Identifying Drug Interaction Candidates in Real-World Data
1 like205 views
The document describes the Real-world Evidence Drug-Drug Interaction (REDDI) Match project. It introduces the multidisciplinary team developing the solution. REDDI Match analyzes real-world data to identify potential drug-drug interactions, including for off-label drug uses like chloroquine for COVID-19. It represents prescription data as a graph to efficiently identify interaction patterns. The team detects communities in the graph and explores interactions within those contexts. A demo of the system is provided with notes on potential future enhancements.
Identifying Drug Interaction Candidates in Real-World Data
- 1. CONFIDENTIAL Kathleen Mandziuk, MPH, RN | Nathan Smith | Kerry Deem 30 June 2020 Real-world Evidence Drug-Drug Interaction (REDDI) Match
- 2. CONFIDENTIAL DATA PATIENT TECHNOLOGY • 360o Patient View • Data-driven decision making • Patient-centric trial design • Pathway • Fit-for-purpose • Integrated Data, Patient, Technology 2
- 3. CONFIDENTIAL PRA’s Layered ApproachTo Real World Evidence Clinical Info • Claims • Electronic Medical Records • Existing Registries • HEOR SE C O N D A R Y D A T A FULLEVIDENCE • Registries • Natural histories • Retrospective Chart Review • Clinical Studies • Long-Term Follow-Up P R IM A R Y D A T A C O L L E C T IO N Patient Experience • Social Listening • Personas • Interviews • Surveys C O M P L E M E N T A R Y D A T A TECHNOLOGY&INNOVATION 3
- 4. CONFIDENTIAL KevinRooney SeniorManager, Information Technology NathanSmith SeniorPrincipal DataScientist Kerry Deem AssociateDirector, Programming AnushaChockalingam Director, Architectureand R&D AshwinKumar Rai SeniorPrincipal Data Scientist DavidQuach SeniorPrincipal DataScientist Meet the REDDI Development Team KathleenMandziuk Vice President, PatientStrategy andDigitalHealth UshitaPalande Machine LearningDeveloper 4
- 5. CONFIDENTIAL “Todaywith the further specializationin medicalcare patients are seeing more than one provider, and likely multiple treatments, this combined with the use of OTC medications and supplements manypatients are on 5 or more products which statisticallyapproaches100% likelihood of a drug-drug interaction, the question is how clinically significantis it?” Frederick T. Lewis, D.O. The Clinical Perspective Dr. Fred Lewis Vice President, Scientific Affairs 5
- 6. CONFIDENTIAL Clinical Implications of Drug-Drug Interactions (DDI) REAL-WORLD EVIDENCE DRUG-DRUG INTERACTION (REDDI) MATCH 6
- 7. CONFIDENTIAL • Clinical Trials • Spontaneous Reporting • Clinical Experience • GAP = Broad, diverse population not measured Current DDI Reporting Methods REAL-WORLD EVIDENCE DRUG-DRUG INTERACTION (REDDI) MATCH That reality check could come in the form of disappointing results when AI products are ushered into the real world. Even Dr. Eric Topol, the author of “Deep Medicine: How Artificial Intelligence Can Make Healthcare Human Again,” acknowledges that many AI products are little more than hot air. “It’s a mixed bag,” he said. https://khn.org/news/a-reality-check-on-artificial-intelligence-are-health-care-claims-overblown/ Post- Marketing "Real World" Clinical Trials In Vitro Testing Spontaneous Reporting 7
- 8. CONFIDENTIAL Nursing Epidemiology Clinical Research Physician Data Science AI Machine Learning Therapeutic Expertise Diverse MultidisciplinaryTeam 8
- 9. CONFIDENTIAL Real-world Evidence Drug-Drug Interaction (REDDI) Match 30 June 2020 The Solution
- 10. CONFIDENTIAL Data for February through April of 2020 160 millionpatients 1.1 billionprescriptions 640 milliondiagnosis codes Subsets −293,000 patientsnationwidetaking chloroquineswithout prior diagnosis for auto- immune diseases −1.7 millionpatientsin New York, New Jersey, and Connecticut for network analysis 10 Data Volume Context CONFIDENTIAL
- 11. CONFIDENTIAL Innovation: Graphical Representationof DDI Representing prescriptions, transactions, and patientsin a graph facilitatesidentification and aggregation of possible DDI patterns Consumable for a wider audience 11
- 12. CONFIDENTIAL Innovation: Focus on COVID-19 An approachthat is especially relevantwhen drugs are used in new waysdistantlyrelated to previousmachine learning training examples 12
- 13. CONFIDENTIAL Aggregate Patient Counts AcrossThe Graph Efficient Way Of Simultaneously Evaluating Multiple Combinations 13
- 15. CONFIDENTIAL Identify communities of people with similar prescriptions and diagnoses Explore drug-drug-interactions in community context Identify potentially undiagnosed patients based on similarity to neighbors Detect Communitiesin the Graph 15
- 16. CONFIDENTIAL • Node similarity algorithm to mutate graph • Louvain community detection algorithm to write community properties • Watch What is GDS and Neo4j's GDS Library session from data science Connections event Applying the Graph Data Science Library 16
- 17. CONFIDENTIAL Real-world Evidence Drug-Drug Interaction (REDDI) Match 30 June 2020 Demo
- 18. CONFIDENTIAL Potential Future Enhancements Confounding factors (age, gender, ethnicity) Impact of co-morbidity Interactions of more than two drugs Broader searches across time and geography Incorporate machine learning methods that infer from known interactions 18
- 19. CONFIDENTIAL Contact Information 19 Nathan Smith Senior Principal Data Scientist Kerry Deem Associate Director, Programming Kathleen Mandziuk Vice President, Patient Strategy and Digital Health MandziukKathleen@prahs.comDeemKerry@prahs.com SmithNathanA@prahs.com