Meta-analysis with rigour

A meta-analysis quantitatively synthesises evidence across studies to produce a single, more precise estimate of an effect — but only when the studies are comparable and the statistical model is chosen correctly. Done well, it is among the most influential forms of evidence in medicine. Done carelessly — pooling studies that should never have been combined, or hiding heterogeneity behind a single headline number — it misleads. We help you do it properly, from effect-size extraction through to the diagnostics that tell you whether your pooled estimate can be trusted.

Choosing the right effect measure

The first decision is what to pool. For binary outcomes that usually means an odds ratio, risk ratio, or risk difference; for continuous outcomes, a mean difference when studies use the same scale or a standardised mean difference when they do not; and for time-to-event data, a hazard ratio. We help you extract or compute the correct effect size and its variance from each study — including the common cases where a paper reports the data in an awkward form and the effect size must be derived.

What we cover

• Effect-size extraction and computation (OR, RR, RD, MD, SMD, HR) from reported study data
• Model selection — fixed-effect versus random-effects, with the rationale documented
• Heterogeneity assessment — the I² statistic, Cochran’s Q, and the between-study variance τ²
• Forest plots with pooled estimates and study weights
• Subgroup and meta-regression analysis to explore sources of heterogeneity
• Publication-bias assessment — funnel plots and Egger’s or Begg’s test
• Sensitivity analysis, including leave-one-out, to test how robust the result is

Fixed-effect or random-effects?

This choice is not cosmetic — it changes both the pooled estimate and how far it can be generalised. A fixed-effect model assumes every study estimates one common true effect; a random-effects model assumes the true effect varies across studies and estimates an average of that distribution. In medical meta-analyses, where populations and protocols differ, a random-effects model is usually more realistic. We help you justify the choice on the basis of your clinical reasoning and the observed heterogeneity, rather than picking whichever gives a tidier result.

Taking heterogeneity seriously

Heterogeneity — real variation in effects across studies — is not a nuisance to be hidden but information to be explained. We quantify it (I², Q, τ²) and, where it is substantial, help you explore why through pre-specified subgroup analyses and meta-regression against study-level characteristics such as dose, population, or risk of bias. Where studies are simply too heterogeneous to pool responsibly, we say so and recommend a structured narrative synthesis instead.

Tools we use

We work in R (the metafor and meta packages) for full flexibility and reproducibility, in RevMan for Cochrane-style reviews, or in Comprehensive Meta-Analysis (CMA) where preferred. You receive publication-ready forest and funnel plots and, on request, the reproducible code or files so your analysis can be re-run and verified.

How the process works

Starting from your included studies and extracted data, we compute the effect sizes, select and fit the model, assess heterogeneity and publication bias, run the planned subgroup and sensitivity analyses, and return the plots and statistics with a written interpretation. We then help you translate that into the results and discussion of your manuscript, with the caveats stated honestly.