Abstract
Preclinical drug development has long relied on animal models to predict safety and efficacy before agents enter human trials, despite critical differences between human and animal model physiology. The withdrawal of rosiglitazone, rofecoxib, and terfenadine due to cardiovascular toxicity exemplifies the translational cost of this mismatch. Alternative, human-based systems enable more accurate modeling of cardiometabolic diseases in a dish; in 2025, the US FDA’s new approach methodologies (NAMs) roadmap authorized the submission of results from human-relevant models. The roadmap encourages utilizing biological and digital twins as part of an integrated, context-specific, fit-for-purpose strategy. A “biological twin” is a human-derived in vitro system that captures the physiology of a patient and can be used to assess potential cardiotoxicity by drug metabolites. A “digital twin” is the computational counterpart trained on clinical drug response results that can further interpret biological twin data at the patient scale and predict pharmacological parameters. NAMs are no longer experimental but are not yet fully validated as replacements for animal models; major challenges remain before they can be effectively incorporated into the cardiometabolic disease drug discovery pipeline. Addressing these challenges head-on is essential for improving drug development and prediction of their cardiovascular safety.