⚗️ Bivariate Benefit-Harm Meta-Analysis v2.0

Rigorous causal inference for DOACs in atrial fibrillation with proper uncertainty quantification

✓ Corrected REML ✓ Hartung-Knapp-Sidik-Jonkman ✓ τ² Confidence Intervals ✓ Honest Conformal Coverage ✓ Proper μ̂ Correlation
⚙️

Analysis Configuration

Typical AF: 1-5%/year depending on CHA₂DS₂-VASc
Must be between 0.1% and 15%
Typical on warfarin: 2-4%/year
Must be between 0.1% and 20%
−0.30
Cannot estimate with k=4; sensitivity analysis recommended
Clinical Decision Thresholds
Must be between 0% and 5%
Must be between 0% and 10%
Estimation Method
📊

Bivariate Predictive Distribution

ℹ️
Heterogeneity Quantification with τ² Confidence Intervals With k=4 studies, heterogeneity estimates have wide uncertainty. I² and τ² confidence intervals computed via Q-profile method.
⚠️
Leave-One-Out Influence Diagnostics Identifies which trials most influence the pooled estimate and acceptability.
Trial Omitted HR Stroke (95% CI) HR Bleeding (95% CI) Acceptability Δ Acceptability
⚠️
Honest Coverage Statement (k=4) With only 4 studies, the maximum achievable conformal coverage is 80% (4/5), not 95%. The displayed region provides 80% finite-sample valid coverage.
🔧
Correlation Sensitivity Analysis Since ρ cannot be estimated with k=4, we examine robustness across plausible values.
📐
Threshold Sensitivity Analysis How acceptability varies with different clinical decision thresholds.
💡
Expected Value of Sample Information (with Monte Carlo SE) Estimates include uncertainty bounds from simulation. Full meta-analysis re-run for each simulated trial.
Future Trial Design N EVSI (95% CI) P(Decision Change)
🌍
External Validity (Approximate) Covariate shift diagnostics. Formal transportability requires IPD.
Target Population Age SMD CHA₂DS₂ SMD Aggregate SMD Assessment
📚

Included Trials

Trial Drug N Stroke HR (95% CI) Bleeding HR (95% CI) Weight (Stroke) Weight (Bleed)