Compare each conclusion sentence with the GRADE certainty rating for the outcome it rests on, then flag any sentence whose wording is stronger than that rating supports. GRADE rates certainty at the outcome level using explicit criteria, so the check has to be done outcome by outcome.
- List every conclusion sentence in the abstract, discussion, and summary, and note which outcome each one describes. Expected result: a short mapping of sentences to outcomes, which exposes conclusions that rest on no rated outcome.
- Record the GRADE certainty rating (high, moderate, low, or very low) for each of those outcomes from the summary-of-findings table. Expected result: each conclusion now has a rating beside it, or is marked as unrated.
- Read the effect estimate and its confidence interval for that outcome, following Cochrane Handbook Chapter 15 guidance on confidence intervals, exact P values, and absolute versus relative effects. Expected result: you can see whether the sentence describes the size and precision of the effect or overstates it.
- Flag mismatches where the wording claims more certainty than the rating supports, such as a firm causal or definitive statement attached to a low-certainty outcome. Expected result: a list of overstatements tied to specific outcomes.
- Write revised wording for each flagged sentence that matches the rating and the effect estimate. Expected result: an alignment table with columns for the original sentence, outcome, certainty rating, effect estimate, mismatch type, and revised wording.
GRADE's explicit, outcome-level judgements give you the reference point for step 2; the Cochrane Handbook's interpretation guidance supports step 3. Where a conclusion has no rated outcome, treat the alignment as unresolved and check the review's methods section for how that outcome was handled before revising the sentence.