Navigating the Uncertainties of Thyroid Cancer Risk in GLP-1RA Users

About This Grant

Glucagon-like peptide 1 receptor agonists (GLP-1RAs) have become pivotal in managing diabetes and obesity, with a prescription surge of 300% from 2020-2022, with more than half of the U.S. adult population eligible for therapy, reflecting their potential to impact patient important outcomes. Despite their benefits, concerns about an association between GLP-1RA use and thyroid cancer have emerged, highlighted by low events rates and inconclusive findings from randomized trials and inconsistent results from real-world data. There is an urgent need to address the risk uncertainty regarding GLP-1RA and thyroid cancer to prevent unnecessary evaluation for thyroid cancer or withholding of therapy in otherwise eligible patients. The overarching goal of this study is to enable evidence-based treatment for type 2 diabetes and obesity by clarifying the thyroid cancer risk associated with GLP-1RA therapy. To achieve this goal we will utilize the EPIC COSMOS database and a trial emulation design with active comparators to comprehensively assess thyroid cancer risk post-GLP-1RA initiation, leveraging advanced statistical methods including machine learning, Bayesian methodology and microsimulation. The use of EPIC COSMOS one of the largest databases in the US including electronic health records from 289 million patients provide a unique dataset to overcome the limitations of previous studies. In Aim 1, we will quantify incidence, relative risk, and timing of thyroid cancer diagnosis after GLP-1RA initiation among adults with type 2 diabetes and/or obesity who initiate GLP-1 RA between 2010 and 2025. We will use a Marginal Structural Model combined with inverse probability treatment weights, inverse probability censoring weights, and inverse probability detection weights to adjust for treatment assignment, censoring, and detection bias. Super-learner high-dimensional propensity score and instrumental variable will be used to adjust for residual confounders. In Aim 2, we will investigate the risk and timing of incident thyroid cancer diagnoses within pre-specified high-risk groups and use a novel hybrid interpretable causal artificial intelligence method to investigate if certain variables can serve as thyroid cancer risk modulators and if subgroups at high risk for thyroid cancer exist. Bayesian methods will be used to integrate prior knowledge to derive probability distributions of the relative risk of incident thyroid cancer under GLP-1 RA use in the overall population and identified subgroups. In Aim 3, we will complete a microsimulation experiment using the Building, Relating, Assessing, and Validating Outcomes (BRAVO) model to evaluate the risk-benefit trade-offs between the potential thyroid cancer risk and metabolic benefits conferred by GLP-1 RAs across population subgroups with diverse risk profile, and identify subgroups for whom the benefits significantly outweigh the risks, or vice-versa.