What was searched: Live PubMed queries on: global burden (GLOBOCAN 2022), MASLD/NAFLD and alcohol etiologic shift, non-cirrhotic NAFLD HCC risk, surveillance RCT + meta-analyses + harms modeling, BCLC 2022, AASLD 2023 and EASL 2025 guidelines, Milan criteria and Metroticket 2.0, SHARP, REFLECT, RESORCE, IMbrave150, HIMALAYA, CheckMate 9DW, exome/mutational landscape (Schulze 2015), TERT in NAFLD-HCC. All PMIDs verified via search + metadata retrieval this session.
Check next time: (1) Full-text primary of CheckMate 9DW global results (only the Japanese-subgroup paper with global interim HR was captured — locate/verify the global primary publication); (2) HIMALAYA long-term (4–5 yr) OS update; (3) adjuvant IO trials (IMbrave050 status after later analyses); (4) GALAD/biomarker validation studies for the surveillance page; (5) TACE+IO combination trials (EMERALD-1, LEAP-012) for locoregional page.
What changed: Completed all 17 canonical wiki pages (3,142 lines; all draft pending separate-session audit). Added 16 new pages and extended the seeded overview. Built literature/BIBLIOGRAPHY.md (105 unique papers with cited-by mappings), five landmark notes (Milan, SHARP, exome landscape, IMbrave150, BCLC 2022), guidelines/REGISTRY.md, statistics/STATISTICS.md (90 table rows), and the four-file patient-voice/ layer. Replaced OPEN-QUESTIONS.md with 26 stable tiered questions and a 12-row “Dots not yet connected” table. Rebuilt INDEX.md with counts, reading paths, live page statuses, and literature-layer links. Updated CONDITIONS-ROADMAP.md from seeded to built.
Live literature searches: PubMed E-utilities searches and record retrieval covered global/etiologic epidemiology; HBV vaccination and antivirals; post-HCV-cure risk; MASLD/non-cirrhotic HCC; alcohol burden; disparities; cirrhosis risk models and surveillance thresholds; ultrasound/AFP harms and utilization; abbreviated MRI and GALAD; LI-RADS CT/MRI/CEUS, biopsy seeding, and response imaging; BCLC/alternative staging and ALBI; exome, molecular/immune classes, CTNNB1 immune exclusion, single-cell atlases, and prospective genotyping; resection, portal hypertension, ablation, and adjuvant therapy; Milan/UCSF/Metroticket/downstaging/RETREAT; TACE/TARE/dosimetry/radiation and EMERALD-1/LEAP-012; systemic phase 3 and second-line trials; AASLD/EASL/APASL/JSH/KLCA-NCC/Chinese guidance; qualitative HCC experience, caregivers, preferences, transplant survivorship, PROs, and palliative care.
ClinicalTrials.gov: Queried API v2 live for HCC active studies and landmark trials. Thirty-eight unique NCT identifiers were written; all 38 returned HTTP 200 from their individual API records in final validation. The page records the 2026-08-28 registry-status snapshot.
Patient organizations: Direct retrieval succeeded for Liver Cancer UK, Blue Faery, and ILCA. American Liver Foundation returned HTTP 403, American Cancer Society HTTP 406, and Macmillan failed TLS verification with the session method; these are documented as access-limited and were not used for substantive claims.
Validation: All 17 pages contain required frontmatter, TL;DR, open questions, related pages, and references; every page is 153–231 lines. No broken local Markdown links or unsourced claim markers were found. All 105 unique wiki PMIDs are represented in the bibliography. A final live ESummary request resolved all 106 unique PMIDs present anywhere in the condition tree (the additional PMID is a superseded guideline in the registry). All 38 NCT records resolved.
Could not verify / follow up: (1) A PubMed-indexed global primary CheckMate 9DW paper was not recovered; only the Japanese subgroup paper reporting the global interim HR was used (PMID 42338692). (2) An exact current ESMO HCC guideline record was not recovered and was omitted. (3) The primary official 2022 Chinese guideline was represented through an indexed review (PMID 37124695), flagged in the registry. (4) Organization sites blocked above should be checked with a browser-capable audit. (5) This was the build session, not the independent citation-to-claim audit; pages intentionally remain draft and the roadmap is built, not audited.
What changed: Extended every one of the 17 canonical wiki pages in place without removing the prior verified content. Added quantitative evidence ledgers, effect sizes and confidence intervals, mechanism detail, trial-design boundaries, guideline divergences, and explicit two-sided controversies. Wiki depth increased from 3,142 to 3,964 lines; every page now carries at least 25 unique live-verified PubMed citations, except the trials page which carries 19 PubMed papers plus 38 live registry records. All pages remain draft because this was a build/deepening pass rather than the separately required audit.
PubMed work: Ran 136 page-specific PubMed E-utilities searches (eight per page) covering randomized trials, meta-analyses, mechanisms, guidelines, adverse effects, disparities, patient priorities, and unresolved questions. Retrieved full metadata/abstracts for 884 unique records and selected 215 new papers supported by the retrieved records. The wiki now cites 320 unique PMIDs, all represented once in literature/BIBLIOGRAPHY.md with page mappings.
ClinicalTrials.gov work: Re-fetched all 38 NCT identifiers individually from API v2 on 2026-08-30. Updated the status note and corrected the live snapshot: HIMALAYA (NCT03298451) and EMERALD-1 (NCT03778957) were active, not recruiting; CheckMate 9DW (NCT04039607), IMbrave050 (NCT04102098), and LEAP-012 (NCT04246177) were completed. Added planned enrollment/status context while explicitly separating registry completion from evidentiary positivity.
Literature layer: Expanded the bibliography from 105 to 321 unique papers: 320 cited by wiki pages plus the superseded 2018 EASL guideline retained in the registry. Strengthened statistics/STATISTICS.md from 90 to 129 table rows with etiologic attributable fractions, surveillance intervals, diagnostic performance, phase-3 biomarkers, downstaging, treatment, transplant rejection, disparities, and patient-preference estimates. Updated guidelines/REGISTRY.md with APASL systemic therapy 2024, the primary indexed China 2024 guideline, Korean modality components, OPTN/LI-RADS allocation context, and China/APASL supersession/overlay chains.
Research agenda: Sharpened OQ-2/3/4/6/7/11/14/15/17 into more explicit prospective or randomized designs, added quantified safety and biomarker thresholds, and expanded “Dots not yet connected” from 12 to 15 junctions.
Final validation: A fresh PubMed ESummary pass resolved all 321 PMIDs present anywhere in the condition tree, including the registry-only superseded EASL 2018 guideline; individual ClinicalTrials.gov API v2 requests resolved all 38 NCT identifiers. Static checks found 17/17 pages at 189–296 lines with draft status, current curation dates, bold TL;DR paragraphs, open questions, related-page links, and reference sections; every inline wiki PMID appears in its page reference list, all condition-tree PMIDs appear exactly once in the master bibliography, and no broken local Markdown links or unsourced claim markers remain.
Follow-up: A separate-session claim-to-citation audit is still required before any page can become curated. High-priority audit targets are numerical claims transcribed from abstracts, ePub/print year differences, current CheckMate 9DW global-primary publication availability, mature IMbrave050/EMERALD-1/LEAP-012 OS updates, and the continuing absence of an exact current PubMed-indexed ESMO HCC guideline recovered by these searches.
2026-08-30 — Independent citation and status audit¶
Scope and counts checked: Audited all 17 canonical wiki pages and all 12 literature artifacts. Re-fetched every identifier currently present in the condition tree in one final live pass: 329/329 unique PMIDs resolved through PubMed E-utilities and 39/39 unique NCT records resolved through ClinicalTrials.gov API v2. Checked 786 inline wiki PMID occurrences, 439 wiki reference entries, 328 one-line bibliography records, 122 statistics rows, all page structures/local links, and every recorded trial status. All 17 pages remained within the 150–400-line target and each carried at least 25 live-verified PMID/NCT records in its evidence set.
Substantive stale-evidence errors found and fixed (5): (1) replaced reliance on the CheckMate 9DW Japanese subgroup with the now-indexed global phase 3 report (Yau 2025, PMID 40349714), including median OS, non-proportional early/late hazards, and treatment-attributed deaths; (2) replaced the outdated “IMbrave050 durability unresolved” wording with the 2026 update showing nonsustained RFS benefit and an unsupported adjuvant benefit–risk profile (Yopp 2026, PMID 41580093); (3) recovered and added the exact current ESMO 2025 HCC guideline (Vogel 2025, PMID 39986353) and supersession chain; (4) updated the post-immunotherapy sequencing gap to incorporate new observational LEVIATHAN and reconstructed-IPD evidence while retaining the dated absence of randomized sequence trials (PMID 41321927; PMID 42128139); and (5) replaced the static abbreviated-MRI mortality-trial absence with the positively stated, dated fact that the 4,700-person PREMIUM mortality RCT is recruiting (NCT05486572; PMID 41624484).
Citation/reference errors found and fixed: Corrected three wrong journal assignments (PMID 35945907, PMID 40058162, PMID 36216350), the PubMed author form for PMID 27511114, the claim-to-source grouping for HCC's approximately 90% share of primary liver cancers, and the under-supported quoted UCSF size criteria by adding the directly matching live record (PMID 12200775). Normalized nine additional PubMed journal qualifiers in page references, repaired the bibliography cited-by mapping for PMID 38126757, removed an unused bibliography-only SBRT commentary, and revalidated all author/year/journal combinations with zero remaining metadata mismatches.
Absence searches rerun: Targeted live searches covered direct first-line IO-combination comparisons, randomized post-IO sequencing, non-cirrhotic MASLD surveillance mortality, abbreviated-MRI mortality trials, transplant bridging randomization, CheckMate 9DW, IMbrave050, and current ESMO guidance. The first-line head-to-head, randomized post-IO, and non-cirrhotic MASLD mortality gaps remain genuine as of 2026-08-30 and are now written as dated positive evidence-gap statements rather than undated absences.
Patient-voice/web audit: Re-fetched every recorded public organization URL. Liver Cancer UK, Blue Faery, ILCA, and Macmillan returned HTTP 200; Macmillan was moved from access-limited to verified. American Liver Foundation (HTTP 403) and American Cancer Society (HTTP 406) remain access-limited and are not used for substantive claims.
What remains flagged: No unresolved verification marker, citation mismatch, broken local link, or substantive page issue remains. Open research gaps remain explicitly labeled: PREMIUM has no outcome data and is estimated to complete in 2031; no direct randomized ranking of current first-line IO combinations or prospective randomized post-IO sequence was found; non-cirrhotic MASLD surveillance lacks prospective mortality validation; EMERALD-1/LEAP-012 OS maturity and future guideline incorporation require continued watch. The two access-limited organization sites above remain retrieval limitations only.
Status outcome: Promoted all 17 wiki pages from draft to curated, marked every literature-layer artifact audited in the index, and advanced CONDITIONS-ROADMAP.md from built to audited.
2026-10-07 — Dot-row grounding pass (grounder: Claude, Opus 5.5)¶
Scope: The 15 uncited rows (D1–D15) of the "Dots not yet connected" table in OPEN-QUESTIONS.md. No other file was touched; OQ-1 to OQ-26 prose is unchanged. All 15 rows carried no identifier, so every row was in scope.
Outcome: Started with 15 rows. Rewrote 6 (D1, D5, D6, D9, D10, D15), so that each Dot states a finding with its figure and PMID and each junction says in prose what has not been done. Deleted 9 (D2, D3, D4, D7, D8, D11, D12, D13, D14). Surviving IDs were not renumbered, and a line under the table records the gap.
Verification: Every PMID now in the table, and every PMID named below, was fetched live this session from PubMed E-utilities (ESummary + EFetch abstract) and read against the number attributed to it. Each was checked for retraction both by PubMed publication type (Retracted Publication/Expression of Concern) and by the Crossref updated-by field. No record carries a retraction or expression of concern. Crossref lists errata (not retractions) for LEAP-012 (PMID 39798578; erratum doi:10.1016/s0140-6736(25)00205-3, still cited in D6) and XXL (PMID 32615109; erratum doi:10.1016/s1470-2045(20)30387-9, cited only below). All 5 NCT IDs in the table were fetched live from ClinicalTrials.gov API v2.
Why each row was deleted:
D2 (post-HCV-cure dynamic fibrosis × surveillance harm/cost → individual stopping rule with competing mortality): junction already published. A microsimulation found biannual surveillance after HCV cure cost-effective (<$150,000/QALY) until age 70 in cirrhosis and until 60 in advanced fibrosis (Mueller 2022, PMID 35157959). Joint modelling of FIB-4/APRI/LSM dynamics with competing death in 3,067 post-SVR cirrhotic patients (228 HCC) found that current values carry risk information but slopes do not, and that NIT dynamics cannot identify patients safe to exclude from surveillance (Nahon 2025, PMID 40020929). A VA cohort proposed an individual rule (post-SVR LSM <5 kPa without diabetes gives <0.4%/year) (John 2024, PMID 38061410), against a cost-effectiveness threshold of 0.7 per 100 PY (Chhatwal 2024, PMID 37302445). What remains, prospective or randomized validation of discontinuation, restates OQ-5 rather than joining two literatures. Found by PubMed S3.
D14 (dynamic stiffness after etiologic cure × time-varying competing mortality → entry/interval/stopping rule): duplicate of D2. It rests on the same evidence (OQ-5/OQ-25 and the same candidates), and Nahon 2025 (PMID 40020929) is precisely the stiffness-dynamics-with-competing-death analysis it asked for.
D3 (Wnt/CTNNB1 immune exclusion × multiple first-line IO combinations → randomized regimen-specific interaction): junction published in exploratory randomized form. CheckMate 459 biomarker analyses found sorafenib OS better with Wnt/β-catenin-mutated tumors, so nivolumab beat sorafenib only in Wnt-unaltered tumors (OS HR 0.66, 95% CI 0.47–0.91) (Neely 2026, PMID 42762856). In IMbrave150, a high mutated-β-catenin gene-signature score was associated with no significant OS/PFS gain for atezolizumab-bevacizumab over sorafenib (Lehrich 2024, PMID 39583094). Both analyses are post hoc, and neither compares IO regimens with each other. The row's premise, that only mouse and retrospective data exist (Ruiz de Galarreta 2019, PMID 31186238; Harding 2019, PMID 30373752, which reported DCR 0% versus 53% in 31 ICI-treated patients), is outdated. Found by PubMed S4.
D4 (single-cell immune classes × routine blood/tissue biomarkers → minimal assay preserving spatial predictive information): the junction restates OQ-20, and Dot B was unsupported. OQ-20 already asks whether spatial immune architecture can be reduced to a deployable assay. The candidate Dot B records (HES V2.0/GALAD, PMIDs: 38899967, 41043723) are surveillance-detection panels with no treatment-predictive role, so they cannot be the "routine biomarker" side of a treatment-prediction junction.
D7 (downstaging response as biology × continuous recurrence models → allocation model combining response trajectory with AFP/morphology): junction already published. In 422 listed patients from 6 European centres, an AFP slope >15 ng/mL/month and mRECIST progression after locoregional therapy were independent predictors of recurrence and death, inside and outside Milan (Lai 2013, PMID 23873764). The TRAIN score (time on list, radiological response, AFP change, inflammation) combines these into one pre-transplant allocation tool and increased potentially transplantable patients by 8.9% and 24.6% over Milan in its training and validation sets (Lai 2016, PMID 27429025). Found by PubMed S9b/S9c; the broader query S9a missed both, which is why the more specific queries were run.
D8 (checkpoint immune memory × transplant graft tolerance → immune/washout predictor of rejection): both sides were the same claim. The condition's only evidence on either side is pre-transplant ICI rejection data: 26.4% rejection, washout aHR 0.92 per week (Rezaee-Zavareh 2025, PMID 38996924); 20.2% rejection, <30-day washout OR 21.3 (Moeckli 2025, PMID 40042053). The wiki holds no separate graft-tolerance literature; the tolerance sentence in red-flags-and-safety-concerns.md is uncited. The junction restates OQ-14.
D11 (MASLD-specific immune surveillance × etiology-specific systemic outcomes → randomized treatment-by-etiology interaction): Dot A and Dot B were one literature, and the junction is published at aggregate level. Pfister 2021 (PMID 33762733) is itself the etiology-specific outcome analysis (a 3-trial meta-analysis of >1,600 patients). An umbrella review of 15 meta-analyses reports OS HR 0.64 (0.55–0.74) in HBV versus 0.91 (0.75–1.11) in nonviral HCC for PD-(L)1 + antiangiogenic combinations, with a treatment-by-etiology interaction P = 0.02 for OS (Petrelli 2026, PMID 41794163). An individual-patient-data network meta-analysis with etiology subgroups is registered (PROSPERO CRD420251126975; protocol PMID 41988192). Found by PubMed S12.
D12 (portal-hypertension measurement × modern minimally invasive resection → physiologic threshold for decompensation): junction already present in the condition's own cited evidence. Aliseda 2024 (PMID 38126757; 9 matched or prospective studies, 1,124 patients) stratifies by both axes. Indirect signs of portal hypertension did not raise complications, post-hepatectomy liver failure or decompensation; HVPG ≥10 mmHg did. Portal hypertension carried no significant excess risk in minimally invasive resection, whereas open resection did. The row's Powers column (OQ-12, a transplant question) was also misassigned.
D13 (phase-3 blood-panel validation × real-world closed-loop recall failures → end-to-end trial): Dot B is unsupported by the condition's evidence. The surveillance page's recall table and the red-flags "closed-loop tracking" sentence cite nothing, and no recall-failure rate is cited anywhere in the condition. The nearest cited figure is a different failure, post-diagnosis treatment delay (14.3% delayed >3 months; HR 1.15 for death) (Wagle 2023, PMID 35933076). Swapping that in would have manufactured a new row, so I deleted the row instead. What remains overlaps D9 and OQ-4.
Searches behind each absence claim (all run 2026-10-07; PubMed counts are the full esearch count):
S1 (D1), PubMed: (ultrasound[tiab]) AND (visualization[tiab] OR visualisation[tiab] OR "visualization score"[tiab]) AND (hepatocellular[tiab] OR HCC[tiab]) AND (noncirrhotic[tiab] OR non-cirrhotic[tiab] OR "without cirrhosis"[tiab]) → 3 records (PMIDs: 38076360, 36253490, 31173745); none measures visualization in non-cirrhotic MASLD.
S2 (D1), PubMed: (ultrasound[tiab] OR "abbreviated MRI"[tiab]) AND (hepatocellular[tiab]) AND surveillance[tiab] AND (MASLD[tiab] OR NAFLD[tiab] OR "fatty liver"[tiab]) AND (noncirrhotic[tiab] OR non-cirrhotic[tiab] OR "without cirrhosis"[tiab] OR "liver stiffness"[tiab]) → 21 records. Abstracts read for PMIDs 40337094, 41868260 and 42700797: a retrospective stage-at-diagnosis study, a stiffness-based risk algorithm with no modality comparison, and a review. None measures ultrasound adequacy in a risk-selected non-cirrhotic stratum; the rest are reviews or out of scope by title.
S3 (D2/D14), PubMed: (hepatitis C[tiab] OR HCV[tiab]) AND (SVR[tiab] OR "sustained virologic response"[tiab] OR "sustained virological response"[tiab] OR cure[tiab] OR eradication[tiab]) AND hepatocellular[tiab] AND surveillance[tiab] AND (discontinu*[tiab] OR stopping[tiab] OR "stop surveillance"[tiab] OR cessation[tiab]) → 10 records; returned the papers that closed D2.
S4 (D3), PubMed: (CTNNB1[tiab] OR "beta-catenin"[tiab] OR "β-catenin"[tiab] OR WNT[tiab]) AND (hepatocellular[tiab]) AND (atezolizumab[tiab] OR durvalumab[tiab] OR tremelimumab[tiab] OR nivolumab[tiab] OR ipilimumab[tiab] OR camrelizumab[tiab] OR pembrolizumab[tiab]) AND (randomized[tiab] OR randomised[tiab] OR "phase 3"[tiab] OR "phase III"[tiab] OR IMbrave150[tiab] OR HIMALAYA[tiab] OR CheckMate[tiab]) → 7 records; returned the papers that closed D3.
S5 (D5), PubMed: (ctDNA[tiab] OR "circulating tumor DNA"[tiab] OR "circulating tumour DNA"[tiab] OR "minimal residual disease"[tiab] OR "molecular residual disease"[tiab]) AND (hepatocellular[tiab]) AND (adjuvant[tiab]) AND (randomized[tiab] OR randomised[tiab] OR "guided"[tiab] OR IMbrave050[tiab]) → 3 records, all reviews (PMIDs: 42692153, 42040612, 41899615).
S6 (D5), ClinicalTrials.gov API v2: query.cond=hepatocellular carcinoma, query.term=(ctDNA OR circulating tumor DNA OR minimal residual disease OR molecular residual disease) AND adjuvant → 12 studies; none randomizes by MRD status (see the D5 row).
S7 (D6), PubMed: (radioembolization[tiab] OR radioembolisation[tiab] OR TARE[tiab] OR SIRT[tiab] OR "yttrium-90"[tiab] OR Y90[tiab]) AND hepatocellular[tiab] AND (dosimetry[tiab] OR "absorbed dose"[tiab] OR "tumor dose"[tiab]) AND (atezolizumab[tiab] OR durvalumab[tiab] OR immunotherapy[tiab] OR "systemic therapy"[tiab] OR lenvatinib[tiab]) AND (intermediate[tiab] OR "BCLC B"[tiab] OR "BCLC-B"[tiab] OR randomized[tiab] OR randomised[tiab]) → 2 records, both reviews.
S8 (D6), ClinicalTrials.gov API v2: query.cond=hepatocellular carcinoma, query.intr=radioembolization OR yttrium-90 OR SIRT OR TARE, AREA[DesignAllocation]RANDOMIZED → 36 studies. The full records of NCT04777851 and NCT05537402 were fetched (termination reasons: slow enrolment, and accrual below 31% of target); neither record contains "dosim".
S9a (D7), PubMed: (mRECIST[tiab] OR "radiological response"[tiab] OR "radiologic response"[tiab] OR "tumor response"[tiab]) AND ("alpha-fetoprotein"[tiab] OR AFP[tiab]) AND ("liver transplantation"[tiab]) AND hepatocellular[tiab] AND (score[tiab] OR model[tiab]) AND ("waiting list"[tiab] OR "waitlist"[tiab] OR "wait-list"[tiab] OR downstaging[tiab] OR "bridging"[tiab]) → 4 records. S9b: "TRAIN score"[tiab] OR ("Time-Radiological-response-Alpha-fetoprotein-INflammation"[tiab]) → 1 (PMID 27429025). S9c: ("alpha-fetoprotein slope"[tiab] OR "AFP slope"[tiab] OR "delta AFP"[tiab] OR "AFP response"[tiab]) AND (mRECIST[tiab]) AND ("liver transplantation"[tiab]) → 1 (PMID 23873764).
S10 (D9), PubMed: (hepatocellular[tiab] OR HCC[tiab]) AND surveillance[tiab] AND (outreach[tiab] OR navigation[tiab] OR "patient navigator"[tiab]) AND (randomized[tiab] OR randomised[tiab] OR trial[tiab]) → 12 records. Abstracts read for the RCTs (PMIDs: 30070379, 34902568, 38099859) and a cost-effectiveness model (PMID 40040860, which modelled, not measured, a 48.4% rise in early-stage detection). Every randomized trial's primary outcome was surveillance uptake. VIGILANT (PMID 40588390) and a VA cluster trial (PMID 40783525) are protocols.
S11 (D10), PubMed: (hepatocellular[tiab] OR "liver cancer"[tiab]) AND (caregiver*[tiab] OR "family carer"[tiab] OR carer*[tiab]) AND (intervention[tiab] OR randomized[tiab] OR randomised[tiab] OR trial[tiab]) AND ("emergency department"[tiab] OR "emergency"[tiab] OR readmission*[tiab] OR hospitali*[tiab] OR "acute care"[tiab]) → 0 records.
S12 (D11), PubMed: (hepatocellular[tiab]) AND (immunotherapy[tiab] OR "immune checkpoint"[tiab] OR atezolizumab[tiab] OR durvalumab[tiab] OR nivolumab[tiab]) AND (etiology[tiab] OR aetiology[tiab] OR "non-viral"[tiab] OR nonviral[tiab] OR NASH[tiab] OR MASLD[tiab] OR MASH[tiab] OR NAFLD[tiab]) AND ("individual patient data"[tiab] OR "individual participant data"[tiab] OR interaction[tiab]) AND (randomized[tiab] OR randomised[tiab] OR "meta-analysis"[tiab]) → 3 records; returned the papers that closed D11.
S13 (D15), PubMed: (hepatocellular[tiab]) AND ("preference-weighted"[tiab] OR "patient-centered endpoint"[tiab] OR "patient-centred endpoint"[tiab] OR "desirability of outcome ranking"[tiab] OR "benefit-risk"[tiab] OR "multi-criteria decision"[tiab] OR "time without symptoms"[tiab] OR "Q-TWiST"[tiab] OR "quality-adjusted time"[tiab]) AND (trial[tiab] OR randomized[tiab] OR randomised[tiab] OR endpoint*[tiab]) → 25 records. The two integrative rankings were read in full abstract (PMIDs: 41634252, 42250662); neither uses patient-elicited weights. The remaining records are trial PRO reports, pharmacology or out of scope by title.
Source boundary: these absences are PubMed and ClinicalTrials.gov absences only. WHO ICTRP, Europe PMC and OpenAlex were not searched in this pass, and Cochrane CENTRAL and SciELO are unreachable (SOURCES.md), so no row claims an absence "anywhere".
Could not resolve / follow-ups for the next sweep:
OQ text is now stale in three places, which I did not edit because they are outside this pass. OQ-5 should absorb Mueller 2022 (PMID 35157959) and Nahon 2025 (PMID 40020929). OQ-10 should absorb the umbrella-review interaction (PMID 41794163) and the registered IPD network meta-analysis (PMID 41988192). OQ-19 should absorb the CheckMate 459 and IMbrave150 Wnt analyses (PMIDs: 42762856, 39583094), and molecular-landscape.md and biomarkers.md describe Wnt evidence as only preclinical or retrospective.
Registry/OQ mismatch: OQ-21 and clinical-trials-landscape.md present NCT07350824 as the vehicle for "randomized ctDNA-guided adjuvant intervention". Its live registry record says allocation is NON_RANDOMIZED, with two cohorts (active monitoring vs adjuvant PD-1), not yet recruiting, last updated 2026-01-20.
New PMIDs not yet in the bibliography: the table now cites PMIDs 30070379, 34902568, 41634252, 42250662 and 40337094, which are not in literature/BIBLIOGRAPHY.md. This pass was restricted to OPEN-QUESTIONS.md and LOG.md, so the next sweep should add them. PMIDs named only in this entry, as reasons for deletion, are listed here for the same purpose: 35157959, 40020929, 42762856, 39583094, 23873764, 27429025, 41794163 and 41988192.
D13 could be rebuilt honestly if a future pass adds a cited, quantified recall-after-abnormal-result failure rate to the surveillance page; the condition currently has none.
Weakest surviving row is D10. Its junction is real (S11 returned 0 records), but OQ-17 already lists acute-care use among a palliative trial's outcomes, so an auditor may judge it close to a restatement. I kept it because it joins a specific qualitative finding (caregivers cannot attribute symptoms) to a specific quantified hazard (7.0% variceal bleeding on atezolizumab-bevacizumab). That is a narrower and more testable claim than OQ-17.
Wording I chose not to rely on: red-flags-and-safety-concerns.md reports Lee 2024 (PMID 39199649) GI bleeding as 5%, but the abstract gives "5% (n = 17)", which is internally inconsistent with its 15% (n = 17) overall bleeding. I did not use it. The wiki's statement that DOSISPHERE-01 improved survival is not visible in that trial's abstract, which reports ORR only, so the D6 row cites ORR alone.
Audit note: I wrote this pass, so I may not audit it. Per protocol, the six rewritten rows and the nine deletion rationales need checking by a different engine, which should re-run S1, S2, S5, S6, S10, S11 and S13 in particular.
Scope and result: I audited all 15 rows in Claude's pass: the 6 surviving rewrites (D1, D5, D6, D9, D10, D15) and all 9 deletion rationales. The table now has 9 rows. I repaired all 6 surviving rows, restored D2, D3 and D13, and left D4, D7, D8, D11, D12 and D14 deleted. No other condition content was edited.
Live identifier and integrity check: All 38 distinct PMIDs now in the table resolved through a fresh PubMed E-utilities EFetch, and their abstracts, authors, years, populations and quoted estimates were read. All 6 NCT IDs now in the table were retrieved from their individual ClinicalTrials.gov API v2 records. Each final-table paper's PubMed publication types and Crossref DOI updated-by field were checked; no retraction or expression of concern appeared. Crossref lists an erratum, not a retraction, for LEAP-012 (PMID 39798578). The 19 PMID records named in Claude's deletion discussion were also re-fetched and checked in both systems; the XXL paper (PMID 32615109) likewise has an erratum, not a retraction.
Citation failures and motivated anchoring:0 invalid or retracted citations; 2 claim-to-source overextensions in surviving rows. D1 treated the non-cirrhotic stiffness/diabetes subgroup as shown to be obese and treated cirrhosis-only imaging estimates as proof that no non-cirrhotic adequacy study exists anywhere; neither follows from the cited papers. D10 treated a qualitative report of general symptom-attribution difficulty as evidence that caregivers failed to recognize bleeding or encephalopathy; the report does not test that. Both rows now state the supported findings and the proposed junction separately. Motivated-anchor count: 2 rows, both repaired. D15's “licensed” wording was also removed because the efficacy trials cited there do not establish regulatory status.
Numerical and population repairs: D6 now attributes the 71% versus 36% DOSISPHERE-01 response result to the 56 treated/evaluable patients (28 per arm), while separately reporting 60 randomized. D1 now includes the 0.46-per-100-person-years subgroup interval (0.29–0.78) and labels the cirrhosis population of the imaging studies. D5 and D6 now include the omitted HR intervals, and D5 identifies the updated OS estimate as immature. D9 now reports the multicentre trial's final early-stage result rather than leaving it “in follow-up.” D13 labels the 46.9% delay as a case-only estimate among patients eventually diagnosed with HCC, not as the failure rate for all abnormal screens. Author/year metadata for restored D13 were checked against PubMed.
Stale or unsupported absences: 2. D9's implication that multicentre early-detection results were still pending is stale: Singal's final randomized outreach report found no statistically significant early-stage increase among 125 detected HCCs (64.2% versus 79.3%; P=0.06; PMID 39288308). D13's deletion rationale said there was no quantified follow-up failure evidence; Choi's VA cohort found that 46.9% of 655 cirrhotic patients eventually diagnosed with HCC had a diagnostic delay exceeding 60 days after a guideline-defined red flag (PMID 32693047). These are distinct measures, so the restored row does not relabel delay as recall failure.
Deletion verdicts:D2 restored: Mueller's age/fibrosis cost model (PMID 35157959) and Nahon's time-updated non-invasive-test/competing-death analysis (PMID 40020929) do not provide a validated individualized surveillance stopping rule; the junction remains. D3 restored: Neely's CheckMate 459 analysis (PMID 42762856) and Lehrich's IMbrave150 analysis (PMID 39583094) are exploratory comparisons against sorafenib, not a prospective Wnt-guided choice between immune combinations. D13 restored: phase 3 HES V2.0 detection evidence (PMID 38899967) and quantified post-red-flag diagnostic delay (PMID 32693047) support distinct dots. D4 remains deleted: its proposed routine panels were detection rather than treatment-prediction assays and the question duplicates OQ-20. D7 remains deleted: Lai 2013 (PMID 23873764) and the TRAIN score (PMID 27429025) already join radiologic response and AFP trajectory for transplant selection. D8 remains deleted: the proposed sides both reduce to ICI washout/rejection evidence (PMIDs: 38996924, 40042053). D11 remains deleted: Pfister 2021 (PMID 33762733) itself joins NASH immune biology with etiology-stratified outcomes, and Petrelli 2026 (PMID 41794163) synthesizes the trial subgroup interaction. D12 remains deleted: Aliseda 2024 (PMID 38126757) already stratifies resection outcomes by portal-hypertension assessment and operative approach. D14 remains deleted as the same stopping-rule junction as D2; Nahon 2025 is pertinent but does not independently justify a duplicate row. D2, D3 and D13 were groundable, rather than impossible to source.
Absence-search replay and boundary: On 2026-10-07, exact PubMed replays returned S1=3, S2=21, S5=3, S7=2, S10=12, S11=0 and S13=25; the exact ClinicalTrials.gov queries returned S6=12 and S8=36. A broader PubMed outreach search recovered the final trial report that S10 missed because it used “screening” rather than “surveillance”; a broader caregiver-intervention search returned 15 titles and no HCC acute-care intervention trial. An additional PubMed search on surveillance follow-up/delay returned 125 records, including the quantified D13 source. Targeted Europe PMC TITLE_ABS searches returned 0 for the narrow D1, D5 and D10 keyword intersections and 2 for D15, but these do not establish database-wide absence. The table therefore names PubMed or ClinicalTrials.gov beside each negative search. Cochrane CENTRAL, SciELO and subscription databases were not searched; no claim of worldwide absence follows.
Novelty-gate judgement:Do not treat this table as a conclusive novelty gate. The repaired rows are source-grounded, distinguish their two dots, and are usable as scoped research leads. Their negative searches are query- and database-bounded; proving that no study has made a junction would require wider registry, conference and citation-chain review, including sources unavailable in this environment. The false D2/D3 closure decisions and stale D9/D13 absences show why a gate must re-run its own searches rather than inherit this table's absence statements.
2026-10-11 — Periodic literature sweep (author: Claude, Opus 5.5; window 2026-08-28 → 2026-10-11)¶
Why this sweep matters: LEAP-012 final OS was posted to ClinicalTrials.gov on 2026-09-03 and is null: median 40.4 vs 40.8 months, HR 0.98 (95% CI 0.76–1.26), p = 0.4442. PFS is unchanged at HR 0.66 (0.51–0.84) (NCT04246177). The registry also changed the trial's status from completed to terminated ("Business Reasons"). This is a registry posting, not a peer-reviewed report. All 25 PubMed records matching LEAP-012 (all dates) were checked by title, and the top 40 of 310 Europe PMC hits for "LEAP-012" AND "overall survival" were reviewed; neither turned up a journal report of the final analysis. It answers part of OQ-6: for one of the two phase-3 TACE-combination trials, a PFS gain did not become an OS gain. EMERALD-1 (NCT03778957) still has no posted OS.
Sources actually queried: PubMed E-utilities. The PubMed MCP tool was disconnected, so esearch/efetch/esummary were called directly. I ran 13 windowed topic queries using [EDAT] 2026/08/28–2026/10/11: RCT/phase 3 (23), meta-analysis/SR (38), Cochrane (2), guidelines/consensus (21), named trials (14), surveillance (87), systemic (72), locoregional (47), transplant (80), resection/adjuvant (45), epidemiology (145), patient experience/disparities (64), and post-IO randomized (0). The overall HCC count in the window was 1,363. Six further all-date queries covered LEAP-012, EMERALD-1, HIMALAYA updates, post-IO RCT[pt] 2025–26, the MSGH consensus and major-society 2026 guidance. Other sources: ClinicalTrials.gov API v2, with all 44 NCT IDs the condition cites re-fetched plus the 2 newly cited (NCT05064553 ALTUS; NCT05883644 SIERRA). Europe PMC: non-MEDLINE sources (39), preprints (81), conference abstracts (0), and full-text availability for 8 key records. Crossref updated-by was checked for 32 records, without mailto. ISRCTN API: "hepatocellular carcinoma", 98 records; the date of last update could not be reliably parsed, so no record was used. Patient-organisation sites were also checked. Not queried: WHO ICTRP (portal is JS-driven), Cochrane CENTRAL and cochranelibrary.com (HTTP 403), SciELO/LILACS (HTTP 403), and Unpaywall (skipped because LIFEBANK_CONTACT_EMAIL is unset). Every absence below is a PubMed (or named-source) absence only. The full query strings, translations, returned IDs and per-record retraction checks are in literature/sweeps/2026-10-11.json.
Cochrane:"Cochrane Database Syst Rev"[jour] crossed with HCC, liver cancer, cirrhosis and hepatitis B/C in the window returned 2 reviews. Neither is about HCC: one covers thymosin-α1 for chronic HBV, the other sickle-cell transfusion. No new or updated Cochrane review touches anything this condition asserts.
Retraction status: All 32 records added or newly cited were checked in PubMed publication types and comments/corrections and in Crossref updated-by. None is retracted or carries an expression of concern.
Full-text boundary (FAILURE-MODES rule 5): Europe PMC reports no open full text for ALTUS, the pooled post-IO IPD analysis, the ICI-downstaging cohort, SiLVER, the JAMA IM risk-score study, the ctDNA meta-analysis or the HIMALAYA 5-year paper; SIERRA was too new to be indexed there. Every number added from these papers is transcribed from the PubMed abstract, and the auditor should check them against tables where it can obtain them. The TACE ± ICI meta-analysis (PMID 42582959) is open access. I read its full text, but it gives per-trial OS inputs only in a forest plot, so I could not confirm which LEAP-012 OS estimate it pooled. Its search ended on 3 June 2026, which predates the final-OS posting. LEAP-012 OS figures come from the registry results table.
Wiki pages changed (8, all demoted curated → draft, last-curated: 2026-10-11):
locoregional-therapy.md: LEAP-012 registry-posted final OS, with an explicit not-peer-reviewed caveat. Added the 6-RCT TACE ± ICI meta-analysis (OS HR 0.85, 0.73–0.99; PMID 42582959) and the note that its search predates the LEAP-012 OS posting. Rewrote the open question.
clinical-trials-landscape.md: LEAP-012 status changed to terminated and its OS added in three tables. NCT06109272 changed from recruiting to active, not recruiting (89 enrolled), a drift since 2026-08-30. NCT04770896 was mislabelled "MORPHEUS-Liver resistance study"; its registry acronym is IMbrave251. NCT07350824 is now described as NON_RANDOMIZED per the registry, closing follow-up 2 of the 2026-10-07 pass.
surveillance-and-early-detection.md and biomarkers.md: ALTUS, the prospective head-to-head of the multitarget blood test against ultrasound (PMID 42800682; NCT05064553). Early-stage sensitivity was 66.7% vs 22.2%, from only 27 early-stage cancers. Specificity was 81.8%, missing the prespecified 82%, vs 98.6% for ultrasound. The study is cross-sectional, so it shows no outcome benefit. biomarkers.md also gained the ctDNA-MRD meta-analysis (PMID 42715861) and the corrected Wnt evidence level (below).
systemic-therapy.md: SIERRA single-arm STRIDE in Child-Pugh B7/B8, ECOG 2 or Vp4 (PMID 42838226), and pooled IPD from three prospective post-IO second-line MKI trials (PMID 42800536). The post-IO randomized-sequence absence was re-run and is now dated 2026-10-11.
liver-transplantation.md: SiLVER post hoc comparison of Milan, UCSF, Up-to-7 and Metroticket 2.0 (PMID 42669573), and the 416-recipient matched ICI-downstaging cohort (rejection 16.3% vs 8.2%; PMID 42821267). Its survivorship structure is stated explicitly: only patients who reached transplant are counted.
mash-masld-hcc.md: decision-curve analysis of 9 clinical scores in 853,131 veterans with non-cirrhotic SLD showing minimal HCC-screening utility (PMID 42766290).
molecular-landscape.md: the Wnt/CTNNB1 evidence level was stale and is corrected. It is now preclinical plus retrospective plus post hoc phase-3: CheckMate 459 (PMID 42762856) and IMbrave150 (PMID 39583094). This closes follow-up 1, the Wnt part of the 2026-10-07 pass.
OPEN-QUESTIONS.md: I sharpened OQ-2, OQ-4, OQ-5, OQ-6, OQ-9, OQ-10, OQ-12, OQ-14, OQ-15, OQ-19 and OQ-21, and added LEAP-012 final OS to dot row D6. OQ-5, OQ-10 and OQ-19 now absorb the evidence the 2026-10-07 pass flagged as stale. OQ-21 no longer presents NCT07350824 as a randomized vehicle. No question was closed. OQ-6 is partly answered for LEAP-012 only, and stays open pending EMERALD-1 OS and a systemic-first comparator.
Literature layer:
BIBLIOGRAPHY.md: 31 PMIDs added in a dated section with cited-by mapping, bringing the total to 359. Every PMID anywhere in the condition tree now appears in the bibliography. This includes the 11 dot-table PMIDs the 2026-10-07 pass left out (follow-up 3) and 41794163 and 41988192, now cited in OQ-10. The LOG-only PMIDs 34902568, 23873764 and 27429025 are cited nowhere in content and were not added.
guidelines/REGISTRY.md: 7 documents added. They are ESMO Pan-Asian adaptation 2026, an overlay on ESMO 2025 (PMID 42660571); AGA Clinical Practice Update 2026 on surveillance in older adults, including stopping (PMID 42820904); ACG 2026 cirrhosis preventive-care guideline, with semiannual HCC surveillance as a strong recommendation (PMID 42831646); KLCA SBRT 2026, published twice (PMIDs: 42810973, 42681828); MSGH Malaysia 2026 (PMID 42601228); the Brazilian consensus 2026 (PMID 42690308); and the HCC-Live trial-design consensus (PMID 42785981). No new AASLD, EASL, APASL, JSH, KLCA-NCC full or China guideline appeared in the window.
statistics/STATISTICS.md: 10 rows added, for 132 in total. They cover US SEER incidence-rate ratios 1992–2022 (PMID 42669752), an HBV-vaccination meta-analysis (PMID 42703321), ALTUS accuracy, LEAP-012 final OS, pooled TACE ± ICI OS, SIERRA Child-Pugh B safety and response, and ICI-downstaging rejection.
INDEX.md: page statuses and counts updated.
Triaged out (deliberately not added):
Global Burden of Liver Disease 2026 (PMID 42790604): narrative review, no new primary estimate.
Real-world atezolizumab-bevacizumab vs durvalumab-tremelimumab meta-regression (PMID 42851630): observational, and its survival summaries are explicitly descriptive.
Real-world second-line lenvatinib vs sorafenib (PMID 42726197): retrospective, and LEVIATHAN is already cited.
Early decompensation on IO in TriNetX (PMID 42777516): a candidate for red-flags next sweep.
THRI vs aMAP low-risk cirrhosis (PMID 42686444): a candidate for cirrhosis-and-risk-stratification / OQ-25.
Randomized phase II HAIC + atezolizumab-bevacizumab (PMID 42665190): primary endpoint not met.
PAL LIVER FACT-Hep baseline (PMID 42678241): a candidate for OQ-17 once outcomes report.
Two Research Square preprints, retrospective second-line durvalumab-tremelimumab vs lenvatinib (doi:10.21203/rs.3.rs-10994277/v1) and an HBV/HCV attribution meta-analysis (doi:10.21203/rs.3.rs-11095113/v1): not peer reviewed, not cited.
Everything else in the windows was basic science, single-centre prognostic modelling, case reports or reviews.
Patient-organisation material: I re-fetched ILCA (200), Blue Faery (200) and Macmillan (200). None showed new HCC-specific material in the window beyond standing pages for Liver Cancer Awareness Month and Love Your Liver. Liver Cancer UK now returns HTTP 403 (it was 200 on 2026-08-30), and the American Liver Foundation still returns 403. No patient-voice file was changed.
Not done / for the auditor (a different engine must audit; I may not audit this):
Re-fetch the NCT04246177 results section and confirm the final-OS numbers, the status and the posting date.
Check abstract-transcribed numbers against full text where obtainable, especially ALTUS (27 early-stage cases; specificity 81.8% vs the 82% target) and Ong 2026 (rejection 16.3% vs 8.2%; HRs).
Re-run S13/S13b for the post-IO randomized absence.
guidelines.md was deliberately left untouched and still curated. Its synthesis does not yet reflect LEAP-012 final OS or the AGA stopping guidance. The next pass should decide whether a synthesis edit is warranted.
red-flags-and-safety-concerns.md still contains the uncited graft-tolerance sentence and the internally inconsistent Lee 2024 GI-bleeding figure flagged on 2026-10-07; neither was touched here.
Next sweep should chase:
(a) A journal publication of LEAP-012 final OS and post-progression therapy data.
(b) EMERALD-1 OS.
(c) IMbrave251 (NCT04770896) results.
(d) ALTUS longitudinal or outcome data, and any guideline response to it.
(e) A peer-reviewed SIERRA survival analysis.
(f) AASLD/EASL responses to LEAP-012 final OS.
(g) Whether Liver Cancer UK's 403 persists.
2026-10-18 — Periodic literature sweep (author: Codex; window 2026-08-28–2026-10-18)¶
Scope and date choice: Read CONVENTIONS.md, CLAUDE.md, SOURCES.md, this log and INDEX.md before writing. Used the requested overlapping publication-date window, not only the week since the last sweep. Fifteen major-topic PubMed E-utilities searches covered epidemiology/etiology, cirrhosis/risk, surveillance, diagnosis/imaging, staging, molecular biology, resection/ablation/adjuvant treatment, transplantation, locoregional treatment, systemic treatment, MASLD, biomarkers, guidelines, patient experience and safety. Each used ("2026/08/28"[DP] : "2026/10/18"[DP]). Added an EDAT search for 2026-10-11–2026-10-18, a broader Cochrane query, and six targeted all-date/windowed queries for named trials, missed guideline/burden documents and randomized post-IO treatment. Retrieved metadata/abstracts for 1,730 unique records in the topic/supplemental-indexing set; screened titles/types and read selected abstracts, not 1,730 full texts. Publication-date searching recovered print-window reports missed by the preceding EDAT-based sweep; older documents recovered through official listings are explicitly labelled historical recoveries.
Sources actually queried beyond PubMed: Fifteen parallel topic searches through the Europe PMC REST API, using FIRST_PDATE:[2026-08-28 TO 2026-10-18], returned 999 distinct source/ID pairs after deduplication: 838 MED, 78 PMC and 83 preprint records. PMC-only entries are not automatically additional papers. No no-PMID record survived citation triage, so no DOI-only citation was introduced. Crossref was queried without mailto for every added record. ClinicalTrials.gov API v2, the ISRCTN API, official society listings (AASLD, EASL, ESMO, APASL and JSH), IARC, and public patient-organisation pages were checked. Full query strings, PubMed translations/returned IDs, registry snapshots, access checks, selected-source evidence and retraction checks are preserved in the sweep manifest.
Cochrane priority check: Ran "Cochrane Database Syst Rev"[jour] with each of the 15 HCC topic queries and the same DP window. The broader liver/HCC/cirrhosis query retrieved two reviews: thymosin-α1 for chronic hepatitis B and a sickle-cell transfusion overview. Neither supplies a new HCC-management review superseding a claim in this condition; the HBV review does not evaluate the nucleos(t)ide-antiviral claims retained here. This is a PubMed CDSR finding only. Cochrane Library/CENTRAL was not searched because of the documented HTTP 403 barrier.
Trial sweep: All 46 previously cited NCT records were re-fetched individually; status, results availability, enrollment and last-posted-update fields matched the 2026-10-11 manifest. LEAP-012's final OS results table was re-read and supports the existing figures; EMERALD-1 and IMbrave251 still have no posted registry results. Three additional live records were retrieved: TORCH (NCT02435953, completed, no posted results), EMERALD-3 (NCT05301842, active/not recruiting, last updated 2026-10-16, no posted results), and RAD-18-TAcE (NCT04803019, unknown status, last update 2021-03-19). The Italian trial's stale registry is not evidence of active recruitment. The ISRCTN API returned 98 records, four updated within the window; the directly HCC-focused update concerns MR elastography/treatment response, not a new comparative clinical outcome. The WHO ICTRP query returned HTTP 403 and the ChiCTR query HTTP 405; neither produced usable search results. No absence of non-US trials is inferred from ClinicalTrials.gov.
What survived triage: Thirteen bibliography entries were added, including one correction notice and two older journal-document recoveries; no landmark note was warranted. RAD-18-TAcE (Mosconi 2026, PMID 42265476) merits a limited locoregional synthesis because its full text exposes under-recruitment, unblinded local imaging and a progressor-only equivalence calculation. TORCH (Lyu 2026, PMID 42530948) and EMERALD-3 (Kudo 2026, PMID 42636832) are important recovered phase 3 reports, but their numerical results were not copied into the wiki: TORCH full text was unavailable and EMERALD-3 has an October erratum (PMID 42805209) whose correction text could not be retrieved. The publisher API returned metadata-only XML despite HTTP 200. These reports are catalogued with a full-text/correction follow-up, not silently treated as absent evidence.
Verified additions and changes: Five wiki pages changed, all dated 2026-10-18 and left draft: epidemiology-and-etiology.md adds the new HCC-specific infection-attribution estimates from Table 1 (Rumgay 2026, PMID 42805198); surveillance-and-early-detection.md adds the London surveillance-delivery denominator (Qurashi 2026, PMID 42866364); locoregional-therapy.md adds the qualified embolization comparison and pending phase 3 reports; clinical-trials-landscape.md refreshes the registry snapshot and adds the three publication-linked records; guidelines.md records newly recovered documents and replaces the stale blanket “awaiting mature OS” framing with LEAP-012's verified registry result. The London paper has internally inconsistent subgroup cells/regression labels, so only the coherent overall attendance counts were adopted. No subgroup odds ratio was added.
Guidelines, statistics and patient sources: The registry adds NCCN Insights v2.2026 (PMID 42874666), IHPBA resource-constrained guidance (PMID 42869716), the older AASLD Critical Update (PMID 39992051) and BEACON-HCC framework (PMID 42565685). JSH's official page separately lists a 2025 Japanese edition and the 2021 English edition; this listing was recorded without extracting recommendations or asserting an exact supersession chain. New recommendation wording/grades await primary full text. The statistics file gains six rows, separating IARC's 2024 all-liver-cancer estimates from the histology-specific HCC estimates and adding the London denominator and RAD response data. Patient-voice sources/organisation access were refreshed: Liver Cancer UK's September surveillance-policy contribution and upcoming 28 October patient-input webinar, and Blue Faery's upcoming 24 October HCC education event. Events were not treated as completed or evaluated interventions. Liver Cancer UK was readable through the web tool while direct requests still returned 403; American Liver Foundation direct requests also returned 403. Blue Faery, ILCA and Macmillan direct retrievals succeeded.
Research agenda and deliberate non-additions: OQ-6 now requires integration of TORCH/EMERALD-3; OQ-16 calls for referral failures to remain in an implementation trial's denominator; OQ-25 flags the THRI/aMAP cohort for full-text review (Moussa 2026, PMID 42686444), without adopting an abstract-derived stopping threshold. The early-decompensation cohort (Rimini 2026, PMID 42777516) was catalogued for safety follow-up without importing its abstract numbers. The European surveillance review (Qurashi 2026, PMID 42819713) supports the patient-source annotation; it is not relabelled a society guideline. Other mechanistic studies, single-centre prognostic/radiomics models, case reports, overlapping reviews and preprints were triaged out. No open question was closed. The targeted PubMed randomized post-IO query returned zero; that query-bounded finding does not establish worldwide absence.
Integrity and validation: Every newly written PMID was retrieved from live E-utilities in this session and all 13 added bibliography records passed both PubMed publication-type/correction and Crossref updated-by checks. No retraction/expression-of-concern flag was found; EMERALD-3's erratum is explicitly retained. Full-text article PMID/DOI identities were matched before extraction, and quotations/table locations are recorded in the manifest. Rechecked the earlier Lee bleeding follow-up against full-text Table 4: 6/112 GI-specific events supports the existing rounded 5% figure (PMID 39199649); no safety-page numerical correction was needed. Static validation checks new identifier provenance, edited-page draft status/date/structure, reference completeness, bibliography uniqueness and local links. Bibliography: 372 records; statistics: 138 data rows; wiki: 7 curated/10 draft. INDEX.md now reflects the actual nine retained junction rows rather than its stale count of fifteen. This is validation of this sweep, not an independent claim-to-source audit.
Access limits: Unpaywall was skipped because LIFEBANK_CONTACT_EMAIL is unset; no personal contact address was sought or transmitted. SciELO and LILACS could not be searched (HTTP 403, per SOURCES.md); transportability across regions therefore remains incompletely covered. Subscription databases were not searched. No git command, publishing script, site-generation step or deployment was run; only this condition's files were changed.
Next sweep / independent auditor: Obtain TORCH full text and EMERALD-3 plus its erratum before extracting outcomes; pursue their applicability and comparator quality alongside LEAP-012 post-progression therapy and EMERALD-1 final OS. Check new NCCN/IHPBA recommendation details, the AASLD Critical Update, BEACON-HCC and the Japanese 2025 guideline/English translation. Review THRI competing-risk/calibration methods and the decompensation study's event definitions. Continue watching IMbrave251 results, ALTUS longitudinal outcomes, SIERRA survival, and guideline responses to registry-posted LEAP-012 OS. Carry forward the uncited graft-tolerance wording on red-flags-and-safety-concerns.md; it was not substantively reworked here. Retry ICTRP/ChiCTR and blocked regional sources; refresh patient events only after any reports become public. A different engine must audit these changes before promotion.
2026-10-25 — Periodic literature sweep (author: Grok; window 2026-08-28–2026-10-25)¶
Scope and date choice: Read CONVENTIONS.md, CLAUDE.md, SOURCES.md, this log and INDEX.md before writing. Used the requested overlapping publication-date window from 2026-08-28, plus an EDAT search 2026-10-18–2026-10-25 for newly indexed records. Fifteen major-topic PubMed E-utilities searches plus high-signal (RCT/phase 3, meta-analysis/SR, Cochrane, guidelines) and named-trial queries (LEAP-012, EMERALD-1/3, TORCH, IMbrave251, ALTUS, SIERRA, HIMALAYA, post-IO RCT, AASLD/EASL, THRI/aMAP). The PubMed MCP tool was unavailable; esearch/efetch/esummary were called directly.
Sources actually queried: PubMed E-utilities; Europe PMC REST (TITLE_ABS / FIRST_PDATE window, non-MEDLINE SRC:PPR, named-trial and Cochrane queries); ClinicalTrials.gov API v2 for all 49 previously cited NCT IDs plus NCT03867084 and NCT04521153; ISRCTN XML API (q=hepatocellular carcinoma, 99 records); Crossref updated-by without mailto; official society listings (AASLD, ESMO, APASL, JSH, NCCN) and IARC GCO; patient-organisation homepages. WHO ICTRP homepage returned HTTP 200 but is JS-driven, so programmatic search was not completed. ChiCTR search page returned HTTP 200 with an unusable body. CTRI timed out. Not queried / blocked: cochranelibrary.com (HTTP 403) so CENTRAL was not searched; SciELO/LILACS (HTTP 403); Unpaywall skipped because LIFEBANK_CONTACT_EMAIL is unset. Every absence below is a named-source absence. Query strings, counts, IDs and registry snapshots are in the sweep manifest.
Cochrane priority check:"Cochrane Database Syst Rev"[jour] with HCC/liver cancer/cirrhosis/hepatitis B/C in the DP window returned 1 review: thymosin-α1 for chronic hepatitis B (Naing 2026, PMID 42713852), an update of CD014610, not an HCC-management review. A 2025–2026 CDSR query added APRI for HCV fibrosis (protocol, PMID 40757482) and Sorafenib with or without co-interventions for hepatocellular carcinoma (PMID 40568833), which is a protocol, not a completed review. No completed Cochrane review supersedes a claim in this condition. This is a PubMed CDSR finding only.
Trial sweep: All 49 previously cited NCT records were re-fetched individually on 2026-10-25. No status, results-availability or last-update change versus the 2026-10-18 snapshot. LEAP-012 remains terminated with the same posted OS; EMERALD-1 and IMbrave251 still have no posted results. KEYNOTE-937 (NCT03867084) is completed, 959 enrolled, results first posted 2026-04-24: RFS HR 1.06 (0.88–1.26), p=0.7190; OS HR 1.08 (0.81–1.43). A PubMed search for NCT03867084 and KEYNOTE-937[ti] did not retrieve a primary journal report. CARES-009 (NCT04521153) is active, not recruiting, 294 enrolled, randomized phase 2/3, last updated 2025-05-11, no posted registry results. ISRCTN listed five records updated since 2026-08-28; none is a new HCC comparative outcome trial (the 2026-10-19 update is maralixibat for intrahepatic cholestasis of pregnancy). No absence of non-US trials is inferred from ClinicalTrials.gov.
What survived triage: No landmark note. Four bibliography entries added (376 unique PMIDs). Wiki numbers were extracted from TORCH and EMERALD-3 PubMed abstracts because that was the outstanding 2026-10-18 follow-up and the abstracts now contain complete primary/secondary estimates; Europe PMC listed no open full text for either. KEYNOTE-937 is cited by NCT ID. CARES-009 is a 2025 historical recovery, labelled as such.
Verified additions and changes: Five wiki pages changed, all dated 2026-10-25 and left draft (surgical-resection-and-ablation.md demoted from curated):
locoregional-therapy.md: TORCH abstract PFS/OS (HR 0.47 / 0.50) and EMERALD-3 abstract PFS/OS (HR 0.70 / 0.84, p=0.18) with an unread-erratum caveat. Inconsistent TORCH TRAE percentages were omitted.
clinical-trials-landscape.md: registry snapshot refreshed; KEYNOTE-937 and CARES-009 added; TORCH/EMERALD-3 rows no longer say numbers await full text.
surgical-resection-and-ablation.md: KEYNOTE-937 registry-null adjuvant pembrolizumab; CARES-009 perioperative EFS 42.1 vs 19.4 months, HR 0.59, with 38% grade ≥3 TRAEs and two neoadjuvant treatment-related deaths.
mash-masld-hcc.md: aMAP in UK Biobank SLD; 15-year high-risk incidence 1.43% and aMAP >64 subgroup 5.35% at 15 years, which do not support the abstract's surveillance-eligibility claim if conventional annual thresholds are used.
guidelines.md: TACE-combination OS pattern now includes EMERALD-3; adjuvant versus perioperative IO distinguished.
OPEN-QUESTIONS.md: OQ-2 (aMAP SLD incidences), OQ-6 (TORCH/EMERALD-3 abstract numbers; EMERALD-3 joins LEAP-012 as PFS-without-OS), OQ-11 (KEYNOTE-937 null; CARES-009 perioperative, not adjuvant-only), OQ-25 (THRI HEDS external validation, C-index 0.77, no stopping rule). No question closed.
Literature layer: Bibliography 376. Guideline registry catalogs the 2026 Chinese conversion-then-surgery consensus (PMID 42765409; English abstract; Crossref 404). Statistics: 9 rows (TORCH, EMERALD-3, KEYNOTE-937, CARES-009, aMAP). Patient-voice access restored: Liver Cancer UK, American Liver Foundation and ACS liver-cancer pages returned HTTP 200 (previously 403/406). No new campaign text was extracted. Blue Faery's 24 October event had passed with no public outcome report; the Liver Cancer UK 28 October webinar had not yet occurred. Liver Cancer UK /news/ returned 404.
Triaged out: Cochrane HBV/sorafenib-protocol records; B-TACE meta-analysis (1 RCT); cfDNA diagnostic case-control MA; AFP-kinetics MA; sleep-disorder MA (I²=96%); phase 2 HFRT+Poly(I:C); second-line lenvatinib phase II (PMID 41038766) because Crossref lists an unread erratum; 102 Europe PMC preprints; mechanistic, single-centre, and overlapping reviews. PubMed IMbrave251[tiab] count=0. LEAP-012 still has no journal final-OS paper (25 records, titles checked).
Integrity: Every newly written PMID was retrieved from live E-utilities this session. Added records were checked in PubMed publication types and Crossref updated-by except PMID 42765409 (Crossref 404). No retraction flag. EMERALD-3's erratum is retained and its unread status is stated wherever numbers are used. KEYNOTE-937 figures come from the live registry results module (OG000 pembrolizumab, OG001 placebo). SciELO and LILACS could not be searched (HTTP 403); regional-transport claims are therefore incomplete. No git command or publish/deploy step was run.
Next sweep / independent auditor: Obtain TORCH full text and the EMERALD-3 erratum text; confirm abstract-transcribed numbers against tables. Watch for a KEYNOTE-937 journal report. Mature CARES-009 OS. EMERALD-1 OS, LEAP-012 journal OS, IMbrave251 results, ALTUS outcomes, SIERRA survival. Full text of THRI HEDS (PMID 41672261) and Moussa THRI/aMAP (PMID 42686444). Second-line lenvatinib phase II plus its erratum (PMID 41038766). JSH 2025 Japanese guideline/English translation and NCCN/IHPBA recommendation grades. Liver Cancer UK 28 October webinar report if posted. Carry forward the uncited graft-tolerance sentence on red-flags-and-safety-concerns.md. Retry ICTRP/ChiCTR structured search. A different engine must audit these changes before promotion.