Negative predictive value of cord-blood procalcitonin for early-onset neonatal bacterial infection: a pre-registered refutation campaign and a prevalence-calibrated decision operator

Pre-registered diagnostic-accuracy meta-analysis and Bayesian decision-analysis of the negative predictive value (NPV) of cord-blood procalcitonin (PCT) for early-onset neonatal bacterial infection (EONI). Six falsifiable hypotheses with GO/NO-GO thresholds were fixed before analysis; four independent cord-blood cohorts and fourteen serum comparator series were synthesised by bivariate random-effects meta-analysis. Pooled cord sensitivity was 0.65 (95% CI 0.53-0.75) and negative likelihood ratio 0.39, refuting a stand-alone discriminating test (H1, H2: NO-GO). NPV is very high at very low prevalence (0.996 at 1%; H3: GO) but falls below 0.98 from 5% onwards (H4: NO-GO). By decision-curve analysis a negative-cord-PCT rule-out spares ~902 empirical antibiotic courses per 1000 neonates at 1% prevalence, dominating 'treat all' above a 0.39% treatment threshold, provided prevalence stays below 2.5%. This deposit contains the article (English and French; PDF, DOCX, sources), the conceptualisation dossier, the audited 2x2 dataset, primary results and analysis code, all figures (English and French), the AlphaFold structure of the CALCA precursor, the audit report and a global SHA-256 manifest. Every reported value is injected from SHA-256-fingerprinted primary files; refuted hypotheses are published on the same footing as the corroborated one.