Recompute the pooled estimate without trials judged high risk of bias, then compare direction, magnitude, and interval width against the main analysis in a sensitivity table.
- Fix the exclusion rule before recomputing. Use the RoB 2 judgement for the specific result you are pooling: exclude trials rated high risk and keep trials rated low risk or some concerns. Record the rule in the table so readers can see which trials left the analysis and why.
- Rebuild the analysis with the same settings. Keep the same effect measure, model, and data-handling choices as the main analysis, and change only the trial set. That isolates the effect of removing high-risk trials from changes caused by switching methods.
- Compare three features side by side. For each outcome, report the main pooled estimate and the restricted estimate with their confidence intervals, then note direction (does the effect still favor the same group?), magnitude (how far did the point estimate move?), and interval width (did the confidence interval widen, and does it now cross the no-effect value?).
- Apply your pre-specified shift rule. State in advance what counts as a conclusion-changing shift, for example a change in direction or a confidence interval that crosses the null. Report the outcome against that rule instead of deciding after seeing the numbers.
- Present the sensitivity table and interpret it. List each outcome, the trials removed, both estimates with intervals, and the shift verdict. If the restricted estimate keeps the same direction with a wider interval, describe the main finding as sensitive to high-risk trials and report both results.
Cochrane Handbook Chapter 10 covers sensitivity analyses and interpretation of pooled results; Chapter 8 describes the RoB 2 domains and judgements that supply the exclusion criterion.