Compare each study's plotted estimate and confidence interval with the analysis dataset to find the mismatch, then correct the plot or the table and recheck the output.
A forest plot and a data table can disagree because one was built from a different dataset, a different effect measure, or a different set of included studies. Cochrane Handbook Chapter 10 treats forest plots as displays of the effect estimates and intervals used in the meta-analysis, so the plot and the analysis dataset should describe the same studies and the same measure.
- Open the analysis dataset and list every study's effect estimate, confidence interval, and effect measure. Compare this list with the table row by row and mark each mismatch.
- Read the plotted point and interval for each study against that list. A point that sits at a different value, or an interval with different bounds, identifies whether the plot or the table carries the error.
- Check that both outputs use the same effect measure and the same included studies. A plot built on a ratio measure and a table reporting a difference measure will not match even when the underlying numbers agree.
- Correct the output that departs from the analysis dataset, then regenerate the plot or table from that dataset so both come from one source.
- Record the source of the conflict, such as a stale dataset, a changed effect measure, or a study added to one output only, in your analysis log.
- Recheck the corrected plot against the table and confirm that every estimate and interval now agrees before the figure goes into the manuscript.
Cochrane Handbook Chapter 15 emphasizes reading confidence intervals and exact P values carefully when interpreting results. After the numbers agree, confirm that the direction of each effect and the position of the pooled estimate match the interpretation in your text.