systematic review

**Definition:** A systematic review (SR) answers a specific, focused question (often framed as PICO: Population, Intervention, Comparison, Outcome) by comprehensively identifying, selecting, appraising, and synthesizing all eligible studies using explicit, reproducible methods.

**Key steps:** 1. Formulate a focused question (PICO) and register a protocol a priori (e.g., PROSPERO). 2. Comprehensive, multi-database search (MEDLINE/PubMed, Embase, Cochrane CENTRAL), plus grey literature and reference lists. 3. Predefined inclusion/exclusion criteria; screening by at least two independent reviewers. 4. Data extraction and risk-of-bias assessment (e.g., Cochrane RoB tool for RCTs; Newcastle-Ottawa Scale for observational studies). 5. Synthesis: qualitative (narrative) and, if studies are sufficiently homogeneous, quantitative (meta-analysis). 6. Assess certainty of evidence (GRADE) and report per PRISMA guidelines (flow diagram of study selection).

**Systematic review vs. meta-analysis:** - Systematic review = the overall methodology of the search and appraisal. - Meta-analysis = the statistical technique of pooling effect estimates (e.g., weighted pooled OR/RR). It is a component of many, but not all, SRs. - Meta-analysis cannot fix poor-quality input data ("garbage in, garbage out").

**Systematic review vs. narrative (traditional) review:** Narrative reviews have no defined search strategy or selection criteria, are prone to selection and author bias, and are not reproducible. SRs are explicit and reproducible.

**Heterogeneity:** - Clinical/methodological differences between studies (populations, interventions, outcomes). - Statistical heterogeneity assessed by Cochran's Q test and I² statistic (higher I² = more inconsistency; roughly >50% is considered substantial). - Fixed-effects model assumes one true effect size; random-effects model assumes the true effect varies across studies (more conservative, wider confidence intervals).

**Forest plot:** Displays each study's point estimate and confidence interval (squares and horizontal lines) plus the pooled estimate (diamond). A line of no effect (RR/OR = 1; mean difference = 0) is shown; a CI crossing it is not statistically significant.

**Funnel plot and publication bias:** A funnel plot graphs effect size against study precision. Asymmetry suggests publication bias (small negative studies are less likely to be published) or other small-study effects. Egger's test is a statistical test for asymmetry.

**Biases and limitations:** - Publication bias, selection bias, and language bias. - Heterogeneity among included studies. - Limited by the quality of primary studies. - Outdated if not updated regularly.

**Evidence hierarchy (high to low):** Systematic reviews/meta-analyses of RCTs > individual RCTs > cohort studies > case-control studies > cross-sectional studies > case series/reports > expert opinion/animal and bench research. Note that the top tier is a synthesis of studies, not a primary study design.

**Board-relevant pearls:** - The Cochrane Collaboration is the best-known producer of SRs. - PRISMA is the reporting guideline for SRs and meta-analyses; CONSORT is for RCTs; STROBE for observational studies. - SRs and meta-analyses increase statistical power and precision and can resolve conflicting study results, but they are not immune to bias. - Clinical practice guidelines are often built on SRs.

**Commonly tested:** Identifying SR/meta-analysis as the highest level of evidence, interpreting a forest plot or funnel plot, recognizing publication bias, and differentiating an SR from a narrative review.