Advertisement

Google AI can predict when people will die with ‘95 per cent accuracy’

Google’s reach into almost all aspects of people’s digital lives – from email and search, to maps and calendars – means there isn’t much it doesn’t know about its 1.17 billion users. And now, the technology giant might even know when you’re going to die.

The company’s Medical Brain team is using a new type of artificial intelligence algorithm to make predictions about the likelihood of death among hospital patients.

A paper published last month in the journal Nature detailed how its predictive algorithm used vast data sets to determine whether inpatients will survive in two different hospitals.

For predicting inpatient mortality, Google’s Medical Brain was 95 per cent accurate in the first hospital and 93 per cent accurate in the second hospital.

“This was significantly more accurate than the traditional predictive model,” the paper stated. “These models outperformed traditional, clinically-used predictive models in all cases. We believe that this approach can be used to create accurate and scalable predictions for a variety of clinical scenarios.”

In one case study, the Medical Brain algorithm gave a woman with metastatic breast cancer a 19.9 per cent chance of dying in the hospital by crunching 175,639 data points from her medical records.

The hospital's Early Warning Score that it typically uses gave her a 9.3 per cent chance of dying. Within two weeks the patient was dead.

Google has already made inroads into the medical industry with its DeepMind subsidiary – considered by some academics to be the leaders in artificial intelligence research.

It has already courted controversy, however, after it was revealed in 2013 that DeepMind had access to 1.6 million medical records of NHS patients at three hospitals run by London's Royal Free Trust.

A leaked letter from Dame Fiona Caldicott, the head of the Department of Health's National Data Guardian, questioned the access deal by suggesting it may be have been made on an "inappropriate legal basis."