Topol Charts AI Path to More Accuracy in Medicine

Dr. Eric Topol

Artificial intelligence (AI) is smart at assessing patterns in medicine, and getting smarter all the time. In some fields, it might already be as or more accurate than your typical seasoned practitioner. Truth is, someday, many years from now, machines may out-detect and out-diagnose physicians, but AI will never out-care human doctors. That caring is the “secret sauce” of the medical profession, emphasized Dr. Eric Topol at a virtual NIH Director’s Wednesday Afternoon Lecture.

Renowned as a cardiologist and award-winning scientist whose forward thinking often challenges the traditional medical community’s status quo, Topol founded Scripps Research Translational Institute at The Scripps Research Institute. 

In 2016, NIH awarded him a $207 million grant to help lead the Precision Medicine Initiative Cohort Program (now known as All of Us Research). However, it may be another pursuit for which he is best known—“world- class Twitter meister.”

That’s how NIH director Dr. Francis Collins described Topol, who has attracted more than 440,000 followers on social media. 

“His tweets are probably the most information-rich of any in biomedicine,” Collins remarked, describing how Topol “with all kinds of creativity is always trying to push the limits of what’s possible into an even more exciting technological space, while retaining this critical aspect of keeping medicine about people and maintaining that compassion and that human focus.”

Topol used his WALS lecture, “Deep Medicine: How Artificial Intelligence Can Make Medicine Human Again,” to point out ways that doctors who combine their expertise with machine learning are in fact exemplifying the care in health care.

Noting a recent National Academy of Medicine report that more than 12 million significant medical diagnostic mistakes are made every year and that every person will experience at least one such error during their lifetime, Topol talked first about precision medicine.

“That’s where AI will have its biggest early impact,” he said. “Precision medicine has become an important objective and a buzz word, but what we really want is not just precision medicine—that is, making the same mistakes consistently—but also we want accurate medicine. That’s the end goal, and that’s where AI can make a big difference.”

Topol said we achieve medical accuracy by plumbing the depths of “deep neural networks,” a subtype of AI. 

Contrasting illustrations of actual neurons with artificial neurons, he described the system of connections, relationships and associations that machines establish using inputs of data—images, speech, text. Deep neural networks can be “trained” with hundreds of thousands, even millions more data points than human diagnosticians. 

More training, more experience with similar information—“augmentation of interpretation”—leads to greater accuracy, Topol said, showing results from studies of chest x-rays and mammography where more tumors were detected routinely by machines using deep neural networks than by their human counterparts.

Topol pointed to images of human retinas from a study on predicting sex. Human retina experts viewing the photos were correct about half the time. The deep neural network had a 97 percent accuracy rate.

“There are better ways to predict sex than by looking at retinas,” Topol acknowledged, “but I think it conveys that we can have machines see things better than humans and certainly the best is a fusion of efforts.”

That growing collaboration between human care providers and well-coded computers or algorithms can revolutionize every aspect of the field, Topol said. 

“Deep learning is going to have its effects across the board,” he said. “There is no specialty or domain in medicine or health care—extending to pharmacists, paramedics and nurses—that will not benefit from some machine support.”

Sharing a forecast by internationally acclaimed neurosurgeon Dr. Antonio Di Leva, Topol said, “Machines will not replace physicians, but physicians using AI will soon replace those not using it.” 

With an illustration that the future is now, Topol also described “momentous AI in the real world”—work that is helping critically ill infants today. 

At Rady’s Children’s Institute in San Diego (and across the nation), physicians are using deep learning algorithms to progress from blood DNA sample of the ailing child to accurate diagnosis and management of the condition within 18 hours. 

The practical result has been that even neonatologists and pediatricians without rare disease experience or expertise can have the best advice for managing very sick newborns.

Topol also acknowledged several of the cons involved with employing AI medicine, including concerns about privacy, accountability, safety and security, fairness, non-discrimination and bias. Many negative effects and results may not be due to algorithm error, he explained, but to problems with data input.

Basically, if the data points entered into the deep neural network are not from diverse populations, then the results AI derives will be neither accurate nor applicable to the widely diverse patients they’re meant to serve.

Another growing trend to be mindful of, Topol suggested, are AI devices—smartwatches, biosensors and other wearables. These popular instruments empower people with their own health data, but that knowledge can have unexpected consequences if not also paired with physician guidance and advice.

Ultimately, Topol concluded, AI returns to doctors the “gift of time.” By allowing machines to carry out some labor-intensive and redundant tasks—synthesis of patient data, keyboarding medical records, primary/screening review of dozens of images and diagnosis of routine non-serious conditions, for instance—doctors can resume what they do best: providing care.

“I think we can all agree there isn’t any algorithm for empathy,” he said. “This is what we are for—the human connection. We aren’t suddenly going to become more intelligent. But machines are. Our charge is to get more humane.”

Topol said a physician’s primary purpose was expressed succinctly by celebrated Harvard professor Dr. Francis Peabody nearly a century ago in a 1927 JAMA article that closed with this line: “‘One of the essential qualities of the clinician is interest in humanity, for the secret of the care of the patient is in caring for the patient.’ If we do this right, AI can help us with this most vital goal.”

Topol’s full lecture is archived at https://videocast.nih.gov/watch=41589. 

WALS kicks off its 2021-2022 season on Sept. 29 with Dr. Sherita Hill Golden, vice president and chief diversity officer at Johns Hopkins Medicine. For a preview of the schedule, visit https://oir.nih.gov/wals/2020-2021/sneak-peek-2021-2022-season-click-here.