Play Live Radio
Next Up:
0:00
0:00
Available On Air Stations
Health, Science & Tech

AI is faster than humans at finding bacterial outbreaks, could save lives and money

UPMC AI Bacteria
UPMC
Dr. Lee Harrison (left) and Alex Sundermann published their work in the journal of Clinical Infectious Disease. Sundermann said the use of machine learning doesn’t render humans obsolete as they will still need to investigate outbreaks; “But it definitely expedites it and finds new routes that we’ve never really looked into before.”

Pittsburgh-based researchers have designed a way to use artificial intelligence to detect infectious disease outbreaks within hospitals.

Researchers published their findings in the journal of Clinical Infectious Disease. They found AI produces quicker results than traditional infection-control methods and could save large hospitals such as UPMC Presbyterian more than $300,000 a year.

The scientists sequenced the DNA of samples taken from Presbyterian patients who were infected with “healthcare-associated” bacterial pathogens. These are infections, such as MRSA or norovirus, that people get when receiving medical care for an unrelated health issue.

If results from samples’ genomic sequencing were highly similar or nearly identical, they almost certainly were part of the same outbreak.

All samples were six months old. This way researchers could determine if the machine learning was able to identify outbreaks that people missed.

After the genomic sequencing, an AI algorithm developed at Carnegie Mellon University used that data along with information from a patient’s health record to identify potentially deadly patterns. The patient data included everywhere someone had been in a hospital, along with which providers they interacted with, and even what equipment was used in their treatment.

What’s even more exciting is that the algorithm analyzes these data sets within minutes, while it might take a busy health care worker hours or even days.

Lead author Alex Sundermann of the University of Pittsburgh said the use of machine learning doesn’t render humans obsolete as they will still need to investigate outbreaks; “But it definitely expedites it and finds new routes that we’ve never really looked into before.”

One of the first outbreaks the AI discovered was related to bronchoscopy, which is a procedure where a lighted tube is put down someone’s nose or mouth into their airways.

“The reason why it wasn't detected otherwise is because there was only one or at most two patients on the same nursing unit [who were infected,]” said lead author Dr. Lee Harrison. “One of these infections in a single nursing [unit] is not going to call attention.”

Because people were still getting sick from this pathogen while the study was taking place, hospital workers were alerted to and then addressed the previously undetected outbreak.

Usually, a hospital wouldn’t perform genomic sequencing on every patient with a bacterial infection, in part because the testing costs roughly $70–$90 per sample. But as addressing an outbreak can cost more than $20,000, the cost-savings are substantial.

Presbyterian staff are now working to fully implement these infection methods. Eventually similar efforts may spread to all 40 UPMC hospitals.