REAL-WORLD OBESITY AND GLP-1RA TOLERABILITY: INCIDENCE AND PROGNOSTIC RISK FACTORS OF SERIOUS GI ADVERSE EVENTS USING FEDERATED LEARNING
Author(s)
Isabella Kaczmarczyk, PhD, MRes1, Milou Brand, PhD1, James T Brash, PhD2, Atif Adam, MPH, PhD, MD3, Eva Oakkar, PhD, MS4.
1IQVIA, London, United Kingdom, 2IQVIA, Brighton, United Kingdom, 3IQVIA, Boston, MA, USA, 4IQVIA, Washington, DC, USA.
1IQVIA, London, United Kingdom, 2IQVIA, Brighton, United Kingdom, 3IQVIA, Boston, MA, USA, 4IQVIA, Washington, DC, USA.
OBJECTIVES: GLP-1 receptor agonists (GLP-1RAs) have advanced obesity management through substantial weight reduction. While serious gastrointestinal (GI) complications are reported in trials, insights into real-world incidence and risk stratification remain limited. Traditional prediction models are constrained by data-sharing barriers and cross-site heterogeneity. Federated learning (FL) enables privacy-preserving analytics on decentralized data, yet remains underused in pharmacovigilance. We estimated the incidence of serious GI adverse events after GLP-1RA initiation and explored prognostic factors using FL across global primary care databases.
METHODS: We conducted a retrospective cohort study using FL across five OMOP Common Data Model-standardized primary care databases in the USA, Belgium, UK, Germany, and Spain. Eligible adults had obesity (BMI ≥30), initiated GLP-1RA from product approval dates (June 2021-March 2022) and latest available data (late 2025) and did not have type 1 or type 2 diabetes. An on-treatment design assessed a composite of serious GI outcomes within 365 days of initiation. Federated descriptive and time-to-event models estimated event patterns and associations with prespecified covariates (age, sex, BMI, prior GI history, comorbidities). Analyses ran locally and were aggregated via Vantage6 (federated learning infrastructure) treatment indicators were evaluated descriptively only.
RESULTS: The cohort included ~390,000 adults with obesity (mean age 51.8 ± 13.4 years, 68% female, mean BMI 36.5 ± 6.8 kg/m². Initiation was predominantly with semaglutide. Approximately 15%-20% experienced a composite serious GI event within one year, with risk concentrated in the first month post-initiation. Component outcomes were consistent with clinical expectations: upper GI and biliary events ~1%-2% and rarer outcomes <1%.
CONCLUSIONS: FL enabled scalable, privacy-preserving estimation of GI event patterns without centralizing data. These findings support improved awareness of early GI risk following GLP-1RA initiation and may support earlier monitoring strategies for GI tolerability following initiation, with scope for future refinement of prognostic modelling approaches.
METHODS: We conducted a retrospective cohort study using FL across five OMOP Common Data Model-standardized primary care databases in the USA, Belgium, UK, Germany, and Spain. Eligible adults had obesity (BMI ≥30), initiated GLP-1RA from product approval dates (June 2021-March 2022) and latest available data (late 2025) and did not have type 1 or type 2 diabetes. An on-treatment design assessed a composite of serious GI outcomes within 365 days of initiation. Federated descriptive and time-to-event models estimated event patterns and associations with prespecified covariates (age, sex, BMI, prior GI history, comorbidities). Analyses ran locally and were aggregated via Vantage6 (federated learning infrastructure) treatment indicators were evaluated descriptively only.
RESULTS: The cohort included ~390,000 adults with obesity (mean age 51.8 ± 13.4 years, 68% female, mean BMI 36.5 ± 6.8 kg/m². Initiation was predominantly with semaglutide. Approximately 15%-20% experienced a composite serious GI event within one year, with risk concentrated in the first month post-initiation. Component outcomes were consistent with clinical expectations: upper GI and biliary events ~1%-2% and rarer outcomes <1%.
CONCLUSIONS: FL enabled scalable, privacy-preserving estimation of GI event patterns without centralizing data. These findings support improved awareness of early GI risk following GLP-1RA initiation and may support earlier monitoring strategies for GI tolerability following initiation, with scope for future refinement of prognostic modelling approaches.
Conference/Value in Health Info
2026-11, ISPOR Europe 2026, Vienna, Austria
Value in Health, Volume 29, Issue 12S
Code
RWD83
Topic
Methodological & Statistical Research, Real World Data & Information Systems
Topic Subcategory
Distributed Data & Research Networks
Disease
Diabetes/Endocrine/Metabolic Disorders (including obesity), Gastrointestinal Disorders