Change-Point Detection Complements Disproportionality Analysis for Emerging Adverse Events in Established Drugs: A Validated Re-analysis of Australia's DAEN

Version 2 (v3.2) — corrected and expanded. 3 June 2026. This version updates the original April 2026 preprint following internal re-analysis from the persisted confirmatory data and alignment to the pre-registered protocol (OSF 10.17605/OSF.IO/E5HBX). The corrected results supersede v1. Substantive changes: Pre-registration-honest primary/secondary structure. The OSF record registers the overall paired CPD-vs-rolling-PRR comparison as the H1 primary (now reported null: median 0 quarters, Wilcoxon p = 0.47) and the genuine-emergence-vs-new-drug subgroup contrast as the H4 directional secondary. v1 had described the genuine-emergence subgroup as the primary. Headline magnitude correction (not a re-analysis of the same test). v1 reported the genuine-emergence advantage as +5.0 quarters / p = 0.003 while labelling the subset "Tier A (n = 14)." That figure was in fact the all-tiers pooled estimate (Tier A+B+C, n = 40). v3 reports the pre-registered Tier A figure as the headline — +3.5 quarters [bootstrap 95% CI 0.0, 8.0] / p = 0.015 vs PRR — and retains the +5.0 q / p = 0.002 pooled result as an explicitly-labelled across-tiers sensitivity analysis. The correction makes the headline more conservative (smaller effect, larger p), and the direction is unchanged. This was a labelling fix, not selection among alternative tests. Retraction of the confidence-PPV inversion. v1 reported a confidence-PPV inversion (Cochran-Armitage trend p = 0.004). This could not be reproduced from the persisted analysis CSVs; the corrected unstratified table gives trend p = 0.50 (non-significant). The claim is withdrawn. Cross-database reframing. v1 framed the DAEN+FAERS pooling benefit as a synergy. The strict synergy test is null (pooled vs better-of-the-two, p = 0.51; pooled = FAERS in 27/31 pairs), so v3 reframes the benefit as a database-size/access effect, not a pooling synergy. Added analyses (post hoc, labelled as such): rolling BCPNN-IC shrinkage comparator (H4 survives: +3.0 q / p = 0.018), truncation-aware Monte Carlo null for the detection-to-action gap, BH-FDR control across the secondary/exploratory family, harmonised-input BOCPD sensitivity. The pre-registration timestamp (OSF, 2026-04-09) precedes all confirmatory analyses. Analysis code is archived at 10.5281/zenodo.19481183.