About this project
EQ-Graph is a research knowledge graph of EuroQol-funded projects. It connects projects and researchers to the instruments, methods, and findings reported in the associated publications.
The aim is to map EuroQol's research portfolio and make it easier to explore, query, and study as one connected body of evidence.
The public data release and code are openly available. Download the data or view the research repository.
The project was conducted by Paul Schneider, Anuja Kulkarni, and Kazik Pogoda, supported by EuroQol seed grant 2582-SG.
Contact: pschneider@abundanceds.com
Beta research release1,024 funded projects ·797 included publications ·798 studies · ontology 0.13
Methods Summary
For more details, see research repository.
From funded projects to structured research evidence
First, we took the public list of EuroQol-funded projects. We extracted the principal investigators, removed duplicates, and matched them to public author profiles on OpenAlex and ORCID where possible.
We then searched OpenAlex, ORCID, and PubMed for publications linked to those investigators or to EuroQol funding. After removing duplicates, 28,600 records remained. Of these, 18,348 journal articles and reviews with an abstract entered screening.
An AI screen of the 18,348 abstracts routed 1,679 papers to full-text assessment. We retrieved 1,607 full texts; 72 were not retrieved. Full-text assessment confirmed EuroQol support or a funded-project origin for 797 papers, which were processed and included in the knowledge graph.
Note on AI use: the human authors planned and designed this study, reviewed its outputs, and take full responsibility for it. Large parts of the implementation, from screening to data extraction, were carried out by AI agents.
Ontology
What the graph can represent
Ontology 0.13
The semantic model
- Entity
- Canonical registry
- Control or assertion
Portfolio
- Project
- title · abstract · investigator · working group · years · status
- Publication
- bibliographic metadata · funding · URLs · citations · editorial status
- Person
- names · ORCID · authorship · membership · project role
- Study
- primary family · purposes · execution state · result state
Study structure
- StudyPart
- separable sample, data source, method, or state
- Design
- component · time · comparison · allocation
- Population + Sample
- population description · role · flow stage · size
- DataUse
- source · origin · level · purpose
- TaskDesign
- profiles · attributes · levels · blocks · randomisation
- Administration
- respondent · channel · setting · language · time point
- StudyFactor
- condition · comparator · determinant · modifier · stage
- StakeholderInvolvement
- group · activity · stage · role · influence
Scientific use
- InstrumentUse
- instrument identity · context · function
- MethodUse
- method identity · context · function
- ProtocolUse
- protocol identity · context · function
- ModelUse
- model identity · context · function · analytic role
- SoftwareUse
- software identity · context · function
- ProductUse
- existing product examined, compared, or synthesised
- ScoringUse
- instrument responses linked to a scoring product
Evidence and control
- Outcome
- family · source label
- Finding
- statement · selected value · unit · subgroup · comparator · uncertainty
- Interpretation
- source-reported meaning, separate from the finding
- Limitation
- source-reported limitation
- Product
- reusable output · type · dated state assertions
- Concept
- cross-cutting theme for discovery
- Gap
- unmapped value · unmodelled aspect · uncertain mapping · not reported
- SourceConflict
- contradictory source statements kept together
Information types: controlled values · canonical registries · open source-faithful text · provenance
Ontology development
Bottom-up ontology creation
We did not assume an ontology of EuroQol research in advance (top-down). Instead, we used AI agents to let one emerge from the papers (bottom-up).
Several AI agents independently reviewed different samples of publications and mapped them against candidate meta-research questions. Each agent proposed ontological structures, which we reviewed and then tested on unseen papers.
The ontology was refined over several rounds until it remained stable. In total, four 15-paper development rounds and two 20-publication calibration batches produced version 0.13.
Contact
Corrections and contributions are welcome.
Both humans and AI make mistakes. Please contact us if you spot an error, or if we missed a paper from a EuroQol-funded project.