Chalk “predict death” off the list of “what computers can’t do.” A supercomputer at Beth Israel Deaconess Medical Center in Boston, Mass., can decipher the likelihood of a patient’s imminent demise with uncanny accuracy.
Patients at the hospital are linked up to the machine, which leverages all available patient data — doctors visits, lab results, medications and vital signs — to generate a rapid diagnostic assessment. This alone would be a tremendous tool for helping doctors provide better treatment, but the real magic comes from applying machine learning jujitsu to an ever-growing body of patient data. By comparing current scenarios with past outcomes from more than 250,000 individuals over the last 30 years, the program is able to determine the likelihood of current and future medical problems, such as heart attacks, infections, cancers — and, yes, even death.
“We can predict with almost a 96 percent confidence that a patient will have this probability of dying,” Beth Israel Project Lead Dr. Steve Horng told BBC News, “so if the computer says you are going to die, you probably will die within the next 30 days.”
In another interview, Horng traces the program’s origins to extensive theoretical machine learning work carried out in collaboration with Stephen Sontag at the Courant Institute at NYU. The algorithm is applied to every single patient that comes into the Beth Israel emergency department, and while the results are used to guide care, the raw information is not shared with patients. In other words, patients are not told their “death number.”
Still, the whole machines-predicting-death-thing is sure to flame concerns about the ethical and societal ramifications of artificial intelligence — concerns shared by such prominent folks as Elon Musk and Stephen Hawking, among others. Horng, for one, appears unphased by the
potential creepiness factor. He is clear that the purpose of the project is to leverage big data to augment doctors’ abilities. One of the ways this is done is by determining just how sick a patient is, which the computer outputs as “probability that death will occur in the next 30 days.”
“Our goal is not to replace the clinician,” said Horng, “this artificial intelligence is really about the doctor’s ability to take care of patients.”
http://www.hpcwire.com/2015/09/25/this-hospital-computer-knows-when-your-days-are-numbered/
You can watch the BBC News video below:
