Multinational pharmacovigilance assessment of drug-associated esophageal ulceration and perforation: evidence from FAERS, JADER, and Canada Vigilance

Figures

• 

  Fig. 1. Study workflow and descriptive characteristics of FAERS reports. (A) Workflow from raw DEMO records through deduplication and identification of reports meeting the prespecified composite case definition. (B) Quarterly numbers of reports from 2004 Q1 through 2025 Q1; the final bar represents 2025 Q1. (C) Age distribution using the categories in Table 1. (D) Retained serious-outcome codes. DEMO, demographic and administrative information; DRUG, drug information; REAC, adverse-event terms; HO, hospitalization; OT, other serious outcome; DE, death; LT, life-threatening; DS, disability; RI, required intervention; CA, congenital anomaly; FAERS, U.S. Food and Drug Administration Adverse Event Reporting System

• 

  Fig. 2. FAERS disproportionality results. (A) RORs with 95% confidence intervals for ranks 1–25 among the 50 primary-suspect drugs with the largest numbers of composite-outcome reports. (B) Corresponding results for ranks 26–50. Complete ROR, PRR, EBGM, EBGM05, IC, and IC025 results are provided in Supplementary Table S2. (C) Volcano plot of log2(ROR) against -log10(Bonferroni-adjusted P); values > 300 are capped and shown as triangles. Labels identify the five LASSO-eligible candidates with the largest adjusted statistical prominence. All estimates describe reporting imbalance, not incidence or causal risk. EBGM, empirical Bayes geometric mean; IC, information component; LASSO, least absolute shrinkage and selection operator; PRR, proportional reporting ratio; ROR, reporting odds ratio

• 

  Fig. 3. JADER disproportionality results. RORs with 95% confidence intervals are shown for ranks 1–25 (A) and 26–50 (B) among suspected-drug entries with the largest numbers of composite-outcome reports. Arrowheads indicate estimates or confidence limits beyond the plotted x-axis range; exact values are printed in the adjacent column. Results are database-specific and are used only for descriptive cross-database comparison. JADER, Japanese Adverse Drug Event Report database; ROR, reporting odds ratio

• 

  Fig. 4. Canada Vigilance disproportionality results. RORs with 95% confidence intervals are shown for product-name-specific ranks 1–15 (A) and 16–30 (B). Brand, biosimilar, and generic entries are retained as separate rows and must not be summed because report-level identifiers were unavailable for reliable alias deduplication. Arrowheads indicate estimates or confidence limits beyond the plotted x-axis range; exact values are printed in the adjacent column. Small-count estimates and wide intervals should be interpreted cautiously; the results do not constitute external validation. ROR, reporting odds ratio

• 

  Fig. 5. Exploratory signal prioritization in FAERS. (A) Outcome-stratified 10-fold cross-validation for binomial LASSO; lambda.min = 1.128109 × 10-5 and lambda.1se = 6.607358 × 10-5. (B) LASSO coefficient paths; all 11 prescreened drug variables remained nonzero at both penalty values. (C) Receiver operating characteristic curve for the valid-age FAERS case/non-case model (3,119,650 reports; 2,136 composite-outcome reports). The apparent AUC was 0.677 in the fitting sample; no internal or external validation was performed. (D) Adjusted odds ratios for age, sex, and the 11 drug variables. Female and male indicators were compared with missing, unknown, or other sex. The model is exploratory and should not be interpreted as a clinical prediction or causal model. AUC, area under the receiver operating characteristic curve; CI, confidence interval; FAERS, U.S. Food and Drug Administration Adverse Event Reporting System; LASSO, least absolute shrinkage and selection operator; OR, odds ratio

• 

  Fig. 6. Grouped recorded time-to-onset patterns in FAERS. (A) Counts and percentages for 5,245 eligible drug-report onset records across prespecified intervals. (B) Corresponding cumulative percentages. The 0-30-day and > 360-day groups were numerically prominent, but they do not define biological risk peaks because records may be non-independent and affected by incomplete dates, long-term treatment, delayed recognition, and reporting delay. FAERS, U.S. Food and Drug Administration Adverse Event Reporting System