Evidence Synthesis

Evidence Synthesis implements meta-analytic methods for combining effect estimates across multiple databases or study sites. When the same analysis (e.g., a cohort method estimation or SCCS) is run against several CDM data sources, evidence synthesis pools the site-level results into a single summary estimate with appropriate uncertainty quantification and heterogeneity assessment.

Parthenon supports both Bayesian random-effects and fixed-effect (inverse variance) meta-analysis, aligned with the OHDSI EvidenceSynthesis R package. This is a critical step in the OHDSI evidence generation pipeline, as regulatory submissions and clinical guideline development increasingly require evidence from multiple independent data sources rather than single-site studies.

Module Architecture

EvidenceSynthesisDetailPage
 |
 |-- Header (name, description, back navigation)
 |-- Execute button + Delete button
 |
 |-- Tab: Design
 |    |-- EvidenceSynthesisDesigner
 |         |-- Basic Information (name, description)
 |         |-- Meta-Analysis Method selector
 |         |    |-- Bayesian: chain length, burn-in, sub-sample
 |         |    |-- Fixed: inverse variance weighting
 |         |-- Site Estimates table
 |         |    |-- Per-site: site name, log(RR), SE(log RR)
 |         |    |-- HR Preview (exp(logRR))
 |         |-- Save / Create button
 |
 |-- Tab: Results
 |    |-- EvidenceSynthesisResults
 |    |    |-- EvidenceSynthesisVerdictDashboard
 |    |    |-- Pooled Summary Cards (HR, 95% CI, method, tau)
 |    |    |-- ForestPlot (graphical per-site + pooled)
 |    |    |-- Per-Site Results table
 |    |    |-- Execution timing
 |    |
 |    |-- Execution History table

Feature Walkthrough

Creating an Evidence Synthesis

Navigate to Analyses and click New Analysis, then select Evidence Synthesis.
Provide a name and optional description.
Select a meta-analysis method.
Enter site-level estimates (minimum two sites required).
Click Create to save.

Meta-Analysis Method

Two methods are available:

Method	Description	When to Use
Bayesian Random-Effects	MCMC-based random-effects model that accounts for between-site heterogeneity	Default. Use when expecting variation across sites due to population differences, coding practices, or data quality.
Fixed-Effect (Inverse Variance)	Assumes a single true effect size; weights sites by inverse variance	Use when sites are highly homogeneous and you believe the true effect is constant.

When Bayesian random-effects is selected, three MCMC parameters become configurable:

Parameter	Description	Default
Chain Length	Total MCMC iterations	1,100,000
Burn-in	Iterations discarded before sampling	100,000
Sub-sample	Thinning interval for posterior samples	100

tip

The default MCMC parameters are suitable for most analyses. Increase chain length and burn-in only if you observe convergence warnings in the results.

Entering Site Estimates

Each site estimate requires three fields:

Field	Description	Example
Site Name	Label identifying the database or study site	"Optum EHR"
Log(RR)	Natural log of the relative risk or hazard ratio	-0.30
SE(Log RR)	Standard error of the log relative risk	0.15

A live HR Preview section below the estimates table displays the exponentiated hazard ratio (exp(logRR)) for each site, allowing quick verification before saving.

You can add additional sites with the Add Site button. A minimum of two sites is required; the save button is disabled until this threshold is met. Sites can be removed when more than two exist.

warning

Ensure that all site estimates use the same outcome definition and time-at-risk settings. Combining estimates from incompatible study designs will produce misleading pooled results.

Executing the Synthesis

Click Execute in the page header. Evidence synthesis does not require selecting a source because it operates on the pre-entered site estimates.
The R runtime runs the EvidenceSynthesis package to compute pooled results.
The page switches to the Results tab upon execution.

Viewing Results

The results tab presents:

Verdict Dashboard -- an interpretive summary of the synthesis outcome.

Pooled Summary Cards -- four metric cards:

Card	Description
Pooled HR	The meta-analytic hazard ratio
95% CI	Confidence interval for the pooled estimate
Method	Bayesian RE or Fixed Effect
Tau (heterogeneity)	Between-site heterogeneity parameter (Bayesian only)

Forest Plot -- a graphical forest plot rendered via the ForestPlot component. Each site is shown as a point estimate with confidence interval, and the pooled diamond appears at the bottom.

Per-Site Results Table -- a detailed table with columns: Site, HR, 95% CI Lower, 95% CI Upper, Log(RR), SE. Hazard ratios above 1.0 are highlighted in red and below 1.0 in teal.

Execution History -- a list of all prior executions. Click a completed row to view its results.

Interpreting Forest Plots

The forest plot is the primary visualization for evidence synthesis results. Key elements to understand:

Individual site estimates are shown as points (squares) with horizontal lines representing 95% confidence intervals. Larger points indicate greater statistical weight.
The pooled estimate appears as a diamond at the bottom of the plot. The diamond width spans the pooled confidence interval.
The vertical reference line at HR = 1.0 represents the null hypothesis (no effect). Estimates entirely to the left suggest a protective effect; estimates to the right suggest increased risk.
Heterogeneity (tau) quantifies between-site variation. A tau value near zero indicates consistent results across sites.

API Reference

Endpoint	Method	Description
`GET /api/v1/analyses/evidence-synthesis/{id}`	GET	Fetch analysis with design and executions
`POST /api/v1/analyses/evidence-synthesis`	POST	Create a new evidence synthesis
`PUT /api/v1/analyses/evidence-synthesis/{id}`	PUT	Update design and metadata
`DELETE /api/v1/analyses/evidence-synthesis/{id}`	DELETE	Delete an evidence synthesis
`POST /api/v1/analyses/evidence-synthesis/{id}/execute`	POST	Run the meta-analysis
`GET /api/v1/analyses/evidence-synthesis/{id}/executions/{execId}`	GET	Fetch execution results

Population-Level Estimation -- source analyses that feed into evidence synthesis
SCCS -- self-controlled case series results can be pooled here
Studies -- multi-database study orchestration

Module Architecture​

Feature Walkthrough​

Creating an Evidence Synthesis​

Meta-Analysis Method​

Entering Site Estimates​

Executing the Synthesis​

Viewing Results​

Interpreting Forest Plots​

API Reference​

Related Pages​