Post on 08-Feb-2017
transcript
Overview of National Learning Health Community
& Learning Health forMichigan LandscapesJoshua Rubin & Timothy Pletcher
October 27th, 2015
Disclosure of Conflicts of Interest-Pletcher
• There are no personal conflicts• Dr. Pletcher has an adjunct faculty
appointment at the University of Michigan Medical School Department of Learning Health Sciences
• Dr. Pletcher also serves as the Executive Director for the Michigan Health Information Network Shared Service
Learning Health for Michigan2
Objectives
1. Understand the national framework for how the Learning Health Community is evolving abroad, in the U.S., and within Michigan
2. Become familiar with the LHS vision and the multi-stakeholder consensus LHS Core Values
3. Learn about other stakeholders spanning the health arena who are working toward collaboratively realizing this shared vision; discover how to join them by participating in the Learning Health Community movement at a national level or participate in Learning Health for Michigan (LH4M) effort
4. Gain insight into how the research and discovery networks are poised to integrate with traditional health care delivery data sharing infrastructure
5. Achieve awareness of the new technology and policy environments and approaches such as PopMedNetTM being used to enable distributed data sharing, as well as rapid learning leveraging the power of analytics
Learning Health for Michigan3
Acknowledgements
This material is based on the work and content provided by:
Charles P. Friedman, PhDJosiah Macy, Jr. ProfessorChair, Department of Learning Health Sciences
&Allen Flynn, Research InvestigatorDepartment of Learning Health Sciences
Learning Health for Michigan4
How Learning Happens : “Virtuous Cycles” of Study and Change
AssembleExperience Data
TakeAction
InterpretResults
AnalyzeData
Tailored Messagesto Decision-Makers
A Problem of Interest
Decision to Study
Learning Health for Michigan5
Members of the Platform Team
• Allen Flynn, University of Michigan
• Chuck Friedman, U-M Medical School
• Johmarx Patton, U-M Medical School
• Jodyn Platt, U-M School of Public Health
• Tim Pletcher, MiHIN • Peter Polverini, U-M School of
Dentistry• Josh Rubin, U-M Medical School
Learning Health for Michigan
CHRT Staff:Leah CorneailBabette Levy Ezinne Ndukwe
7
How to Learn Routinely: A Single Platform Supports Multiple Simultaneous “Virtuous Cycles”
DifferentProblems
Rapid Cycle
Slower Cycle
PLATFORM
Learning Health for Michigan8
In Other Words…
• Without a platform, each learning cycle develops its own, sub-optimal methods for learning; no economy of scale
• With a platform, all cycles share & benefit from a common infrastructure; costs are distributed
Learning Health for Michigan9
So What’s in a Complete Platform?
Mechanisms for managing
communities of interest
Learning Health for Michigan11
Platform Development
Domain
Story
Collection
Use cases
Standard “widgets”
Refinement/ Testing
Synthesis
Collection of use cases (across domains)
Domain
Story
Collection
Use cases
Learning Health for Michigan12
The Afferent and Efferent Sides of the Learning Cycle
A Problem of Interest
Afferent(BD2K)
Efferent(K2P)
Learning = BD2K + K2P
Learning Health for Michigan13
The LHS and Big Data
• The LHS is bigger than Big Data• Big Data addresses only the blue side of
the learning cycle• The LHS infrastructure must support
complete learning cycles
Learning Health for Michigan14
The LHS Must Do This
AssembleRelevant Data
Take Action to Change Practice
InterpretResults
AnalyzeData
Deliver Tailored Message
A Problem of Interest
Decision to Study
Learning Health for Michigan15
Not This
AssembleRelevant Data
Take Action to Change Practice
InterpretResults
AnalyzeData
Deliver Tailored Message
A Problem of Interest
Decision to Study
Journals?
Learning Health for Michigan16
Record decisions
Communicateadvice
Store knowledge
Formulate advice
Icon: Brain by Eovaro Atli Birgisson, The Noun Project, 2015.Flynn, 2015
LHS componentsto organize, manage and
provide access towhat is learned,
i.e., to knowledge.At scale, the Brain is a
Digital Library of Learning. There can be one such
library, or many.
The LHS Needs a Brain to Drive the Efferent Side
Objective:
Design and build a LHS brain
Store knowledge
Formulate advice
Learning Health for Michigan18
Specific Functions of a Brain
Basic Brain Functions
Organize knowledge to know what is known
Manage knowledge to know about what is known
Represent and provide knowledge for use
Advanced Brain Functions
Formulate tailored advice
Infer what is NOT YET known
Predict an individual’s immediate knowledge needs
Learning Health for Michigan19
No brain. Slow gain.• Publications are NOT ready for use. The knowledge they
contain has to be transformed into actionable knowledge.
• Evidence-based guideline development is slow. Guidelinedissemination is inadequate.
• RCT-level evidence is NOT available to guide most health caredecisions so learning from experience is a necessity thus a capacity to manage experiential knowledge is a necessity.
• Generating up-to-date, individualized, relevant, clear advice remains a difficult task
• “Inventory principle” - It is difficult to know what is known and NOT YET known unless knowledge can be assessed in aggregate
Learning Health for Michigan20
A brain contains knowledge
Examples of knowledge are…
Regression Equation
Clinical Calculation
Checklist
Template
Guideline
Predictive Model
Decision Model
Learning Health for Michigan21
Store knowledge
Formulate advice
- Now I can learn!
With a brain to contain, organize, and manage knowledge, our health system can be responsive, adaptive, effective and efficient.
What is a Digital Knowledge Object?
Attribution, versioning, and context comes from metadata.
Transactional capabilities afford (i) access and authorization controls, and (ii) direct interaction with executable code.
The knowledge contained in a DKO can be generally modeled using terms and relations amongst them – forming its ontology.
The knowledge can be specifically represented in one of more computable formats – R code, javascript, GEM, etc.
A digital knowledge object takes an instance of knowledge-in-the-world and adds digital metadata and transactional capabilities to it.
Learning Health for Michigan23
DKOs can be explicitly related or linked
Step 3InterrelatePieces of
Digital Knowledge in a
Knowledge Network
DKO networks afford new capabilities.Learning Health for Michigan
24
Types of Digital Knowledge Pieces
Interactivity
Agen
cy Logic in AutonomousSystems
Logic in Apps
Digital Knowledge Objects (DKOs) with Logic
Static Websites & PDFs
Icons: App by Garrett Knoll, Website by buzzyrobot, PDF by Laurent Canivent, The Noun Project, 2015.Flynn, 2015
consumed by
contain, link toor reference
evolution fromPDF to DKO
Learning Health for Michigan25
Digital Knowledge Object Maturity Levels
Static: A digital document (e.g., PDF file) that a person can
read
InteractiveAn “APP” that accepts inputs and provides outputs
Self-describingA DKO that describes its role and uses in metadata
SemanticA DKO with explicit links to known terms or concepts
NodalA DKO node that has defined relations to other DKOs
passive - narrative
active - transactional
automatically disseminable
Learning Health for Michigan26
Store Knowledge and Formulate Advice
• Semantically aware queries of DKOs• Automated queries based on individual features• Inference over any DKO space to identify what is NOT YET known• Digital DKO libraries online
- Versioning- Governance- Curation
• Knowledge stored and linked in various forms, includingstatic forms and transactional, coded, computable forms
• A platform to create advice-giving systems of all kinds• A necessary component of a Learning Health System
Fedora Repositories of Digital Knowledge Objects (DKOs)
Learning Health for Michigan27
Fedora is a digital knowledge repository
• An open source management system for digital content
• Scalable knowledge engineering and management system
• Ready solution that speeds up LHS “brain” development
• Proven system already in use by libraries worldwide
• Fedora’s creators are faculty and staff now at U-M
Precursor A
CreateDigital
KnowledgeRepositories
https://wiki.duraspace.org/display/FF/Fedora+Repository+Home
We are not starting from scratch.
Learning Health for Michigan28
How to create a “brain” for the LHS
Precursor A Precursor B Step 1 Step 2 Step 3
CreateDigital
KnowledgeRepositories
Make DigitalKnowledge as
Explicit andTransactional
as Possible
Wrap Piecesof Digital
Knowledge inDescriptiveMetadata
Associate Pieces of
Digital Knowledge
with Terminologies & Ontologies
InterrelatePieces of
Digital Knowledge in a
Knowledge Network
Manage knowledge to know about what is known
Formulate Tailored Advice
Infer what is NOT YET stored
Represent and provide knowledge for use
Predict knowledge needs
Organize knowledge to know what is known
Store what is known in a way that it persists and is always accessible
Learning Health for Michigan29
Acknowledgements
The PopMedNet content was made available courtesy of:
Jeffrey Brown, PhDMichael Klompas, MD, MPHMDPHnet Research Team
Learning Health for Michigan30
Connection to the Blue Side
Learning Health for Michigan
The PopMedNet™ software application enables simple creation, operation, and governance of distributed health data networks.
31