Methodology questions, answered directly
Every guide opens with the short answer, then explains what it depends on and where it stops being true. They are written for people who will be asked to defend the choice at peer review — so they take a position, and they say when the usual rule of thumb is wrong.
Effect measures
Risk ratio or odds ratio: which should I use?
Risk ratios are easier to interpret; odds ratios are required by some designs and exaggerate when outcomes are common. How to choose, and how to report either honestly.
How do I pool outcomes measured on different scales?
When trials measure the same construct with different instruments, the standardised mean difference lets you pool them — with real caveats about interpretation and sample spread.
How do I meta-analyse hazard ratios?
Pool hazard ratios on the log scale using inverse-variance weighting, recovering the standard error from the published confidence interval — and check proportional hazards before you do.
How do I meta-analyse prevalence?
Pooling proportions is not the same as pooling comparative effects. Why raw proportions misbehave near 0 and 1, which transformation to use, and why heterogeneity is expected.
Models & heterogeneity
Fixed-effect or random-effects: which should I use?
Choose between fixed-effect and random-effects meta-analysis based on the question you are asking, not on the heterogeneity you observe. A practical guide.
Is my I² too high?
I² is not a measure of how much effects differ, and there is no threshold that makes a meta-analysis invalid. What to look at instead of the percentage.
Network meta-analysis
When should I use a network meta-analysis?
Network meta-analysis compares three or more interventions at once, including pairs never tested head to head — but only when the network is connected and transitivity holds.
Can I trust network meta-analysis rankings?
Treatment rankings and SUCRA values are the most over-read output of a network meta-analysis. Why a rank of first is not evidence of superiority.
Certainty & bias
What does GRADE certainty actually mean?
GRADE rates confidence in an effect estimate per outcome, not study quality and not effect size. What high, moderate, low and very low actually claim.
How do I assess risk of bias with RoB 2?
RoB 2 assesses a specific result from a randomised trial across five domains, using signalling questions. It is not a quality score and it is not assessed per study.
Question design & process
PICO, PEO or SPIDER: which framework fits my question?
Question frameworks are not interchangeable. Choose by what your question is about — an intervention, an exposure, or an experience — before you write a search.
How do I do a systematic review, end to end?
The complete path from a research question to a synthesised, reportable result: protocol, search, screening, extraction, appraisal, synthesis and reporting.
See the whole process on a real review
A worked systematic review, open and read-only — the search, the screening decisions, the extracted data and the pooled result, with every number traceable to its source. No account needed.