Skip to content

Leave-One-Out Meta-analysis

Re-pool a meta-analysis leaving out each study in turn to test robustness.

Model
Confidence level

Enter your values to see the result.

Ratios are pooled on the log scale with inverse-variance weights; the random-effects model uses DerSimonian–Laird τ². A study whose removal changes the conclusion or I² a lot deserves a closer look, but do not drop studies on this basis alone.

Continue your work

Share this tool

Found it useful? Send it to a friend or teammate.

Rate this tool

0.0

0 ratings

  • 5 stars 0
  • 4 stars 0
  • 3 stars 0
  • 2 stars 0
  • 1 star 0

Click a star to rate this tool

Clear instructions

Find steps, examples and limitations below.

Use online

Open the tool in a supported web browser.

Free to use

No sign-up required. Tool-specific limits may apply.

How to use the Leave-One-Out Meta-analysis

Re-pool a meta-analysis leaving out each study in turn to test robustness.

  1. 1 Paste one study per line with its ratio and CI, difference and CI, or effect and SE.
  2. 2 Choose a random-effects or fixed-effect model.
  3. 3 Review the table of pooled results with each study left out.

Example and practical tips

Six trials pool to an odds ratio of 1.52 (1.22 to 1.89). Leaving out the largest effect shows whether the conclusion depends on that single trial.

Frequently asked questions

What is a leave-one-out analysis for?

It is a sensitivity analysis that shows whether one study, often a small or high-risk one, drives the pooled result or the heterogeneity.

Should I remove an influential study?

Not on this basis alone. Report the analysis and look for a reason, such as risk of bias or a different population.

Report an issue

Something broken or not quite right? Tell us and we will look into it.