The Top 3 Use Cases for Generative AI in Healthcare

Could generative AI revolutionize healthcare as we know it? Technologists have long predicted AI will transform the massive and complex medical industry—only to have actual products fall short—remember IBM's Watson, the AI who conquered Jeopardy in 2011 but sputtered in the doctor's office?

Last week, in an A.Team webinar titled "Demystifying AI in Healthcare," a panel of experts parsed generative AI's medical prospects, discussing the new technology's role in everything from personalized medicine to drug discovery to the tedium of medical red tape, distilling the imminent use cases from the moonshots.

The three panelists—Ed Kopetsky, the former chief information officer at Stanford Children's Health and founding member of the College of Healthcare Information Management Executives; Ohad Zadok, co-founder and CTO of Alike.Health, and Mida Pezeshkian, founder of consultancy STEMA_cg—emphasized one central tension: Generative AI can improve doctors' quality of life, the drugs they prescribe, and the care they provide patients—but only if industry leaders and regulators find ways to navigate the thorny legal and compliance challenges.

"How are we going to protect patient data? That's the million-dollar question," said Kopetsky.

The stakes are enormous. Here are the most feasible—and most urgent—near-term targets for those implementing AI in health systems:  

Personalizing primary care

Personalized medicine is the holy grail of primary care. After all, our genes, our personal and familial histories, even the way our bodies break down drugs—these all differ from person to person. Yet with few exceptions (think Dr. Kevin O'Connor tending personally to President Joe Biden), doctors rarely provide such personalized care. It simply isn't feasible: The research is time-consuming, and the relevant data is scattered across many years of medical records.

Up to 80% of patient data is unstructured, Kopetsky estimated, and often gets lost or overlooked. Even experts get neglected, he said: Kopetsky's father died of kidney disease at age 42. "I've been treated for 14 years, and none of that has been factored into my treatment," he said.

How are we going to protect patient data? That's the million-dollar question.

Large language models can rapidly digest and interpret such unstructured information, helping overloaded doctors provide personalized medicine to patients even if their face-to-face time averages just eight minutes.

"We have a huge store of data that can be used way more optimally—almost magically—in 30 seconds," said Kopetsky. "Today's noise is tomorrow's critical information." 

Reducing administrative burden

Clinicians are often swamped with administrative duties that eat into their valuable time with patients. AI is already being developed to ease this burden. Amazon's HealthScribe and the partnership between Epic Systems and Microsoft's Nuance Communication lead the way with so-called Ambient Documentation, which listens to doctors' interactions with patients, makes notes, and even suggests diagnoses. AI allows doctors to refocus on patient care by taking over these routine tasks.

Goal number one, according to Ohad Zadok, is helping practitioners "spend less time in front of a computer and more time in front of patients."

Streamlining drug discovery

"Better prediction and prevention of diseases" was what the webinar's audience overwhelmingly found the most exciting potential of AI—with over 50% giving that answer in a mid-discussion survey question that asked, "What aspect of your healthcare experience do you think could be most improved with AI?"

Drug discovery holds the most exciting potential, agreed the panelists. Just as large language models could help doctors sift through unstructured personal data on behalf of their patients, AI tools can also analyze vast and complex data sets connected to pharmaceutical trials, identifying potential drug targets and predicting their interactions with possible medications in new ways. This analysis could lead to the rediscovery of many such "difficult, orphan, and rare drugs and therapeutics," noted Mida Pezeshkian—and at rapid speed.

"You can achieve things 5 to 10 times faster than traditional methods,” Zadok said.

"A lot of data scientists will need to look for a job," said Zadok. "AI cannot replace doctors, though, at least for now," he added with a chuckle. 

That All Sounds Great—But What About HIPAA?

While generative AI promises numerous benefits in healthcare, it's not without its challenges. One of the critical hurdles is ensuring data privacy and proper data sharing. Patient's medical data is sensitive and highly personal, and any use of it needs to respect privacy laws, ethical considerations, and the patients' rights. AI systems need robust safeguards to prevent unauthorized access and misuse, Kopetsky noted. There's also the challenge of regulatory restrictions on clinicians using only the information they “need to know.”

Today's noise is tomorrow's critical information.

While AI could potentially provide a wealth of patient data, doctors and nurses must be careful not to violate these regulations. Plus, public perception of AI can be a stumbling block. Many view AI as "doomsday scenarios," said Pezeshkian—with Big Brother invading their privacy and controlling their healthcare. This perception could hinder the adoption of AI in healthcare. The healthcare industry and AI developers must address these concerns, showing that AI can enhance care, not compromise it.

"There's a fear of the unknown," Pezeshkian said. "But the more we talk to each other, the more fear goes down, and our ability to execute goes up," she added. 

As with all healthcare IT, the key to success, stressed Pezeshkian, is always the same: New technology must help real people with real problems on their terms.