Analysis features: complete breakdown

Analysis starts with project configuration

The Analyze page adapts to the project's analysis configuration. Pairwise, NMA, and single-group experiences are selected from project settings and available extracted outcomes.

• Confirm outcome names before analysis.
• Confirm arm/treatment labels are consistent.
• Confirm effect direction and unit conventions.
• Validate extracted data before final analysis.
• Save analysis runs when results should be reused, exported, or audited.

Basic pairwise meta-analysis

Basic pairwise analysis is available without premium access. It covers the core comparative workflow for eligible extracted data.

• Continuous outcomes: mean, SD, N, and compatible effect measures such as MD/SMD depending on configuration.
• Binary outcomes: events and totals with odds ratio, risk ratio, or risk difference style measures where configured.
• Forest plot preview and rerendering.
• Primary model selection and common/random effect display settings.
• Continuity-correction controls for sparse binary data.
• Study sorting and direction controls.

Statistical environment

Supported analysis workflows run against an R 4.3.3 statistical-service environment. The environment is documented in a public Zenodo methods record for citation and reproducibility.

• Pairwise and general meta-analysis: meta 8.5-0 and metafor 5.0-1.
• Network meta-analysis: netmeta 3.5-0 and gemtc 1.1-1 for supported frequentist and Bayesian workflows.
• Specialized workflows: mada 0.5.12 for diagnostic test accuracy, metamedian 1.2.2 for median-reported outcomes, bayesmeta 3.5 and rjags 4-17 for Bayesian methods, and brglm2 1.1.0 for selected model support.
• Plotting and output support: ggplot2 4.0.3, robvis 0.3.1, scales 1.4.0, svglite 2.2.2, png 0.1-9, jsonlite 2.0.0, dplyr 1.2.1, base64enc 0.1-6, and plumber 1.3.3.
• Citation: Masood, M., & Masood, S. (2026). Coventra v1.0.0: Statistical Methods and R Environment Manifest (1.0.0). Zenodo. https://doi.org/10.5281/zenodo.20691893

Saved analysis runs

Saved runs preserve the analysis type, outcome, relevant settings, status, and result. They make it possible to reopen prior outputs instead of relying only on the current screen state.

• Statuses include queued, running, succeeded, and failed.
• Saved analysis types include pairwise, subgroup, leave-one-out, cumulative, funnel, diagnostic, meta-regression, network, proportion, rate, mean, survival HR, and DTA where supported.
• The saved record identifies the analysis family and settings needed to review the result.

Influence diagnostics

Leave-one-out influence diagnostics rerun the model while omitting each study in turn. This helps identify whether one study strongly changes the pooled estimate or heterogeneity.

• Useful when results appear unstable.
• Useful when one study has a very large sample or extreme effect.
• Available from the analysis workflow and saved as a leave-one-out run.

Cumulative meta-analysis

Cumulative analysis adds studies in an ordered sequence, commonly by year or another protocol-relevant order, to show how the pooled estimate changes as evidence accumulates.

Publication-bias and funnel views

Funnel and publication-bias views help inspect small-study effects and asymmetry where the data are appropriate.

• Funnel plot outputs are available with premium publication-bias tools.
• A funnel plot is not meaningful for every outcome, especially with very few studies.
• Use visual inspection together with methodological judgment.

Diagnostic plots

Diagnostic plots are premium tools intended to help assess model fit, outliers, and analysis assumptions where supported by the selected analysis family.

Meta-regression

Meta-regression is premium-gated and is used to explore whether study-level covariates explain variation in effects.

• Requires appropriate covariate data.
• Should be pre-specified where possible.
• Can be underpowered with few studies.
• Exploratory findings should be reported carefully.

Advanced single-arm and special analyses

• Proportion meta-analysis: for single-arm event proportions, requiring studies with events and totals.
• Rate analysis: for incidence-rate style synthesis where event and denominator/time data are appropriate.
• Mean/single-group support: represented in saved analysis kinds where configured.
• Survival HR analysis: premium analysis for hazard ratios with confidence intervals.
• Diagnostic test accuracy: premium bivariate DTA analysis requiring TP, FP, TN, and FN values from at least two studies.

Network meta-analysis

Frequentist NMA is available to eligible premium projects configured for network analysis. The workflow covers session setup, validation, analysis, and result review.

• Create sessions from extracted network-ready outcomes.
• Support arm-level and contrast-level data modes where available.
• Validate treatment network shape and connectivity before running.
• Inspect network graph, forest outputs, league table, rankings, inconsistency, small-study or incoherence plots, and available reporting checks.
• Network diagnostics can include netsplit behavior where the network structure supports it; star networks and networks without closed loops may skip inconsistency tests.

Bayesian analyses

Supported Bayesian pairwise and network analyses are premium tools. Availability depends on the active project's analysis configuration.

• Confirm priors and model assumptions before running the analysis.
• Keep the final settings with the saved analysis record.
• Outputs should be reviewed for convergence, assumptions, priors, and model fit before reporting.

GRADE and generated text

Coventra can generate GRADE table rows and human-reviewable GRADE profiles from saved analysis runs where premium access permits it. It can also generate deterministic result sentences, figure captions, and statistical-methods paragraphs from saved settings/results.

NNT, ARR, and adverse-event transforms

For eligible binary effect measures, Coventra can display NNT/ARR-style information. Adverse-event display transforms help present safety outcomes more clearly where the data shape supports them.