Skip to content

ECM Disorder Claim–Evidence Review (2026-07-25)

Correctness review of ten extracellular-matrix (ECM) disorder entries, with the emphasis on claim–evidence match: does each cited snippet actually support the claim it is attached to, at the strength and scope the entry asserts?

Scope and method

Ten ECM disorders spanning the major matrix compartments were reviewed:

Entry ECM compartment / protein Evidence items
Marfan_Syndrome fibrillin-1 microfibril 118
Vascular_Ehlers-Danlos_Syndrome collagen III 36
Ehlers-Danlos_Syndrome_COL5A1-related collagen V 60
Osteogenesis_Imperfecta_Type_I collagen I 33
Pseudoxanthoma_Elasticum elastic fibre mineralisation (ABCC6/PPi) 44
Alport_Syndrome collagen IV basement membrane 60
Kniest_Dysplasia collagen II 55
Stickler_Syndrome_Type_1 collagen II 58
Arterial_Tortuosity_Syndrome GLUT10 arterial wall matrix 49
Dystrophic_Epidermolysis_Bullosa collagen VII anchoring fibril 87

~620 evidence items were checked in two passes:

  1. Mechanical — every snippet was re-verified as a substring of its references_cache/ file (ellipsis-split, whitespace-normalised).
  2. Semantic — each snippet was read against the claim's own description / frequency / supports / presence values, which is the layer no validator covers.

The mechanical pass was effectively clean. Six snippets differ from cache only by Greek-letter transliteration or quote style (β→beta, α→alpha, ⩽→≤, "→') — all faithful quotes that the fuzzy reference validator accepts. No fabricated snippets, no wrong-paper PMIDs, and no Named Entity Confusion were found. Every finding below comes from the semantic pass.

Verdict

Three entries — Vascular_Ehlers-Danlos_Syndrome, Pseudoxanthoma_Elasticum, Arterial_Tortuosity_Syndrome — are clean and should be treated as the reference standard. ATS is the best of the set: it retains a contradicting PARTIAL item (TGF-β is not elevated in skin or end-stage vascular tissue) with the note "included to represent the mechanism faithfully", and it states the limits of single-case treatment evidence rather than overselling it.

Twenty-six findings in the other seven entries follow, ordered by severity.

High severity

M1 — Marfan_Syndrome: four REFUTE items whose explanations all affirm support

biochemical > Fibrillin-1 Protein (lines 1048–1075) carries presence: Abnormal and four evidence items, all marked supports: REFUTE. Every one of the four explanations states the opposite:

"The literature indicates that fibrillin-1 protein abnormalities are directly linked to Marfan syndrome…" (PMID:8180508, REFUTE)

"The literature highlights the role of fibrillin-1 in the pathogenesis of Marfan syndrome…" (PMID:22705998, REFUTE)

All four papers support presence: Abnormal. If the intent was to refute the secondary context: claim ("not routine diagnostics"), that is nowhere stated, and PMID:22705998 / PMID:37688493 do not address diagnostics at all. As it stands the supports field is inverted relative to its own explanations.

M2 — Marfan_Syndrome: prevalence_class contradicts the record's own rate

Prevalence record 1 (lines 106–111):

prevalence_class: BAND_1_5_PER_10000   # = 10–50 per 100,000
rate_low: 5.0
rate_high: 10.0                        # = 0.5–1 per 10,000

BAND_1_9_PER_100000 is the band consistent with the recorded rate. The cited source (PMID:21308160, "one in 10,000 to 20,000") matches the rate, not the class.

M3 — Marfan_Syndrome: a REFUTE that the record actually supports

In the same record, PMID:26631233 (max prevalence 6.5/100,000) is marked REFUTE. 6.5 lies inside the recorded rate_low: 5.0 – rate_high: 10.0 interval, so it supports the record. The explanation compares against "the 0.01-0.02 percentage range stated in the question" — a value that appears nowhere in the record (percentage: 0.005-0.01). Both explanations in this block are stale against a superseded earlier version of the claim.

The Myopia phenotype asserts frequency: OCCASIONAL on two evidence items, neither of which supports it:

  • PMID:36237549 (PARTIAL) — explanation: "does not specifically mention myopia"
  • PMID:30246406 (NO_EVIDENCE) — a feline case report; explanation: "does not mention myopia"

A phenotype with a frequency band and zero supporting evidence should either gain a real citation or be removed (CLAUDE.md §4, Option C).

genetic > COL5A1 lists Frameshift Mutations as a variant class (line 479). The only evidence bearing on it is PMID:2683783 (NO_EVIDENCE, a 1989 paper), whose explanation reads:

"This reference doesn't support the existence of frameshift mutations in COL5A1, making the specific part of the statement regarding frameshift mutations unsubstantiated."

The entry states a claim its own evidence explicitly marks unsubstantiated.

treatments > Supportive Care cites PMID:33741806 as SUPPORT. That paper is "Practical management strategies for benign hypermobility syndromes" — hEDS/HSD, a different subtype from COL5A1 classic EDS. The same block cites PMID:32941194 ("Current management of the vascular subtype of Ehlers-Danlos syndrome"), whose own explanation concedes it "is different from COL5A1-related EDS". Three of the five items in this block are NO_EVIDENCE. Separately, treatments > Gastrointestinal Management rests on a single NO_EVIDENCE item.

A1 — Alport_Syndrome: two human phenotypes supported only by mouse data

GBM Lamellation and Focal Segmental Glomerulosclerosis are human phenotype claims whose sole evidence is PMID:34029143, a Col4a5 mouse study (evidence_source: MODEL_ORGANISM). This contradicts the standing rule that "model organism evidence should not be the only support for human phenotypes." GBM basket-weave lamellation is abundantly documented in human biopsy series, so this is a citation gap rather than a factual error.

A2 — Alport_Syndrome: digenic claim quotes the hypothesis, not the result

The Digenic inheritance block quotes PMID:25575550's BACKGROUND section:

"Therefore, we explored the possibility that Alport syndrome is under digenic control."

The explanation says this "explicitly establishes digenic inheritance". Exploring a possibility does not establish it. The same abstract's CONCLUSIONS contain a directly supporting sentence:

"This pedigree analysis provides evidence for digenic inheritance of Alport syndrome."

as does METHODS ("we identified 11 patients who had pathogenic mutations in two collagen IV genes"). This matters because CLAUDE.md names Alport as a worked digenic exemplar. Straightforward snippet swap.

Medium severity

M4 — Marfan_Syndrome: Aortic Aneurysm cites a paper about a different disease

PMID:7911041 is a 1994 review, "Cardiovascular molecular genetics". The quoted sentence concerns supravalvular aortic stenosis, Williams syndrome and elastin; it never mentions aortic aneurysm. It is marked SUPPORT with the explanation "supporting the frequent occurrence of aortic aneurysm and dissection." The claim is well supported by the other five items in the block, so this one can simply be dropped. (PMID:36058493 and PMID:32290873 in the same block are also generic disease-definition sentences that do not speak to aneurysm frequency.)

M5 — Marfan_Syndrome: exercise-benefit trials cited to support activity restriction

environmental > Physical Activity Restrictions claims "Avoidance of strenuous isometric exercise and contact sports". Two of its three SUPPORT items show the opposite direction of effect:

  • PMID:36453629 — a 10,000-steps/day intervention slowed aortic root Z-score growth (−0.24 vs +0.008/yr, P=0.01)
  • PMID:28947563 — exercise blunted aortic root dilation in Marfan mice

The explanations strain to reconcile this ("indirectly supporting the idea that physical activity restrictions can help"). The nuance is real — moderate dynamic exercise is beneficial while isometric/contact activity is restricted — but the claim text does not carry it, and these items should not be SUPPORT as written.

Chronic Joint Pain (frequency: FREQUENT) cites PMID:10906878:

"Thirty patients with Type III Ehlers-Danlos syndrome reported joint pain more frequently than did patients with Types I, II, or IV."

Types I and II are classic EDS — the subtype this entry curates. The quote therefore reports that classic EDS patients had less joint pain than hypermobile-type patients, which does not support a FREQUENT band here.

Delayed Wound Healing asserts frequency: VERY_FREQUENT (99–80%) while its ORPHA:287 evidence item — marked SUPPORT — reads HP:0001058 | Poor wound healing | Frequent (79-30%). Either the band should be FREQUENT, or the conflict should be handled the way Marfan_Syndrome handles spontaneous pneumothorax: keep the curated band, mark the Orphanet item PARTIAL, and record the disagreement in notes:.

All 56 literature items in this file omit evidence_source. Three are misclassifiable by default:

  • PMID:30246406 — feline COL5A1 EDS case → MODEL_ORGANISM
  • PMID:2728341 — wound healing in dogs and cats → MODEL_ORGANISM
  • PMID:15095409 — cultured patient fibroblasts → IN_VITRO

Per CLAUDE.md, veterinary observations are MODEL_ORGANISM. This is the only file of the ten with the field systematically unset.

O1 — Osteogenesis_Imperfecta_Type_I: evidence_source inconsistent with the rest of the KB

All six ORPHA:666 items use evidence_source: HUMAN_CLINICAL. Every other file in the sample tags Orphanet structured-database rows OTHER (Marfan 32/32, Stickler 38/38, vEDS 2/2, cEDS 4/4). Orphanet rows are curated database records, not primary clinical observations.

O2 — Osteogenesis_Imperfecta_Type_I: spectrum-level bands applied to a subtype entry

ORPHA:666 is "Osteogenesis imperfecta" — the whole spectrum, not type I. Five phenotypes (Osteoporosis, Bone Pain, Progressive Hearing Impairment, Wormian Bones, Bruising Susceptibility) adopt the spectrum band as the type I frequency: and mark it SUPPORT. Hyperhidrosis, in the same file, explicitly declines to do so:

"No type I-specific frequency is asserted here because the Orphanet band reflects the whole OI spectrum rather than type I alone."

The reasoning is right; it should be applied consistently. Since type I is the mildest OI form, importing spectrum bands is likely to overstate severity-linked phenotypes.

K1 — Kniest_Dysplasia: quantitative treatment claim with no evidence

treatments > Cleft Palate Repair asserts "Clefting abnormalities are present in approximately 70% of Kniest dysplasia cases" and carries no evidence block. The supporting datum exists in the file — PMID:25592122's "five of the seven patients exhibited clefting abnormalities" (71%), cited under the Cleft Palate phenotype — and just needs to be attached here. Otherwise Kniest is one of the strongest entries reviewed, with quantified, disease-specific evidence on nearly every phenotype.

S1 — Stickler_Syndrome_Type_1: three unreconciled prevalence figures

Three prevalence records disagree by up to an order of magnitude, with no note reconciling them:

  • PMID:40146061 — 1 in 21,844 live births ≈ 4.6/100,000
  • ORPHA:828 — 1-9 / 100 000 | Europe | Point prevalence
  • ORPHA:828 — 1-5 / 10 000 | Worldwide | Prevalence at birth ≈ 10–50/100,000

Records 1 and 3 are both birth-referenced and differ ~2–10×. Contrast the same file's careful handling of the type-1-vs-grouping issue elsewhere. (Incidentally, the second Orphanet row quotes PMID:2012 — a year captured as a PMID in the upstream Orphanet data. The snippet is a faithful quote; the artifact is upstream.)

S2 — Stickler_Syndrome_Type_1: grouping-level frequencies on a subtype entry

Roughly 18 phenotypes (Arachnodactyly, Kyphosis, Pectus Carinatum, Spondylolisthesis, Hypotonia, Chronic Otitis Media, …) take ORPHA:828 — the Stickler syndrome grouping, all types — verbatim as the STL1 frequency. The entry does this well twice: Cataract is downgraded to FREQUENT with stated reasoning, and Membranous Vitreous is upgraded with stated reasoning. Given the entry's own statement that STL1 skeletal features are milder than STL2/STL3, the remaining ~18 deserve the same treatment. Same root cause as O2.

A3 — Alport_Syndrome: progression explanation describes content not in its snippet

progression record 1 claims "Microscopic hematuria is typically the first clinical manifestation, present from early childhood in X-linked males." Its evidence (PMID:32712016) is a generic disease-definition sentence about collagen IV genes and basement membranes; the explanation reads "Emphasizes the importance of early diagnosis, supporting childhood onset" — describing the paper's title rather than the quoted text.

A4 — Alport_Syndrome: over-read screening recommendation

Microscopic Hematuria is described as "the earliest and most consistent finding", supported only by "screening programs for glomerular hematuria in children and young adults could benefit from inclusion of genetic testing" — a recommendation about screening programme design, not a statement about earliness or consistency.

D1 — Dystrophic_Epidermolysis_Bullosa: severe-RDEB frequencies asserted at disease level

Seven phenotypes carry a disease-level frequency: (mostly VERY_FREQUENT, i.e. 80–99% of patients) restricted only by a free-text note:

Phenotype frequency note
Pseudosyndactyly VERY_FREQUENT "Characteristic of severe RDEB"
Esophageal Stricture VERY_FREQUENT "Primarily in RDEB"
Dysphagia VERY_FREQUENT "Secondary to esophageal strictures in RDEB"
Growth Retardation VERY_FREQUENT "Primarily in severe RDEB"
Cutaneous SCC VERY_FREQUENT "Leading cause of death in severe RDEB"
Microstomia FREQUENT "Primarily in severe RDEB"
Osteoporosis FREQUENT "Primarily in severe RDEB"

The entry's scope includes DDEB, where — by its own subtype description — "nail dystrophy may be the only manifestation." None of these occur in 80–99% of all DEB patients. The entry defines four subtypes (DDEB, RDEB-sev gen, RDEB-intermediate, RDEB-Inversa) but uses the subtype: foreign key zero times, though the schema supports it and tests/test_data.py enforces it. Moving these to subtype: RDEB-sev gen would make the scoping machine-readable instead of prose-only.

Low severity

  • M6 — Marfan_Syndrome progression: two explanations argue a source "does not confirm that the onset of Marfan syndrome itself is restricted to childhood-adolescence", but the record now reads age_range: All ages. Stale text from a superseded claim.
  • O3 — Osteogenesis_Imperfecta_Type_I: three snippets are mid-sentence line-wrap fragments rather than clean quotes — "year, one-half of which affected the tibia/fibula. Long-bone fracture rate was" (Osteoporosis), "findings (presence of Wormian bones, platybasia, basilar impression (McGregor's" (Wormian Bones, unbalanced parentheses), "Other areas included pain, gastrointestinal problems," (Bone Pain). Each validates as a substring but cuts off before the clause the explanation relies on.
  • S3 — Stickler_Syndrome_Type_1: Orphanet rows that conflict with the curated band are marked SUPPORT (Cataract, Membranous Vitreous). Marfan_Syndrome marks the analogous conflicting pneumothorax row PARTIAL — the better pattern.
  • D2 — Dystrophic_Epidermolysis_Bullosa: PMID:19700011 on Esophageal and Mucosal Blistering quotes "…bone marrow, musculoskeletal system, heart, kidney, and teeth" — no esophageal or mucosal content. (Its second use, on Dental Caries, is apt.)
  • D3 — Dystrophic_Epidermolysis_Bullosa: Anemia (VERY_FREQUENT), Osteoporosis and Constipation (FREQUENT) rest on management-list fragments — e.g. "management of constipation" — that establish the complication is managed, not how often it occurs.

Cross-cutting patterns

  1. Grouping/spectrum frequency leakage (O2, S2, D1). The most common systematic problem. A subtype entry inherits frequency: from a parent-level source — ORPHA:666 (all OI), ORPHA:828 (all Stickler) — or asserts a severe-subtype rate at disease level. Three files already contain the correct pattern in miniature (OI's Hyperhidrosis note, Stickler's Cataract downgrade, Marfan's pneumothorax PARTIAL + notes:); it is applied inconsistently within the same files.

  2. supports drifting from explanation (M1, M3, M4, C3, C4, M5). Where an entry is wrong, it is usually the enum and the prose disagreeing rather than a bad quote. REFUTE items whose explanations affirm support, and SUPPORT items whose explanations concede non-support, are both present. Worth a lint: flag any REFUTE/NO_EVIDENCE whose explanation lacks negation, and any SUPPORT whose explanation contains "does not", "but only", or "different from".

  3. NO_EVIDENCE items retained as if they were support (C1, C2, C3). Concentrated almost entirely in Ehlers-Danlos_Syndrome_COL5A1-related. In two cases a claim's only evidence is NO_EVIDENCE, leaving a frequency-banded phenotype and a variant class standing on nothing.

  4. Stale explanations surviving claim edits (M3, M6). Explanations referencing values ("0.01-0.02", "childhood-adolescence") no longer present in the record. Because explanations are free text, no validator catches this.

  5. What is working. No fabrication was found anywhere in ~620 items. The three clean entries plus Kniest show the target standard: disease-specific quantified quotes, honest PARTIAL grading, contradicting evidence preserved rather than dropped, and limits of single-case data stated explicitly.

Disposition

Sixteen of the 26 findings were fixed in place on claude/ecm-disorders-review-ky690u (PR #6961); the remaining ten are tracked as issues because they need new literature, clinical input, or a curation-policy decision.

Fixed

Finding Entry Fix
M1 Marfan 4 REFUTE → SUPPORT/PARTIAL; dropped one non-bearing item
M2 Marfan prevalence_class → BAND_1_9_PER_100000, matching the recorded rate
M3 Marfan REFUTE → SUPPORT (6.5/100,000 is inside the interval); stale explanations rewritten
M4 Marfan Removed PMID:7911041 from Aortic Aneurysm
M6 Marfan Rewrote two stale progression explanations
C1 cEDS Removed unsupported Myopia phenotype + its causal edge
C2 cEDS Removed the NO_EVIDENCE item contradicting its own claim
C3 cEDS Dropped 4 NO_EVIDENCE items; hEDS-cohort item SUPPORT → PARTIAL
C5 cEDS Delayed Wound Healing VERY_FREQUENT → FREQUENT
C6 cEDS evidence_source added where unambiguous (5 items)
O1 OI type I 6 ORPHA rows HUMAN_CLINICAL → OTHER
O3 OI type I 3 truncated snippets replaced with clean sentences
A2 Alport Digenic snippet swapped for the paper's METHODS + CONCLUSIONS
K1 Kniest Attached the 5/7 clefting evidence to Cleft Palate Repair
S3 Stickler 2 conflicting Orphanet rows SUPPORT → PARTIAL
D2 DEB Removed the mismatched citation from the mucosal-blistering node

All seven edited files pass linkml-validate; the full tests/test_data.py suite passes (1648 tests); and every snippet still substring-matches its cache file.

Tracked as issues

Issue Findings Why deferred
#6951 O2, S2, D1 Grouping/spectrum frequency: leakage — needs a curation policy plus a DEB subtype: migration
#6952 A1, A3, A4 Alport phenotypes needing human citations
#6953 M5 Exercise direction-of-effect — needs clinical input on dynamic vs isometric framing
#6954 C4, C6 (bulk) ~45 remaining evidence_source tags; uncited frameshift variant class; cross-subtype pain evidence
#6955 cross-cutting QC proposal: lint supports enums against their own explanation
#6956 S1 Stickler prevalence reconciliation — needs epidemiology judgment

Note on tooling

linkml-reference-validator reports Total checks: 0 in this environment, including on files untouched by these edits — an environmental issue, not a content one. Snippet verification here was done with an offline substring checker against references_cache/ implementing the same contract (ellipsis-split, whitespace- and Unicode-normalised).