Post-Marketing Safety Assessment for Glucagon-Like Peptide-1 and Dual Incretin Therapies in Diabetes and Obesity

Abstract

Glucagon-like peptide-1 receptor agonists (GLP-1RAs) have reshaped the treatment landscape for type 2 diabetes and obesity. Introduction of dual GLP-1 and glucose-dependent insulinotropic polypeptide (GIP) receptor agonists, such as tirzepatide, has further expanded therapeutic possibilities. However, interpreting potentially unreported adverse drug reaction (ADR) signals and long-term complication risks for these novel therapies requires careful methodological consideration.

Accurately characterizing novel side effect profiles remains a challenge. Historically, financial conflicts of interest may limit clinicians reporting of novel serious ADRs. Clinicians with high prescribing rates of drugs of potential interest are often key opinion leaders for the relevant drugs and also clinicians most likely to identify first cases of novel ADRs. Second, reliance on voluntary reporting systems populated by reports from clinicians, pharmacists, attorneys, patients, attorneys, and pharmacovigilance programs of pharmaceutical manufacturers has contributed to widespread reporting of incomplete side effect data, as observed in prior medication classes such as opioids and in a report from a large-scale National Institute-funded pharmacovigilance program.¹ Furthermore, aggregation of data by therapeutic class makes it difficult to distinguish individual drug-specific versus class-specific risks among agents sharing presumed therapeutic mechanisms, as was the case with thienopyridine-associated thrombocytopenic purpura caused by ticlopidine (antibody-mediated) and clopidogrel (vascular toxicity);² pure-red cell aplasia caused by Johnson and Johnson-manufactured epoetin, but not by Amgen-manufactured or Roche-manufactured epoetin.³ Without drug-specific post-marketing evidence, clinicians face uncertainty when counseling patients about potential drug-related versus class-related side effects and risk profiles.

Unlike liraglutide and semaglutide, which are pure GLP-1RAs, tirzepatide's dual agonist mechanism, engaging both GLP-1 and GIP receptors, interacts with distinct physiological pathways. GIP co-activation may mitigate gastrointestinal and potentially metabolic events typically seen with GLP-1RAs.⁴ Therefore, grouping tirzepatide with pure GLP-1RAs in pharmacovigilance analyses risks obscuring crucial mechanistic differences, particularly regarding potentially unrecognized toxicities. While mechanism-aware pharmacovigilance can guide hypothesis generation about expected toxicities, mechanism-agnostic approaches are equally crucial. The inherent complexity of human biochemistry—characterized by redundant, interconnected, and often unpredictable signaling pathways—means that unanticipated adverse events may emerge outside of expected pharmacologic models.

Formulation differences further complicate safety interpretation within the GLP-1RA class. For instance, oral semaglutide undergoes distinct absorption processes compared to injectable semaglutide, including co-administration with an absorption enhancer. This alters its pharmacokinetic profile and potentially its tissue distribution and toxicity profile. Such formulation-specific variation could influence the occurrence or detection of certain adverse effects, including gastrointestinal or hepatic toxicities. Therefore, grouping all semaglutide formulations (injectable and oral) together in pharmacovigilance analyses may lead to missed safety signals that are unique to a specific formulation, underscoring the importance of disaggregating these drugs in real-world analyses.

Temporal factors must also be considered. Tirzepatide's more recent approval in 2022 results in limited cumulative exposure compared to earlier GLP-1RAs. For instance, while injectable semaglutide was approved in 2017, oral semaglutide's approval in 2019 also means a shorter cumulative observation period. This shorter market presence for newer agents, compared to therapies like liraglutide with a longer market history, provides a less extensive safety reporting database for evaluating potential novel safety signals. Rare toxicities, such as pancreatitis, thyroid malignancies, or certain psychiatric conditions, often require years of post-market surveillance to identify. Consequently, early differences in adverse drug event rates may simply reflect variations in observation time.⁵ Differences in patient demographics for newer therapies compared to earlier GLP-1RAs also introduce potential confounding factors in safety comparisons. This demographic variation can be attributed, in part, to the distinct FDA approval timelines for various indications, including type 2 diabetes, cardiovascular disease in type 2 diabetes, and weight loss.

Clinicians who prescribe GLP-1RAs daily often report perceived differences in tolerability and side effects between semaglutide and liraglutide, reinforcing the need for agent-specific pharmacovigilance rather than broad class generalizations. Patient experiences suggest that seemingly minor structural variations between drugs within the same class may have meaningful clinical implications, as was the case noted above with pure red cell aplasia that occurred with only one of three marketed epoetin alfa formulations.

Reporter type further complicates signal interpretation. Healthcare professional reports typically offer greater diagnostic specificity, especially for complex outcomes like pancreatitis or psychiatric events.⁶ A greater proportion of patient-generated reports for newer therapies could introduce differential misclassification, affecting signal robustness. Additionally, while social media and direct-to-consumer marketing may influence patients' awareness and expectations, subtly biasing self-reported adverse event patterns, social media sources have been incorporated into 21st-century pharmacovigilance approaches as reported by the Southern Network on Adverse Reactions (SONAR) program.^{7, 8}

Beyond traditional spontaneous reporting, post-marketing surveillance efforts developed in the 21st century expanded to include large-scale data aggregation from deidentified electronic medical records (EMRs) through initiatives such as the FDA Sentinel Program.⁹ Since 2008, Sentinel has linked administrative claims and EMR data nationwide, allowing advanced analytic strategies to identify potential safety signals more rapidly.¹⁰ However, EMR-based analyses are not immune to biases inherent in clinical documentation, including diagnostic ambiguity, inconsistent coding practices, and incomplete data capture. While artificial intelligence and machine learning offer exciting opportunities for pattern recognition, they also risk amplifying existing biases if not rigorously validated. Thus, Sentinel and similar initiatives represent critical advancements developed in the 21st century, but not infallible solutions, in post-marketing safety science.

Expansions in the supply chain, particularly the rise of compounding pharmacies to meet unprecedented demand for GLP-1RAs, further underscore the need for vigilant pharmacovigilance using 21st-century methodologies. Compounded formulations vary in purity, potency, and excipient profiles, and traditionally have been subject to less stringent regulatory oversight than branded medications, although this distinction is diminishing over time. These factors could introduce novel adverse effects or alter existing risk profiles, complicating attribution in post-marketing surveillance.

Improving our ability to detect and differentiate adverse effects in new therapies is critical for patient safety and restoring trust in the patient-physician relationship. Transparent, timely, accurate safety data empower patients to make informed choices, especially as use expands across diverse populations. As incretin-based treatments proliferate, rigorous, unbiased post-marketing surveillance is essential to ensure equitable, trustworthy diabetes care.

The authors declare no conflicts of interest.

This research was supported in part by the National Cancer Institute, grant numbers: 1R01CA102713 and 1R01CA165609 (Charles L. Bennett).

Data sharing is not applicable to this article as no new data were created or analyzed in this study.