grade-evidence

name	grade-evidence
description	Apply the GRADE framework (Grading of Recommendations, Assessment, Development, and Evaluations) to label evidence quality and recommendation strength. Use when the user wants to weight competing recommendations or label confidence on a proposal.

GRADE evidence

When to use this

• You're combining multiple sources (RCTs + cohort studies + guidelines) and need a single confidence label.
• You're recommending a question wording change based on imperfect evidence and want to be honest about how strong the basis is.
• The user asks "how confident are you?" about a specific proposal.

Quality of evidence (4 levels)

• High — RCTs of low risk of bias, consistent, direct, precise.
• Moderate — downgraded once for one of: risk of bias, inconsistency, indirectness, imprecision, publication bias.
• Low — downgraded twice, or non-randomized design as base.
• Very low — case reports / expert opinion.

Upgrade observational data on: large effect, dose-response, plausible confounders biasing toward null.

Strength of recommendation (2 categories)

• Strong — most informed patients would choose this option; ratio of benefits to harms is clearly favorable; high confidence.
• Conditional / weak — patient values or circumstances change the right answer.

How to surface it

In reply.md and in any proposed-edit note:

Adding this red-flag screen is a strong recommendation based on moderate-quality evidence (downgraded for imprecision in the pooled CI).

In findings.json, attach a gradeAssessment field per finding:

{
  "pmid": "12345",
  "gradeAssessment": {
    "qualityOfEvidence": "moderate",
    "downgradedFor": ["imprecision"],
    "recommendationStrength": "strong"
  }
}

Pitfalls

• Calling something "strong" because the effect size is large; the GRADE strength rating considers values + harms, not just magnitude.
• Mixing GRADE with USPSTF letter grades or NICE strengths. They're parallel systems; cite the source's native grading and add GRADE on top only if the source didn't.