Canonical text (the exact bytes fingerprinted)

TRUVACE RECORD VERSION
record: TRV-2026-2170
version: 1
kind: certified
reason: Certified into the record
timestamp: 2026-10-22T18:12:54.576377Z
status: published
lens: g_space
sector: health
headline: From Divergent Guidelines to Delphi-Ready Items: Tutorial for Large Language Model-Assisted Guideline Harmonization Using Penile Cancer as a Worked Example
dek: Background: Clinical practice guidelines are central instruments of evidence-based care; yet, in many domains, particularly rare or complex diseases, multiple guidelines coexist and frequently offer conflicting advice. Existing strategies for resolving such divergences are often labor-intensive, opaque, and insufficiently reproducible.
Objective: This tutorial aimed to develop and demonstrate a reproducible workflow for transforming divergent clinical guideline recommendations and literature-derived white areas…
gain_title: An LLM-assisted workflow condensed 157 recommendations from 5 penile cancer guidelines into 68 harmonized items and flagged where guidelines diverge, making consensus preparation more reproducible.
problem_title: (none)
trace_subject: (none)
gain_reading: An LLM-assisted workflow condensed 157 recommendations from 5 penile cancer guidelines into 68 harmonized items and flagged where guidelines diverge, making consensus preparation more reproducible.
gain_evidence: the workflow extracted 157 clinical recommendations from 5 guidelines and consolidated them into 68 harmonized guideline-derived items
problem_reading: (none)
problem_evidence: (none)
quick_read: Researchers published a step-by-step tutorial in the Journal of Medical Internet Research showing how three large language models can help reconcile conflicting clinical guidelines. Using penile cancer as a worked example, they extracted recommendations from five guidelines, merged overlapping statements, and classified each guideline's stance on every item.

The work matters because rare and complex diseases often have several guidelines that disagree, and manual reconciliation is slow and hard to reproduce. The authors present the method as scalable and transferable, but the evidence is a single worked example in one disease, and the later clinical consensus voting had not yet taken place.
limitation: The method was demonstrated on a single disease as a worked example, and its output is only candidate items awaiting clinical consensus voting, so clinical impact is not yet shown.
tag: Evidence-backed gain
key_points: 157 recommendations were extracted from 5 guidelines and consolidated into 68 harmonized guideline-derived items. | Stance classification found 15 uniformly supported items and 53 divergent, ambiguous, or incompletely addressed items. | Literature processing produced 95 harmonized white-area items, of which 9 were true white areas not addressed by the guidelines. | The final output was 62 human-reviewed Delphi-ready candidate items for later consensus voting.
rundown: The workflow began with manual selection of guidelines and a systematic PubMed search for candidate issues the guidelines might not address. Three complementary LLMs then ran predefined prompts for multimodal recommendation extraction, semantic harmonization, and item-level stance classification.

Evaluation combined a consistency check across the three-model ensemble with an independent pre-Delphi appraisal by four penile cancer experts. They used a predefined five-point suitability scale with an acceptance threshold.

Most harmonized guideline items, 53 of 68, were divergent, ambiguous, or incompletely addressed, which suggests substantial inconsistency among the guidelines. The output is a set of candidate items for subsequent clinical consensus voting, not final clinical recommendations.
sources:
- peer_reviewed | Journal of Medical Internet Research | https://doi.org/10.2196/96886 | 2026-10-20
prev: 0000000000000000000000000000000000000000000000000000000000000000

sha256

eecce01bf788578837c2682bd8d373e9a3f4878c6609eac1e00460593f915f0f

previous

0000000000000000000000000000000000000000000000000000000000000000