The first supercomputer for artificial intelligence in medicine was developed at Stanford University in the early 1970s. So it makes sense that the health system affiliated with the university recently named its inaugural chief data scientist to integrate AI into patient care.
But before unveiling cool, innovative tools, Nigam Shah, PhD, has been spending his first few months in the role at Palo Alto, Calif.-based Stanford Health Care trying to prove that AI is actually useful in healthcare.
"The chatter tends to be about: Is it fair? Is it going to impact people? Is it going to increase disparities? Is it going to put people out of work?" Dr. Shah told Becker's. "Very few places, if any, ask the question that if we use AI to guide care, will we get any value?"
Dr. Shah and some colleagues recently analyzed assessments of 15 healthcare AI models. They found that of the 220 items of information requested, more than half focused on the model itself, while only four pertained to fairness and none looked at usefulness.
"I was like, 'Damn, that's not good,'" he said.
One early use case of AI at Stanford Health Care involves advanced care planning. The project uses an algorithm to flag the charts of patients who might need the service and then tracks how often providers follow through with it.
"One of the things we decided early on was to keep a human in the loop. So the algorithm doesn't have autonomy; it doesn't decide whether someone does or doesn't get advanced care planning," Dr. Shah said. "And we keep the threshold secret. Because we don't want people to say, 'Oh, this person has so-and-so probability of dying, so let me do XYZ.' By choice, we don't tell the threshold to the frontline care providers."
Dr. Shah was previously Stanford Health Care's associate CIO of data science. He's also a professor of medicine and biomedical data science at Stanford University School of Medicine. He said he got to write the new role himself.
He's an advocate for other hospitals and health systems creating chief data scientist roles. He said medicine also needs a subspecialty dedicated to data science.
"When X-rays and MRI machines and CT scanners came around, we got radiology. When we started radiation therapy for cancers, we got radiation oncology, we got nuclear medicine," he said. "Just like we don't have people running their own MRI scanners and zapping patients with radiation by themselves, AI is going to need specialists to make sure that its use is useful, reliable and fair."
He said healthcare AI could become a subspeciality of clinical informatics, or its own new specialty. He also said medical schools should teach data science so physicians will be able to, as he put it, "tell the nonsense from the good stuff."
AI should undoubtedly be used now for mundane healthcare tasks like scheduling, intake and billing, but it still largely has to prove its usefulness for patient care, Dr. Shah said.
Either way, he sees AI not as a luxury in healthcare but as a necessity, especially with looming health professional shortages and an aging population.
"It takes a decade-plus to produce a physician in the United States. There are only about a million physicians total across all specialties," he said. "We will need technology as a force multiplier. There is no other way. And so it's still unclear exactly how all of this will play out, with the intersection of AI and digital and telehealth. But whenever there's chaos, there's opportunity."