An Algorithm to Predict the Death of Terminally Ill Patients

A thoughtful, riveting piece from author and oncologist Siddhartha Mukherjee in the New York Times Magazine touches upon the current, general inaccuracies in predicting how much time remains in the lives of patients with terminal illnesses. Mukherjee, however, goes into explaining a “dying algorithm” that might just be able to help doctors with palliative care. Plugging a hospital’s medical records for 160,000 deceased patients into a “deep neural network,” Anand Avati of Stanford’s computer-science department developed such an algorithm that aimed to predict which living patients would die in the next three to 12 months. This algorithm was applied to 40,000 living patients, and was accurate with 9 out of ten patients predicted to die within three to 12 months, doing so. Inversely, 95% of patients assigned low probabilities survived longer than 12 months. Read more about the algorithm and Mukherjee’s thoughts on it at the NYTIMES Magazine.