A recent issue of the New York Times Sunday Magazine featured two articles on artificial intelligence. One story described the use of a deep neural network in determining when to switch from curative care to palliative care. This decision, to switch from healing to comfort, is agonizing and emotional, for patients and families that must consent, for those who provide care and realize that there will be no cure. The timing of the decision is critical, for there is a sweet spot, a window between three and twelve months before death, when making the right decision makes a world of difference, can make for a good death.
Enter Anand Avati, a graduate student at Stanford. His team started with data on 200,000 deceased patients. They entered all of the data for the first 160,000, including medical records of diagnoses, procedures, hospitalizations, and treatments, as well as the actual date of death, then used the remaining 40,000 patients to test the accuracy of their algorithm. The results were remarkable. Not only did the algorithm successfully predict those who would die within the nine-month palliative care window with 90% accuracy, but it also was 95% accurate in identifying those who would live longer than twelve months, in other words, those for whom it was definitely too early to switch from curative to palliative care. Continue reading →