The Integrity Ratio and the Transparency Discount: a registry-based framework for correcting and pricing publication bias in evidence synthesis and HTA

Abstract

Background: Evidence synthesis assumes the published literature is an unbiased sample of all conducted research, yet selective publication -- positive trials published, negative trials left in the file drawer -- is pervasive. Standard corrections such as trim-and-fill and Copas selection models rest on symmetry assumptions and treat the missing studies as a theoretical quantity inferred from the shape of what was published. ClinicalTrials.gov now offers what those methods do not use: a near-census of intended research, from which the proportion of evidence that actually reached print can be measured directly. Methods: We define a single registry-derived quantity, the Integrity Ratio lambda -- the fraction of enrolled patient data visible in the published literature -- and use it twice. As a statistical correction, the Ghost-Weight Adjustment Model (GWAM) classifies registered, completed trials as published, results-posted, or ghost (completed but never reported), treats the ghost layer as latent data, re-weights the pooled effect by lambda, and propagates the uncertainty about unobserved trials through a Monte-Carlo layer. As an economic price, the Transparency Discount converts the same lambda into a corrected ratio: discounted ICER = Cost / (Effectiveness x lambda). Results: Across 5,000 Monte-Carlo replications, a standard random-effects estimator carried a bias of +0.168 in a high-suppression scenario, while GWAM reduced this to -0.050 -- trading a large optimistic bias for a small conservative one. On the Turner et al. (2008) antidepressant dataset, where the published literature claimed an effect size of 0.41 against an FDA all-trials ground truth of 0.31, a registry scan showed 27.5% of patient data missing (lambda = 0.725); GWAM recovered 0.31 and the simpler multiplicative discount returned 0.30, both within rounding of the FDA value. In a worked HTA example, a drug priced at an ICER of $40,000/QALY rose to $80,000/QALY once a lambda = 0.5 discount was applied -- crossing a conventional threshold from fundable to low-value. Conclusion: One registry-anchored metric corrects the effect size for statisticians and prices the missing evidence for payers; making non-publication financially visible turns transparency from an ethical appeal into a budget line.