Artificial Intelligence Outsmarts Past Practices that Predict Patient Outcomes

Google has been playing a guessing game with the human race for quite some time now. Its Smart Replies, which is built into Gmail, reads the message our colleague has sent to us and suggests a number of short, plausible answers. Google Photos uses artificial intelligence to start applying edits to photos we’ve taken even before we start fiddling with them. But who would have thought one of Google’s most recent predictive algorithms could be a matter of life or death?

According to a paper published in the journal Nature last spring, Google is training machines to predict with — so far — 95 percent accuracy, whether hospital patients will die 24 hours after being admitted. It can also estimate the length of a patient’s hospital stay and their chances of being readmitted better than previous models.

AI “Medical Brain” Assesses More Patient Information

Google’s “Medical Brain” outsmarts the way hospital doctors and other health care providers use stockpiles of electronic and health records and other patient information. In the past, inputting and analyzing every single record of a patient was too time-consuming and too expensive, resulting in much of the information being tossed aside. The”Medical Brain’s” enhanced accuracy is in part due to its ability to analyze even doctors’ notes scribbled on old charts, or memos scratched into the margins of PDFs.

According to the findings published in Nature, Google’s model used that data to predict — with 86 percent accuracy — how long patients had to remain in the hospital for treatment ( a 10 percent increase over past practices). Its 95 percent accuracy in determining whether patients would die during their stays was 9 percent higher than traditional computerized models.

Eighty percent of the time spent on the usual predictive models goes to the “scut work” of making the data presentable, co-author of the report Nigam Shah told Bloomberg News.  Google’s articficial intelligence circumvents this problem. “You can throw in the kitchen sink and not have to worry about it,” said Shah, an associate professor at Stanford University.

In one of the report’s case studies, the “Medical Brain” algorithm gave a woman with metastatic breast cancer a 19.9 percent chance of dying in the hospital after crunching 175,639 data points from her medical records. The Early Warning Score system typically used by the hospital gave her a 9.3 percent chance of dying. Within two weeks, the patient succumbed, making the new Google system the better predictor.

Critical AI Thinking

Google AI chief Jeff Dean told Bloomberg News that the IT giant’s next step is to move this predictive system into clinics. Dean’s health research unit is working on a medical bag full of artificial intelligence tools that can predict symptoms and disease with a level of accuracy that is being met with hope as well as concern.

For all its promise, “Medical Brain” protocol would add to the vast amounts of information Google already has on us. “Companies like Google and other tech giants are going to have a unique, almost monopolistic, ability to capitalize on all the data we generate,” said Andrew Burt, chief privacy officer for the data company Immuta. He and pediatric oncologist Samuel Volchenboum wrote a recent column arguing governments should prevent this data from becoming “the province of only a few companies.”

Thus, Bloomberg News reports “Google is treading carefully when it comes to patient information, particularly as public scrutiny over data-collection rises.” Google and its hospital partners insist the data used in the Nature study is anonymous, secure and used with patient permission. Volchenboum contends the company may have a more difficult time maintaining that data standard if it expands to smaller hospitals and health-care networks.