Downloaded 64 times
Uploaded byHealth Catalyst
The Health Catalyst Data Operating System (DOS™): Lessons Learned and Plans for the Next Evolution
The document discusses the Health Catalyst Data Operating System (DoSTM) and its evolution since its launch in January 2017, highlighting its significance in healthcare data utilization and the ongoing challenges in integrating various data systems. It outlines the goals of DoSTM, including interoperability, data analytics, and AI, while emphasizing the importance of a unified data architecture for improving healthcare outcomes. Additionally, it provides insights into lessons learned and future plans for enhancing the platform and its applications.
In this document
Powered by AI
Introduction to Health Catalyst DOS, discussing its vision, evolution since Jan 2017, and objectives.
Explains how DOS fits in tech stack, emphasizing the importance of data utilization and interoperability.
Discusses future learning from experiences, achieving measurable improvements, and integrating insights into practice.
Highlights challenges in integrating disparate systems and the limitations of specific modeling-focused platforms.
Insights from other industries and case studies (like Netflix) on data utilization to drive business performance.
Introduces frameworks for data engineering, emphasizing the evolution of data operations and pipeline management.
Showcases data sources available in DOS and engages the audience with a Q&A session.
More Related Content
Similar to The Health Catalyst Data Operating System (DOS™): Lessons Learned and Plans for the Next Evolution
More from Health Catalyst
The Health Catalyst Data Operating System (DOS™): Lessons Learned and Plans for the Next Evolution
- 1. The Health CatalystData Operating System (DOS™): Lessons Learned and Plans for the Next Evolution Bryan Hinton, CTO Dale Sanders, Senior Advisor September 30, 2020
- 2. © 2019 Health Catalyst • Dale:Looking back • Review the “why” behind DOS… the concept and vision • Bryan: Looking forward • How far have we carried the vision? • What’s left and will it ever end? Let’s Dive In… 2
- 3. © 2019 Health Catalyst • Changeof the guard, old to new… • DOS was a very ambitious vision and endeavor that we publicly launched less than four years ago… Jan 2017 • Here’s the status update • DOS is about Health Catalyst, but it is also about new concepts, and thinking differently about the utilization of data in healthcare and life sciences, and laying the foundation for digital health • Has the concept been validated or not? Why are we giving this webinar today… 3
- 4. © 2019 Health Catalyst Today mightsound more “sales-y” than usual for one of my webinars, so apologies are offered, and forgiveness appreciated. We will do our humanly best to give you the truth… you can decide whether it’s a sales pitch or not Today Might Sound ”Sales-y” Klaus Voormann 4
- 5. © 2020 Health Catalyst Generally speaking,computer science hasn’t addressed the last and critically important layer in the technology stack, especially for the incredibly complex world of healthcare and life science data. Where Does DOS Fit in the Tech Stack? User Interface Application Software Operating System Infrastructure 5
- 6. © 2020 Health Catalyst • Applicationswill come and go, but data is here forever • Free your data, free your mind • There’s a big opportunity to locally innovate at the application layer, by leveraging the data you already paid handsomely to collect • Separate technology infrastructures for managing the same data is a violation of common sense • It’s expensive and leads to all sorts of inconsistencies in data management • There are, essentially, three missions for data… A Few Assertions 6
- 7. © 2020 Health Catalyst InteroperabilityData Analytics &AI Data-First Application Dev Three Missions of Data The Health Catalyst Data Operating System (DOS™) is a single, cloud- based, API-based architecture with a common, consistent layer of data content, to support the Three Missions of Data… Workflow transactions and ambient analytics in the same software user experience Interoperability, portability, and source-to-subscriber, transaction-level exchange of data 7
- 8. © 2020 Health Catalyst • Theorigin of the Data Operating System (DOS™) concept is this cartoon • The digital world of health is largely stuck in the lower left two clouds • Our digital understanding of the patient requires the entire data ecosystem • Digital health is just beginning, and DOS is the enabling platform The Human Health Data Ecosystem 8
- 9. © 2020 Health Catalyst The HealthcareAnalytics Adoption Model Level 8 Direct-to-Patient Analytics & Artificial Intelligence Analytics and AI tools available to patients to empower their health decisions. Digital Twin and “Patients Like Me” is a core capability, including social networking with similar patients as extended members of the care team. Level 7 Personalized Medicine & Prescriptive Analytics Patient care informed by analytics; decision support based on data from SDOD, genomics, and patient activation (willingness x ability) in their care. Population-based protocols customized per patient. Level 6 Population Health Management & Predictive Analytics Using predictive models and clinical suggestive analytics to support risk intervention. Risk reduction models are informed by economic models that reward prevention over treatment. Level 5 Waste & Care Variability Reduction Reducing variability in care processes. Focusing on internal optimization and waste reduction; strong focus on reducing low-value care and adverse events. Level 4 Automated External Reporting Efficient, consistent production of reports required for reimbursement, compliance, accreditation, etc.; adaptability to changing requirements. Level 3 Automated Internal Reporting Efficient, consistent production of basic financial and clinical operations reports; widespread availability in the organization. Level 2 Standardized Vocabulary & Patient Registries Curating, organizing, and standardizing core data content. Reusability of data models and analytics logic balanced with late binding. Level 1 Enterprise Data Operating System Collecting and integrating the core data content. Level 0 Fragmented Point Solutions Tolerating inefficient, inconsistent versions of the truth; cumbersome internal and external reporting. 9
- 10. © 2020 Health Catalyst Our SolutionsUse Data, Analytics, & Expertise to Catalyze Improvement Analytics Applications: A robust set of applications, built on top of DOS, that generate meaningful insights for improvement Professional Services 2 The Data Operating System (DOSTM ): A healthcare-specific, open, scalable platform for analytics, application development, and interoperability 1 Activity- based Costing (CORUSÔ Suite) Patient Safety (Patient Safety MonitorÔ Suite) Foundational Software Applications Domain-Specific Software Suites Services Expertise: Analytical, clinical, financial and operational experts facilitate and accelerate measurable improvement3 Clinical, Financial, and Operational Domain Experts Analysts, Data Scientists, and Data Engineers Strategic Consulting ž Readiness Assessment ž Opportunity Analysis ž Governance ž Outcomes Improvement ž Population Health ž Technical Support ž Training Terminology Services Text Processing Real-time Streaming & Interoperability Machine Learning Big DataClosed-Loop EHR IntegrationData Warehouse Reusable Data Logic Cloud-based Source Connectors Tailored Analytics Accelerators Care Management Benchmarking (TouchstoneÔ Suite) Registries & Measures Authoring (Population Builder) Financial & ACO (Payment Model Analyzer, Financial Management, Revenue Cycle, Hierarchical Condition Categories, etc.) Operational (Supply Chain, Patient Flow, Surgical Services, Labor Management, Practice Management, etc.) Population Health Foundatio ns Clinical & Patient Safety (Sepsis, Readmissions, Heart Failure, Joint Replacement, CLABSI,COPD, etc.) Quality and Regulatory Measures Dashboards & Reporting (Leading WiselyÔ) 10
- 11. © 2020 Health Catalyst AI Algorithmsare Commodities, Data Platforms are Not “…it is dangerous to think of these quick wins as coming for free. Using the software engineering framework of technical debt, we find it is common to incur massive ongoing maintenance costs in real-world ML systems.” Neural Information Processing Systems (NIPS) Advances in Neural Information Processing Systems 28 (NIPS 2015) 11
- 12. © 2020 Health Catalyst The MachineLearning Code, in the Black Box, is a Small Fraction of the Machine Learning Investment and Ecosystem 12
- 13. AI/Data Science Commoditize access to AI models IDEA Addmissing data DOS Operations Console Schedule, define, monitor, and troubleshoot ETL processes Atlas Data Governance Catalog, manage, and explore existing analytics, data, and building blocks Metadata & Task Management Health Catalyst DOS Ecosystem Touchstone (Benchmarking) Leading Wisely (Visualization) Data to the Edges: Rapid Response Analytics Closed Loop EHR Decision Support Data published back to DOS Population Builder Client- developed analytics & apps 3rd-party apps Data published back to DOS Patient Safety Monitor Population Health Foundations CORUS Suite (Activity-Based Costing) Able Health (Regulatory Measures) 45+ Analytic Accelerators Data published back to DOS Data to the Domains Subsets of data for specific analytic use cases & standardized terminology Level 1: DOS Marts Clinical, cost, claims, etc. Level 2: Population SAMs Sepsis, diabetes, CHF, COPD, etc. Customized SAMs D O S H o m e p a g e i n A t l a s t i e s i t a l l t o g e t h e r Measures Manager View and manage all measures in one place Hundreds of compulsory & internal measures definitions Requirements drive data content needs Data Acquisition, Movement, Curation, Governing, and Monitoring >300 data sources Text Processing Integrate text & discrete data DOS Data Lake Data Processing Engine Subject Area Mart Designer Aggregate & manage data Data Processing Engine Source Mart Designer Enable real-time data ingestion Raw text Data of all types HC Interoperability Community-based data integration Community data Precision Medicine Suite
- 14. © 2020 Health Catalyst From aJanuary 2017 lecture 14
- 15. © 2020 Health Catalyst One ofMy Observations in Life The greatest leaders, companies, and organizations in history exhibit chronic, constructive dissatisfaction. Bryan Hinton personifies this trait. 15
- 16. Learnings and FuturePlans
- 17. © 2020 Health Catalyst Healthcare AnalyticsAdoption Model Variation Reduction Improve Health Efficiency 17
- 18. © 2020 Health Catalyst The DataOperating System Data Ingest Real-time Streaming Source Connectors Catalyst Analytics Platform Fabric Data Services Real-time Processing Health Catalyst Applications Data Quality Data Governance Pattern Recognition Hadoop/ Spark Data Export Population & Registry Builder Leading Wisely Care Management Atlas Client-built Applications NLP Touchstone Benchmarks CORUS Cost Accounting Patient Safety Measures Manager ACO Financials Patient Engagement HL7 Data Pipelines Metadata Data Lake Reusable Content AI Models Third-party Apps Artificial Intelligence Pipelines Marketplace SAMD & SMD Fabric Application Services Terminology & Groupers EMR Integration Security, Identity & Compliance Patient & Provider Matching Value Sets & Measures Standard, Extensible Data Models RegistriesFHIR HL7 Analytic Accelerators 18
- 19. Healthcare Analytics AdoptionModel Data First Application Platform & Late-Binding Capabilities Terminology Infrastructure & Standard Data Models “Interoperability Foundation” Advanced Analytics, Data Science, Big Data
- 20. © 2020 Health Catalyst Learning andEvolving with Documented Results with Improvement ON THE GROUND EXPERIENCE From 2018 to 2020 YTD • 1,000 documented financial, operational, and clinical improvements • Estimated Economic valuation of $407M 20
- 21. © 2020 Health Catalyst Learning andEvolving from Industry Analysts INDUSTRY ANALYSTS 21
- 22. © 2020 Health Catalyst Tool Integrationand Scaling Challenges Disparate Systems for: • Streaming • Machine Learning • Traditional ETL • Data Curation • In-memory Analytics Causing: • Focus on tool integration and interoperability and difficult to scale operations rather than value creation. Providing: • Technical capability with limited healthcare knowledge built-in Resulting in: • Large, sustained custom technology support costs • With limited business success 22
- 23. © 2020 Health Catalyst • Technologyfocused • Limited to domain specific content • Very flexible, difficult to scale Open Data Platforms and Tools 23
- 24. © 2020 Health Catalyst • Tendto be focused around a specific model (large or small) • Sometimes can ingest multiples sources but are limited for non- model focused analytics Examples: • Source System Analytic Capabilities (EMR, Finance, ERP) • Population Health Platforms Closed or Niche Data Systems 24
- 25. © 2020 Health Catalyst Learning andEvolving from Other Industries OTHER INDUSTRIES 25
- 26. © 2020 Health Catalyst Many businessesuse data to support their decisions instead of drive their actions. But why? After all, data is really only valuable if you can translate it into actionable insights. Gaining these insights starts with figuring out what you want from your data finding its value. Here we talk about the questions you should ask regarding the context, need, vision and outcome of your data, and we offer a helpful framework for turning that data into meaningful stories and business successes. In 1910, Scottish writer and poet Andrew Lang said, "He uses statistics as a drunken man uses lampposts—for support rather than illumination." Decades later, many modern businesses still do just that, using data to support rather than drive their decisions. Daniel Waisberg https://www.thinkwithgoogle.com/marketing-resources/data- measurement/data-to-insights-blueprint-for-your-business/ From data to insights 26
- 27. © 2020 Health Catalyst Learning andEvolving from Other Industries OTHER INDUSTRIES 27
- 28.
- 29. © 2020 Health Catalyst Data Mesh DataOrchestration Layer 29
- 30. © 2020 Health Catalyst Software EngineeringData Engineering Domain Driven Design Data Mesh Service Mesh Data Orchestration Layer Declarative programming Declarative Dataflows Reactive Programming Data Driven Execution Unit Testing Data Quality and Business Rule Testing DevOps DataOps Interfaces Schema Dynamic Programming Late-Binding and Schema on Read Static Typing Schema on Write Code Centric Code Centric Bytecode Directed Acyclic Graphs (DAGs) Software Engineering applied to Data Engineering 30
- 31. © 2020 Health Catalyst Learning andEvolving ON THE GROUND EXPERIENCE INDUSTRY ANALYSTS SOLUTIONS FROM OTHER INDUSTRIES 31
- 32. © 2020 Health Catalyst Data SuccessFramework How to turn data into insight-driven decisions at scale Generate Govern Deliver Acquisition Analysis & Engineering Operations Literacy Quality Access Reporting & Analytics Modality Any Location 32
- 33. © 2020 Health Catalyst Data SuccessFramework Capabilities Generate Govern Deliver Acquisition Analysis & Engineering Operations Literacy Quality Access Reporting & Analytics Modality Any Location Structured & Unstructured Streaming/IOT Ambulatory Master Data Patient Matching Terminology Self-Service Individual Desktop Clinical Workflows EMR Notification (mobile, email, phone) Industry Standards (HL7v2, CCD, FHIR, Direct) API Support (FHIR, etc.) 33
- 34. Level 9 Direct-to-patient Analytics& Artificial Intelligence User Generated Content Streaming/IoT API Support De-Identification Intelligent Execution Mobile Level 8 Personalized Medicine & Predictive Analytics SQL, R, Python, Spark Support Records API Level 7 Clinical Risk Prevention & Predictive Analytics NLP & Data Science Extensibility Automated Optimization Recommendations Profiling Personalized & Automatic Viz Level 6 Population Health Mngt & Suggestive Analytics Ambulatory Executable Content Declarative Dataflows Measure Catalog Analytics Catalog Population Catalog Business Rules Scalable Access Management Guided Analytics Self Service Notifications Level 5 Waste & Care Variability Reduction Integrated Data Quality Declarative Dataflows Adaptive Data Modeling Measures Framework Alerting 24x7 Support Data Lineage Training Data Platform Integration Data Quality Reporting Stewardship Analytic Applications Limited Viz Layer Business Logic Business Workflows Clinical Workflows Level 4 Automated External Reporting 3rd Party Organizations Level 3 Automated Internal Reporting DataOps Support Scriptable Column Level Security Burstable Reporting Web Desktop Apps Conference Room EMR Level 2 Standardized Vocabulary & Patient Registries Structured & Unstructured Source System Connectors Patient Matching Pragmatic Data Binding Business Glossary Terminology Level 1 Enterprise Data Operating System ELT RBDMS & Data Lake Support Master Data Common Operations Console Data Catalog Row Level Security Auditing Dashboards Industry Standards Individual Desktop Level 0 Fragmented Point Solutions Excel Support Flat File Extracts
- 35.
- 36.
- 37. © 2020 Health Catalyst Anatomy ofa Data Pipeline Business Logic Data Persistence Code Persistence Optimization Schema Creation Data Validation Framework Integration Incremental Loading Management Full Load Path Dependency Management Logging Schema Evolution Index Management Execution Optimization Notifications Restartability 37
- 38. © 2020 Health Catalyst Declarative natureof DOS Bindings Business Logic SQL, R, XQuery, Delimited Python, Spark Data Persistence Code Persistence Optimization Schema Creation Data Validation Framework Integration Incremental Loading Management Dynamic Switching between Full/Incremental Dependency Management Logging Schema Evolution Index Management Execution Optimization Notifications Restartability What users do What DOS does Significant code reduction for both development and maintenance and improved SLA due to optimized execution 38
- 39.
- 40.
- 41. © 2020 Health Catalyst 320+ Pre-mappedData Sources 41
- 42.
- 43.
- 44.
- 45.