ProportionMA: a single-file browser tool for meta-analysis of disease prevalence and event proportions, validated from first principles against R metafor and meta::metaprop

Abstract

Can a single-file browser tool pool prevalence from study-level event counts and denominators as faithfully as a dedicated R package? We implemented three transformations — Freeman–Tukey double-arcsine, logit, and raw proportion — with DerSimonian–Laird and REML pooling in one HTML file. Each was coded from first principles: Hartung–Knapp–Sidik–Jonkman with a max(1, Q/(k−1)) variance floor, prediction intervals on t with k−1 df, Miller's 1978 harmonic-mean back-transform, and Clopper–Pearson exact intervals. Applied to a published dataset (neurological improvement after cerebral vasospasm; 14 studies, 326 patients), it pools a Freeman–Tukey prevalence of 0.802 (95% CI 0.705–0.887; I²=68%). Against R's metafor and metaprop, closed-form estimators agreed to 4×10⁻¹³ and Clopper–Pearson bounds to 5×10⁻¹⁴, independently reproduced. Forest and funnel plots with subgroup stratification show the pooled proportion depends on the transform — logit pools 0.757 — so it must be pre-specified. The tool fills a gap in browser-based prevalence synthesis but is limited to single-group proportions, without within-study clustering or meta-regression.