GLP-1 Receptor Agonists Safety Analysis

What this research found

Do the three most-discussed GLP-1 safety signals — pancreatitis, thyroid cancer, and gastroparesis — fall harder on any one drug? Pairwise tests across 657 FDA adverse event reports for semaglutide, tirzepatide, and liraglutide found no significant differences, and the biology behind each event maps onto expected GLP-1 receptor activity rather than an unexpected off-target effect. The conclusion drawn is a class effect rather than a drug-specific liability, including for tirzepatide's dual GIP/GLP-1 mechanism.

• None of the nine drug-by-event comparisons reached significance even before multiple-testing correction. The largest signal, semaglutide versus tirzepatide for pancreatitis, gave an odds ratio of 2.24 (95% CI 0.748–6.705, p = 0.1715), well short of the Bonferroni-adjusted threshold of 0.005556.
• Pancreatitis dominates the record: 622 of the 657 reports, against 33 for thyroid cancer and 2 for gastroparesis. With two gastroparesis reports in total, those odds ratios range from 0.07 to 5.52 with confidence intervals spanning 0.003 to 89.5, so they carry no useful information.
• Reporting is climbing across the class. Semaglutide pancreatitis reports rose 55.4% between 2024 and 2025 (74 to 115) and tirzepatide's rose 60.0% (20 to 32), while semaglutide thyroid cancer reports went from 1 to 6.
• Genetic constraint data argues that the GIP arm of tirzepatide adds little inherent risk: GLP1R is highly constrained against loss of function (observed/expected 0.311) whereas GIPR tolerates it (1.321), which points to functional redundancy at GIPR.
• Hospitalization was the most common reported outcome for every drug — 60.6% of semaglutide reports, 50.8% of tirzepatide, and 46.7% of liraglutide — with deaths at 3.6%, 5.1%, and 1.6% respectively.

How it was done

Adverse event reports for the three drugs were pulled from the FDA Adverse Event Reporting System covering Q3 2010 through Q4 2025, giving 657 records that matched the three event categories of interest. Every drug pair was compared on every event using Fisher's exact test where an expected cell count fell below five and a Yates-corrected chi-square test otherwise, then Bonferroni-corrected across the nine comparisons. Open Targets Platform entries for GLP1R and GIPR supplied safety liabilities and genetic constraint scores, which were used to classify each adverse event as on-target or off-target and to reason about tirzepatide's second receptor. Findings were assembled into a 23-page report with a forest plot of odds ratios, a significance heatmap, temporal trend charts, and 18 literature citations.

Data sources

• FDA Adverse Event Reporting System (FAERS) — 657 matched reports for semaglutide, tirzepatide, and liraglutide, Q3 2010 to Q4 2025
• Open Targets Platform — GLP1R (ENSG00000112164) and GIPR (ENSG00000010310) safety liabilities and genetic constraint scores
• PubMed — 18 peer-reviewed citations, 2019–2025

Figures from this analysis

Outputs produced

• 01_data_acquisition.py

  Data acquisition script for FAERS and Open Targets APIs

• faers_raw.json

  Raw FDA FAERS adverse event records (655 records for 3 drugs)

• opentargets_safety.json

  Open Targets safety liabilities and genetic constraint data for GLP1R and GIPR

• 01_acquisition_summary.txt

  Summary of data acquisition results and success criteria check

• 02_data_analysis.py

  Data preprocessing and quantitative signal analysis script

• 02_analysis_results.json

  Processed analysis results with temporal trends, severity metrics, and cross-tabulation

• 02_analysis_summary.txt

  Statistical summary of key findings from adverse event analysis

• 03_statistical_tests.py

  Statistical significance testing script with pairwise comparisons