Recheck whether the analysis combined multiple arms from one trial or ignored clustering, then recompute the pooled estimate with an adjustment and compare the revised estimate with the original. The Cochrane Handbook's meta-analysis guidance covers effect measures, fixed and random effects, heterogeneity, missing data, sensitivity analyses, and forest plots, and its interpretation chapter covers confidence intervals, exact P values, absolute and relative effects, applicability, and certainty. Use those chapters to structure the repair.
Repair sequence
- Identify the unit-of-analysis problem in the original analysis. Check whether the same trial contributed more than one arm to the same comparison, or whether clustered participants were entered as independent observations. Record which trials and comparisons are affected, because that list defines the scope of the recomputation.
- Choose an adjustment that matches the problem. For multi-arm trials, combine or select arms so each trial contributes once per comparison. For clustered data, apply an adjustment that accounts for the clustering before pooling. State the method you used and the inputs it required.
- Recompute the pooled estimate with the adjustment and produce a forest plot alongside the original result. Compare the point estimate, its confidence interval, and the heterogeneity statistics between the two versions.
- Run a sensitivity analysis that removes or reweights the affected trials. Report whether the revised estimate changes the direction, magnitude, or certainty of the finding.
- Document the correction in the review record: the original analysis, the problem found, the adjustment applied, and the revised estimate with its confidence interval. Report both versions so readers can see what changed.
What to report
Present the revised estimate with its confidence interval and exact P value, and describe the absolute and relative effects. Discuss applicability and certainty in light of the correction. If the revised and original estimates disagree materially, say so and explain which analysis you consider appropriate for the review's conclusions.