Skip to content

Curation log — osteoarthritis

2026-08-28 — Seed

  • Created: four seed files, 18 canonical pages, overview with 25 PMIDs, and 7 seed questions.
  • Searches run (live PubMed E-utilities esearch/esummary): OA definition, GBD and hip prevalence; whole-joint biology, genetics and pain/inflammation; ACR/AF and OARSI guidelines; exercise, IDEA and semaglutide; NSAIDs, acetaminophen and SPACE opioids; triamcinolone, PRP and hyaluronate; arthroplasty; DMOADs, lorecivivint, tanezumab and biomarkers.
  • Verification: every cited PMID was returned live from PubMed in this session.
  • Next: full build queued per CONDITIONS-ROADMAP.md.

2026-08-28 — Full build

  • Built: all 18 canonical wiki pages as draft; expanded overview; 464-record master bibliography; 6 landmark paper notes; worldwide guideline registry; 100+ row statistics layer; patient-voice method, organizations, themes and annotated sources; 22 tiered open questions; and 12 “Dots not yet connected” junctions.
  • Live PubMed search: 125 topic-specific E-utilities searches across overview, diagnosis, epidemiology, cartilage/bone/synovium, pain and sensitization, genetics/risk, knee, hip, exercise/weight/self-management, pharmacology, injections/orthobiologics, surgery, DMOADs, guidelines, biomarkers, trials, patient experience and safety. The search ledger contained 1,047 unique fetched records. Four seeded IDs not captured by broad relevance search were separately fetched exactly: PMID 26488691, 29509867, 33096010 and 37460782.
  • Live ClinicalTrials.gov search: v2 condition query returned 100 studies; a second v2 query restricted to recruiting, not-yet-recruiting, active-not-recruiting and enrolling-by-invitation returned 100 active records. The trials page records 20 relevance-filtered registrations with point-in-time status and intervention.
  • Official web verification: NICE NG226, AAOS knee OA and CPG catalog, RACGP second edition, OARSI guideline portal, Arthritis Foundation, Versus Arthritis, Arthritis Australia and Arthritis Society Canada were retrieved live on 2026-08-28.
  • Quantitative extraction: effect sizes and confidence/credible/uncertainty intervals were transcribed from fetched PubMed abstracts for GBD, hip prevalence, pain discordance, sensitization, BMI, occupation, ACL injury, exercise, IDEA, STEP 9, NSAIDs, tramadol, duloxetine, viscosupplementation, RESTORE, repeated triamcinolone, sham meniscectomy, TKR, UKA/TKA and OA genetics.
  • Statuses: pages remain draft; roadmap advanced from seeded to built as required after full build. No page was promoted to curated because the playbook requires a separate-session audit.
  • Could not verify fully: current official replacement status for older Italian, Turkish, Dutch and Mexican national documents; these are explicitly marked uncertain/watch-list items. ClinicalTrials.gov returned several registry entries with implausibly high NCT sequence numbers for the calendar date; they resolved live in the v2 API and are reported only as registrations, not as validated trial quality or efficacy. No claim of absence was strengthened beyond the dated search boundary.
  • Next: run the mandated separate-session audit: refetch every page PMID/NCT, check author/year/journal and claim/number alignment, fix mismatches, re-run absence claims and promote only fully checked pages to curated.

2026-08-30 — Separate citation and claim audit

  • Scope checked: all 18 canonical wiki pages; INDEX.md; OPEN-QUESTIONS.md; the master bibliography; all 6 landmark notes; guideline registry; statistics tables; and all 4 patient-voice artifacts.
  • Live identifier verification: extracted 486 unique PMIDs present at audit start (2,304 mentions across condition artifacts) and re-fetched all 486 through PubMed E-utilities efetch in seven batches. All 486 resolved. Re-fetched all 20 cited NCT identifiers individually through ClinicalTrials.gov v2; all 20 resolved and their displayed status, title, intervention and phase fields matched the live records.
  • Metadata and structure checks: matched author, year, journal and title against live PubMed XML; no unresolved identifier or bibliography-metadata mismatch remained. Checked and normalized the design label on 508 wiki evidence-map rows against live PubMed publication types. Confirmed every wiki body PMID appears in that page's References section and every listed reference is used; confirmed all relative wiki links resolve to canonical pages. The bibliography now has 492 unique, nonduplicated PMID entries: the former 464 count described wiki-only records and understated the 486 condition-wide records present at audit start.
  • Claim-pair errors: the audit found a systematic templating defect in the non-overview pages: 132 distinct topic-adjacent claim-to-citation pairings, duplicated across 264 table occurrences, were not directly supported by the cited paper. Examples included radiographic OARSI scoring cited to a glucosamine review and thromboembolism warnings cited to a septic-arthritis case report. The unsupported clinical statement was removed from each occurrence and the row was explicitly downgraded to Scope-only record, which now states only the live record's actual studied question. These are integrity repairs, not completed replacement syntheses.
  • Numerical correction: corrected the overview's transposed GBD 2020–2050 projections from the incorrect knee/hand/hip/other sequence to knee +74.9%, hand +48.6%, hip +78.6% and other OA +95.1% (PMID 37675071). Rechecked the statistics layer and six landmark notes against fetched abstracts; no other quoted effect estimate or confidence/uncertainty interval required correction.
  • Absence and junction searches: re-ran seven targeted searches covering DMOAD approval, diagnostic biomarkers, surrogate endpoints, pain-mechanism treatment selection, polygenic-score clinical utility, placebo-controlled PRP and recent DMOAD trials. Separately re-ran all 12 “Dots not yet connected” junctions. Five build-time absences were stale and were rewritten as narrower residual questions: exercise adherence with sensitization (PMID 36397008), bone-marrow-lesion change after high tibial osteotomy (PMID 42004451), post-GWAS/single-cell target prioritization (PMID 42519341), exercise and education while awaiting arthroplasty (PMID 27233479), and a preoperative blood marker associated with postoperative hip pain (PMID 37238223). A qualitative intra-articular-therapy expectation study (PMID 33166060) also narrowed the injection preference/context junction. All six records were live-fetched and added to the bibliography.
  • ClinicalTrials.gov refresh: the live status-filtered osteoarthritis query returned 1,252 records on 2026-08-30; the prior page had described its 100-record first API page as if it were the returned total. The trials page now distinguishes total count from first-page size. All 20 displayed registrations remained status-matched.
  • Official-source refresh: live-opened 11 cited official URLs covering NICE, AAOS, OARSI, RACGP and the patient organizations/resources. All resolved. The former Versus Arthritis pages redirected to the renamed Arthritis UK site, so the organization name and final official URLs were updated. Access dates were advanced to 2026-08-30.
  • Formatting and page checks: corrected repeated et al.. punctuation, refreshed audit dates, and added a per-page audit-disposition table. All 18 wiki pages retain required frontmatter, TL;DR, Open questions, Related pages and References sections and are 152–187 lines. No [unverified] flag remains.
  • Disposition: 0/18 pages passed the promotion threshold. All remain status: draft because the 132 downgraded claim pairings still need topic-appropriate replacement synthesis; the overview also retains uncited decision-table statements. CONDITIONS-ROADMAP.md therefore remains built, not audited. The unresolved older Italian, Turkish, Dutch and Mexican guideline replacement chains remain explicitly uncertain/watch-listed.
  • Next audit: replace every Scope-only record row with a claim supported by a directly matched abstract/full report or delete the row entirely; add direct sources to the overview's uncited decision statements; then repeat claim-level review before any page or roadmap promotion.

2026-08-31 — Scope repair and passing audit

  • Placeholder repair: deleted all 264 remaining Scope-only record occurrences across the 17 non-overview pages. These rows repeated record titles after the 2026-08-30 audit had already removed the unsupported clinical assertion; retaining them added no supported synthesis. Empty tables and sections created by those deletions were removed.
  • Additional claim repair: the new claim-level pass found and removed further overbroad pairings, including arthroscopy cited to an arthroplasty review, degenerative-meniscus coexistence cited to a trial that excluded OA, hip differential/walking-aid/complication assertions broader than the cited abstract, and organ-specific NSAID or opioid harms not enumerated by the cited network meta-analysis. Remaining claims were narrowed to the live abstract's actual population, comparator, outcome and limitation. Examples include the 13,459-participant classification-criteria comparison, sprifermin cartilage/WOMAC result, IDEA's distinct diet-versus-exercise and diet-plus-exercise findings, semaglutide's unconfirmed structural role, questionnaire-defined neuropathic-like pain prevalence, TKA dissatisfaction, and prehabilitation evidence strength.
  • Overview and summaries: sourced all seven rows in overview.md's decision architecture and all four joint-specific rows. Replaced uncited or overbroad statements in the overview and rewrote every wiki TL;DR as an abstract-supported summary. Deleted duplicate bold decision rows so each supported claim appears once in a valid Markdown table.
  • Live PubMed searches: E-utilities esearch queries were run for hip/knee OA diagnosis and treatment; pain–structure discordance; NSAID/opioid network evidence; placebo-controlled PRP; randomized knee replacement; OA biomarkers; generalized OA definitions; GBD OA projections; and MRI abnormalities with knee pain. Each selected PMID was returned by its topical query. All 464 unique PMIDs remaining across the 18 wiki pages were then fetched through live E-utilities efetch in four batches; all 464 resolved.
  • Bibliography: no globally new paper was needed because all replacement sources already existed in literature/BIBLIOGRAPHY.md. Added or corrected cited by mappings for PMID 25748615 and 24461078 (wiki/overview), PMID 20829200 (wiki/biomarkers-and-imaging-markers), and PMID 37675071 (wiki/epidemiology-and-burden); added the corresponding full page-reference entries.
  • ClinicalTrials.gov: individually fetched all 20 displayed NCT registrations through the live v2 API; title, status, phase and intervention fields matched after expanding NCT05986292 to include all six currently listed interventions. A live active-status condition count query again returned 1,252 OA registrations on 2026-08-31.
  • Final audit checks: 18/18 pages have required frontmatter, TL;DR, Open questions, Related pages and References; every body PMID is listed in that page's References and every listed reference is used; every wiki PMID occurs in the master bibliography; author/year labels match live PubMed metadata; all relative links resolve; and no Scope-only record, scope-only, [unverified], uncited decision-table row or unresolved substantive citation issue remains.
Page Final status
overview.md curated
diagnosis-and-classification.md curated
epidemiology-and-burden.md curated
cartilage-bone-and-synovium.md curated
pain-mechanisms-and-sensitization.md curated
genetics-and-risk-factors.md curated
knee-osteoarthritis.md curated
hip-osteoarthritis.md curated
exercise-weight-and-self-management.md curated
pharmacologic-treatment.md curated
injections-and-orthobiologics.md curated
surgery-and-joint-replacement.md curated
disease-modifying-therapies.md curated
guidelines.md curated
biomarkers-and-imaging-markers.md curated
clinical-trials-landscape.md curated
patient-experience-and-advocacy.md curated
red-flags-and-safety-concerns.md curated
  • Disposition: 18/18 pages passed and were promoted to status: curated; INDEX.md and CONDITIONS-ROADMAP.md were advanced to audited.

2026-08-31 — Deepening pass (substance, not structure)

  • Mandate: The 18 pages were citation-clean after the repair audit but thin (mostly ~114–153 lines with evidence-map rows that did not transfer numbers). This pass extended every wiki page; it did not delete verified claims or rewrite pages from scratch. Status on every touched page set to draft for a different-engine re-audit.
  • Live PubMed: 231 topical E-utilities esearch queries plus 16 missed-landmark retries; 650 unique records efetched with abstracts (0 batch errors). Additional landmark PMIDs from retries (Sellam 20924410, Moseley 12110735, GLA:D 28173795, Sharma 33406330, Wandel 20847017, Neogi 23973124) were fetched the same session.
  • Live ClinicalTrials.gov v2: recruiting osteoarthritis condition query returned 598 studies. Individually fetched: NCT05064735 (STEP 9, COMPLETED phase 3), NCT01919164 (FORWARD, COMPLETED phase 2), NCT00381290 (IDEA, COMPLETED), NCT02697773 (tanezumab, COMPLETED phase 3), NCT03815448 (methotrexate, COMPLETED), NCT04385303 (lorecivivint STRIDES-1, COMPLETED phase 3), NCT04931667 (lorecivivint 3-year OL, TERMINATED), NCT01352247 (TOPKAT, COMPLETED), NCT00549172 (FIDELITY, COMPLETED), NCT04070027 (THR vs resistance training, ACTIVE_NOT_RECRUITING). Lorecivivint query total 12; tanezumab+OA query total 24. NCT03491188 (RESTORE) and NCT01558727 (SPACE) did not resolve on filter.ids and were not cited as NCT IDs.
  • Official web: NICE NG226 recommendations and AAOS knee OA CPG portal opened 2026-08-31.
  • What was deepened (by page):
  • overview — lifetime risks, paracetamol WMD, Moseley/Kirkley/MeTeOR/Kise arthroscopy block, Frydendal THR increment, FORWARD/lorecivivint P3, explicit controversies table.
  • epidemiology — GBD 2010/2017/2021 not averaged; Losina 13.83% lifetime knee; Dillon NHANES 37.4%/12.1%; Murphy 25.3% lifetime hip; Wallace 2.1-fold residual increase; BMI/occupation/ACL magnitudes.
  • diagnosis — ACR 1986/1991 operating characteristics; EULAR 2010 six-feature ladder to 99%; Guermazi 89% MRI+ in KL 0; Hannan discordance; Dell'Isola six phenotypes as hypotheses.
  • cartilage/bone/synovium — ADAMTS5 knockout and IGD-resistant aggrecan knock-in; Sellam/Kapoor/Sanchez-Lopez synovitis loop; subchondral initiation vs progression still unresolved.
  • pain — Felson BML 77.5% vs 30%; Finan high-pain/low-KL sensitization; Arendt-Nielsen TS/DNIC; Previtali/Zolio instrument dependence.
  • genetics — arcOGEN → Zengini → Tachmazidou → Boer 100 variants → Hatzikotoulas 962 associations/700 genes; SMO OR 2.8; BMI causal, T2D/triglycerides not.
  • knee — five arthroscopy RCTs one direction; TOPKAT OKS 1.04 (−0.42 to 2.50); Skou TKR 15.8; dissatisfaction 10% not 20%.
  • hip — cam OR 2.11, dysplasia OR 2.19; Kjeldsen PRT vs NEMEX null; Frydendal 11.4-point THR increment; hip precautions RR 1.8 (0.6–5.2).
  • exercise/weight — Uthman sequential analysis by 2002; Lawford 8.70/100 pain with MID-relative uncertainty; Holden IPD baseline-pain moderator; Yan aerobic SUCRA; Weng exercise ≈ NSAID; Christensen 5% threshold; GLA:D n=9,825.
  • pharmacology — da Costa 2017/2021 dose MCID probabilities; Machado/Leopoldino paracetamol; SPACE 3.4 vs 3.3; GAIT vs Wandel; CNT and Bally naproxen discrepancy.
  • injections — Jüni low-quality short-term steroids; McAlindon repetition harm; Rutjes large-trial HA −0.11 vs overall −0.37; Pereira −0.08; RESTORE vs Tang; Gazendam hip saline.
  • surgery — Skou/Frydendal increments; TOPKAT/Wilson revision trade-off; Gademan 10-year revision; precautions; two different decision aids.
  • DMOADs — FORWARD 0.05 mm; lorecivivint OA-11/OA-10 misses; tanezumab analgesia vs 38.3–71.5/1000 PY joint-safety; MIV-711 structure without pain; doxycycline JSN without pain; methotrexate null.
  • guidelines — class/level table across ACR/AF, OARSI (1A–5), NICE offer/consider/do-not, AAOS, EULAR 2023, ESCEO algorithm.
  • biomarkers — BIPEDS qualification gap; WORMS; Guermazi MRI background rate; FNIH mediation not a treatment rule.
  • trials — landmark NCT table replacing over-reliance on a random recruiting first page.
  • patient experience — GLA:D implementation, work/hand/stiffness themes, two decision-aid RCTs that do not agree.
  • red flags — 0.08% septic arthritis after IA GC; 6% RPOA after hip CSI (heterogeneous); NGF RPOA rates; CNT/Bally NSAID MI.
  • Literature layer: appended deepening bibliography block; added lifetime-risk, NSAID-vascular, landmark-trial and conflict rows to STATISTICS.md; added EULAR 2023 update, Fernandes 2013 supersession and ESCEO 2019 profile to REGISTRY.md; sharpened OQ-6, OQ-8 and added OQ-23 (methotrexate inflammatory phenotype closed-negative).
  • Deliberately left alone: evidence-map tables (verified records; numbers were added in new synthesis sections rather than by deleting maps); patient-voice ethics files; landmark notes/ (already cover GBD, ACR/AF, IDEA, triamcinolone, RESTORE, TKR); no new paper-note files (none of the added trials independently outranked those six for a seventh note). Patient-experience page remains the shortest (~126 lines) because further qualitative padding would not add decision-relevant quantities.
  • Could not verify: NCT03491188 and NCT01558727 did not return on filter.ids this session; they are not cited as NCT IDs. Cui 2020 global knee-prevalence meta-analysis, Loeser 2012 Nature Reviews aging review, and Gala 2016 dysplasia review did not hit on title search and were not cited. Sharma 2021 NEJM knee seminar abstract was empty and was not used.
  • Ambiguous decisions (no one to ask): kept evidence maps rather than deleting “scope-only” rows, because the instruction was EXTEND not delete verified content; used NICE “offer/consider/do not offer” language rather than inventing GRADE classes NICE does not publish; treated lorecivivint OA-10/OA-11 as failed primaries with a KL 2 post-hoc, not as a new enrichment standard.
  • Next: different-engine claim-level re-audit of newly added PMIDs and numbers; do not promote off this session.

2026-08-31 — Independent audit of Grok deepening pass

  • Authorship and independence: Grok authored the 2026-08-31 deepening additions; Codex performed this separate claim-level audit as the independent auditor.
  • Scope: audited the new synthesis, quantitative tables, landmark-trial material and absence statements across all 18 wiki pages that Grok returned to status: draft. Previously curated evidence-map content was not re-synthesized, but every identifier now present on the draft pages was included in the resolution and structural checks.
  • Live PubMed verification: re-fetched all 527 unique PMIDs present across the 18 pages through PubMed E-utilities; all resolved. The explicitly appended deepening bibliography block contained 55 unique PMIDs and all 55 resolved to the claimed papers. Author, year and journal labels were checked against the live records, with ePub/print year offsets tolerated.
  • Live ClinicalTrials.gov verification: individually re-fetched all 31 unique NCT registrations now cited across the pages through the v2 API; all resolved. Displayed title, status, phase and intervention fields matched. The recruiting OA count remained 598; the active-status OA count refreshed from 1,252 to 1,243, and the tanezumab-plus-OA query refreshed from 24 to 22; the lorecivivint query remained 12.
  • Claim and number repairs: corrected the 2024 exercise review's pain evidence from moderate- to low-certainty; identified GLA:D as an uncontrolled registry before presenting its before-after changes; removed an unsupported diagnostic-criteria prescribing rule; removed an unverified current ADAMTS5-approval statement and an unsupported anticytokine comparison; removed the unsupported assertion that the Skou trial remained the only published randomized comparison; and narrowed molecular-endotype and semaglutide-MRI absence language to what the cited reports actually establish.
  • Padding removed or replaced: replaced the citation-only Dutch revision sentence with its 1.6%–13% stratified 10-year range; removed the uncontextualized zero-replacement sprifermin-arm observation; replaced a vague septic-arthritis citation with its 0.18% absolute risk and matched −0.07% risk difference; removed an unsupported claim about generic instruments missing hand-OA burdens; and replaced editorial “marketing”/“folklore” wording with source-bounded comparisons.
  • Guideline repair: re-opened NICE NG226 and the official AAOS knee nonarthroplasty third-edition guideline. Replaced the AAOS column's vague scope placeholders with the document's actual Strong, Moderate or Limited recommendation directions, including the AAOS positions that conflict with ACR/AF or NICE.
  • Absence searches repeated: ran live PubMed searches for approved disease-modifying OA treatment, ADAMTS5 inhibitor trials, sprifermin symptom benefit, imaging/biochemical surrogate validation, molecular-endotype treatment selection, semaglutide MRI mediation and arthroplasty-versus-nonsurgical randomized evidence. Broad current absences were either narrowed to the retrieved report's boundary or removed; no unresolved absence claim remains from the deepening pass.
  • Reference and structure repair: added three page-reference entries omitted by the deepening pass, consolidated one duplicated bibliography PMID record, and updated every page's audit-disposition identifier totals. Final checks found every body PMID represented in that page's References, no unused page reference, every wiki PMID present in the master bibliography, no duplicate bibliography PMID, all required sections present and all relative wiki links resolving.
  • Disposition: all new claims and identifiers resolved after repair. All 18 pages were promoted from status: draft to status: curated. No [unverified], Scope-only record, citation mismatch or substantive audit issue remains.