Presentation (CTI)

OBJECTIVES: Prior observational studies of glucagon-like peptide-1 receptor agonists (GLP-1 RAs) and pancreatitis have largely been negative or shown small, inconsistent risk increases. Given heterogeneity in study design, residual confounding, and limited data on predominant GLP-1 RAs, uncertainty remains regarding pancreatic safety. The objective of this study was to compare the self-controlled case series (SCCS) design with cohort methods, including propensity score matching (PSM) and inverse probability weighting (IPW), to evaluate pancreatitis risk with GLP-1 medications.
METHODS: We conducted a Monte Carlo simulation study with 500,000 person-level observations and 1,000 replications. Baseline covariates and latent time-invariant confounding were generated once and fixed across replications. Each replicate simulated 72 months of GLP-1 initiation and pancreatitis events, targeting a true incidence rate ratio (IRR) of 1.10. SCCS was implemented using fixed-effects Poisson regression with a 1-month exposure lag. PSM and IPW cohort analyses used Poisson regression with 1:1 nearest-neighbor matching (caliper 0.2) and stabilized, trimmed weights, respectively. Bias, precision, mean squared error, and empirical coverage of nominal 95% confidence intervals were evaluated.
RESULTS: Across 1,000 replications, mean effect estimates were SCCS IRR 1.099, PSM IRR 0.993, and IPW IRR 0.971. SCCS exhibited minimal bias (-0.06%) compared with substantial attenuation for PSM (-9.7%) and IPW (-11.7%). Root mean squared error was lowest for SCCS (0.066) relative to PSM (0.113) and IPW (0.136). Empirical coverage of 95% confidence intervals was near nominal for SCCS (92.9%) but markedly poor for PSM (15.1%) and IPW (25.2%), despite excellent covariate balance in cohort analyses.
CONCLUSIONS: In scenarios with rare outcomes, such as GLP-1-associated pancreatitis, and the presence of unmeasured confounding, SCCS provided substantially less biased estimates and near-nominal confidence interval coverage compared with standard cohort designs. These findings highlight the importance of study design choice for post-marketing drug safety evaluation and for regulatory and HTA evidence generation involving GLP-1 receptor agonists.