A knowledge graph of utility values in eye disease
Studies, health states, instruments and utility values across eye diseases, linked and ready to query.
Countries are counted only where the abstract names them.
Cell counts show reported values; shading shows the share within each condition group. Direct elicitation dominates the older retina and glaucoma literature, while population surveys rely on EQ-5D.
Utility falls as disease or vision loss worsens, across conditions and instruments.
Each line is one study following one instrument from its least to its most severe state. In AMD, TTO utility falls from 0.89 with good better-eye vision to 0.40 at counting fingers or worse; thyroid eye disease vignettes start low even in the mildest state.
Rows are studies that valued the same patients with more than one method. HUI3 tends to sit lowest because it captures vision directly; standard gamble against blindness sits highest because few patients will risk sight.
Explore the utility graph.
Ask a question about the evidence, or browse the connections yourself.
What do you want to investigate?
Answers come only from the 75 extracted values and cite PMIDs. Charts are drawn from the stored values, not from numbers Claude writes. Questions use your own Claude usage; the first one asks for permission.
Data
Value browser
Click a dot to see its study, instrument, respondents, population and any note.
Decrements and gains
Included studies
Full text needed
Relevant, but the abstract has no extractable value.
Excluded at abstract review
How this graph was built
Record flow
Method
The workflow follows EQ-Graph by Abundance Decision Systems, with the funded-project anchor replaced by a topic search. Europe PMC was searched for any eye condition (about 70 condition and vision terms, from glaucoma and AMD to uveitis, thyroid eye disease and visual impairment), combined with a utility term block (utility, EQ-5D, time trade-off, standard gamble, HUI, SF-6D, CHU-9D, AQoL, QALY).
Rules removed letters, editorials and similar article types, records that are not about an eye condition, and records whose title and abstract contain no utility-type number. The remaining 177 abstracts were screened one at a time and tagged to one of 16 condition groups. A study was included when its abstract reports a preference-based utility value, decrement or gain for people with the condition.
Every extracted value and sample size was checked by machine against the abstract text; all 304 checks matched, with eight sample sizes written as words in the abstract.
Graph model
Nodes: study, condition, health state, instrument (with method class), country. Edges: a study addresses a condition, reports values for health states, uses instruments, and is set in a country.
Limits
Values come from abstracts only, so subgroup values reported only in full-text tables are missing. Screening and extraction were done by Claude without a second human reviewer. Four values are derived and marked. Vignette, carer, clinician and student valuations are labelled by respondent type. Two myopia papers report the same sample of 2,198 adolescents and the same CHU-9D mean, so they likely share one cohort. Cost-effectiveness models and reviews were set aside for a later phase.
Credit
Structure and workflow adapted from EQ-Graph (MIT licence). This is an independent pilot and is not affiliated with EuroQol or Abundance Decision Systems.