How do I run a sensitivity analysis plan across a review's pooled results?

Prespecify the exclusion rules, the outcomes they apply to, and the threshold for a material change before you recompute any pooled result, then run each scenario and compare it against the primary analysis.

A sensitivity analysis plan is a written set of alternative analyses that test whether the pooled estimate depends on specific decisions. Cochrane Handbook Chapter 10 treats sensitivity analyses as a way to examine how robust pooled results are to assumptions and choices, alongside its guidance on effect measures, heterogeneity, and missing data. GRADE rates certainty at the outcome level, so the plan should name the outcomes it covers and record how each scenario affects that rating.

Build the plan

1. List each exclusion rule with the study-level criterion it targets, such as a high risk-of-bias judgement from RoB 2, a specific trial design, or an imputed outcome. RoB 2 assesses a specific randomized-trial result across bias domains and returns low risk, some concerns, or high risk, so cite the domain judgement that triggers exclusion.
2. Map each rule to the outcomes it applies to. State whether it covers the primary outcome only or every pooled outcome, and note any outcome where the rule cannot apply because too few studies remain.
3. Set the threshold for a material change in advance. Define it in the effect measure you pooled, such as a shift in the pooled estimate or in the confidence interval that would alter your interpretation, and state what happens to the conclusion if the threshold is crossed.
4. Run each scenario as a separate recomputation and record the number of studies, the pooled estimate, the interval, and the heterogeneity result for each.

Comparison table template

| Scenario | Rule applied | Outcomes covered | Studies included | Pooled estimate (CI) | Heterogeneity | Change vs primary | Threshold crossed? | Conclusion impact |
|---|---|---|---|---|---|---|---|---|
| Primary | None | All | | | | Reference | — | — |
| Scenario 1 | | | | | | | | |
| Scenario 2 | | | | | | | | |

Decision rule

Compare each scenario against the primary result using the prespecified threshold. If no scenario crosses it, report the pooled result as stable across the tested rules and keep the conclusion. If a scenario crosses it, report that scenario alongside the primary analysis, describe which rule drove the change, and revise the conclusion or the certainty rating for the affected outcome. Record the decision in the review so readers can trace which rule changed the interpretation.