GALLBLADDER EVENTS FOLLOWING INITIATION OF GLP-1 RECEPTOR AGONISTS VERSUS LIFESTYLE INTERVENTION FOR OBESITY: A CLAIMS-BASED COMPARATIVE ANALYSIS USING TARGETED MAXIMUM LIKELIHOOD ESTIMATION
Author(s)
Ryan Ross, MS, Michelle Jerry, MS, Kristin Evans, PhD, Robert Fowler, MS, Liisa A Palmer, PhD.
Merative, Ann Arbor, MI, USA.
Merative, Ann Arbor, MI, USA.
OBJECTIVES: Glucagon-like peptide‑1 (GLP-1) receptor agonists are increasingly prescribed for obesity but may be associated with gallbladder events (GBE), though real-world comparisons versus non-pharmacologic obesity management are limited. This study compared GBE incidence between patients initiating GLP‑1 versus lifestyle intervention (LSI) using targeted maximum likelihood estimation (TMLE).
METHODS: Adults with obesity initiating GLP-1 (semaglutide; tirzepatide) or LSI (behavioral counseling; nutrition therapy), 6/4/2021-6/30/2024, were identified in the MarketScan® Databases; index was the first treatment claim. Incident GBE diagnoses/procedures in a hospital/ER were identified during 12-month follow-up. Unlike conventional propensity score approaches, TMLE uses flexible machine learning algorithms within a doubly robust framework for treatment and outcome models while maintaining valid statistical inference. TMLE estimated the average treatment effect on the risk ratio (RR) scale for GBE comparing GLP-1 versus LSI, adjusting for baseline patient characteristics. Confidence intervals (95% CIs) were obtained using the efficient influence function. Data-adaptive learners, including penalized regression (GLMNET) and Bayesian additive regression trees (BART), were incorporated into a Super Learner ensemble, which optimally combined candidate algorithms based on cross-validated performance to capture non-linearities and interactions.
RESULTS: 78,939 GLP-1 and 74,095 LSI patients were included. Mean age was similar between groups (45.6 vs. 46.6); GLP-1 users were more often female (74.1% vs. 69.1%). Unadjusted GBE incidence was similar between GLP-1 and LSI (1.26% vs. 1.23%; p=0.603). TMLE showed no significant difference in GBE risk, with near-zero risk difference (1.23% vs 1.24%) and RR of 0.997 (95% CI 0.923-1.078). Findings were consistent across models; in the final Super Learner ensemble, treatment and outcome models were dominated by BART (weight=0.80) and penalized regression (weight=0.77), respectively.
CONCLUSIONS: This analysis showed no association between initiation of a GLP‑1 for obesity and risk of GBE. TMLE findings were robust to flexible modeling approaches, supporting the safety of GLP‑1s with respect to gallbladder outcomes in this population.
METHODS: Adults with obesity initiating GLP-1 (semaglutide; tirzepatide) or LSI (behavioral counseling; nutrition therapy), 6/4/2021-6/30/2024, were identified in the MarketScan® Databases; index was the first treatment claim. Incident GBE diagnoses/procedures in a hospital/ER were identified during 12-month follow-up. Unlike conventional propensity score approaches, TMLE uses flexible machine learning algorithms within a doubly robust framework for treatment and outcome models while maintaining valid statistical inference. TMLE estimated the average treatment effect on the risk ratio (RR) scale for GBE comparing GLP-1 versus LSI, adjusting for baseline patient characteristics. Confidence intervals (95% CIs) were obtained using the efficient influence function. Data-adaptive learners, including penalized regression (GLMNET) and Bayesian additive regression trees (BART), were incorporated into a Super Learner ensemble, which optimally combined candidate algorithms based on cross-validated performance to capture non-linearities and interactions.
RESULTS: 78,939 GLP-1 and 74,095 LSI patients were included. Mean age was similar between groups (45.6 vs. 46.6); GLP-1 users were more often female (74.1% vs. 69.1%). Unadjusted GBE incidence was similar between GLP-1 and LSI (1.26% vs. 1.23%; p=0.603). TMLE showed no significant difference in GBE risk, with near-zero risk difference (1.23% vs 1.24%) and RR of 0.997 (95% CI 0.923-1.078). Findings were consistent across models; in the final Super Learner ensemble, treatment and outcome models were dominated by BART (weight=0.80) and penalized regression (weight=0.77), respectively.
CONCLUSIONS: This analysis showed no association between initiation of a GLP‑1 for obesity and risk of GBE. TMLE findings were robust to flexible modeling approaches, supporting the safety of GLP‑1s with respect to gallbladder outcomes in this population.
Conference/Value in Health Info
2026-11, ISPOR Europe 2026, Vienna, Austria
Value in Health, Volume 29, Issue 12S
Code
MSR69
Topic
Clinical Outcomes, Methodological & Statistical Research, Real World Data & Information Systems
Topic Subcategory
Artificial Intelligence, Machine Learning, Predictive Analytics, Confounding, Selection Bias Correction, Causal Inference
Disease
Diabetes/Endocrine/Metabolic Disorders (including obesity), No Additional Disease & Conditions/Specialized Treatment Areas