Keratoconus — Claude Code literature sweep
Disease: Keratoconus (MONDO:0015486, HP:0000563)
KB entry: kb/disorders/Keratoconus.yaml
Provider: Claude Code literature sweep (PubMed E-utilities, esearch + esummary)
Run date: 2026-07-31
Context: PR #7128, review item 🟡 #3 ("no deep-research artifact for a new entry")
Method
No third-party deep-research provider (Falcon / Asta / OpenScientist / Perplexity) was available in this runner, so this is a Claude Code literature sweep rather than a DR provider report. Ten independent PubMed queries were issued, one per coverage axis, each retrieving the top 6 relevance-ranked hits. This is a multi-modal sweep — each axis is blind to what the others surface, which is what makes it a coverage audit rather than a single-query snapshot.
Because the axes were chosen to probe the specific coverage gaps the PR review named (phenotypes, histopathology, comorbidities, clinical trials, genetics depth), a "nothing new" result on an axis is itself a finding.
Provenance discipline. Every PMID below came back from a live PubMed query — none was
recalled from model memory. Every PMID that was mined into the KB entry was first cached
with just fetch-reference, and every snippet was verified as an exact
whitespace-normalized substring of that cache before commit. Titles for PMIDs that were
surfaced but not mined are reproduced from the esummary payload and have not been
independently re-verified — treat them as leads, per the CLAUDE.md DR guidance.
Queries
Table (click to expand)
| Axis | PubMed query |
|---|---|
| biomechanics | keratoconus AND corneal biomechanics AND (review[pt]) |
| tear proteomics / inflammation | keratoconus AND (tear OR proteomic) AND (cytokine OR MMP OR inflammation) |
| atopy / eye rubbing | keratoconus AND (eye rubbing OR atopy OR allergic conjunctivitis) |
| comorbidity | keratoconus AND (comorbidity OR association) AND (sleep apnea OR Down syndrome OR connective tissue) |
| hydrops management | acute corneal hydrops keratoconus management |
| keratoplasty outcomes | keratoconus AND (keratoplasty OR DALK OR penetrating) AND outcomes |
| cross-linking evidence | corneal collagen crosslinking keratoconus AND (randomized OR meta-analysis) |
| progression definition | keratoconus progression definition criteria |
| pediatric | pediatric keratoconus progression |
| genetics | keratoconus genetics AND (GWAS OR polygenic risk score) |
Findings acted on in this PR
Four sweep hits were material enough to mine into the entry. All four were cached with
just fetch-reference and their snippets verified.
1. PMID:42228627 — Global Consensus on Keratoconus and Ectatic Diseases, Edition 2 (Cornea, 2026)
The most consequential finding of the sweep. The entry cited the 2015 first edition
(PMID:25738235). Edition 2 was published 2026-07 and supersedes it: 128 ophthalmologists
from 6 continents and 12 international societies, 4 Delphi rounds, covering definition,
diagnosis, staging, and progression criteria. Curating a brand-new entry against a
superseded consensus would have been a real defect.
→ Mined into diagnosis as the current framework, with the 2015 edition retained
(not replaced) because it is the consensus underlying most of the intervening literature
this entry cites.
2. PMID:37374145 — Systemic Associations with Keratoconus (Life, 2023)
Directly addresses the review's comorbidities gap. Names atopy, Down syndrome and
connective tissue disease as the most frequently cited associations, and reports diabetes
mellitus as a candidate protective factor.
→ Mined as a KNOWLEDGE_GAP discussion (gap_kc_systemic_associations) rather than as
comorbidity assertions. Reasoning: the three "positive" associations are not
mechanistically equivalent — the atopy association is plausibly mediated entirely by
itch-driven eye rubbing (already modelled as both an environmental factor and a
pathophysiology node), so asserting it as an independent comorbidity would double-count a
single causal pathway. The review itself states the mechanisms "are largely unknown". The
gap carries two proposed experiments (rubbing-adjusted mediation analysis; MR for the
diabetes inverse association) that would resolve which pairs deserve kb/comorbidities/
entries.
3. PMID:33649486 full text (already cached) — per-gene GWAS mining
Not a new PMID, but the sweep's genetics axis prompted a re-read of the cached full text (not just the abstract) of the multi-ethnic GWAS. Review item 🟡 #4 was correct that this source was under-mined: it supports named per-gene records with quotable association statistics.
→ Added COL12A1 (strongest coding signal: rs35523808 p.Glu2160Val, and collagen XII
localises to Bowman's layer — connecting the locus directly to the "Anterior Limiting
Membrane Rupture" node), FNDC3B, ZNF469, ALDH3A1, ITGA2; and added genuine
association evidence (Table 1 rows) to the pre-existing COL5A1 and LOX records.
The LOX/VSX1 typing inconsistency the review flagged is resolved by this: LOX now rests
on a genome-wide significant association (rs840464) plus an eQTL effect on its own
transcription, whereas VSX1 still rests only on the shared "analysed markers" sentence —
so SUSCEPTIBILITY vs DISPUTED no longer rest on the same evidence.
4. HP:6001232 Corneal iron line / HP:0012040 Corneal stromal edema
Not literature, but found while closing the phenotype gap: HPO does have a term for the
Fleischer ring (HP:6001232, whose own definition names keratoconus as a typical setting)
and for the stromal oedema of acute hydrops (HP:0012040). Both verified present in the
local sqlite:obo:hp adapter.
→ Both added as phenotypes; a histopathology section was added for stromal thinning,
Bowman's-layer breaks and epithelial basement-membrane iron deposition (all unbound —
NCIT has no corneal morphologic finding terms for these).
Findings NOT acted on, and why
Recording these explicitly so the next curator does not re-run the same searches.
Table (click to expand)
| Lead | Axis | Why deferred |
|---|---|---|
~~PMID:39448666 Keratoconus (Nat Rev Dis Primers, 2024)~~ |
atopy/rubbing | Now mined (round 3) — supplied the UV-light-exposure and contact-lens-wear environmental: records. |
~~PMID:38830186 The Pathophysiology of Keratoconus (Cornea, 2025)~~ |
biomechanics | Now mined (round 3) — supplied two new pathophysiology nodes (Corneal Oxidative Stress, Keratocyte Apoptosis), a biomechanical-primacy evidence item, and two REFUTE items strengthening the inflammatory-status controversy. |
~~PMID:39535071 polygenic risk score (Hum Mol Genet, 2025)~~ |
genetics | Now mined (round 3) — added as two evidence items on the Polygenic susceptibility architecture genetic record. The cohort ambiguity flagged here was handled by quoting the biobank by name in each snippet (EstBB OR 2.28; UKB AUC 0.84→0.88) rather than stating a bare AUC. |
~~PMID:39671084 Definition of Progressive Keratoconus: systematic review (Cornea, 2025)~~ |
progression | Now mined (round 3) — supplied a "Documented progression (the cross-linking indication)" progression: record; typed the lack-of-unified-criteria conclusion PARTIAL, and cross-referenced the Edition 2 consensus in the diagnosis section rather than presenting it as superseded. |
PMID:39681212 Corneal cross-linking (Prog Retin Eye Res, 2025); PMID:37938377 epi-on vs epi-off meta-analysis; PMID:36094374 oxygen in CXL |
CXL evidence | The entry's CXL treatment is already evidenced. These would deepen protocol-level detail, which is below the abstraction level this KB curates. |
PMID:39943883 PK vs DALK meta-analysis (27,018 eyes); PMID:27802912 |
keratoplasty | Would support a surgical-outcomes elaboration of the existing keratoplasty treatment. Not a mechanism claim. |
PMID:38317314 Management of acute corneal hydrops (2024) |
hydrops | Management-level; the hydrops mechanism is already curated from PMID:24491416 and was split into two atomic nodes in this PR. |
PMID:37227479 Ocular manifestations of OSA meta-analysis |
comorbidity | A genuine candidate comorbidity (OSA-keratoconus) not covered by PMID:37374145. Left as a lead; it belongs in the systemic-associations gap discussion's scope. |
PMID:33463562, PMID:39396644, PMID:36973341 pediatric keratoconus |
pediatric | Pediatric-onset keratoconus progresses faster and is a distinct management context. Not curated — the entry does not currently model onset-stratified subtypes, and adding one is a scoping decision beyond this review round. |
Clinical trials (clinical_trials section) |
— | Deliberately not added. The review suggested an NCT record for the US multicentre CXL trial (PMID:28495149). ClinicalTrials.gov queries returned several plausible CXL trials, but none could be confirmed as that trial (the closest enrollment/design match, NCT00567671, is an Emory single-site study — not the multicentre trial). Guessing an NCT here would be exactly the misattribution failure the repository's evidence SOP exists to prevent. Left uncurated pending a verifiable registry link. |
Coverage assessment after this round
Table (click to expand)
| Section | Before | After |
|---|---|---|
phenotypes |
4 | 6 |
histopathology |
absent | 3 findings |
genetic |
5 records, 1 real source | 10 records, GWAS-anchored |
diagnosis |
1 | 2 (current consensus) |
discussions |
2 | 3 |
comorbidities (separate files) |
absent | still absent — deliberately, see above |
clinical_trials |
absent | still absent — deliberately, see above |
Round 3 — mining pass over cached-but-unconsumed references
Prompted by the second PR review, which flagged that four references shipped in
references_cache/ in this PR were consumed by zero evidence items. All four are now
mined; no new literature search was run and no new cache files were fetched.
Table (click to expand)
| Section | After round 2 | After round 3 |
|---|---|---|
pathophysiology |
12 nodes | 14 nodes (Corneal Oxidative Stress, Keratocyte Apoptosis) |
environmental |
2 | 4 (UV exposure, contact lens wear) |
progression |
1 record | 2 records (progression definition / CXL indication) |
genetic evidence on the polygenic record |
1 item | 3 items (PRS validation in two biobanks) |
| verified snippets | 73/73 | 85/85 |
| cached-but-unconsumed references | 4 | 0 |
Both new pathophysiology nodes carry a notes: stating which upstream edge was
deliberately not asserted and why — the source reports oxidative stress and keratocyte
apoptosis without establishing what drives either, and "along with" is co-observation, not
causation. The Anterior Limiting Membrane Rupture → Loss of Corneal Biomechanical
Rigidity edge was also recalibrated from DIRECT to INDIRECT_UNKNOWN_INTERMEDIATES
(review suggestion 2): the snippets establish the lesion histopathologically, not its
biomechanical consequence, and Bowman's layer is ablated wholesale in PRK without
inducing ectasia.
Open follow-ups worth issues
keratoconus_corneal_ectasiamechanism module. Keratoconus carries both an HP and a MONDO id and is used as a phenotype by at least five other entries (Ehlers-Danlos_Syndrome,Arterial_Tortuosity_Syndrome,Spondylodysplastic_Ehlers-Danlos_Syndrome,CRB1_Retinal_Dystrophies,GUCY2D-Related_Retinopathy). That is the "disease-like phenotype" pattern that glaucoma, cataract and osteoporosis each model as a disorder entry plus akb/modules/module. Raised by the PR reviewer (suggestion 13) and endorsed here.- MAXO NTR for corneal collagen cross-linking. Neither MAXO nor NCIT codes CXL; the entry uses
NCIT:C15301Phototherapy +CHEBI:17015riboflavin as an interim composition. - OSA-keratoconus (
PMID:37227479) as a candidate comorbidity pair.