Researchers at the University of Pennsylvania have developed a new artificial intelligence model capable of predicting cardiac arrest up to 15 minutes in advance. The project, led by cardiologist Rajat Deo from the Perelman School of Medicine and computer scientist Rajeev Alur from the School of Engineering and Applied Science, leverages vast amounts of unused electrocardiographic (ECG) data routinely collected by hospitals. For decades, these electrical traces of heart activity have served primarily as real-time diagnostic tools but have rarely been analyzed retrospectively for predictive purposes. By bridging clinical expertise with advanced data science, the team aims to transform these archives into a proactive safety net. The resulting system, named the Cardiac Autoregressive Model for ECG Language-Modeling, or CAMEL, represents a shift from classifying existing abnormalities to forecasting future events. Traditional AI models typically analyze short 10-second snippets of ECG signals to identify immediate irregularities. In contrast, CAMEL treats continuous heart rhythm data like language, analyzing extended telemetry records that span hours. This approach allows the model to recognize subtle, evolving patterns that often precede dangerous arrhythmias such as ventricular fibrillation or ventricular tachycardia. By converting ECG waveforms into a format compatible with clinical text, the AI can reason about how minor variations in rhythm may indicate a patient's deteriorating condition. Testing has shown promising results in identifying high-risk patients with normal sinus rhythms who are at risk of in-hospital cardiac arrest. The model detects indicators that are often too faint or complex for conventional monitoring tools to interpret. The research team published their findings on the arXiv preprint server, highlighting the collaborative nature of the work across the university's medical and engineering faculties. Despite the technological success, the researchers emphasize the need for extreme caution before clinical deployment. Deo noted the dangers of alarm fatigue in hospital settings, where false alerts divert limited medical resources away from other critical patients. Consequently, the team plans to conduct silent trials first, processing real-time patient data in the background to verify predictions without triggering immediate alerts. This phase will focus on weighing the model's foresight against actual patient outcomes to ensure it exceeds the current standard of care. Beyond the hospital environment, the researchers envision broader applications for this technology. They are optimistic about adapting CAMEL for consumer wearable devices, potentially enabling early warning systems for the general population. By converting historical, underused data into predictive insights, this project offers a significant advancement in cardiac care, promising to give medical teams crucial time to intervene before life-threatening events occur.