data Operating SystemData WarehouseData Operating System1Collects data from EHR and claims.Collects data from many sources.2Enables creating reports.Enables creating reports and web/mobile apps.3Enables SQL queries.Enables SQL, machine learning (R/Python) queries.4Data is updated nightly.Data is updated in real-time.5Not available in the EHR workflow.Insights are easily available in the EHR workflow.6Requires replacing your existing EDW.Works with your existing EDW (or use our EDW).7Proprietary schemas.Industry standard schemas (e.g., FHIR).8Text analytics is a separate process.Text analytics is built-in.9Works with rows and columns.Works with rows, columns and reusable healthcare logic like registries, measures,risk, insights.10Provides centralized security at app and data levels.11Makes machine learning as easy to use as SQL.12Content Marketplace to share executable content with other health systems.
The Health Catalyst Data Operating System (DOS)MarketplaceHealth Catalyst lasACOFinancialsRegistryBuilderSAMD & SMDReusableContentMeasureLibraryML ModelsClient BuiltApplicationsand more 3rd Party AppsFabric Application ServicesEHRIntegrationFHIRSecurity, Identity& ComplianceCatalyst Analytics PlatformData IngestSourceConnectorsRegistriesTerminology& GroupersPatient & ProviderMatchingFabric Data ServicesData PipelinesData ExportData ic Real Time ServicesStandardData ModelsData QualityPatternRecognitionNLPReal-timeStreamingReal timeProcessingMachineLearningPipelinesHL7
Fabric - Building Blocks for Healthcare Microservices. Open APIs. FHIR-based. Open-source. Reusable clinical and financial logic. Fast access to ALL aggregated data. Interoperable and portable apps. Install only the components you want. Replace any component with your own.Monolithicsystem
Who Needs a Data Operating System?CliniciansHospital stratorsDataScientists
Clinicians Need a Data Operating SystemMergers & Acquisitions have left health systems with fragmented sources ofdata.“I have to login to multiple EHRs just to see bits of patient data. Howcan I make a good decision about this patient in 10 minutes?”What if you could access the data in a single place from all the fragmentedsources of data?
Data Aggregation Reusable Logic Open APIsReportsRegistriesWeb AppsMeasuresMobile AppsCostAccountingInsightsEHR ilPatientSatisfactionClaimsG/L Cost(ERP)DevicePatient Social DeterminantsGeneratedof HealthNotesReusable logicData AggregationData WarehouseEHRsClientsCare MgmtData Sources
Clinicians Need a Data Operating System (cont.)Extend life and value of current EHR investments.“Our clinicians are over-worked and over-measured, butunder-informed”What if the EHR was not just a place to record data but a place whereclinicians see synthesized data and are informed?
How Can We Be Non-intrusive?No more “black box” yes or no recommendationsFocus on data synthesis and letclinicians choose the right path
How likely are clinicians to followcare recommendations if they are inan EHR vs reports.15xRefer to Pain ClinicSend Summary to PCP
Evolution from a Data Warehouseto Data Operating System for CliniciansEHRData Operating System1Contains EHR data.Contains aggregated data from all sources.2Designed for data entry.Designed for data synthesis.3Each EHR is an island.Works with all the EHRs in your health system.4Apps are not portable.Apps are portable: standard APIs and data models.5No access to machine learning.Has built-in machine learning.6No reusable logic.Has reusable logic like registries, measures, risk.7Each integration requires makingchanges in the EHR.Integrate once into your EHR. Insights from many apps can show up withoutreconfiguring the EHR.8Information overload for clinicians.Hospital IT can centrally define rules to control what is shown in EHRworkspace.
Fabric.EHR, the GPS for CliniciansAnalyst(SQL)Apps(FHIR)Data Scientists(REST)Hospital IT(Rules)Fabric.EHRTarget by Patient, User, and EncounterEpicCernerAllscriptsOther
Poll Question #1How many EHRs does your health system use?1)2)3)4)5)6)ZeroOneTwoThreeFour or moreUnsure or not applicable
Hospital IT Needs a Data Operating SystemScaling existing data warehouses“We have the data, but it is impossible for us to scale to morethan a few reports or apps with the team we have.”What if you could enable lots of reports and apps without having to hiremore people?
SQL or Big Data?SQL is great for most cases Simpler. Existing skills. Existing code.You need big data Volume (typically 5TB). Velocity (typically 1/day). Variety (typically 10 sources).False choice: You don’t need to choose!A data operating system allows you to use the same toolswith SQL or big data.
Evolution from Data Warehouseto Data Operating System for ITData WarehouseData Operating System1Source Mart Designer (SMD)creates source marts.Source Mart Designer (SMD) creates source marts and data lakes.2Subject Area Mart Designer(SAMD) creates SQL bindings.Subject Area Mart Designer (SAMD) creates SQL bindings andHadoop/Spark bindings.3SAM Engine runs pipelines viaSQL Server.SAM Engine runs pipelines via SQL Server or via Hadoop.4Atlas tracks meta-data for SQLdata.Atlas tracks metadata for SQL data and Hadoop data with smartsearch across both.5Machine learning is a separateprocess.Create machine learning (R/Python) bindings and SAM engine runsthem at scale.6Only deals with data.myCatalyst portal to access all your analytical assets: data, reports,apps, re-usable logic, permissions with smart search across all of them.
But I Already Have aData Warehouse
The Health Catalyst DOSRuns on Any Data WarehouseMarketplaceHealth Catalyst lasACOFinancialsRegistryBuilderSAMD & SMDReusableContentMeasureLibraryML ModelsClient BuiltApplicationsand more 3rd Party AppsFabric Application ServicesEHRIntegrationFHIRSecurity, Identity& ComplianceData WarehousesHC EDWHome abric Real Time ServicesStandardData ModelsData criptsDataGovernanceOraclePatient & ProviderMatchingFabric Data ServicesMetadataEpicTerminology& GroupersIBMReal timeProcessingMachineLearningPipelinesHL7
Hospital Leaders Need a Data Operating SystemProviders need to manage risk.Machine learning is too hard.“How can I manage risk if I don’t have allthe data and can’t easily calculate risk?”“I would love to use machine learning, but Iwould have to build a huge infrastructure andhire a bunch of expensive data scientists.”What if it was easy to calculate risk andtrend it over time both at population leveland patient level?What would a data platform look like if it wasdesigned to make machine learning as easyas SQL?
Machine Learning as Easy as SQLTools for analysts tocreate “features” usingSQL (or R/ Python).Feature library for ata operatingsystem tools makeit easy to trainmodels.Train andModelSelectionAccess data aslongitudinal patientrecords.Point-and-click wizards forcommon machinelearning algorithms.Access registries,measures, terminologyhierarchies, costaccounting, etc.For advanced scenarios,include R, Python anddeep learning scripts.Real DataThe data operatingsystem takes careof running yourmachine learningalgorithms at scalein production.ModelPredictionsPredictions can berouted to EHRworkflow via thedata operatingsystem.
Poll Question #2How does your health system calculate per-member, per-month costs for aninsurance contract?a)b)c)d)e)f)Machine learning modelsExtrapolating from last yearWe negotiate up from what the insurance company offersRough calculationsWe need to but can’t todayUnsure or not applicable
Data Administrators Need a Data Operating SystemSecurity that is secure, but allows access tothe data.“I need to be HIPAA compliant so I can’t allowanyone to access the data.”Are you protecting the right thing?What if you could easily implement commonhealthcare security models in a central place?
We Don’t Need More LocksWe needintelligentsecurity Imagine your hospital having theseon every door and passing out keys
Security Should Not Be an AfterthoughtSecure by design At app level: Fabric.Identity and Authorization-Integrates with existing identity providers like AD.-Centralized permissions.At data level: Fabric.FHIR data service-Implements permissions at the data layer.-At conceptual layer (patient, medications, diagnosis,claims) instead of table and column level.Centralized Security Console Set permissions for all your applications in oneplace.Implements HL7 SecurityReference ModelBuilt-in support for commonhealthcare use cases: By affiliates.By role.By care program.By sensitivity of data.By insurance company.Define your own securitytemplate.
Lessons Learned Healthcare industry faces significant challenges Hospital margins are at risk. Clinicians are over-worked and over-measured. One monolithic app (EHR) model is not working. Machine learning has potential, but is not scaling. The future of healthcare is an app ecosystem Best-of-breed applications from
Apr 22, 2017 · 11.Epic - Acute EMR Clinicals 12.Epic - Ambulatory EMR Clinicals 13.GE (IDX) Centricity - Ambulatory EMR Clinicals 14.McKesson Horizon - Acute EMR Clinicals 15.McKesson Horizon Enterprise Visibility 16.Meditech 5.66 EHR w/DR 17.NextGen - Ambulatory Practice Management 18.Quality Systems (Next Gen) - Ambula
