How do I fix a direction-of-effect error in my forest plot?

Fix a direction-of-effect error by rechecking each study's event coding and comparison direction against the protocol, then recomputing the pooled estimate and revising any interpretation that depended on the original sign.

A forest plot places each study's effect estimate on a common scale, so a single study entered with the comparison reversed can pull the pooled estimate and its confidence interval in the wrong direction. The Cochrane Handbook's meta-analysis chapter covers effect measures, forest plots, and pooled results, and its interpretation chapter covers confidence intervals and review conclusions. Use those chapters to confirm what your chosen effect measure represents before you change any numbers.

1. Reopen the protocol and write down the intended comparison, the event or outcome definition, and the direction that counts as benefit for each outcome. Expected result: one reference statement you can check every study against.
2. For each study, compare its coded events and group labels with that statement. Expected result: a list of studies whose coding or comparison direction conflicts with the protocol.
3. Correct the coding or the sign for the flagged studies, and record the change and its reason in your analysis log. Expected result: every study now points in the same direction on the forest plot.
4. Recompute the pooled estimate and its confidence interval with the corrected data, and compare the new result with the original. Expected result: a revised pooled effect and interval you can report.
5. Review every sentence that cited the original sign, including abstract, results, and discussion text, and rewrite those that no longer match. Expected result: conclusions consistent with the corrected estimate.

Run a sensitivity analysis that repeats the pooled estimate with and without the corrected studies. The Cochrane Handbook describes sensitivity analyses as a way to examine how robust results are to analytic choices, so the comparison shows how much the error affected your conclusion. When the corrected interval crosses the null value, describe the result as uncertain and adjust the strength of your claims accordingly.