Evidara · Meta-Analysis — Statistical Evidence Synthesis

Forest plots from your
evidence base. In minutes.

Evidara extracts HR, OR, RR, and MD effect sizes from study texts, pools them using fixed-effects and DerSimonian-Laird random-effects methods, and generates publication-quality forest plot SVGs — automatically, on every Full SLR run.

Request a Demo → See Full SLR

HR/OR/RR

Hazard ratio, odds ratio, risk ratio, MD and SMD — all supported with 95% CI

I²+τ²

Cochran's Q, I², and τ² calculated — heterogeneity fully reported on every run

SVG→PDF

Forest plot exported as vector SVG — publication-quality at any resolution

GRADE linked

Meta-analysis outputs feed GRADE certainty assessment per outcome automatically

The meta-analysis bottleneck

Three statistical evidence problems
that slow every HTA submission.

Meta-analysis is a required component of most HTA submissions and systematic reviews — but it sits between the evidence gathering and the submission, requiring specialist statisticians and software that most HEOR teams don't have on standby.

01 — Effect Size Extraction

Extracting consistent effect sizes from heterogeneous study reports

Primary studies report effect sizes in inconsistent formats — some as HR (0.63; 95% CI 0.48–0.82), others as "hazard ratio of 0.63 with a confidence interval from 0.48 to 0.82", others as tables without explicit HR labelling. Manual extraction from 20–50 studies requires days of careful review and data entry.

~2–3 days per meta-analysis data extraction

02 — Statistical Synthesis

Running meta-analyses requires specialist software and statisticians

Fixed-effects and random-effects pooling, Cochran's Q heterogeneity testing, and τ² calculation typically require R packages (metafor, meta) or commercial software (RevMan). HEOR teams without an in-house biostatistician wait weeks for an analysis that should take hours.

~1–2 week turnaround for external statisticians

03 — Publication-Quality Output

Forest plots for HTA submissions need to meet publication standards

HTA bodies and journal peer reviewers require forest plots that are publication-quality — correct diamond sizes for pooled estimates, confidence interval whiskers, I² and τ² annotations, and proper labelling of study weights. Generating these in Excel or PowerPoint is not acceptable for NICE submissions.

Rework loops on forest plot formatting

How Evidara solves it

Automated statistical synthesis
built into every SLR run.

Effect Size Extraction

Automatic parsing from free-text results

Evidara's meta-analysis engine parses HR, OR, RR, hazard ratio, odds ratio, risk ratio, MD, SMD, WMD, and mean difference from primary result strings in free text — handling varied formatting across journals, including bracket-enclosed CIs, em-dash separators, "to" notation, and plain-text descriptions. No manual data entry required.

HR, OR, RR, MD/SMD/WMD extraction — all formats handled

95% CI extracted alongside point estimate

Extraction confidence flagged — uncertain parses disclosed

Every extraction links to source PMID and result text

Statistical Pooling

Fixed-effects and random-effects meta-analysis

Two pooling methods are run in parallel on every meta-analysis: inverse-variance fixed-effects pooling and DerSimonian-Laird random-effects pooling (log-transformed for ratio measures, linear for MD/SMD). Cochran's Q, I², and τ² are calculated and reported. Heterogeneity is never hidden.

Inverse-variance fixed-effects pooling

DerSimonian-Laird random-effects with τ² estimation

Cochran's Q statistic and p-value

I² and τ² reported with interpretation (low/moderate/high)

Forest Plot SVG Export

Publication-quality forest plots in one click

Every meta-analysis produces a publication-quality forest plot as an inline SVG — with correct diamond sizing for pooled estimates, confidence interval whiskers proportional to weight, study labels, I² annotation, and a vertical line of no effect. Exported to PDF within the SLR report for direct HTA submission inclusion.

Vector SVG — scales to any resolution without pixelation

Pooled diamond with fixed + random estimates both shown

Study weights by inverse variance — shown as box size

Embedded in PDF SLR report for HTA submission

Real result · computed by the pooling engine

GLP-1 MACE reduction — 6 cardiovascular outcome trials pooled

Six published GLP-1 receptor agonist cardiovascular outcome trials, their primary 3-point MACE hazard ratios taken from the source abstracts (PMIDs below), pooled through Evidara's meta-analysis engine. The estimates, heterogeneity statistics and forest plot are the engine's actual output — not an illustration. Two further CVOTs (HARMONY, AMPLITUDE-O) are not included in this set.

Run a Full SLR with meta-analysis →

6 GLP-1 RA cardiovascular outcome trials · primary 3-point MACE · inverse-variance pooling

SUSTAIN-6 HR 0.74 (0.58–0.95) · PMID 27633186 PIONEER-6 0.79 (0.57–1.11) · 31185157

LEADER 0.87 (0.78–0.97) · 27295427 REWIND 0.88 (0.79–0.99) · 31189511

EXSCEL 0.91 (0.83–1.00) · 28910237 ELIXA 1.02 (0.89–1.17) · 26630143

Fixed-effects: HR 0.897 (0.851–0.946) · Random-effects (DerSimonian–Laird): HR 0.895 (0.838–0.955)

Heterogeneity: Cochran's Q 6.79 (df 5, p = 0.24) · I² 26.4% (low–moderate) · τ² 0.0017 — fixed and random estimates agree

Meta-Analysis Module · effect sizes from published abstracts, pooling computed by the engine · PROVISIONAL — human statistician review required before submission · live literature updates, so the trial set is a point-in-time selection

Forest plots from yourevidence base. In minutes.

Three statistical evidence problemsthat slow every HTA submission.

Automated statistical synthesisbuilt into every SLR run.