Which effect measure for which data?
| What the study reports | Effect measure | Where to enter it |
|---|---|---|
| Events and totals per arm (binary) | Odds ratio, risk ratio, risk difference | Forest plot, “Events / total” input |
| Mean, SD and n per arm | Mean difference, standardised mean difference (Hedges’ g) | Forest plot, “Mean / SD / n” input |
| Median with IQR or range | Estimated mean and SD, then MD or SMD | Median to Mean & SD Calculator → forest plot |
| Hazard ratio with 95% CI | Hazard ratio (log scale) | Forest plot, “Effect + CI” input |
| Kaplan–Meier curve but no hazard ratio | Hazard ratio from reconstructed patient data | Kaplan–Meier Curve Digitizer → forest plot |
| Any estimate with 95% CI | The reported ratio or difference | Forest plot, “Effect + CI” input |
Pool one effect measure per analysis. Ratios are pooled on the log scale; the standardised mean difference lets you combine outcomes measured on different scales, at the cost of interpretability.
Worked examples
- PRISMA 2020 flow diagram for a new review
- RoB 2 traffic-light plot
- Forest plot: BCG vaccine risk ratios
- Funnel plot with Egger’s test and trim and fill
- Two-panel figure: forest and funnel plot
Citing the methods
The tools that calculate statistics draft a methods paragraph with its references (for example DerSimonian and Laird, Egger et al., Luo et al., Guyot et al.), and the forest plot, funnel plot, risk-of-bias, median-to-mean and digitizer tools also give an R script that reproduces the results. See how to cite CleanFigure andhow we check the numbers.
Frequently asked questions
Common-effect or random-effects model?
A common-effect (fixed-effect) model assumes every study estimates the same effect; a random-effects model allows the true effect to vary between studies and is the usual choice when studies differ in populations or methods. The forest plot tool fits both (DerSimonian–Laird for τ²) and shows I², τ² and the prediction interval, which describes where the effect of a new study is likely to lie.
How many studies do I need for a funnel plot?
The Cochrane Handbook advises tests for funnel plot asymmetry only with at least 10 studies; with fewer, the tests have too little power to tell asymmetry from chance. The funnel plot tool says so when you have fewer studies.
Can I pool studies that report medians with studies that report means?
Yes, if the data are roughly symmetric: estimate means and SDs from the median with the IQR and/or range (Luo et al. 2018, Shi et al. 2020 or Wan et al. 2014), then pool as usual and say so in the methods. The calculator warns when a median lies far from the middle of its quartiles, which suggests skewed data where the estimates can be biased.
A trial shows a Kaplan–Meier curve but no hazard ratio. What can I do?
Reconstruct the individual patient data from the curve and its numbers at risk with the Guyot et al. (2012) algorithm, then calculate the hazard ratio. The Kaplan–Meier Curve Digitizer does both in the browser and gives the hazard ratio with its 95% CI for the forest plot, with an R script to reproduce it.
Do the results match RevMan, R or Stata?
The statistics are checked against R packages on published datasets: metafor for the forest and funnel plots, meta for the median conversions, IPDfromKM for the curve reconstruction and survival for Kaplan–Meier analyses. The forest plot, funnel plot, risk-of-bias, Kaplan–Meier and median-to-mean tools and the digitizer export an R script that reproduces their results. RevMan also pools with inverse-variance weights and DerSimonian–Laird, so its pooled results agree when the settings are the same. The details are on the page about how we check the numbers.
Is my data uploaded?
No. Every tool runs in your browser, including the image reading and the statistics, so unpublished data and extracted results never leave your computer. Nothing requires an account.