Building Data Capacity for Patient-

Centered Outcomes Research:

Perspectives from the OHDSI community
Patrick Ryan, PhD
Vice President, Observational Health Data Analytics, Janssen
Research and Development
Assistant Professor, Adjunct, Department of Biomedical
Informatics, Columbia University Medical Center
 OHDSI to advance the science of observational research

GLOBAL OHDSI NETWORK - Industry, Academia, Government

>2,100 >150 >2B
Researchers in Databases Patient records
epidemiology, Included Represented
statistics, informatics,
health policy, and
• OHDSI is an open science community, founded by clinical sciences
scientists at J&J, Columbia University, Stanford,
UCLA, Regenstrief Institute, and Iqvia OMOP DRIVING RWE
EHR
• OHDSI is an open multi-stakeholder, Clinical
A
interdisciplinary collaborative with a mission to OMOP Characterization
improve health by empowering a community to COMMON
collaboratively generate the evidence that promotes DATA MODEL
Claims
better health decisions and better care
+ B
Population-level
• OHDSI is driving development and adoption of open effect estimation
community data standards, open source analysis OPEN
software, and open science best practices within SOURCE
Clinical Data
ANALYSIS
regulators, academia, industry, payors, and health CODE Patient-level
C
systems prediction

2
 OHDSI Community
OHDSI data network
OHDSI collaborations
OHDSI Data partner 1
Open community
data standards
(OMOP CDM)

Firewall
Source data in Standardized Standardized
local structure ETL patient-level analytics
database Open source
and vocabularies (OHDSI tools)
(OMOP CDM) development
(OHDSI tools)

OHDSI Data partner 2 Methodological

Firewall
research
Source data ETL OMOP CDM OHDSI tools

Clinical evidence
OHDSI Network studies generation
OHDSI Data partner 3 Pre-specified

Firewall
protocol with
Source data ETL OHDSI tools analysis specification
OMOP CDM

OHDSI Data partner n Standardized

summary Collaborative Evidence

Firewall
statistics results Interpretation dissemination
Source data ETL OMOP CDM OHDSI tools
repository

Hripcsak G, Schuemie MJ, Madigan D, Ryan PB, Suchard MA. Drawing Reproducible Conclusions from Observational Clinical Data with OHDSI. Yearb Med Inform. 2021 Apr 21.
 Data standards are a means to an end, not an end in itself

• What data standards could make • What evidence would be useful to

the PCOR data infrastructure improve health policy and health care,
more useful for research and which could be reliably generated by
other data needs? the PCOR data infrastructure?

• What data standards are likely to • How can data standards enable real-
become more relevant looking world analytics to meet the relevant
forward? evidence needs moving forward?

• What needs to be prioritized? • What needs to be prioritized?

 Desired attributes for reliable evidence
Desired Question Researcher Data Analysis Result • Data standards are
attribute
necessary to enable
Repeatable Identical Identical Identical Identical = Identical replicability,
generalizability, and
Reproducible Identical Different Identical Identical = Identical robustness
• Data standards without
Replicable Identical Same or Similar Identical = Similar
standardized analytics
different are not sufficient to
ensure reliable evidence
Generalizable Identical Same or Different Identical = Similar • Standards limited to US
different data are harmful for US
Robust Identical Same or Same or Different = Similar population; public
different different health questions require
global data to generate
Calibrated Similar Identical Identical Identical = Statistically global evidence
(controls) consistent
 Delineating the roles of standards to enable evidence
What evidence would be
How can data standards enable real-world analytics to meet
useful to improve health
the relevant evidence needs moving forward?
policy and health care?

Characterization:
Data standards Data standards Analytics Who are the patients exposed to
medical interventions?
to enable data to harmonize standards How often do outcomes occur
exchange data structure to generate and amongst those patients?
and enable disseminate
analytics evidence Estimation:
Is the risk of outcome causally
related to exposure?
How does the risk compare with
alternative interventions?

Vocabulary standards
to harmonize data content and enable analytics
Prediction:
Which risks can be actionably
predicted with available data?
Which patients are at highest risk of
adverse events?
Open community data standard to enable analytics:
OMOP Common Data Model Standardized health
Person Standardized metadata
Observation_period system data
CDM_source
Location
Visit_occurrence Metadata
Location_history
Visit_detail
Care_site
Standardized
Condition_occurrence
Provider vocabularies
Standardized clinical data

Drug_exposure Concept
Standardized derived
Procedure_occurrence elements Vocabulary
Condition_era
Device_exposure Domain
Drug_era
Measurement Concept_class
Dose_era
Note Concept_relationship
Results Schema
Note_NLP Cohort Relationship

Cohort_definition Concept_synonym
Survey_conduct

Observation Standardized health Concept_ancestor

economics
Specimen Source_to_concept_map
Cost

Fact_relationship Payer_plan_period Drug_strength

 OHDSI’s standardized vocabularies
to harmonize data content and enable analytics
• 164 Vocabularies across 41 domains
– MU3 standards: SNOMED, RxNorm, LOINC
– Disparate sources: ICD9CM, ICD10(CM), Read, NDC, Gemscript, CPT4,
HCPCS…
• >9.5 million concepts
– >3.4 million standard concepts
– >5.4 million source codes
– >674,000 classification concepts
• >57 million concept relationships
• >86 million ancestral relationships

As of v5.0 26-FEB-21 Publicly available for download at: http://athena.ohdsi.org/

 Common data model and standardized vocabularies to
enable standardized analytics
Source 1 raw data
Source 1 CDM

Transformation to OMOP common data model

Open-source
analysis code
Electronic health
records
Source 2 raw data Source 2 CDM

Administrative claims

Source 3 raw data Source 3 CDM Open

evidence

Clinical data
 Standards enabling evidence for regulators:
Safety of hydroxychloroquine (HCQ)
• Evidence needed around the safety of HCQ alone and in combination with azithromycin (AZ)
• Standardized analytics: comparative cohort design with large-scale propensity adjustment
• Standardized data: 14 databases (claims and EHR) from 6 countries, all using OMOP CDM
• Findings: In RA population, HCQ alone is generally safe but in combination with AZ it shows a
doubling of risk of 30-day cardiovascular mortality.
Standards enabling evidence for policy:
COVID-19 treatment utilization patterns
 Data standards are part of methodological
standards for reliable evidence
• Patients with cardiovascular diseases and hypertension treated with angiotensin
converting enzyme inhibitors (ACEs) angiotensin-II receptor blockers (ARBs) may
influence susceptibility to COVID-19 and worsen its severity.
 Take-home messages
• Data standards should be driven by evidence needs
• Complementary standards required across the evidence generation lifecycle
– Data standards for exchange
– Data standards for structure for analytics
– Vocabulary standards for data content and analytics
– Analytic standards for evidence generation and dissemination
• Reliable evidence is strengthened by disparate data and analytics
standardized within a common shared framework
– Data network should focus on depth and breadth as required for evidence needs
– Open science enables adoption of data standards and analytic best practices
– International collaboration is possible with international standards