All Docs Need More Tech Training

Artificial intelligence (AI) is already changing healthcare, but data science training for clinicians and medical students is still largely elective. The vast majority of doctors believe that needs to change.

According to a soon-to-be-published Medscape report, AI Adoption in Healthcare, 85% of practicing physicians agreed: The use of AI in medicine will require significant changes in medical education and training.

And experts interviewed for this story concur. Wielding AI-assisted tools well will require a massive educational push in medical school and beyond. To steward this new technology, maintain their clinical judgment, and protect patients, physicians will need more training on how AI is built and the data it feeds on.

A doctor that can man the machine is in everyone’s best interest.

Keeping Doctors in Control

More AI education is crucial for doctors to maintain their sense of autonomy when using AI, Shauna Overgaard, PhD, senior director for AI Strategy & Frameworks, Mayo Clinic Center for Digital Health in Rochester, Minnesota, told Medscape Medical News.

“It’s never about just trusting AI,” Overgaard said. “We want physicians to maintain their judgment, their empathy, their perception of control — all things that contribute to [their] skepticism.”

Because AI is not a silver bullet, it has the potential to make significant improvements in healthcare, but it will also make mistakes. Patients and the healthcare system need doctors who know how to use AI to their benefit but also know when to be critical of its solutions.

Just as physicians are taught to scrutinize a study’s methods and statistics or evaluate a new drug for their patient, they’ll need the skills to be skeptical of an algorithm, Overgaard said. Does the AI’s output make sense? What data were used to form the prediction? For which patients does the algorithm perform well, and where is there a lack of data?

“Clinicians don’t need to outsmart the system,” Overgaard added, “but they do need to know when to call BS.”

At the end of the day, the care decisions are still the physicians to make. “We can’t put in the chart, AI told us to make this decision,” said Oren Mechanic, MD, MPH, a private practice physician in Miami and a health tech advisor.

To strike this delicate balance of AI collaboration and criticism, everyone needs more training.

A Data Science Foundation

Ideally, the foundation of AI education would be laid early in undergraduate medical training. “I see data science becoming a core part of medical school curriculum,” said Peter Steel, MD, emergency medicine physician and AI researcher at Weill Cornell Medical School in New York City. Doctors-in-training will get a fundamental understanding of AI. They’ll learn basics like data analytics, machine learning, limitations, and bias, Steel said.

Once students have a solid understanding of the architecture that drives AI, they’ll be able to “interpret its outputs critically and crosscheck with their own clinical judgment,” Steel said. With such training they’ll be more likely to spot bias, recognize hallucinations, and know what to do when the AI disagrees with their own clinical decisions.

It could take as little as one course to lay the foundation, Mechanic added. Then, as students engage with AI in different training and clinical settings, their data science skills will continually be refined.

Doctors will also need training on how to deliver AI-based insights to patients, said Steel. They’ll need to be able to explain where the AI output came from and how reliable it is. Sometimes, these solutions can be very complex and not entirely explainable, he added. So, physicians will need the skills to engender trust.

To date, data science training has mostly been “elective offerings largely driven by interested students,” Kim Lomis, MD, vice president of Medical Education Innovations at the American Medical Association, said in an American Medical Association interview. Lomis noted that it has “now become urgent” that medical leadership put AI training content into the required curriculum.

However, the prospect of adding emerging AI content may challenge educators: The content is new to everyone, including leadership. “Many sites feel that they don’t have the expertise,” Lomis said in the interview. And medical schools are already tasked with teaching students an enormous amount of curriculum in just 4 years.

The answer will likely come outside of medical school. Steel said that to meet students’ educational needs, medical schools will likely need to take on or partner with advanced data science faculty.

Training Across the Medical Spectrum

Trainees aren’t the only ones who need education on using AI. The entire clinical workforce needs a better understanding of how to offer AI-augmented, clinician-led care, experts say.

Much like undergraduate medical education, graduate and postgraduate training in AI is largely elective. Several institutions like Mayo Clinic and University of Louisville, Louisville, Kentucky, offer master’s programs in AI specifically for medical professionals.

But “that may not be helpful for everyone,” Overgaard said. Those degrees are more concentrated on doctors planning to direct an AI program in their specialty, she said.

A number of institutions, like Stanford, Dartmouth, Harvard, and MIT, offer short-term courses or certificates in healthcare AI. These may be good options for physicians who want a deeper understanding of AI tools — especially early on.

But all of our experts agree that outside education won’t be a requirement. Healthcare systems will ultimately be responsible for educating their workforce as they test and launch AI tools.

If the workforce is apprehensive or feels underequipped, they won’t adopt the AI no matter how good it is, Steel said. Healthcare systems will have to invest in equipping their frontline workforce if they want to see these big AI purchases pay off.

Overgaard added, “Many clinicians I’ve spoken to are gaining the education right now already.” Mayo [Clinic] has already invested pretty heavily in AI education, building it into some of their training programs and forming partnerships with developers like Google.

And mid-size and smaller systems will likely benefit from a trickle-down effect. Larger institutions will test and implement AI for a myriad of healthcare use cases and pass along what works. “We have a responsibility as this organization with an incredible amount of funding and intellectual capacity not just to serve ourselves,” Overgaard said. “…but set up systems, make recommendations, and create frameworks other [health] systems can adopt and improve upon.”

Donavyn Coffey is a Kentucky-based journalist reporting on healthcare, the environment, and anything that affects the way we eat. She has a master’s degree from NYU’s Arthur L. Carter Journalism Institute and a master’s in molecular nutrition from Aarhus University in Aarhus, Denmark. You can see more of her work in Wired, Teen Vogue, Scientific American, and elsewhere.