How do I repair a narrative synthesis that contradicts its own evidence table?

Repair the synthesis by matching each narrative claim to its evidence table row, correcting the text wherever the direction, size, or certainty of the effect differs, then rechecking the conclusions for consistency.

Reconciliation sequence

1. Build a claim-to-row map. List every sentence in the synthesis that states an effect, then write the table row, study, or pooled estimate it cites beside it. A claim with no matching row is the first candidate for revision or removal.
2. Compare the effect measure and direction. Check that the narrative uses the same measure as the table (for example, risk ratio versus odds ratio) and reports the same direction of effect. Cochrane Handbook Chapter 10 covers effect measures and the interpretation of pooled results, so use its definitions to confirm what the table value represents.
3. Compare precision and certainty. Check that confidence intervals, exact P values, and any certainty rating in the narrative match the table. Cochrane Handbook Chapter 15 addresses confidence intervals, exact P values, absolute and relative effects, and certainty, which are the fields where narrative text most often drifts from the table.
4. Revise the synthesis text. Rewrite each mismatched sentence to state the table's value and its uncertainty, and remove claims the table does not support. Record the change, the row it was reconciled against, and the reason in a reconciliation log.
5. Recheck the conclusions. Read the revised conclusions against the corrected claims and confirm that each conclusion follows from the reconciled evidence, including any applicability or indirectness limits noted in the table.

What to record

Keep a short log with four fields per change: the original claim, the table row and value, the revised wording, and the reason for the change. This log lets co-authors and reviewers trace how the synthesis was reconciled and spot claims that were dropped.

Where contradictions usually sit

Mismatches concentrate in effect direction, measure type, confidence interval bounds, and certainty statements. Cochrane Handbook Chapter 15's guidance on absolute and relative effects and on responsible review conclusions helps when a narrative overstates what a pooled result supports. Cochrane Handbook Chapter 10's guidance on heterogeneity and sensitivity analyses helps when the narrative describes consistency the table does not show.