Last updated on October 16th, 2025 at 01:22 pm
Introduction
Systematic Review versus meta analysis—what is the difference? A systematic review is a rigorous, protocol-driven method to identify, appraise, and synthesize all relevant studies for a focused question, minimizing bias through predefined methods and transparent reporting (PRISMA 2020).
A meta-analysis is a statistical technique—often conducted within a systematic review—that pools comparable quantitative results to produce a single, more precise effect estimate and assess heterogeneity. (PRISMA statement)
This article defines the differences between the two in depth.
Table of Contents
Distinct Definitions
1. The Systematic Review: A Rigorous Framework for Evidence Synthesis
A systematic review is a structured, protocol-driven process used to identify, critically appraise, and synthesize all relevant studies addressing a specific research question (Page et al., 2021; PRISMA 2020).
Unlike narrative reviews, which may rely on author judgment, systematic reviews follow predefined, transparent procedures—from search strategy to study inclusion—to minimize bias and ensure reproducibility. They are guided by international standards such as PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) or Cochrane Handbook recommendations.
Key Features
- Pre-registered protocol: Usually published on registries such as PROSPERO to prevent selective reporting.
- Comprehensive search: Uses structured database queries (e.g., PubMed, Scopus, Cochrane Library) plus grey literature to reduce publication bias.
- Eligibility criteria: Clearly defined inclusion/exclusion rules (e.g., study design, population, intervention, outcome, language).
- Quality appraisal: Employs standardized tools (e.g., Cochrane RoB 2.0, ROBINS-I) to assess risk of bias.
- Synthesis: Summarizes findings narratively or statistically (through meta-analysis, if appropriate).
Example
Consider a systematic review investigating “the effectiveness of mindfulness-based interventions for reducing anxiety in university students.” The authors would:
- Formulate a precise question (using the PICO model—Population, Intervention, Comparator, Outcome).
- Search databases using controlled vocabulary (e.g., “mindfulness,” “anxiety,” “college students”).
- Screen and appraise all relevant studies following transparent inclusion criteria.
- Synthesize the findings—either narratively (if results are diverse) or quantitatively (if data are homogeneous).
The strength of the systematic Review lies in its methodological transparency: another researcher could replicate the same process and arrive at a similar dataset.
2. The Meta-Analysis: A Statistical Extension of the Systematic Review
A meta-analysis is a quantitative technique that may be incorporated into a systematic review when the included studies are sufficiently comparable in design, outcomes, and measures (Borenstein et al., 2009; Higgins et al., 2023).
It statistically combines individual effect sizes from multiple studies to produce a single pooled estimate—often represented through a forest plot. This aggregation enhances statistical power, increases precision, and allows exploration of between-study variation (heterogeneity).
Key Features
- Data extraction: Collects numerical outcomes (e.g., odds ratios, mean differences) from eligible studies.
- Model selection: Uses fixed-effect or random-effects models depending on whether study effects are assumed identical or variable.
- Heterogeneity analysis: Employs statistics such as Q or I² to quantify differences among studies.
- Publication bias assessment: Uses funnel plots or Egger’s test to detect asymmetry in reporting.
- Subgroup or sensitivity analysis: Tests the robustness of findings by removing outliers or analyzing specific study subsets.
Example
Using the same topic—mindfulness-based interventions and anxiety—if 15 trials report comparable quantitative outcomes (e.g., mean reduction in anxiety scores), a meta-analysis can statistically combine them to yield a pooled standardized mean difference (SMD).
This pooled estimate might reveal, for example, an overall effect size of –0.45 (95% CI: –0.62 to –0.28), indicating a moderate, statistically significant reduction in anxiety. The forest plot visually displays each study’s effect and confidence interval alongside the combined result.
While the systematic Review provides context and comprehensiveness, the meta-analysis delivers numerical precision—transforming qualitative evidence into measurable effect sizes.
3. How They Interrelate
Every meta-analysis should be grounded in a systematic review, but not every systematic review includes a meta-analysis (Ahn & Kang, 2018).
A systematic review may remain qualitative when:
- Studies use different designs, populations, or outcome measures.
- Data are missing, inconsistent, or non-comparable.
- The Review aims to explore conceptual frameworks or theoretical diversity rather than estimate effect sizes.
Conversely, a meta-analysis without a preceding systematic review is methodologically unsound—it risks selection bias by pooling non-representative studies.
The relationship can therefore be summarized as:
“A systematic review is the map of the evidence; a meta-analysis is the measurement of that map.”
4. Why the Distinction Matters
Understanding the distinction is vital for scholars, practitioners, and policymakers because:
- Systematic reviews support evidence-based decisions by ensuring all relevant data are considered.
- Meta-analyses quantify uncertainty, enabling confidence in estimated effects.
- Recognizing their complementary roles prevents misinterpretation—especially when readers assume that all systematic reviews must yield pooled results.
For instance, in health sciences, policy guidelines (e.g., WHO, NICE) often rely on systematic reviews with meta-analyses for quantitative recommendations but also value standalone systematic reviews for contextual insights.
5. Illustrative Analogy
A helpful analogy is to think of a systematic review as the architectural blueprint of evidence—it defines what studies exist, how they were built, and how reliable they are.
The meta-analysis is the mathematical model that tests the strength of that structure, quantifying its stability and variance.
Both are essential: the Review ensures completeness and transparency, while the meta-analysis ensures precision and inferential strength.
Systematic Review vs Meta-Analysis — Side-by-Side
| Aspect | Systematic Review | Meta-Analysis |
|---|---|---|
| Primary goal | Identify, appraise, and synthesize all eligible evidence for a focused question. | Statistically pool comparable quantitative results for a precise overall effect. |
| Process | Protocol (e.g., PRISMA-aligned) → comprehensive search → screening → data extraction → risk-of-bias appraisal → synthesis (narrative and/or quantitative) | Determine feasibility → compute effect sizes/variances → choose model (fixed/random) → assess heterogeneity (Q, I²) → sensitivity/subgroup/meta-regression → forest/funnel plots. |
| Synthesis method | Narrative and/or quantitative | Quantitative only |
| Output | Transparent summary of evidence; can be narrative or mixed. | Pooled effect size with CIs, heterogeneity, and bias diagnostics. |
| Relationship | Can stand alone; provides the framework for possible meta-analysis. | Usually embedded within a systematic review (rarely credible alone without systematic identification). |
| When used | Heterogeneous designs/outcomes or when qualitative insights are needed. | When data are sufficiently comparable to justify pooling. |
Sources: PRISMA 2020; Cochrane Handbook; Ahn et al., 2018. (PRISMA statement)
Recommended Reads:
How To Do A Systematic Literature Review: 7 Steps
How To Write A Lit Review For A Research Paper
What is a Systematic Review in Research?
Processes & Best Practices
Both systematic reviews and meta-analyses require rigor, transparency, and reproducibility, but they differ in focus and level of analysis.
A systematic review establishes the methodological foundation—how evidence is identified, appraised, and synthesized—while a meta-analysis statistically integrates comparable quantitative findings from that evidence base.
Below are detailed step-by-step processes and practical considerations for each, drawn from authoritative guidelines such as PRISMA 2020 (Page et al., 2021), the Cochrane Handbook (Higgins et al., 2023), and the Wiley Online Library best practices for quantitative synthesis.
A. Systematic Review — Step-by-Step Framework
A systematic review is not simply a comprehensive literature search; it is a methodologically transparent investigation guided by pre-specified objectives and replicable procedures.
1. Pre-register the Protocol
Before beginning, researchers are strongly encouraged to pre-register their protocol on a registry such as PROSPERO (for health and social sciences) or the Open Science Framework (OSF).
Pre-registration defines the Review’s scope, inclusion criteria, and analytical plan in advance, reducing bias and ensuring accountability.
If substantial deviations occur (e.g., adding a database, adjusting criteria), these should be transparently reported in the final manuscript (Page et al., 2021).
Example: A review on “the effects of gamified learning interventions in higher education” might pre-register its protocol in PROSPERO, clearly stating the objective, databases, inclusion criteria, and outcome measures.
2. Define the Research Question and Eligibility Criteria (PICO/PEO)
The PICO (Population, Intervention, Comparison, Outcome) or PEO (Population, Exposure, Outcome) frameworks provide structure to the research question.
They help determine:
- Who or what is being studied (Population),
- What is being tested or compared (Intervention/Exposure),
- What the comparator or control is, and
- What outcome measures will be synthesized?
Eligibility criteria then operationalize the scope by specifying acceptable study designs, publication years, languages, and contexts.
This step ensures consistency during screening and minimizes subjective inclusion.
3. Conduct a Comprehensive Literature Review
A systematic review requires a multi-academic database and search engine strategy, typically involving sources such as PubMed, Scopus, Web of Science, PsycINFO, and Embase.
Beyond formal databases, grey literature (e.g., dissertations, conference proceedings, policy documents) and citation chasing (tracking references forward and backward) enhance completeness and mitigate publication bias (Higgins et al., 2023).
Researchers should document the exact search strings, Boolean operators, and date limits used for each database. This level of transparency allows replication and demonstrates methodological rigor.
Example: The PRISMA 2020 flow diagram records the number of sources retrieved, screened, excluded, and included at each stage—a hallmark of systematic reporting.
4. Apply Dual Screening and Data Extraction
To reduce human bias, two independent reviewers should perform both title/abstract screening and full-text inclusion decisions.
Discrepancies are resolved through consensus or arbitration by a third reviewer.
Similarly, data extraction should be conducted independently using pre-tested extraction forms that capture key study characteristics (e.g., author, year, design, sample size, outcomes).
This duplication not only enhances reliability but also provides an audit trail if decisions are questioned later.
5. Assess Risk of Bias (Quality Appraisal)
All included studies must undergo quality assessment using standardized tools:
- Cochrane Risk of Bias (RoB 2.0) for randomized trials,
- ROBINS-I for non-randomized designs, or
- JBI Critical Appraisal Tools for qualitative and mixed methods.
The goal is not to exclude all imperfect studies, but to evaluate the confidence in cumulative evidence.
Researchers often present biased summaries visually (traffic-light or bar plots).
Example: A high risk of bias in several small studies might explain heterogeneity or inform a sensitivity analysis later.
6. Synthesize the Evidence (Narrative ± Quantitative)
After appraisal, results are synthesized—either narratively (qualitative integration) or quantitatively (via meta-analysis).
If data are too diverse to combine numerically, reviewers use narrative synthesis, grouping results by population, intervention type, or outcome domain.
Where quantitative pooling is possible, studies progress to the meta-analysis stage.
Either way, transparency in grouping logic and synthesis reasoning is essential (Higgins et al., 2023).
7. Report Using PRISMA 2020
The PRISMA 2020 statement provides a standardized checklist and flow diagram for reporting.
It ensures clarity regarding search procedures, study selection, and synthesis decisions.
Following PRISMA promotes transparency and comparability across systematic reviews and is now a requirement for most high-impact journals.
Example: The flow diagram typically includes counts for records identified, screened, excluded, and analyzed—each supported by reasons for exclusion.
B. Meta-Analysis — Statistical Integration Process
Once the systematic Review identifies and appraises eligible studies, the meta-analysis aggregates their quantitative data into a single pooled estimate.
This approach increases precision, allows for exploration of variation (heterogeneity), and can reveal overall patterns that individual studies are too small to detect (Borenstein et al., 2009; Higgins et al., 2023).
1. Confirm Appropriateness
Before conducting a meta-analysis, researchers must verify that studies are sufficiently homogeneous in design, population, and outcome measurement.
Pooling incompatible data risks producing meaningless averages.
Thus, this stage involves conceptual and statistical evaluation of comparability.
Example: Studies measuring “depression improvement” with entirely different scales (e.g., PHQ-9 vs. Beck Inventory) may need standardized conversion before pooling.
2. Select the Effect Metric and Compute Variances
Researchers choose an effect size metric appropriate to their data type:
- Odds ratio (OR) or risk ratio (RR) for dichotomous outcomes,
- Mean difference (MD) or standardized mean difference (SMD) for continuous outcomes.
Each study’s effect estimate and variance are computed, allowing weights to be assigned based on sample size and precision (Borenstein et al., 2009).
3. Choose the Model (Fixed vs. Random Effects)
Two statistical models dominate meta-analysis:
- Fixed-effects model: Assumes a single “true” effect across all studies (variation due only to chance).
- Random-effects model: Assumes actual effects differ among studies (accounts for between-study heterogeneity).
The random-effects model is generally preferred for social and health sciences, where study contexts often vary (Higgins et al., 2023).
Example: A meta-analysis of interventions across multiple countries likely employs a random-effects model to account for contextual variability.
4. Assess Heterogeneity and Explore Moderators
Heterogeneity—the degree of variation among study results—is quantified using:
- Q statistic (significance test for heterogeneity) and
- I² statistic (percentage of total variation due to heterogeneity).
If heterogeneity is substantial (e.g., I² > 50%), subgroup analyses or meta-regression can identify moderators such as population type, intervention duration, or methodological quality (Higgins et al., 2023).
Example: A high I² in a review of online learning interventions may be explained by age differences among participants or differences in digital platform types.
5. Evaluate Publication Bias
Publication bias arises when studies with significant findings are more likely to be published.
To detect it, analysts use funnel plots (scatterplots of effect size vs. precision) and statistical tests such as Egger’s regression.
Asymmetry in the funnel plot suggests the presence of missing small or negative studies, prompting sensitivity or trim-and-fill analyses.
6. Present Results with Visual and Sensitivity Analyses
The final step involves visually summarizing findings using forest plots, where each line represents a study’s effect and confidence interval, and the pooled effect appears as a diamond.
Funnel plots assess bias visually, while sensitivity analyses (e.g., excluding high-risk studies) test robustness.
Comprehensive reporting includes effect estimates, heterogeneity indices, confidence intervals, and bias assessments.
Example: A forest plot showing a pooled OR = 0.75 (95% CI: 0.62–0.90) indicates a consistent reduction in risk, with moderate heterogeneity (I² = 45%).
When to Use Which?
- Choose a systematic review if evidence is conceptually diverse, outcomes differ, or your aim is a comprehensive, unbiased narrative of what is known and where gaps remain. (PRISMA statement)
- Add a meta-analysis when studies report sufficiently similar quantitative data, making pooling meaningful and assumptions defensible, which yields a more precise overall effect estimate. (Cochrane)
Strengths & Limitations
Both systematic reviews and meta-analyses sit at the top of the evidence hierarchy because of their methodological rigor and integrative nature. Yet, each comes with unique strengths and limitations. Understanding these distinctions allows researchers to apply each method judiciously and to interpret results appropriately.
A. Strengths of Systematic Reviews
1. Strong Bias Control and Transparency
Systematic reviews are designed to minimize bias through predefined protocols, comprehensive search strategies, and transparent reporting (Page et al., 2021; PRISMA 2020).
Every step—from question formulation to inclusion/exclusion criteria—is documented, allowing readers to trace decisions.
This transparency ensures reproducibility, making systematic reviews the gold standard for evidence synthesis in clinical, social, and educational research.
Example: A systematic review on “the effects of digital learning on academic performance,” which documents its PROSPERO registration, inclusion criteria, and screening process, gives readers confidence that its findings are not selectively reported.
2. Comprehensive and Structured Evidence Base
Systematic reviews capture the breadth of existing research, including both published and unpublished (“grey”) literature, reducing publication bias (Higgins et al., 2023).
Because they integrate multiple databases and search strategies, they provide a holistic understanding of a topic—identifying consistencies, contradictions, and knowledge gaps.
Implication: Policymakers or educators relying on such reviews can make evidence-informed decisions based on the totality of available research rather than cherry-picked results.
3. Quality Appraisal and Weight of Evidence
Unlike narrative or scoping reviews, systematic reviews critically assess methodological quality using validated instruments (e.g., RoB 2.0, ROBINS-I).
This process allows the reviewer to distinguish between high-confidence evidence and studies that carry a significant risk of bias, giving readers a more nuanced view of reliability.
Example: A systematic review on mindfulness-based therapy for anxiety might find that while many studies show positive results, the highest-quality trials demonstrate only moderate effects—an insight lost in less rigorous reviews.
4. Adaptability to Both Qualitative and Quantitative Data
Although systematic reviews often underpin meta-analyses, they can also stand alone as qualitative syntheses when data cannot be pooled (Grant & Booth, 2009).
This makes them applicable across diverse fields—education, psychology, and public policy—where conceptual frameworks are as important as numerical estimates.
Note: A systematic review summarizing the conceptual evolution of inclusive education may remain purely narrative but still achieve high credibility through transparent synthesis.
Limitations of Systematic Reviews
1. Time-Intensive and Resource-Heavy
Because systematic reviews demand extensive database searches, dual screening, and detailed documentation, they are labor-intensive and slow (Page et al., 2021).
High-quality reviews often take six months to two years to complete, making them less suitable for rapidly evolving fields such as artificial intelligence or COVID-19 research.
2. Limited by Available Data
A systematic review’s conclusions are only as firm as the studies it includes.
If the existing evidence base is small, outdated, or methodologically weak, the Review’s synthesis will reflect these limitations (Higgins et al., 2023).
Example: A systematic review on virtual-reality-based therapy for stroke rehabilitation may find few randomized trials, limiting its ability to reach definitive conclusions.
3. Potentially Qualitative Outcomes
When studies are too heterogeneous to pool, the Review must rely on narrative synthesis, which—while valuable—provides no quantitative effect estimate.
Such qualitative summaries may be viewed as less authoritative in evidence-based disciplines.
Example: If outcome measures vary (e.g., “student motivation” assessed by self-reports, grades, or attendance), a quantitative meta-analysis is impossible, leaving only descriptive comparison.
B. Strengths of Meta-Analysis
1. Enhanced Precision and Statistical Power
By statistically pooling data from multiple studies, meta-analysis produces a more precise effect estimate than any single study alone (Borenstein et al., 2009).
This increased statistical power allows detection of small but consistent effects that might be missed in individual trials.
Example: A single small study might show no significant improvement from a teaching method, but a meta-analysis of ten similar studies could reveal a meaningful pooled benefit (e.g., SMD = 0.35, p < 0.01).
2. Objective Quantification of Evidence
Meta-analysis converts qualitative findings into quantitative estimates, helping policymakers and practitioners make data-driven decisions.
Effect sizes (e.g., standardized mean differences or odds ratios) provide standardized units of comparison across diverse studies, allowing clear interpretation of magnitude and direction.
Implication: A pooled effect size of –0.45 on anxiety reduction provides a tangible, interpretable estimate for clinicians and educators alike.
3. Explicit Assessment of Heterogeneity
Unlike qualitative reviews, meta-analysis can formally measure variability between study results using statistical tools such as Q and I² (Higgins et al., 2023).
This helps identify whether observed differences stem from random variation or from actual contextual differences in study populations or methods.
Example: A meta-analysis of exercise interventions might find substantial heterogeneity (I² = 70%) due to differences in program duration or participant demographics, prompting subgroup analysis.
4. Ability to Explore Moderators and Sources of Variation
Through meta-regression and subgroup analyses, researchers can examine how moderators—such as age, gender, intervention type, or methodological quality—affect outcomes.
This level of statistical granularity provides theoretical insight beyond average effects.
Example: A meta-analysis might show that mindfulness interventions are more effective among graduate students than undergraduates, offering direction for targeted implementation.
Limitations of Meta-Analysis
1. “Garbage In, Garbage Out” Problem
A meta-analysis is only as reliable as the data it aggregates.
Pooling results from methodologically flawed or heterogeneous studies can lead to misleading conclusions (Higgins et al., 2023).
High statistical precision does not compensate for poor study quality.
Example: Combining low-quality observational studies on diet and mental health could yield a spurious pooled association if confounding variables were not controlled.
2. Risk of Over-generalization
While meta-analysis aims to produce an overall estimate, it may obscure critical contextual nuances among studies.
If underlying populations, interventions, or outcomes differ substantially, the pooled result might lack real-world applicability.
Example: A pooled effect across global COVID-19 vaccination studies may not reflect regional differences in vaccine type or public health infrastructure.
3. Sensitivity to Bias and Publication Effects
Meta-analyses are susceptible to publication bias—the tendency for journals to publish studies with significant results.
If unpublished null studies are missing, pooled effects may be inflated.
Although funnel plots and Egger’s tests help detect bias, they cannot eliminate it (Borenstein et al., 2009).
Example: A meta-analysis on cognitive-behavioral therapy (CBT) might overestimate efficacy if small negative trials remain unpublished.
4. Statistical and Interpretive Complexity
Conducting and interpreting a meta-analysis requires advanced statistical skills.
Misapplication of models, incorrect weighting, or failure to account for heterogeneity can distort results.
Moreover, readers unfamiliar with statistical concepts may misinterpret findings, assuming pooled results imply uniform effectiveness across contexts.
C. Comparative Summary: Strengths and Limitations
| Aspect | Systematic Review | Meta-Analysis |
|---|---|---|
| Primary Strength | Minimizes bias through structured, transparent protocol (PRISMA). | Increases precision and power through data pooling. |
| Analytical Approach | Qualitative and/or quantitative synthesis. | Quantitative statistical aggregation. |
| Bias Control | Comprehensive search and appraisal reduce selection bias. | Publication bias tests (e.g., funnel plots) increase reliability. |
| Main Limitation | Time-intensive; may remain qualitative if the data is too diverse. | Quality-dependent; heterogeneity can distort findings. |
| Interpretive Risk | Subject to reviewer judgment in narrative synthesis. | Risk of “garbage in, garbage out” if poor studies are pooled. |
| Practical Output | Comprehensive evidence map. | Pooled effect estimate with precision metrics. |
| Scholarly Value | Strong foundation for evidence-based reviews. | Provides actionable, measurable outcomes for policy and practice. |
FAQs
Can you do a meta-analysis without a systematic review?
Best practice is no. Meta-analysis relies on a comprehensive, unbiased study set; without systematic identification, pooled results risk selection bias and are less credible. (Cochrane).
Do all systematic reviews include a meta-analysis?
No. If studies differ substantially in design, outcomes, or metrics, reviewers should present a narrative synthesis instead of forcing an invalid pooled estimate. (Cochrane)
What is PRISMA and why is it important?
PRISMA 2020 is a reporting guideline (checklists + flow diagrams) that improves the transparency and completeness of systematic reviews and meta-analyses. Many journals expect PRISMA-compliant reporting. (PRISMA statement)
Which model should I use—fixed or random effects?
Use fixed effects when studies are estimating a common true effect; random effects when true effects plausibly vary across studies. Check heterogeneity (Q, I²) to inform the choice. (meta-analysis.com)
Conclusion
A systematic review is the scaffold of trustworthy evidence synthesis: it plans and documents how studies are found, screened, appraised, and synthesized, minimizing bias with transparent methods (PRISMA). A meta-analysis is the quantitative engine that can be attached to that scaffold when studies are sufficiently comparable: it converts multiple estimates into a pooled effect, quantifies uncertainty, and examines heterogeneity (Q, I²) and potential biases (e.g., publication bias).
Not every systematic Review should force a meta-analysis; when designs, outcomes, or contexts diverge too far, a rigorous narrative synthesis is more defensible. Conversely, when assumptions are met, a well-executed meta-analysis increases precision and explanatory power via moderator analyses and sensitivity checks.
In practice, begin with a well-specified systematic review protocol, commit to comprehensive search and duplicate screening, adopt validated risk-of-bias tools, and then decide—based on comparability and statistical diagnostics—whether a meta-analysis adds valid insight. Used together and reported transparently, they deliver both breadth and precision, helping scholars, clinicians, and policymakers make sound decisions. (PRISMA statement)
Key Takeaways
- Systematic review = rigorous, transparent synthesis framework; narrative and/or quantitative.
- Meta-analysis = statistical pooling within a systematic review when data are comparable.
- Use heterogeneity tests (Q, I²) and risk-of-bias tools to decide on pooling.
- Report with PRISMA 2020 for credibility and completeness.
- Combine both judiciously to gain breadth + precision.
References (APA)
- Ahn, E., & Kang, H. (2018). Introduction to systematic Review and meta-analysis. Korean Journal of Anesthesiology, 71(2), 103–112. (PMC)
- Borenstein, M., Hedges, L. V., Higgins, J. P. T., & Rothstein, H. R. (2009). Introduction to Meta-Analysis. Wiley. (Wiley Online Library)
- Borenstein, M. (2010). A basic introduction to fixed-effect and random-effects models in meta-analysis. (White paper). (meta-analysis.com)
- Cochrane Handbook (v6, current chapters). Chapter 10: Analyzing data and undertaking meta-analyses. Cochrane. (Cochrane)
- Page, M. J., et al. (2021). The PRISMA 2020 statement: An updated guideline for reporting systematic reviews. BMJ, 372, n71. (BMJ)
- PRISMA. (2021–2025). PRISMA 2020 statement & resources. PRISMA website / EQUATOR Network. (PRISMA statement)
- Thorlund, K., et al. (2012). Evolution of heterogeneity (I²) estimates and their 95% confidence intervals. BMC Medical Research Methodology, 12, 61. (PMC)
