When a positive procalcitonin is not diagnostic of early-onset neonatal infection: a systematic diagnostic-accuracy review and Bayesian decision analysis

Background. Early-onset neonatal bacterial infection (EOS) is rare but fulminant; fear of missing it drives antibiotic overtreatment, and no biomarker alone guides antibiotics reliably. What can cord-blood procalcitonin (PCT) decide? Predictive values depend on prevalence. Methods. We pre-registered six falsifiable hypotheses with fixed thresholds. From four cord-blood cohorts and fourteen serum series, we pooled accuracy (bivariate, bootstrap) and re-analysed results as a Bayesian decision chain — accuracy, pre-test probability, posterior probability, threshold, antibiotic action — triangulated against an independent meta-analysis. Results. Sensitivity was 0.65 (95% CI 0.53–0.75); specificity 0.91 (0.69–0.98); negative likelihood ratio 0.39. Stand-alone use is refuted. At 1% prevalence, a positive result carries 6.8% posterior probability: it cannot rule in EOS. Negative predictive value is 0.996 at 1% prevalence — the sole corroborated hypothesis, robust under bootstrap, worst-case bounds and external syntheses. The safe window closes at 2.5% prevalence (1.5% worst case). At 1% prevalence, rule-out spares 902 antibiotic courses per 1,000 newborns for 3.5 deferred infections (255:1), and dominates treat-all above a 0.39% threshold. Conclusions. Cord-blood PCT cannot rule in EOS at any realistic prevalence. Its only corroborated property is a strictly prevalence-gated rule-out: the actionable unit is the bounded decision rule, not the biomarker. Version 3.0 note. This version restores and updates the package matching the manuscript submitted to Pediatric Research (PR-2026-2379, under review): corrected title page with full corresponding-author address, Impact Statement in five bullet points, three journal-format figures (PDF), and the PRISMA 2020 checklist. It supersedes version 2.0, which contained a different analysis package.