Text

Background/Aim: Glucagon-like peptide-1 receptor agonists (GLP-1RAs) have been linked to changes in the risk of multiple cancers through observational studies with varying design choices. These changes have ranged from protective to harmful effects even for the same cancer type, with sometimes unexpectedly large and early effects. In this study, using current pharmacoepidemiological standards, we estimate the risk of four obesity- and hormone-related cancers following GLP-1RA initiation and evaluate how study design choices influence risk estimates.

Methods: Emulation of the design of a previous randomised controlled trial comparing the GLP-1RA liraglutide to placebo, using dipeptidyl peptidase-4 inhibitors (DPP4i) as placebo proxy. Based on U.S. claims data, the rate of new endometrial, pancreatic, prostate, and thyroid cancer diagnoses is estimated for liraglutide alone and all GLP-1RAs, using propensity-score matching for confounding adjustment. Cancer-specific subdistribution hazard ratios (sHRs) were estimated using Fine-Gray models. Sensitivity analyses evaluated the influence of comparator choice, implementation of a latency window, specificity of the outcome definition, confounding assessment window, eligibility criteria, and intercurrent event handling.

Results: GLP-1RA initiation was not associated with increased risks of endometrial, pancreatic, or prostate cancer. A modest increase in thyroid cancer risk was observed, with a stronger association for liraglutide than for all GLP-1RAs, corresponding to small absolute risk increases. Sensitivity analyses demonstrated substantial variations in effect size depending on design choices, particularly implementation of a latency window.

Discussion/Conclusion: GLP-1RA initiation was not associated with meaningful increases in the risks of endometrial, pancreatic, or prostate cancer and was associated with a small increase in thyroid cancer risk although confidence intervals were wide. Study design choices substantially influence effect sizes and may explain discrepancies across previous studies.