meta-analysis

**Definition:** Statistical synthesis of data from multiple studies (usually RCTs, but can include cohort/case-control) to generate a pooled estimate of effect. A meta-analysis is usually performed as part of a systematic review (a structured, reproducible literature search with predefined inclusion/exclusion criteria), but a systematic review does not always include a meta-analysis.

**Steps:** 1. Formulate a focused question (PICO: Population, Intervention, Comparison, Outcome). 2. Comprehensive, systematic literature search (multiple databases, grey literature, unpublished data). 3. Apply predefined inclusion/exclusion criteria. 4. Assess study quality and risk of bias (e.g., Cochrane Risk of Bias tool). 5. Extract data and statistically pool results, weighting larger/more precise studies more heavily (inverse-variance weighting). 6. Assess heterogeneity and publication bias; perform sensitivity/subgroup analyses.

**Key concepts:** - **Forest plot:** Displays each study's effect estimate (square, size proportional to weight) with its 95% CI (horizontal line) and the pooled estimate (diamond). Vertical line = null value (RR/OR = 1; mean difference = 0). If the CI or diamond crosses the line of no effect, the result is not statistically significant. - **Funnel plot:** Scatterplot of effect size vs. study precision/size, used to detect publication bias. Symmetric inverted funnel = low bias; asymmetry suggests publication bias (small negative studies missing). - **Heterogeneity:** Variability among studies' results beyond chance. Assessed with Cochran's Q test and the I² statistic (% of variability due to heterogeneity rather than chance; roughly >50% = substantial). - **Fixed-effects model:** Assumes one true underlying effect; differences are due to chance. Used when heterogeneity is low. - **Random-effects model:** Assumes the true effect varies across studies; gives wider CIs. Used when heterogeneity is present.

**Advantages:** - Increased statistical power and precision (narrower CI) - Can detect small effects or effects in subgroups that individual trials were underpowered for - Resolves conflicting results across studies - Generalizability across populations - Can identify gaps in the literature and generate hypotheses

**Limitations:** - Publication bias (positive studies preferentially published) - Heterogeneity in populations, interventions, outcomes, or methodology ("apples and oranges") - "Garbage in, garbage out": pooling low-quality studies yields unreliable results - Selection bias from search strategy or inclusion criteria - Confounding from observational studies - Can be outweighed by a single large, well-designed RCT in some cases

**Board-relevant points:** - Evidence hierarchy (highest to lowest): systematic reviews/meta-analyses of RCTs → individual RCTs → cohort studies → case-control studies → cross-sectional studies → case series/case reports → expert opinion/animal/in vitro studies. - Meta-analysis is not a primary study design; it is a secondary analysis of existing data. - It does not eliminate bias inherent in the underlying studies. - Related terms: Cochrane Collaboration (source of high-quality systematic reviews), PRISMA (reporting guideline for systematic reviews/meta-analyses), intention-to-treat analysis, NNT/NNH, relative risk, odds ratio. - Typical USMLE question stem: "A researcher combines data from 12 randomized trials to estimate the overall effect of a drug on mortality" → meta-analysis. Or: "Which tool is used to assess publication bias?" → funnel plot.