Data Synthesis and Meta-Analysis of Binary Data
by Shemsedin Amme
Published on August 27, 2026
Guides: Data Synthesis & Meta-Analysis
Quantitative Synthesis
AI-Powered Interpretation
90 minutes
Guides Objectives:
By the end of this session, users will be able to:
By the end of this session, users will be able to:
- Navigate the RevProHub Reporting Phase and Data Synthesis Dashboard
- Configure and run meta-analyses (effectiveness, prevalence, continuous outcomes)
- Map extraction form fields to analysis parameters (N, events, effect measures)
- Interpret forest plots, heterogeneity statistics (I², Q-test), and subgroup analyses
- Use meta-regression to explore sources of heterogeneity
- Generate funnel plots and interpret publication bias (Egger's test)
- Export results and add AI-generated interpretations to manuscript
Part 1: Reporting Phase Dashboard
Navigate to: Main Project Dashboard → Reporting Phase → Reporting Phase Dashboard
PRISMA
Flow Diagram
Data Synthesis
Meta-Analysis
Report Generator
Templates
Manuscript Editor
Auto-Injection
📊 PRISMA Chart Generator: Automatic flow diagram from screening data (redirects to Screening → PRISMA). Manual builder also available. Diagram auto-inserts into Manuscript Editor.
📝 Manuscript Writer: Real-time editor with auto-injected content (search query, PRISMA chart, QA results, synthesis outputs). Export as DOCX/PDF for journal submission.
📄 Report Generator: Create comprehensive reports using pre-designed templates or custom builds; export as PDF/DOCX.
Part 2: Data Synthesis Dashboard
Navigate to: Reporting Phase Dashboard → Data Synthesis → Synthesis Dashboard
Start New Analysis
Effectiveness Meta-Analysis
2x2 table data (OR, RR, RD)
2x2 table data (OR, RR, RD)
Prevalence Meta-Analysis
Single-group proportions
Single-group proportions
Continuous Meta-Analysis
Mean, SD, N data
Mean, SD, N data
Meta-Aggregation (Qualitative)
Synthesize themes/findings
Synthesize themes/findings
General Synthesis (Charting)
Scoping/narrative reviews
Scoping/narrative reviews
Descriptive Analysis
Summarize characteristics
Summarize characteristics
Umbrella Synthesis
Multiple reviews synthesis
Multiple reviews synthesis
Past Analysis Runs: Each configured analysis is saved with name, type, date, and actions (view, clone, delete, add to manuscript). Bulk delete available.
Prerequisites for Data Synthesis:
- ✅ Completed data extraction for at least 3-5 studies (Session 3.3 / Extraction Module)
- ✅ Extracted numeric data (N, events, means, SDs) for quantitative synthesis
- ✅ All conflicts resolved (if dual extraction) – only consensus data is used
- ✅ Understanding of effect measures (OR, RR, RD, mean difference)
Part 3: Step-by-Step Configuration Guide (Meta-Analysis of Effectiveness)
Step 1: Basic Information
📝 Why: The analysis name helps identify this specific run in the Past Analyses table. The extraction form determines which fields are available for mapping.
Step 2: Select Studies
📚 Why: Only studies marked as "Included" after full-text screening appear. Select only those with complete data for the outcome of interest.
Step 3: Data Mapping & Settings
⚙️ Why mapping matters: Correct field mapping ensures RevProHub calculates accurate effect sizes. For 2x2 tables, you must map N (total users) and Events (outcome occurrences) for both intervention and control groups.
Step 4: Methodological Options (Handling Data Issues)
Guidance for Novice Users:
For most systematic reviews with normal data (no zero events in any study arm), all available methods yield very similar results. The Standard Inverse-Variance method is the current default and recommended choice.
Only consider alternative methods (Constant Correction, Peto, Log Odd, GLMM) when you have sparse data (studies with zero events in one or both arms, uncommon outcomes, or very small sample sizes). In such cases, consult with a statistician or the AI Assistant for guidance.
For most systematic reviews with normal data (no zero events in any study arm), all available methods yield very similar results. The Standard Inverse-Variance method is the current default and recommended choice.
Only consider alternative methods (Constant Correction, Peto, Log Odd, GLMM) when you have sparse data (studies with zero events in one or both arms, uncommon outcomes, or very small sample sizes). In such cases, consult with a statistician or the AI Assistant for guidance.
⚠️
When to change from default: The Standard Inverse-Variance method automatically excludes studies with zero events from the analysis. If your review includes several such studies and you want to retain them, consider switching to Constant Correction (adds 0.5 to all cells) or GLMM (handles zeros via likelihood estimation). For typical systematic reviews with event rates >5%, the default method works perfectly.
Step 5: Advanced Options
📊 Subgroup & Meta-Regression: Subgroup analysis compares effect sizes across categorical variables (e.g., intervention type). Meta-regression explores continuous moderators (e.g., publication year) to explain heterogeneity.
Part 4: Results Page – Complete Breakdown
4.1 Forest Plot
| Study / Group | Int (n/N) | Ctrl (n/N) | Weight (%) | OR [95% CI] |
|---|---|---|---|---|
| intervention_desc: Semaglutide | ||||
| Lingvay L (2020) | 22/30 | 15/30 | 2.3 | 2.75 [0.93, 8.10] |
| Wilding JPH (2021) | 850/1306 | 340/655 | 9.0 | 1.73 [1.43, 2.09] |
| Davis D (2022) | 1400/2000 | 600/1000 | 9.2 | 1.56 [1.33, 1.82] |
| Subgroup Summary | 6447/10354 | 4400/7940 | 100.0 | 1.38 [1.14, 1.67] |
| intervention_desc: Liraglutide | ||||
| Marso SP (2016) | 1500/2487 | 800/1244 | 9.3 | 0.84 [0.73, 0.97] |
| Kumar K (2017) | 1091/4668 | 1234/4672 | 9.6 | 0.85 [0.77, 0.93] |
| Subgroup Summary | 7451/18222 | 5039/14928 | 100.0 | 1.37 [1.15, 1.63] |
| intervention_desc: Tirzepatide | ||||
| Jastreboff AM (2022) | 450/611 | 200/305 | 18.4 | 1.47 [1.09, 1.97] |
| Garcia G (2023) | 1100/1500 | 1000/1500 | 40.5 | 1.38 [1.18, 1.61] |
| Subgroup Summary | 2410/3226 | 1750/2620 | 100.0 | 1.52 [1.31, 1.76] |
| Total (random) | 21896/42485 | 15764/34744 | 100.0 | 1.42 [1.27, 1.58] |
Forest Plot Interpretation:
- Each row shows a study's effect size (OR) and 95% confidence interval.
- Squares represent point estimates; size corresponds to study weight.
- Horizontal lines = confidence intervals (crossing 1.0 = non-significant).
- Diamonds = pooled effect sizes (subgroup and overall).
- The overall diamond at the bottom shows the summary OR = 1.42 [1.27, 1.58].
- Each row shows a study's effect size (OR) and 95% confidence interval.
- Squares represent point estimates; size corresponds to study weight.
- Horizontal lines = confidence intervals (crossing 1.0 = non-significant).
- Diamonds = pooled effect sizes (subgroup and overall).
- The overall diamond at the bottom shows the summary OR = 1.42 [1.27, 1.58].
4.2 Model Summary & Heterogeneity
Model Type: Random-Effects
Heterogeneity (I²): 86.6%
Q-statistic: 327.99 (df=44), p < 0.001
τ² (between-study variance): 0.0952
📖 Interpretation: I² = 86.6% indicates substantial heterogeneity (≥75% = considerable). The Q-test p < 0.001 rejects homogeneity, confirming that study differences are not due to chance alone. τ² = 0.095 represents the variance of true effect sizes across studies.
🔧 Action: High heterogeneity requires subgroup analysis or meta-regression to identify moderators. A random-effects model (already selected) is appropriate as it incorporates between-study variance.
4.3 Subgroup Analysis Table
| Subgroup | k (studies) | OR [95% CI] | I² (%) |
|---|---|---|---|
| Semaglutide | 14 | 1.38 [1.14, 1.67] | 83.8 |
| Liraglutide | 16 | 1.37 [1.15, 1.63] | 84.2 |
| Tirzepatide | 6 | 1.52 [1.31, 1.76] | 21.7 |
| Co-agonist X | 1 | 0.83 [0.77, 0.90] | 0.0 |
📊 Test for subgroup differences: Q = 136.32, df = 3, p = 0.000 → significant differences between subgroups. Tirzepatide shows the largest effect (OR=1.52) and lowest heterogeneity (I²=21.7%).
4.4 Meta-Regression Analysis
Moderator: publication_year
Regression Model: y = -97.380 + 0.048 × publication_year
Coefficient (Slope): 0.0483 (p = 0.0327)
R² (Heterogeneity Explained): 10.2%
📈 Interpretation: The positive slope (0.048) indicates that more recent studies report slightly higher ORs. The p-value (0.033) suggests a statistically significant relationship, but the low R² (10.2%) means only 10% of heterogeneity is explained by publication year – other factors remain important.
4.5 Publication Bias – Funnel Plot & Egger's Test
Egger's Test: Intercept = 5.703, t = 6.28, p = 0.000 (df=43)
Interpretation: A significant Egger's test (p < 0.05) suggests funnel plot asymmetry, which may indicate publication bias, small-study effects, or heterogeneity. Examine the funnel plot visually – missing studies in the bottom-left corner would suggest selective reporting of positive results. RevProHub generates interactive funnel plots with trim-and-fill optional adjustment.
Interpretation: A significant Egger's test (p < 0.05) suggests funnel plot asymmetry, which may indicate publication bias, small-study effects, or heterogeneity. Examine the funnel plot visually – missing studies in the bottom-left corner would suggest selective reporting of positive results. RevProHub generates interactive funnel plots with trim-and-fill optional adjustment.
4.6 AI Assistant: Interpretation & Methodological Feedback
Critical Interpretation: The meta-analysis, encompassing 45 studies, demonstrates a statistically significant positive effect (OR 1.419, 95% CI: 1.274–1.582), suggesting that the intervention is associated with higher odds of the desired outcome compared to control. However, the interpretation is heavily caveated by the extremely high I² value of 86.6%, which indicates substantial inconsistency across studies. Given the high heterogeneity, the pooled OR should be interpreted as an average effect rather than a universal one.
Methodological Feedback: An I² of 86.6% indicates significant heterogeneity, which mandates that you move beyond simple reporting of the pooled estimate. First, investigate sources of heterogeneity via subgroup analyses or meta-regression. Second, consider using REML estimator for τ². Finally, perform a sensitivity analysis by excluding outliers.
Zero Cell Handling Advice: The analysis used the "constant" method for the "MA-SZ (Single Zeros)" data pattern. AI recommends Continuity Correction (0.5 adjustment) or Mantel-Haenszel with zero-cell handling.
Methodological Feedback: An I² of 86.6% indicates significant heterogeneity, which mandates that you move beyond simple reporting of the pooled estimate. First, investigate sources of heterogeneity via subgroup analyses or meta-regression. Second, consider using REML estimator for τ². Finally, perform a sensitivity analysis by excluding outliers.
Zero Cell Handling Advice: The analysis used the "constant" method for the "MA-SZ (Single Zeros)" data pattern. AI recommends Continuity Correction (0.5 adjustment) or Mantel-Haenszel with zero-cell handling.
4.7 Clone & Edit Analysis – Sensitivity Analysis
🔁 Clone & Edit Function: Allows you to duplicate an existing analysis, modify parameters (e.g., change model type, zero-cell method, or exclude studies), and compare results. This is essential for sensitivity analyses.
Sensitivity Analysis Comparison: "Sabit Ababor (Clone) with glmm" vs. base analysis "Sabit Ababo"
Pooled Effect (OR): 1.419 → 1.419 (0.0% change)
95% CI: Stable [1.274, 1.582]
Egger's Test p-value: 0.000 → 0.000
📌 When to use: Test robustness by (a) changing the model (fixed vs. random), (b) applying different zero-cell corrections, (c) excluding individual studies (leave-one-out), or (d) restricting to high-quality studies.
4.8 Export Results & Add to Manuscript
Export as PDF
Full report with forest plot, tables, and AI interpretation
Full report with forest plot, tables, and AI interpretation
Export as DOCX
Editable Word document for manuscript integration
Editable Word document for manuscript integration
Export as CSV
Raw effect sizes and weights for external software (R, Stata, SPSS)
Raw effect sizes and weights for external software (R, Stata, SPSS)
Add to Manuscript
Auto-injects forest plot, summary table, and AI interpretation into the Manuscript Editor
Auto-injects forest plot, summary table, and AI interpretation into the Manuscript Editor
Part 5: Other Synthesis Methods
5.1 Descriptive Analysis – Overview & Configuration
What is Descriptive Analysis?
Descriptive analysis summarizes the characteristics of your included studies and extracted data without calculating pooled effect sizes. It generates frequencies, means, medians, and ranges to help you understand the nature of your evidence base.
Descriptive analysis summarizes the characteristics of your included studies and extracted data without calculating pooled effect sizes. It generates frequencies, means, medians, and ranges to help you understand the nature of your evidence base.
Configure: Descriptive Analysis
Generates descriptive statistics (frequencies, means, medians) for study characteristics and extracted data.
Step 1: Basic Information
Step 2: Select Studies
Step 3: Data Mapping & Settings
A. Study Characteristics (Select fields to summarize)
B. Extracted Data Fields
Descriptive Analysis – Results Example
| Study Characteristics | Category | Frequency (n) | Percentage (%) |
|---|---|---|---|
| Source Database | PubMed | 6 | 31.6 |
| Embase | 8 | 42.1 | |
| Cochrane Library | 5 | 26.3 | |
| Study Design | RCT | 14 | 73.7 |
| Cohort | 5 | 26.3 |
| Numeric Field | Count | Mean | Median | SD | Min | Max |
|---|---|---|---|---|---|---|
| Intervention N | 19 | 1092.05 | 350.00 | 1497.42 | 40 | 4860 |
| Follow-up (years) | 19 | 1.47 | 1.00 | 0.51 | 1 | 2 |
📊 When to use Descriptive Analysis: Ideal for scoping reviews, characterizing your evidence base, identifying gaps in the literature, or presenting study characteristics before meta-analysis. Results can be exported as tables for your manuscript.
5.2 General Data Synthesis (Tabulation) – Overview & Configuration
What is General Data Synthesis?
General Data Synthesis creates custom summary tables (charting) for scoping reviews, narrative syntheses, or any review where data is presented in a structured table rather than statistically pooled. You select which extracted fields become columns in your table.
General Data Synthesis creates custom summary tables (charting) for scoping reviews, narrative syntheses, or any review where data is presented in a structured table rather than statistically pooled. You select which extracted fields become columns in your table.
Configure: General Data Synthesis (Tabulation)
For Scoping Reviews, Narrative Syntheses, or any review where data is presented in a summary table.
Step 1: Basic Information
Step 2: Select Studies
Step 3: Data Mapping & Settings
📋 How it works: Each selected field becomes a column in your output table. Each study becomes a row. You can reorder columns, rename headers, and export the table as CSV, DOCX, or add directly to your manuscript. This is ideal for presenting evidence tables in scoping or narrative reviews.
General Data Synthesis – Results Example
| Study | Design | Intervention | Sample Size (I/C) | Key Outcome |
|---|---|---|---|---|
| Wilding JPH (2021) | RCT | Semaglutide 2.4mg | 1306 / 655 | ≥5% weight loss: 86.6% vs 47.6% |
| Jastreboff AM (2022) | RCT | Tirzepatide | 611 / 305 | ≥5% weight loss: 73.6% vs 32.6% |
| Marso SP (2016) | RCT | Liraglutide | 2487 / 1244 | ≥5% weight loss: 60.3% vs 64.3% |
📌 When to use General Data Synthesis: Scoping reviews, systematic reviews with heterogeneous outcomes that cannot be meta-analyzed, narrative syntheses, or any situation where you need a structured evidence table rather than a statistical summary.
Part 6: PRISMA 2020 Flow Diagram Generator
Automatic PRISMA Chart: RevProHub generates a PRISMA 2020 flow diagram using screening stage data:
- Identification: Records from database searches + other sources
- Screening: Duplicates removed, records screened
- Eligibility: Full-text articles assessed, exclusions with reasons
- Included: Studies included in synthesis
Part 7: Manuscript Editor with Auto-Injected Content
📖 Navigate to: Reporting Phase → Manuscript Writer
Real-time Editor Features:
- Auto-injected search strategy (from Search & Query Management)
- Auto-injected PRISMA flow diagram (from Screening)
- Auto-injected Risk of Bias summary (from Quality Assessment Visualization)
- Auto-injected data synthesis results (forest plots, tables, AI interpretation)
- Citation management integration
- Export as DOCX/PDF for journal submission
🧪 Group Hands-On Activity: Data Synthesis (50 min)
- Navigate: Reporting Phase → Data Synthesis Dashboard.
- Configure New Analysis: Click "Start New Analysis" → "Effectiveness Meta-Analysis". Name it "Training MA - Weight Loss".
- Select Form & Studies: Choose your extraction form. Select 5-10 completed studies.
- Map Fields: Map intervention_n, intervention_events, control_n, control_events.
- Set Model: Random-Effects, Odds Ratio. Select zero-cell method (e.g., GLMM).
- Run Analysis: Click "Run Analysis". Review forest plot, I², and overall effect.
- Add Subgroup: Clone the analysis, add subgroup by "intervention_desc". Compare subgroup I² values.
- Meta-Regression: Run analysis with moderator "publication_year". Interpret p-value and R².
- Export: Export results as PDF and DOCX. Click "Add to Manuscript".
- AI Assistant: Review AI interpretation. Copy methodological feedback for your report.
- PRISMA & Manuscript: Generate PRISMA diagram and verify auto-injection into Manuscript Editor.
Debrief Questions:
- How do you interpret an I² of 75% or higher? What actions should you take?
- When would you choose a fixed-effect vs. random-effects model?
- What does a significant Egger's test indicate about your meta-analysis?
- How can subgroup analysis help explain high heterogeneity?
- What is the role of the AI Assistant in improving the transparency and reproducibility of your synthesis?
Reporting Phase Completion Checklist:
✔ Meta-analysis configured and run (effectiveness/prevalence/continuous)
✔ Fields mapped correctly from extraction form
✔ Heterogeneity assessed (I², Q-test)
✔ Subgroup analysis performed (if applicable)
✔ Meta-regression explored (if applicable)
✔ Publication bias assessed (funnel plot + Egger's test)
✔ Results exported and/or added to manuscript
✔ PRISMA diagram generated and verified
✔ AI interpretation reviewed and incorporated
✔ Meta-analysis configured and run (effectiveness/prevalence/continuous)
✔ Fields mapped correctly from extraction form
✔ Heterogeneity assessed (I², Q-test)
✔ Subgroup analysis performed (if applicable)
✔ Meta-regression explored (if applicable)
✔ Publication bias assessed (funnel plot + Egger's test)
✔ Results exported and/or added to manuscript
✔ PRISMA diagram generated and verified
✔ AI interpretation reviewed and incorporated
Next: Final Report Generation and Manuscript Writing.