Marco Kolo

Sudden cardiac death claims hundreds of thousands of lives every year, often striking people who feel completely healthy. Unlike a typical heart attack caused by blocked blood flow, sudden cardiac arrest happens when the heart's electrical system unexpectedly malfunctions. While implanted defibrillators can easily save lives, doctors have long struggled to predict who actually needs one. Standard screening tests measure how well the heart pumps blood, an approach that unfortunately misses the vast majority of people who eventually suffer a fatal cardiac arrest. ​In a new Nature study, researchers Ziad Obermeyer, Alexander Schubert, James Ross, Sendhil Mullainathan, and Markus Lingman used artificial intelligence to solve this medical blind spot. By linking a massive Swedish database of routine electrocardiogram recordings directly to death certificates, their deep learning model learned to spot hidden electrical warning signs on its own. Remarkably, the algorithm identified high-risk patients across Sweden, the United States, and Taiwan, catching over 86 percent of vulnerable individuals who had been completely overlooked by standard heart tests. ​Even more exciting, the scientists translated the computer's findings into a clear visual signal that doctors can actually see with the naked eye. In high-risk patients, a specific part of the ECG tracing loses its normal sharp downward dip and instead shows a slow, slurred wave. The team traced this unusual pattern back to microscopic scarring inside the heart muscle that quietly delays electrical signals at the end of each heartbeat. Because standard ECGs are cheap, painless, and performed millions of times every day, spotting this hidden warning sign during routine checkups could help doctors protect vulnerable patients long before their heart suddenly stops. Article: Obermeyer, Z., Schubert, A., Ross, J., Mullainathan, S. & Lingman, M. An ECG biomarker for sudden cardiac death discovered with deep learning. Nature 655, 210–218 (2026). DOI: 10.1038/s41586-026-10674-6