RapidMeta Cardiology: A Browser-Based Living Meta-Analysis Platform Validated Against Finerenone Phase III Trial Data

Introduction

We developed and validated RapidMeta Cardiology, a browser-native platform for hosting living meta-analyses, using Finerenone Phase III trial data as the benchmark dataset. Finerenone is a non-steroidal mineralocorticoid receptor antagonist that has been evaluated in three landmark Phase III RCTs: FIDELIO-DKD (n = 5,674), FIGARO-DKD (n = 7,437), and FINEARTS-HF (n = 6,001).^1–3 The resultant evidence base that is comprised of two pre-specified individual patient data (IPD) pooled analyses (FIDELITY, FINE-HEART) and more than 13 independent literature-based meta-analyses, makes this an ideal benchmark for software validation.^4,5

Living meta-analyses are becoming recognised as increasingly essential for rapidly evolving clinical fields, yet maintaining one is intensely labour heavy. Existing tools require professional programming expertise (R metafor, Stat), software installation (RevMan), or server infrastructure which raises data privacy concerns (Shiny-based web apps). We have developed RapidMeta Cardiology to address these limitations, producing a standalone HTML/JavaScript single-page application (~7,800 lines) that executes entirely within an internet browser, requiring no installation, no server dependency and without transmitting data externally.

Methods

The statistical engine implements DL and restricted maximum likelihood (REML) random-effects meta-analysis for OR, RR and HR. For ORs and RRs, per-study effects are computed from 2x2 event-count tables. HR is pooled via generic inverse-variance using published point estimates and 95% confidence intervals. The Hartung-Knapp-Sidk-Jonkman (HKSJ) adjustment, prediction intervals, Egger’s regression test, Bayesian analysis and Trial Sequential Analysis are also implemented alongside 15-plus interactive Plotly visualisations (forest plot, funnel plot, L’Abbé plot, cumulative meta-analysis, influence analysis and others) All trial-level data were extracted exclusively from open-access sources: ClinicalTrials.gov results API v2, PubMed abstracts (PMIDs: 33264825, 34449181, 39225278), and FDA regulatory documents (NDA 215341). Each of the 63 unique data points carries an embedded evidence record that documents source, extracted text, and highlighted values for reviewer verification.

We assessed numerical accuracy by comparing application output against R metafor (V4.8.0) on R 4.5.2 across 14 analyses (five outcomes x two to three effect measures). External concordance was then assessed against 17 published comparisons from finerenone meta-analyses and IPD pooled analyses, defining agreement as an absolute difference of ≤0.03 in pooled point estimates.

Results

All 14 DL pooled estimates were identical to the R metafor output to six decimal places (maximum absolute difference < 10–6). DL and REML estimators produced identical results for all 14 analyses, consistent with the low heterogeneity observed (τ² = 0 for four of five outcomes). Key pooled hazard ratio estimates generated by RapidMeta Cardiology are summarised in Table 1.

Discussion

This study demonstrates that a browser-native JavaScript implementation of random-effects meta-analysis can achieve numerical accuracy indistinguishable from R metafor outputs across five clinical outcomes and a total of three effect measures. Here, the finerenone results are well-established findings, and are used solely as a benchmark dataset for software validation. Our methodological contribution proves that client-side evidence synthesis without server dependency or installation is viable for cardiology research applications.

The platform offers several advantages over existing tools. Unlike R metafor, this web-based program requires no programming expertise, and unlike RevMan requires no installation. The application also operates in its entirety without an internet connection, transmitting no data and directly addresses data privacy concerns relevant in clinical and institutional settings. Evidence provenance is embedded at the data-point level, bridging the common disconnect between raw data and computed results in meta-analysis software.

The finerenone dataset confirms what the broader trial programme established, with consistent and homogeneous treatment effects across cardiovascular and renal outcomes (I² = 0% for MACE, all-cause mortality, and renal composite), with modest heterogeneity for HF hospitalization (I² = 20%), likely reflecting differing heart failure prevalence between the FIDELIO-DKD and FIGARO-DKD populations. The HKSJ adjustment is methodologically appropriate, but produces very wide confidence intervals with only two studies, underscoring a well-known limitation of k = 2 meta-analyses.

Key limitations include a reliance on published HR estimates, rather than using reconstructed time-to-event data, restriction to three Phase III trials, and limited statistical power for publication bias diagnostics with fewer than ten studies. The application is currently pre-configured for finerenone, although the statistical engine is general-purpose and adaptable to other clinical questions.

Conclusion

RapidMeta Cardiology provides a validated and privacy-preserving, installation free platform for living meta-analyses. The statistical engine reproduces R metafor to six decimal places and matches published finerenone meta-analyses across five outcomes. The platform, its source code, trial data and R validation are freely available for independent verification and adaptation.

Software availability

Software available from: https://github.com/mahmood726-cyber/rapidmeta-finerenone

Source code available from: https://github.com/mahmood726-cyber/rapidmeta-finerenone

Archived source code at time of publication: https://doi.org/10.5281/zenodo.19714771

License: MIT.

Artificial intelligence assistance

AI-based tools were used only to support language editing.