Study-level predictors of varying effect sizes
Meta-regression extends meta-analysis by regressing effect size on study characteristics (moderators) like publication year, sample size, intervention dose, or population age. Each study becomes a dot, sized by precision, scattered across a moderator axis. A trend line slopes upward or downward, revealing whether the moderator predicts effect magnitude.
Unlike subgroup analysis, which divides studies into categories, meta-regression fits a continuous relationship. A moderator showing strong association (steep trend, tight scatter) suggests it explains heterogeneity; a flat trend suggests the characteristic is not predictive. This technique helps answer 'what type of study or population sees the largest effect?'
Understanding variation through study context
A positive slope on treatment duration might show that longer interventions produce larger effects, directing future research and practice toward optimal exposure. A zero slope on population age suggests treatment efficacy is age-independent, simplifying clinical application. Multiple moderators can be examined together or separately.
Meta-regression power depends on number of studies. With fewer than ten studies, precision is low and spurious associations arise. Fitted trends look clean but confidence intervals widen. Reporting both univariable and multivariable models, along with confidence intervals, prevents over-interpretation of weak patterns.