NCT-bridge methodology is structurally blind to PACTR-registered African trial publications

Abstract

How well does NCT-bridge methodology, used by every existing ClinicalTrials.gov-style synthesis audit, capture publications from PACTR-registered African trials? We scraped 5,878 PACTR trials (snapshot 2026-05-04, sha256 44f99b14), assigned 768 to ten high-burden conditions, and tracked four gates by algorithm and a blinded 30-trial spot-check. Algorithm and auditor applied identical gate definitions: gate 2 via NCT-bridge to Europe PMC (algorithm) or direct title/PACTR-ID search (auditor), and gate 3 via the Pairwise70 index. The algorithm found 3.6% peer-published and 0.5% Cochrane-cited; the blinded spot-check found 53.3% peer-published (16/30) and one Cochrane match the algorithm missed (WOMAN trial, CDE coverage gap). Algorithm publication-detection sensitivity was 6.2%; 96.0% of trials carry no NCT cross-reference and are structurally unreachable through NCT-bridge. NCT-bridge methodology is blind to PACTR publications even when published, undercounting trial-to-publication conversion by an order of magnitude. The preregistered 30-trial spot-check returned 0/30 verdict-level agreement and the threshold was not lowered — the disagreement is the finding.