Meta-Analysis

The statistical pooling of results from multiple studies (within a systematic review) into a single, weighted summary estimate. Studies are weighted by their precision, combined with a fixed-effect or random-effects model, displayed on a forest plot, and their consistency assessed with heterogeneity (I²).

Each square is one study (sized by its weight); the gold diamond is the pooled estimate. Pooling borrows strength across studies, so the diamond’s confidence interval is narrower than any individual study’s — greater precision, provided the studies are similar enough to combine.

First check heterogeneity (I²): low (<25%) supports a fixed-effect pool; higher values demand a random-effects model and caution. Then read the diamond — its centre is the best estimate, and if its interval excludes the null line the pooled effect is statistically significant.

A meta-analysis can settle a question that individual underpowered trials left uncertain — or it can manufacture false confidence, turning a heap of biased or dissimilar studies into one deceptively precise number. Precision is not the same as truth.

• Weight by precision · fixed- or random-effects model
• Always report I² — high heterogeneity undermines pooling
• Pooled CI is narrower → more precise (if valid to pool)

Corticosteroids in community-acquired pneumonia — a 2023 systematic review & meta-analysis (search updated to March 2023, MEDLINE/Embase/Cochrane Library) pooled randomised trials and found corticosteroids reduced all-cause mortality, RR ≈ 0.69 (95% CI 0.53–0.89), with the benefit concentrated in severe CAP and hydrocortisone — consistent with the CAPE COD trial. The pooled diamond is only as good as its inputs and its homogeneity — here low I² and pre-specified subgroups support the estimate; high heterogeneity or publication bias would render a tight pooled CI falsely reassuring.

• Combining apples and oranges — pooling clinically dissimilar populations, interventions or outcomes.
• Ignoring heterogeneity (I²) and publication bias (funnel-plot asymmetry / small-study effects).
• Over-precision — mistaking a narrow pooled CI for certainty when the inputs are biased.