Researchers at Imperial College London built CardioKG, an artificial intelligence tool that connects heart scans from the UK Biobank with information about genes, diseases, and medicines. The system identified possible uses for existing drugs, including methotrexate for heart failure and diabetes medicines called gliptins for atrial fibrillation. It also found a possible protective link involving caffeine, although the researchers warned patients against changing what they consume. CardioKG produces candidates for further study. It does not write prescriptions or replace clinical trials.

The distinction can shrink quickly once a result reaches a headline. A medicine already sitting in a pharmacy carries the authority of prior approval, even when its proposed use for another disease remains untested. The graph gives researchers a faster way to connect thousands of observations that no physician could review at once. It also gives a statistical suggestion the visual form of a path, as if the route from scan to drug had already been cleared. Doctors will still need trials, dosage studies, records, and conversations with patients. The useful output is a hypothesis with an address. The dangerous one is the same hypothesis presented without the waiting room, consent form, and years of evidence that must come before a new prescription.