Claim–Evidence Review: 10 Cardiac Disorder Entries (2026-07-25)

Correctness review of ten cardiac entries in kb/disorders/, focused on whether each cited reference actually supports the claim it is attached to.

Status: remediation applied. Findings 1–7 have been fixed in the KB; see Remediation applied at the end of this report for exactly what changed and what deliberately remains open.

Scope and method

Entries reviewed: Brugada_Syndrome, Long_QT_Syndrome, Hypertrophic_Cardiomyopathy, Dilated_Cardiomyopathy, Arrhythmogenic_Right_Ventricular_Cardiomyopathy, Atrial_Fibrillation, Heart_Failure, Myocardial_Infarction, Coronary_Artery_Disease, Peripartum_Cardiomyopathy.

Three passes:

Mechanical (snippet fidelity). Every one of the 490 evidence items was checked offline against its references_cache/ file. Result: no fabricated or misquoted snippets. Twelve initial mismatches all resolved to transliteration or formatting artifacts, not curation errors — Greek letters written out (β-blockers → beta-blockers), ± → +/-, smart-vs-straight quotes around "strain", a trailing period on a truncated quote, and one line-wrap artifact in the cache (definitive ( MYBPC3). The anti-hallucination layer is doing its job.
evidence_source audit. Cross-checked each item tagged (or defaulting to) HUMAN_CLINICAL against what the cached paper actually reports.
Semantic (claim ↔ evidence). Read every claim with its snippets and explanations, asking whether the quote is probative for the specific assertion.

Pass 1 is clean. All substantive findings come from passes 2 and 3.

Headline: the ten entries fall into two sharply different tiers

Entry	Evidence items	`evidence_source` unset	`NO_EVIDENCE`	`PARTIAL`
Brugada_Syndrome	30	0	0	0
Long_QT_Syndrome	79	0	0	0
Myocardial_Infarction	24	0	0	0
ARVC	79	0	0	3
Peripartum_Cardiomyopathy	24	1	0	0
Dilated_Cardiomyopathy	82	15	0	0
Coronary_Artery_Disease	16	13	2	8
Atrial_Fibrillation	23	14	1	3
Heart_Failure	18	15	2	7
Hypertrophic_Cardiomyopathy	115	82	0	9

Tier A — Brugada, Long QT, MI, ARVC, Peripartum CM. These are model entries. Tight claim–evidence coupling, honest hedging, correct source typing. Highlights:

Peripartum_Cardiomyopathy splits a single paper (PMID:17289576) into two evidence items — MODEL_ORGANISM for the STAT3-knockout mouse result and HUMAN_CLINICAL for the patient serum result — exactly the discipline CLAUDE.md prescribes for mixed-source papers.
ARVC marks the Wnt/plakoglobin branch as provisional because its support is experimental, and flags the T-wave-inversion edge PARTIAL because the mechanistic bridge is genuinely unresolved.
Myocardial_Infarction — every snippet is directly probative of its node. No filler.
Brugada correctly types a bioinformatic hotspot study as COMPUTATIONAL and a patch-clamp study as IN_VITRO.

Tier B — HCM, DCM, Atrial Fibrillation, Heart Failure, CAD. These carry real claim–evidence defects, described below. The pattern suggests an older, more automated curation pass that was never revisited.

Finding 1 — `NO_EVIDENCE` items whose `explanation` argues the opposite

The most serious recurring defect. An evidence item is flagged supports: NO_EVIDENCE — correctly, because the snippet is non-probative — but the explanation then asserts the claim as established fact. A reader (or downstream export) sees a confident justification attached to a citation that does not support it.

Location	Reference	Snippet is about…	Explanation asserts…
`Coronary_Artery_Disease.genetic[0]` (APOE)	PMID:40594772	a methods sentence: "GWAS-significant CVD risk genes were used to calculate risk gene scores"	"GWAS studies have identified APOE as a significant cardiovascular disease risk gene"
`Coronary_Artery_Disease.genetic[1]` (LDLR)	PMID:40594772	same methods sentence	"LDLR is a well-established cardiovascular disease risk gene identified through GWAS"
`Atrial_Fibrillation.pathophysiology[3]` (Atrial Thrombus Formation)	PMID:38255832	"AF is an arrhythmia that affects the left atrium, cardiac function, and the patients' survival rate"	"AF affects cardiac function including atrial contraction, which contributes to thromboembolism risk"
`Heart_Failure.pathophysiology[1]` (Neurohormonal Activation)	PMID:33432192	difficulty of modelling HFpEF in animals	"highlights the complexity of neurohormonal effects extending beyond the heart"
`Heart_Failure.pathophysiology[3]` (Fluid Retention)	PMID:36769308	need for new treatments and biomarkers	"supporting the mechanism of fluid retention through impaired kidney function"

Neither APOE nor LDLR is named anywhere in the quoted text. Both gene–disease relationships are trivially citable; the same entry could use the ClinGen (CGGV:) pattern already used successfully in HCM and DCM.

Finding 2 — non-probative citations marked `SUPPORT`

Cases where the explanation itself concedes the quote does not establish the claim, yet supports: SUPPORT is retained.

Hypertrophic_Cardiomyopathy.treatments[1] (Calcium Channel Blockers), PMID:3515244. Snippet is about CCBs for angina. Explanation: "Even though the focus is on angina, the acknowledged use of calcium channel blockers reinforces their role in cardiovascular conditions including HCM." That is a non-sequitur. Note the adjacent item (PMID:36044874, hypertensive LVH) was correctly downgraded to PARTIAL with an honest rationale — so the curator knew the pattern and missed this one.
Hypertrophic_Cardiomyopathy.treatments[2] (Septal Myectomy), PMID:22687587. Explanation: "this reference does not directly state septal myectomy… Therefore, it indirectly supports the use."
Hypertrophic_Cardiomyopathy.phenotypes[3] (Arrhythmias), PMID:34969871. Snippet is a prevalence trend ("Between 2010 and 2018, prevalence increased for ARVC by 180% and HCM by 9%") used to support arrhythmia as an HCM phenotype. The same PMID is used appropriately in prevalence[0].
Hypertrophic_Cardiomyopathy.phenotypes[4] (Arrhythmias), PMID:29203161. A generic textbook definition of cardiac arrhythmia that never mentions HCM; the explanation supplies "including hypertrophic cardiomyopathy."
Heart_Failure.treatments[3] (SGLT2 Inhibitors), PMID:41110921. Claim: "Reduce hospitalizations and mortality across HF spectrum." Snippet: "Recommendations are complemented by practical tips to guide the initiation, titration, and maintenance of these foundational treatments." The quote names neither SGLT2 inhibitors nor any outcome. One of the most citable results in cardiology (DAPA-HF, EMPEROR) is effectively uncited.
Dilated_Cardiomyopathy.phenotypes[2] (LV systolic dysfunction), PMID:39298146. Snippet: "Left ventricular ejection fraction (LVEF) (per 1%) was not associated with all-cause mortality." This is a null prognostic finding cited to support a near-defining DCM phenotype; the explanation concedes it "shows it is less prognostic than fibrosis markers."
Dilated_Cardiomyopathy.phenotypes[0] — phenotype named identically to the disease ("Dilated cardiomyopathy"), supported by "Accurate risk stratification of NIDCM remains challenging."

Finding 3 — semantic conflation: same word, different claim

Hypertrophic_Cardiomyopathy.biochemical[0] (Troponin), PMID:15631686. The claim is troponin as a circulating biomarker of myocardial injury. The snippet is about mutations in troponin genes causing cardiomyopathy. Two unrelated senses of "troponin"; the explanation bridges them by inference ("leading to myocardial damage where elevated troponin can be expected").
Coronary_Artery_Disease.biochemical[0] (LDL) and [2] (C-Reactive Protein). Both cite the same snippet from PMID:39518492 — "inflammation, lipid accumulation, and smooth muscle cell proliferation" — which names neither LDL nor CRP.

Finding 4 — mechanism inferred backwards from treatment response

Dilated_Cardiomyopathy.pathophysiology[1] (Neurohormonal Activation) is supported entirely by two snippets about therapy (RAS inhibitors and beta-blockers improve prognosis), with the explanation reasoning "…validating neurohormonal activation as a key pathophysiological mechanism." Drug efficacy is weak evidence for mechanism, and direct evidence for neurohormonal activation in HF/DCM is abundant. (By contrast, Heart_Failure.pathophysiology[1] cites PMID:37895150 directly measuring SNS/RAAS/AVP activation — the right pattern.)

Finding 5 — `evidence_source` mis-typing (animal/in-vitro tagged as human)

Per CLAUDE.md, unset evidence_source defaults to HUMAN_CLINICAL. HCM has 82 of 115 items unset and DCM 15 of 82, and several of those papers are explicitly not human clinical:

Location	Reference	What the paper actually is
`HCM.pathophysiology[2].evidence[0]`	PMID:1414892	spontaneously hypertensive rats; also a hypertensive-LVH paper, not HCM
`HCM.pathophysiology[2].evidence[2]`	PMID:29522370	mouse pressure-overload SPARC/macrophage study
`HCM.pathophysiology[0].evidence[1]`, `genetic[0].evidence[3]`	PMID:36797478	base editing in humanized mice + iPSC-CMs
`HCM.treatments[0].evidence[2]`	PMID:37850394	R-carvedilol in Myh6R403Q mice + iPSC-CMs
`Heart_Failure.pathophysiology[2]`, `phenotypes[5]`	PMID:38636927	isoproterenol-induced animal model (Lilrb4a)
`Atrial_Fibrillation.pathophysiology[0]`, `environmental[1]`	PMID:39146015	DIO mice + hiPSC-aCMs + human tissue (mixed — should be split)
`Coronary_Artery_Disease` (5 separate items)	PMID:40594772	scRNA-seq of mouse carotid arteries + human carotid plaques (mixed)

Two consequences worth calling out:

Heart_Failure.phenotypes[5] (Cardiomegaly) is supported solely by the mouse isoproterenol snippet. CLAUDE.md states model-organism evidence should not be the only support for a human phenotype.
Dilated_Cardiomyopathy.pathophysiology[6] (Mitochondrial Dysfunction) is supported solely by PMID:35418250, a doxorubicin cardiotoxicity mouse study — a distinct etiology — correctly tagged MODEL_ORGANISM but still the only support.

HCM demonstrates the inconsistency is local, not conceptual: PMID:25573453 (feline HCM) is correctly tagged MODEL_ORGANISM while the neighbouring rodent papers are not.

Finding 6 — claims that outrun their evidence

Dilated_Cardiomyopathy.pathophysiology[4] (Immune and Inflammatory Activation). Claim asserts "viral myocarditis or autoimmune reactions can trigger or accelerate DCM"; the sole evidence is a COMPUTATIONAL immune-cell deconvolution study that addresses neither viral myocarditis nor autoimmunity.
Dilated_Cardiomyopathy.pathophysiology[3] (RNA Splicing Dysregulation). Detailed claim about RBM20 splicing of TTN/CAMK2D/CACNA1C and toxic biomolecular condensates; the sole snippet only states RBM20 variants are linked to aggressive DCM. The entire mechanistic content is uncited.
Atrial_Fibrillation.pathophysiology[0] (Atrial Electrical Remodeling). Description asserts the classic rate-dependent "AF begets AF" mechanism (shortened refractory period, loss of rate adaptation); the evidence is about obesity/NOX2-mediated remodeling. The canonical claim is uncited.
Heart_Failure.pathophysiology[0] covers systolic and diastolic dysfunction, but both evidence items are HFpEF/diastolic — the systolic half is unsupported.
Heart_Failure.phenotypes[0] (Dyspnea) gives a detailed mechanism (elevated left atrial and pulmonary venous pressure driving fluid into the interstitium) but cites an HFpEF fibrosis/exercise-tolerance snippet that never mentions dyspnea. Same shape for phenotypes[1] (Peripheral Edema), cited to a generic cardiorenal sentence.

Finding 7 — structural and scope defects

Duplicate phenotype node. Hypertrophic_Cardiomyopathy.phenotypes[3] and phenotypes[4] are both Arrhythmias / Cardiovascular / HP:0011675 / FREQUENT. These should be merged.
HCM.environmental[0] is self-refuting. The node is named None Applicable and carries a single supports: REFUTE item whose explanation states "The statement that hypertrophic cardiomyopathy is not influenced by environmental factors is incorrect." The KB is retaining a claim it knows to be false rather than curating the real environmental factors (strenuous exertion, heat) that the cited paper describes.
CAD MONDO mapping is too broad. Coronary_Artery_Disease.disease_term is bound to MONDO:1060134 atherosclerotic cardiovascular disease ("Any cardiovascular disease resulting from atherosclerosis" — includes stroke and PAD). The entry's own description defines CAD as coronary narrowing, so a coronary-specific class is needed. (Resolved independently on main in #7187, which bound it to MONDO:0021661 coronary atherosclerosis — a child of MONDO:0005010 coronary artery disorder and a close match to this entry's atherosclerosis-centred description.)
Carotid evidence used for coronary claims. CAD's two main pathophysiology sources (PMID:38639096, PMID:40594772) are carotid-plaque studies applied to coronary disease without qualification. Defensible for general atherogenesis, but it should be stated.
Single-source overload. PMID:40594772 carries 5 separate CAD claims; PMID:38255832 carries 3 AF claims from near-identical generic sentences.
Curation process notes leaking into explanation. Dilated_Cardiomyopathy.pathophysiology[0].evidence[3] ends with "From the Stanford iPSC-cardiomyocyte 'clinical-trials-in-a-dish' program (Joseph Wu group) presented at the NAMeRS 2026 symposium (issue #4873)." Provenance belongs in notes or a history record, not in a scientific justification field.
HCM datasets section is dominated by spaceflight data — 7 NASA/GeneLab entries (ISS heart-on-a-chip, hiPSC-CMs in microgravity, mouse hearts on ISS, the NASA Twins Study). Relevance to HCM is asserted in the descriptions rather than demonstrated; the Twins Study item justifies itself via carotid intima-media thickness as "context for understanding how spaceflight stress may interact with HCM susceptibility." This reads as a bulk import into a disease entry where it does not belong.
Deprecated percentage field still populated in HCM.prevalence[0] (percentage: 0.2) alongside the structured rate_per_100000.

Finding 8 — large unevidenced sections in the four common-cardiology entries

Blocks carrying zero evidence items:

Entry	Section	Unevidenced
Heart_Failure	treatments	7/8 — ACE-I/ARB, beta blockers, ARNI, MRA, diuretics, CRT, ICD
Heart_Failure	genetic	3/3 — TTN, MYH7, LMNA
Coronary_Artery_Disease	treatments	5/6 — antiplatelets, ACE-I, beta blockers, PCI, CABG
Coronary_Artery_Disease	genetic	2/4 — PCSK9, 9p21
Atrial_Fibrillation	treatments	5/7 — rate control, rhythm control, cardioversion, LAA closure, risk-factor modification
Atrial_Fibrillation	genetic	4/5 — KCNQ1, KCNE2, KCNJ2, SCN5A
Dilated_Cardiomyopathy	treatments	6/7 — ACE-I/ARB, beta blockers, SGLT2i, CRT, transplant, ICD

This is the largest volume gap found. These are among the best-evidenced interventions in medicine, and Heart_Failure's genetic section could be filled immediately from ClinGen using the CGGV: pattern already working in HCM and DCM.

Recommended remediation, in priority order

Fix the contradictory NO_EVIDENCE/SUPPORT metadata (Findings 1–2). Either replace the citation with a probative one or delete the evidence block and keep the claim in notes, per the CLAUDE.md SOP. Highest priority: CAD APOE/LDLR, Heart_Failure SGLT2 inhibitors, DCM LV systolic dysfunction.
Backfill evidence_source on the 139 unset items across HCM/DCM/AF/HF/CAD, splitting mixed-source papers (Finding 5). Remove or supplement the two phenotype nodes resting solely on model-organism evidence.
Merge the duplicate HCM Arrhythmias phenotype and replace HCM.environmental[0] = "None Applicable" with the real exertion/heat risk factors its own citation describes.
Re-map CAD to a coronary-specific MONDO class.
Evidence the empty treatment and genetic sections (Finding 8); ClinGen covers the genetic ones mechanically.
Narrow the over-broad claims in DCM pathophysiology[3]/[4] and AF pathophysiology[0] to what their citations actually support.
Re-assess the HCM spaceflight datasets block for scope.

Tier A entries (Brugada, Long QT, MI, ARVC, Peripartum CM) need no remediation and are the right templates for the Tier B rewrites.

Remediation applied

All five Tier-B entries were edited. Post-fix state:

Entry	Evidence items	`evidence_source` unset	`NO_EVIDENCE`
Hypertrophic_Cardiomyopathy	108	0	0
Dilated_Cardiomyopathy	83	0	0
Atrial_Fibrillation	26	0	0
Heart_Failure	23	0	0
Coronary_Artery_Disease	19	0	0

Contradictory metadata removed (Findings 1–2). The five NO_EVIDENCE-with-affirming- explanation items and the non-probative SUPPORT items were replaced or deleted:

CAD APOE → PMID:17878422 (apoE genotype–coronary risk meta-analysis, 121 studies).
CAD LDLR → PMID:28444290 (EAS consensus: reduced LDL-receptor function raises ASCVD risk).
CAD PCSK9 (was unevidenced) → PMID:16554528 (ARIC: PCSK9 nonsense variants, −88% CHD).
CAD 9p21 (was unevidenced) → PMID:17478681 (original 9p21 GWAS).
CAD LDL Cholesterol → PMID:28444290; CAD C-Reactive Protein → PMID:20031199 (ERFC meta-analysis).
CAD Myocardial Infarction phenotype → PMID:24902970 (plaque rupture → acute coronary syndrome).
Heart_Failure SGLT2 inhibitors → PMID:31535829 (DAPA-HF outcome data added alongside the guideline).
Atrial_Fibrillation thrombus node → PMID:8572814 (91% of nonrheumatic AF thrombi in the LAA).
DCM LV systolic dysfunction and Dilated cardiomyopathy phenotypes → PMID:39519012; the null-LVEF result is retained but demoted to PARTIAL and reframed as a qualifier.
HCM: six non-probative items deleted (the prevalence-trend and generic-arrhythmia citations, the troponin gene/biomarker conflation, the CCB-for-angina and indirect-myectomy citations, and the LVH-in-diverse-conditions subtype citation), plus one item whose "snippet" was just an article title.

evidence_source backfilled (Finding 5). 139 items across the five entries, classified per reference: narrative reviews/guidelines/expert curation → OTHER; original human studies → HUMAN_CLINICAL; in vivo animal work → MODEL_ORGANISM. Mixed-source papers (PMID:39146015, PMID:40594772) were assigned per snippet.

Structural fixes (Finding 7). The duplicate HCM Arrhythmias phenotype was merged (richer node kept, evidence folded in, 11 → 10 phenotypes); HCM.environmental[0] changed from the self-refuting None Applicable/REFUTE to Strenuous exertion and heat stress with SUPPORT; the deprecated percentage: 0.2 was dropped; CAD re-mapped to MONDO:0021661 coronary atherosclerosis; the carotid-to-coronary extrapolation is now stated explicitly in the two affected explanations; the NAMeRS/issue-number curation note was removed from a DCM evidence explanation.

Over-broad claims narrowed (Finding 6). DCM RNA Splicing Dysregulation and Immune and Inflammatory Activation descriptions now separate what the cited evidence shows from what is proposed; the DCM neurohormonal explanation now states it is therapeutic rather than direct mechanistic evidence; the "AF begets AF" claim gained its actual source (PMID:7671380, Wijffels).

Partial progress on Finding 8. Heart_Failure genetic (TTN/MYH7/LMNA) now uses ClinGen CGGV: assertions; HF ACE-I/ARB, beta blockers and MRA anchor on PMID:37254024 (GDMT); AF rhythm control and risk-factor modification anchor on PMID:40526576; CAD antiplatelet therapy on PMID:11786451.

Still open

Treatment and phenotype blocks that remain unevidenced — a curation-volume task, not a correctness defect: DCM treatments (6/7), CAD treatments (4/6: ACE-I, beta blockers, PCI, CABG), HF treatments (4/8: ARNI, diuretics, CRT, ICD), AF treatments (3/7: rate control, cardioversion, LAA closure), AF genetic (KCNQ1, KCNE2, KCNJ2, SCN5A), and assorted phenotype/biochemical blocks. The HCM spaceflight datasets scope question (Finding 7) is a curation-policy judgement and was left for maintainers.

Verification

All 495 evidence items across the ten entries re-checked offline against references_cache: 0 snippet mismatches. (The single flagged HCM item is the pre-existing definitive ( MYBPC3 line-wrap artifact in the cache, not a curation error.)
linkml-validate against the Disease class: no issues on all five edited entries.
linkml-term-validator with --labels: passes on all five. Note that the configured ols:mondo adapter cannot perform ancestor traversal, so cache/enums/diseaseterm_*.csv is the effective membership source for MONDO terms; a new binding needs its CURIE there, verified against the local sqlite:obo:mondo adapter.
pytest tests/test_data.py: 28 global structural tests and all 55 per-file tests for the edited entries pass. (Remaining warnings about missing subtype_term are pre-existing.)
Eight new reference cache files were fetched with the validator, never hand-written. Three PMIDs initially recalled from memory turned out to be unrelated papers and were discarded — the correct sources were located via PubMed search and verified before use.