From update alert to useful briefing
What changed in a medical guideline? Build a traceable comparison with AI
Compare two dated guideline versions with an evidence trail. Use a change table that preserves wording, population, grading and source locations.
The short answer
A useful guideline comparison shows the old wording, the new wording, where each appears and what still needs interpretation. Give AI the two actual versions and ask it to map changes before explaining them.
Make sure you are comparing the right documents
An announcement that a guideline has changed is the beginning of a reading task. Obtain both versions from the issuing organization and record their titles, dates, jurisdictions and scope. Check whether the newer document replaces the whole guideline, updates one section, or is an addendum. An executive summary and a full guideline are different documents; label them accordingly.
Store the documents in MedSafeAI’s AI Library with clear names such as ‘Society — topic — previous version — date’ and ‘Society — topic — current version — date.’ Include relevant appendices when available. If you only have an abstract or announcement, limit the comparison to that material and keep the full-document review pending.
Use one row per recommendation or topic
Begin with extraction rather than a broad ‘what is new?’ summary. The following is our comparison worksheet, not a claim that a particular guideline changed. Keep source locations even when the AI reports no change. A row without a readable source remains unverified.
| Field | Record for each version | Why keep it? |
|---|---|---|
| Document identity | Title, publication date and section | Prevents mixing an update with a different publication |
| Population and setting | The group and situation the wording addresses | Shows whether the scope is actually comparable |
| Recommendation | Short exact passage or faithful paraphrase plus location | Makes the claimed change checkable |
| Strength and certainty | The guideline’s own labels, where reported | Preserves the authors’ distinctions |
| Change status | Changed, unchanged, added, removed or not located | Separates findings from gaps in retrieval |
Ask for a comparison that can be audited
‘Compare [previous version and date] with [current version and date] for [topic and population], using only these supplied documents. Create one row per recommendation. Include the wording, source section or page, population, setting, strength and certainty labels where stated. Classify each row as changed, unchanged, added, removed or not located. Do not translate “not located” into “removed.” Separate textual changes from your interpretation of their significance. List the passages a clinician should inspect before drawing a practical conclusion.’
Check the largest claimed changes first. Confirm that a relocated paragraph has not been labelled a deletion and that differently numbered sections address the same question. Preserve qualifiers such as setting, exceptions and eligibility; shortening those away can change the meaning of a recommendation.
Read the rationale before writing the headline
After verifying the wording, inspect the guideline’s explanation and supporting references. Evidence Search can help you locate cited research; Strict Mode can support a citation-focused review. Distinguish ‘the authors explain the change with this evidence’ from ‘this is our interpretation.’ AGREE’s reporting checklist is a useful companion when checking whether a guideline describes its scope, methods and other essential information.
If you use Ask All AI and Consensus, give each model the same two documents and scope. Investigate differences in extraction or interpretation by returning to the source. Agreement between models does not resolve an ambiguous passage or establish that the update applies to a particular patient.
Publish a briefing with a visible boundary
Your team briefing can now state the versions reviewed, confirmed changes, unchanged points that matter, and outstanding questions. Link to both issuing-body documents and date your review. Keep local implementation decisions separate from this document comparison. When a correction or addendum appears, update the relevant rows and explain what changed in your own briefing.
Sources & further reading
Follow the original sources to check the details. Product pages describe their own products; they are not independent clinical evaluations.
- AGREE Enterprise: AGREE Reporting Checklist agreetrust.org
- MedSafeAI: medical AI workspace and product features medsafeai.com
For healthcare professionals. AI responses and agreement between models do not replace source verification or clinical judgment.