Skip to content

Clinical Care Guidelines Registry (seed)

Tracking issue: #4878 — collect clinical care guidelines

Goal. Scale the approach used for the Fanconi Anemia entry — deep integration of a disease-specific clinical care guideline — to other disorders where comparable high-quality guidelines exist. This file is the seed registry plus a reusable search / prioritization / gap-assessment methodology. It is a worklist, not curated KB content.

Provenance & verification policy. Every PMID in the "Verified guideline" tables below was checked with just fetch-reference PMID:<id> and its real PubMed title is reproduced verbatim. Do not add any PMID to a KB entry from this file without re-running just fetch-reference + just validate-references on the entry, per CLAUDE.md. The registry records which guideline document to mine, never a pre-validated snippet.

The Fanconi Anemia reference model

kb/disorders/Fanconi_Anemia.yaml is the deepest entry in the KB (4,789 lines, ~246 PMID references). It was built from the 5th edition Fanconi Anemia Clinical Care Guidelines, anchored on two verified sources:

Source PMID (verified title)
FA GeneReviews chapter PMID:20301575 — Fanconi Anemia.
FA endocrine screening recommendations PMID:25575015 — Endocrine disorders in Fanconi anemia: recommendations for screening and treatment.

The guideline drove (a) phenotype coverage (screening-indicator tables → individual phenotype entries) and (b) treatment/surveillance content (e.g. annual TSH/free-T4, biennial cancer surveillance), with notes: fields tying phenotypes back to specific guideline tables. That "guideline table → phenotype/surveillance rows + frequency data" pattern is the template to replicate.

⚠️ Anti-hallucination audit of the auto-generated issue summary

The auto-summary on #4878 proposed a Tier-1 table of guideline PMIDs. 8 of the 10 suggested PMIDs are hallucinated / misattributed — they resolve to unrelated papers. This mirrors the DR-failure pattern documented in CLAUDE.md (§2a) and the #4873 precedent (both PMIDs there were also wrong). The verified replacements are in the next section. Do not use any of the debunked PMIDs below.

Disease (as claimed) Suggested PMID Actually resolves to Verdict
Tuberous Sclerosis Complex PMID:24053983 2012 International TSC Consensus surveillance/management ✅ correct
Ehlers-Danlos Syndrome PMID:28306229 2017 international classification of the EDS ✅ correct
Duchenne Muscular Dystrophy PMID:29940086 "Barriers… Long-Acting Reversible Contraceptives in Massachusetts Community Health Centers" ❌ hallucinated
Spinal Muscular Atrophy PMID:28528957 "Proteasomes in corneal epithelial cells…" ❌ hallucinated
Spinal Muscular Atrophy PMID:29681584 "Fate of Acute Heart Failure Patients With Mid-Range Ejection Fraction" ❌ hallucinated
Neurofibromatosis type 1 PMID:33860374 plant pathology (Xylella fastidiosa in Spartium junceum) ❌ hallucinated
Turner Syndrome PMID:28049635 "Cost… clinical decision support systems for cardiovascular disease prevention" ❌ hallucinated
Prader-Willi Syndrome PMID:26483315 "…21-Gene Recurrence Score Assay on Chemotherapy Delivery in Breast Cancer" ❌ hallucinated
Phenylketonuria PMID:33808622 "…Microbial Consortia for Improving Gluten Digestion…" ❌ hallucinated
Marfan Syndrome PMID:28286880 "The [5+5] route to the phenanthrene skeleton" (organic chemistry) ❌ hallucinated

Takeaway: auto-summary PMID suggestions for this issue ran ~80% hallucinated. Treat all machine-suggested citations as leads to verify, never ground truth.

Tier 1 — verified disease-specific foundation guidelines

Each PMID below was fetched and title-verified. The dismech entry and cites guideline? columns were checked against kb/disorders/ on main (this is the gap-assessment).

Disease Verified guideline PMID & title dismech entry Already cites guideline?
Duchenne Muscular Dystrophy PMID:29395989 — Diagnosis and management of Duchenne muscular dystrophy, part 1… Duchenne_Muscular_Dystrophy.yaml No → mine
Spinal Muscular Atrophy PMID:29290580 — Diagnosis and management of spinal muscular atrophy: Part 1… Spinal_Muscular_Atrophy.yaml No → mine
Tuberous Sclerosis Complex PMID:24053983 — Tuberous sclerosis complex surveillance and management: recommendations of the 2012 International TSC Consensus Conference. Tuberous_Sclerosis_Complex.yaml No → mine
Neurofibromatosis type 1 PMID:31010905 — Health Supervision for Children With Neurofibromatosis Type 1. Neurofibromatosis_Type_1.yaml No → mine
Marfan Syndrome PMID:37389507 — 2022 ACC/AHA guideline for the diagnosis and management of aortic disease… Marfan_Syndrome.yaml No → mine
Phenylketonuria PMID:24385074 — Phenylalanine hydroxylase deficiency: diagnosis and management guideline. (ACMG) Phenylketonuria.yaml Yes (already integrated)
Ehlers-Danlos Syndrome PMID:28306229 — The 2017 international classification of the Ehlers-Danlos syndromes. Ehlers-Danlos_Syndrome.yaml (+ 4 subtype entries) No → mine
Turner Syndrome PMID:28705803 — Clinical practice guidelines for the care of girls and women with Turner syndrome: proceedings from the 2016 Cincinnati International Turner Syndrome Meeting. none — new-entry gap n/a

Gap summary: 6 of 7 already-curated entries do not yet cite their disease's foundation guideline (only PKU does). Turner syndrome has a high-quality guideline but no dismech entry at all — the single new-entry opportunity in this seed batch.

Leads needing confirmation (not yet verified — do not cite as-is)

  • Prader-Willi Syndrome — PMID:22237428 resolves only to the GeneReviews stub ("Prader-Willi syndrome."), not the management consensus. The true diagnosis/management consensus PMID still needs to be located and verified before use.

Methodology (reusable)

Search strategy (tiers, highest yield first)

  1. Tier 1 — disease-specific foundation/consensus guidelines. Patient-foundation or international-consortium documents with tabular screening indicators and surveillance schedules (the FA model). Highest curation yield.
  2. Tier 2 — GeneReviews chapters. Standardized Management / Surveillance sections across >900 chapters; the highest-coverage resource for inherited-disease management. Cross-reference existing dismech entries first.
  3. Tier 3 — Orphanet / European Reference Network (ERN) guidelines. Best for rare diseases with little other structured guidance.

Prioritization criteria

  1. HPOA density gap — diseases whose canonical phenotype.hpoa annotations are sparse relative to what a guideline covers (cross-ref #3179, #3846).
  2. Entry-completeness gap — diseases with a stub dismech entry but few phenotypes/treatments (compliance dashboard).
  3. Guideline richness — explicit tabular screening indicators (like FA Table 1) map far more cleanly than narrative-only guidelines.
  4. Frequency data present — guidelines with frequency columns ("80% of patients") directly address the HPOA frequency-annotation gap (#3179).
  5. Surveillance/treatment novelty — content not otherwise capturable from individual papers.

Gap-assessment procedure (deterministic, repeatable)

  1. Verify a candidate guideline PMID with just fetch-reference PMID:<id>; confirm the title is the guideline (the anti-hallucination step — ~80% of unverified leads failed here).
  2. Check kb/disorders/ for an existing entry (grep -ril <disease>); if absent, route to a new-entry curation issue (high-effort tier).
  3. If an entry exists, grep <PMID> it to see whether the guideline is already integrated.
  4. Score with the criteria above; open a bounded per-disease child issue for the highest-ranked targets.

Suggested next actions

  • [ ] Confirm the Prader-Willi consensus-guideline PMID; extend the registry with verified Tier-1 rows for additional foundation guidelines (CF, sickle cell, MPS subtypes, etc.).
  • [ ] Open bounded per-disease child issues to mine the 6 "No → mine" Tier-1 guidelines into their existing entries (start with the richest tabular guidelines: TSC, DMD, SMA).
  • [ ] Route Turner syndrome to a new-entry curation issue (verified guideline available, no entry yet).
  • [ ] Consider a schema follow-up to capture guideline provenance distinctly from individual PMID evidence items (e.g. a [GuidelineCareRecommendation] reference tag, mirroring the existing [GeneReviews] tag convention).