Strategies for Unlocking Knowledge Management in Microsoft 365 in the Copilot...
Elsevier's Healthcare Knowledge Graph: An Actionable Medical Knowledge Platform to Power Diverse Applications
1. Maulik R. Kamdar, PhD
Senior Data Scientist, Health and Commercial Markets
Knowledge Graph Conference, 3rd
- 6th
May 2021
Elsevier’s Healthcare
Knowledge Graph
An Actionable Medical Knowledge
Platform to Power Diverse Applications
2. About Elsevier
2
Publish Physical
Books/Journals Digital
Analytics
& Decision Support
Research
Intelligence
Clinical
Solutions
Education
Research
Platform
Books &
Journals
R&D
Solutions
A global provider of information analytics
and decision support tools for professional
and business customers
4. Themes of this talk
4
Introduction
Medical challenges and
the Elsevier’s Healthcare
Knowledge Graph
Methods
Graph enrichment,
curation, and projection
methods
Applications
Medical applications
powered by the graph
and the road ahead
5. Themes of this talk
5
Introduction
Medical challenges and
the Elsevier’s Healthcare
Knowledge Graph
Methods
Graph enrichment,
curation, and projection
methods
Applications
Medical applications
powered by the graph
and the road ahead
6. Questions asked by clinical experts
Entity Types:
• Drugs
• Diagnoses
• Diseases
• Phenotypes
• Symptoms
• ...
Del Fiol, G., Workman, T. E., & Gorman, P. N. (2014). Clinical questions raised by clinicians at
the point of care: a systematic review. JAMA internal medicine, 174(5), 710-718. 6
7. Elsevier’s Healthcare Knowledge Graph
400k concepts
8m relationships
75k diseases
46k drugs
63k procedures
90k symptoms
Elsevier’s Healthcare Knowledge Graph
connects the world’s healthcare concepts and
relationships supported by evidence in content,
and unlocks the knowledge through scalable,
easily-navigable information services.
7
8. Elsevier’s Healthcare Knowledge Graph
• Related medical concepts
• Mappings to external
Terminologies
• Clinical relations between
different concepts from
multiple sources
• Supporting literature and
external references
• Multi-lingual content
The Visualizer is a prototype of synoptic content in Elsevier’s Healthcare Knowledge Graph. 8
9. Poly-hierarchical taxonomy and rich medical relations
9
What is the drug of choice for asthma?
What is the cause of physical finding tachypnea?
10. Capturing and representing more information
• Capture additional information and context
surrounding medical names and semantic
relations between different medical entities.
• Context may include :-
• Supporting snippets in medical textbooks
and other Elsevier documents
• Cohort information (age, sex, ethnicity)
10
11. Medical knowledge is continuously increasing …
11
Densen P. Challenges and opportunities facing medical education. Trans Am Clin Climatol Assoc. 2011;122:48-58.
https://www.clinicalkey.com/#!/browse/books
12. Themes of this talk
12
Introduction
Medical challenges and
the Elsevier’s Healthcare
Knowledge Graph
Methods
Graph enrichment,
curation, and projection
methods
Applications
Medical applications
powered by the graph
and the road ahead
14. Elsevier-Stanford collaboration to develop the
open-source WebProtégé platform to accommodate
the size and complexity of the knowledge graph
WebProtégé knowledge curation platform
14
https://webprotege.stanford.edu/
Code: https://github.com/protegeproject/webprotege
15. Continuous
enrichment
and curation
15
Medical Literature
Summaries, Textbooks,
Images, Journals, etc.
Curation
Interface
ML/NLP
Pipelines
Graph Database Projections
and Services
ML/NLP methods for tagging and extraction
• Relation identification methods
• Relation extraction methods
• Topic models
• Tagging images with medical concepts
16. Establishing inter-reviewer consensus is difficult
16
No Consensus Achieved
Relation Excerpt
Asthma
has drug
Epinephrine
Chapter: Asthma
Section: Basic Information
Status asthmaticus, or acute severe asthma, is a
refractory state that does not respond to standard
therapy such as inhaled beta-agonists or
subcutaneous epinephrine.
Diabetes Mellitus
has clinical
finding Nausea
Chapter: Diabetes Mellitus
Section: Treatment - General Rx
Nausea is its major side effect.
Kamdar, Maulik R., et al. "Text snippets to corroborate medical relations: an unsupervised approach using a
knowledge graph and embeddings." AMIA Summits on Translational Science Proceedings 2020 (2020): 288.
Expected Actual
17. 17
Medical Literature
Summaries, Textbooks,
Images, Journals, etc.
Curation
Interface
ML/NLP
Pipelines
Graph Database Projections
and Services
Manual
Tagging
Subject matter experts manually tag content with concepts
and relations from the knowledge graph through an interactive
authoring interface. These manually tagged excerpts are
ingested back in the graph with provenance information.
Continuous
enrichment
and curation
18. 18
Medical Literature
Summaries, Textbooks,
Images, Journals, etc.
Curation
Interface
ML/NLP
Pipelines
Graph Database Projections
and Services
Legacy Databases
Drugs, Statistics, etc.
Manual
Tagging
ETL
Pipelines
Continuous
enrichment
and curation
19. 19
Medical Literature
Summaries, Textbooks,
Images, Journals, etc.
Curation
Interface
ML/NLP
Pipelines
Graph Database Projections
and Services
Legacy Databases
Drugs, Statistics, etc.
Manual
Tagging
ETL
Pipelines
Build and Deployment Automation
Continuous
enrichment
and curation
20. Knowledge projections and services
• Information stored in the knowledge graph is made available through
projections (i.e., extracted subgraphs in different formats) and API services
• API services provide JSON and JSON-LD content and enable developers to
provide different parameters for different representations and locales
• Remove the steep learning requirements for other developers to learn
graph queries and the underlying model!
20
Build and Deployment Automation
DeJong, Alex, et al. "Elsevier's Healthcare Knowledge Graph and the Case for Enterprise Level Linked Data
Standards." International Semantic Web Conference (P&D/Industry/BlueSky). 2018.
21. Themes of this talk
21
Introduction
Medical challenges and
the Elsevier’s Healthcare
Knowledge Graph
Methods
Graph enrichment,
curation, and projection
methods
Applications
Medical applications
powered by the graph
and the road ahead
22. Applications powered by the knowledge graph
Search Services (e.g., ClinicalKey Search Service)
• Reference searches, focused clinical queries at point of care, question answering
Recommendation Services
• Recommend clinical guidelines, clinical calculators, and algorithms for readers
Clinical Decision Support Services
• Provide knowledge on the diagnosis or treatment steps given patient trajectory
22
And several other challenges in authoring platforms, education, pharmacovigilance, differential diagnosis, etc.
Kamdar M.R, et al. Focused Clinical Search through Query Intent Interpretation and a Healthcare Knowledge Graph (November 5,
2020). Proceedings of the 4th Annual RELX Search Summit, Available at SSRN: https://ssrn.com/abstract=3775468
23. Challenges for adoption of the knowledge graph
23
• Making the healthcare knowledge graph tangible
to diverse stakeholders
• Informing developers and engineers on how to
use and query the graph in their applications
• Informing informaticists and product owners on
the value of adopting the graph in their product
• Keeping the graph complexity simple
(reification!), while emphasizing the benefits of a
knowledge graph over conventional approaches
• Knowledge graph platforms need constant
maintenance and improvements
24. Unlocking knowledge to drive discovery
24
Elsevier has vast amounts
of knowledge locked up in
human readable form
We use machine learning
and innovative curation
tools to extract and
represent this knowledge
We generate and provide
projections and services for use
of this knowledge to power
diverse clinical applications in
different regions
Medical Literature
Clinical Applications
Adding context data
such as EHR or usage
data for personalization
and predictive analytics
25. Integrated knowledge graphs in the future
25
Structured and
unstructured content
across Elsevier
Several other knowledge graphs
such as Entellect, Omniscience,
etc. developed for other domains
Entellect: https://www.elsevier.com/solutions/entellect
Malaisé, Véronique, et al. "OmniScience and Extensions–Lessons Learned from Designing a Multi-domain,
Multi-use Case Knowledge Representation System." European Knowledge Acquisition Workshop. 2018.
Knowledge-driven
applications
26. Elsevier: Maulik R. Kamdar, Linda Wogulis, Cailey Fitzgerald, Will
Dowling, Danielle Walsh, Doug Anderson, Alex Ausio, David Childs,
Sravanthi Tummala, Tan Nguyen, Connor Skiro, Chris Stoces,
Katie Scranton, Veronique Moore, Paul Snyder, David Conrad,
Craig E. Stanley Jr., Rinke Hoekstra, Steve Ross, Alex De Jong,
Mev Samarasinghe, Dru Henke, Rhett Alden
Acknowledgments: Matthew Horridge, Rafael Goncalves, Mark Musen
Contact:
Maulik R. Kamdar
Senior Data Scientist, Elsevier Health
Email: m.kamdar@elsevier.com
Twitter: @maulikkamdar