based decision making based on structured data
sources
Gert van Valkenhoef
Section 1
About me
MSc Artificial Intelligence (2009)
Researcher and lead developer, ADDIS project (2009-now)
PhD Medical Sciences (2009-2012) Funded through 2016
Based in the Netherlands
Visiting scholar @ Brown, Oct-Dec 2014
Acknowledgements
PhD supervisors: Prof. Hans Hillege, Prof. Bert de Brock Key collaborators: Dr. Tommi Tervonen, Dr. Douwe
Postmus, Prof. A.E. Ades, Dr. Sofia Dias, Dr. Nicky Welton, Guobing Lu, Dr. Byron Wallace, Dr. Tom Trikalinos
Programmers and students: Jo¨el Kuiper, Dr. Daan Reid,
ADDIS 1.x: Project Escher
Escher (2007-2013) was a national research project of the Dutch Top Institute Pharma aiming to improve drug regulation through science
16 PhD students and 4 PostDocs working in 5 universities (RUG/UMCG, UU/UMCU, Erasmus MC) in collaboration with industry (MSD, GSK, Amgen, WINap)
ADDIS 1.x: Escher WP 3.2 Goals (2009-2013)
Develop a drug information system:
Effective knowledge access and management Answer drug efficacy and safety questions
in an efficient, transparent and accountable way within and across compounds
for a broad audience (including regulators)
Improve consistency in regulatory decision making Based on systematic review and meta-analysis
ADDIS 1.x: Escher WP 3.2 Results
ADDIS decision support system for health care policy:
Database of clinical trials
Evidence synthesis (network meta-analysis) Decision aiding (multi-criteria benefit-risk analysis)
Research output:
7 journal articles + PhD thesis
+ additional journal and conference papers
Primary limitations:
Gathering data is time consuming
ADDIS 2.x: IMI GetReal (2014-2016)
GetReal is a European project of the Innovative Medicines Initiative (IMI) that aims to integrate randomized and observational data to best inform relative effectiveness 5 work packages, 13 academic partners, 15 industry partners, ties with regulatory and reimbursement networks
EUR 16mln funding, 130 person-years of effort over 3 years, primarily senior scientists
ADDIS 2.x plans
Web-based multi-user system Collaborative database building
Flexible (ad hoc) data integration / harmonization Predictive modeling / relative effectiveness
Current status
ADDIS 1.x no longer developed ADDIS 2.x progressing
Most key components in place
But functionality is rough / incomplete
Section 2
ADDIS: Aggregate Data Drug Information System
ADDIS is a decision support system For health care policy decision making
Bridging the gap between aggregated clinical data and
evidence-based drug regulation using state of the art methods for benefit risk decision making
Software should (eventually) also apply to HTA, hospital, pharmacy, etc. decision making
Evidence-based health care policy
Basing policy on evidence is challenging: Data acquisition
Evidence synthesis Decision aiding / making
ADDIS: Aggregate Data Drug Information System
How could evidence-based decision making be supported or improved if clinical trials data were available in a structured format?
Case: EMA EPAR – Edarbi (Azilsartan Medoxomil)
Dossier investigates three doses: 20, 40, 80 mg/day
With various populations, comparators
Is there a benefit of 80 mg/day over 40 mg/day? If so, does that benefit outweigh additional harms?
ADDIS 2
Can the availability of structured clinical trials data be improved through an on-line collaborative platform for sharing and improving data extractions?
ADDIS 2 status
Most components in place
Closing in on ADDIS 1.x feature parity
Data entry is ‘next big thing’
Section 3
ClinicalTrials.gov import in ADDIS 1.x
ClinicalTrials.gov import is a key feature
Helped show feasibility of ADDIS concept
Import was remarkably easy to achieve
Data models similar, despite independent development
Some stumbling blocks
Key advantages
Huge time saver: well-reported CT record saves many hours Loads of data available, also helps when papers are unclear Most key dimensions represented: easily maps
User input is required mainly for “harmonization”
Typically good trade-offs between text and structured data Links to literature and other IDs
Lack of referential integrity
There are no key/keyref constraints in the XML schema:
Duplicate IDs
References to undefined IDs
Especially arm/group references become confusing
IDs can be (re-)defined in each section
Use of XML enum could clarify range of some attributes
e.g. Number, Mean, ...
Categorical as a catch-all
A category can mean:
A true categorical variable Stratified reporting
Reporting at multiple time points This is complex to disentangle.
Proper support for time points is #1 on my wish list!
Reporting thresholds
Political issue
Investigator-set reporting thresholds for AEs seriously reduce the value of datasets
We’ve modelled regulatory dossiers where the key events discussed were not reported on ClinicalTrials.gov
Further wish-list items
XML elements instead of structured text for study design, eligibility criteria, etc.
Distinguish uses of Number: ‘count’ versus ‘percentage’ Add MedDRA IDs when MedDRA is used
Section 4
Discussion
Your feedback to the ADDIS concept & system!
Are changes to the schema planned / expected / possible? Is it / will it be possible to flag problems in records? Are there further plans for cross-linking other services? Best way to poll for new and recently changed records?
Thank you!
Thank you!