GLP-1 Adverse Events FAERS Analysis

What this research found

Adverse event reports for five GLP-1 drugs were mined to separate genuinely emerging safety signals from the reporting growth that naturally follows a drug becoming popular. Thyroid cancer, the concern that dominates sell-side commentary, behaved like volume growth: its share of reports stayed flat (Mann-Kendall p = 8.74e-02). Gastroparesis, discussed far less, was the one proportion that rose significantly over time (p = 1.21e-03). The result was packaged as a 17-slide brief for a position-sizing meeting.

• Thyroid-related reports appear to grow only because total GLP-1 reporting grows. From 2015Q1 to 2025Q4 overall report volume rose steeply (Mann-Kendall tau = 0.8668, p = 2.22e-16) while the thyroid share stayed flat (tau = 0.1797, p = 8.74e-02, R² = 0.0024 on a linear fit).
• Gastroparesis is the exception. Its share of reports increased significantly (tau = 0.3393, p = 1.21e-03) at 0.000054 per quarter with R² = 0.2275, the pattern expected of an emerging signal rather than of rising usage.
• No single drug stands out for either event within the class. The highest thyroid cancer reporting odds ratio was liraglutide at 1.52 (95% CI 0.91–2.52, 22 cases) and the highest gastroparesis figure was tirzepatide at 5.19 (95% CI 0.69–39.32) resting on one case; no lower confidence bound cleared 1.
• Both events are rare in absolute terms. In the quarterly series, thyroid events accounted for 403 reports (0.60%) and gastroparesis for 130 (0.19%), while the strict gastroparesis term set matched only 3 reports in the 20,307-report disproportionality dataset.
• Seriousness rates diverge sharply between drugs, from 87.7% of semaglutide reports down to 29.7% of exenatide reports (chi-square = 23173.82, df = 4). Age distributions also differ significantly between the drug populations (ANOVA F = 547.63), so between-drug comparisons carry a confounder.

How it was done

Reports for semaglutide, tirzepatide, liraglutide, dulaglutide, and exenatide were pulled from the FDA Adverse Event Reporting System through the openFDA interface, giving 20,307 unique reports and 87,154 reaction-level observations from February 2004 to December 2025. Reporting odds ratios and proportional reporting ratios were computed for each drug against the rest of the class, with a signal declared only where the lower bound of the 95% interval exceeded 1. Quarterly event proportions across 66,824 reports from 2015Q1 to 2025Q4 were then tested with Mann-Kendall trend tests and linear regression to tell true acceleration apart from volume growth. The output was a 17-slide presentation with an adverse event heatmap, a forest plot of the target signals, and dual-axis volume-versus-proportion charts.

Data sources

• FDA Adverse Event Reporting System (FAERS) via openFDA — 20,307 unique reports and 87,154 reaction-level observations, February 2004 to December 2025
• Reports by drug: semaglutide 6,027, liraglutide 5,082, exenatide 4,913, dulaglutide 2,181, tirzepatide 2,104
• Quarterly FAERS series of 66,824 GLP-1 reports, 2015Q1 to 2025Q4

Figures from this analysis

Outputs produced

• glp1_adverse_events.json

  Raw GLP-1 adverse event data from OpenFDA FAERS

  1736.77 MB

• data_acquisition_summary.txt

  Summary of data acquisition with record counts per drug

• 01_data_acquisition.py

  Python script for querying OpenFDA API

• glp1_cleaned.parquet

  Cleaned and deduplicated adverse event data (flattened)

  0.8 MB

• glp1_cleaned.csv

  Cleaned data in CSV format

  8.9 MB

• eda_summary.txt

  Comprehensive EDA summary report

• 02_eda_cleaning.py

  Python script for data cleaning and EDA

• disproportionality_stats.csv

  Full disproportionality statistics for all drug-event pairs