AI alone is not the answer

With their state-of-the-art facilities and histories of groundbreaking clinical advances, American medical schools have built reputations as global leaders in scientific innovation – but does that reputation match reality?

In one episode Improved Healthcare Podcasts recorded in 2023, Deep Medicine Author Eric Topol points out a major oversight in medical education.

“It’s pretty embarrassing,” he said. “You look at 150 medical schools and not one of them has AI as a core part of their curriculum.”

A year later, most U.S. medical schools have responded by incorporating AI into their programs. But a closer look reveals that most syllabus — a mix of theoretical applications, ethical considerations, and using AI to streamline routine tasks (billing, coding, charting) — falls short in equipping future physicians with the training they need to improve care and save lives.

Medical education is failing patients and doctors

In 1910, education reformer Abraham Flexner published his groundbreaking indictment of American medical schools. His controversial findings revealed serious flaws in the training of future physicians.

The Flexner Report led to the closure of approximately half of the nation’s medical schools and a restructuring of medical education to create more scientifically rigorous and clinically relevant curricula. Flexner’s goal was not only to standardize clinical training but also to stop thousands of unnecessary deaths caused by substandard medical practice.

More than a century later, American medicine faces a similar crisis: hundreds of thousands of people die each year from largely preventable chronic diseases, misdiagnosis, medical errors, and a lack of research.

Having researched and written extensively about the future of generative AI in healthcare, I believe this technology can revolutionize healthcare—but this will only happen if medical schools fundamentally redesign their curricula to teach students how to use GenAI to fundamentally improve healthcare practices and processes.

The time to reinvent medical education is now. Here are three GenAI-enabled opportunities students should master to improve clinical outcomes and save patient lives.

1. Chronic disease management: from episodic to ongoing

Traditional medical education focuses on students memorizing tens of thousands of facts and memorizing diagnostic and treatment “algorithms.”

By the time they graduate, doctors are expected to apply these memorized facts and algorithms to help patients manage chronic diseases and prevent complications. Clinicians are taught the steps to follow, including which medications to prescribe, what lifestyle changes to recommend, and how to schedule regular follow-up appointments three to four months later.

The problem with this “standard” episodic approach is that it doesn’t provide doctors with actionable data between visits. Lack of continuous monitoring leads to problems such as:

• Medication adjustments are delayed.
• Inconsistent adherence to treatment plan.
• The disease goes unnoticed until the next appointment and is poorly controlled.

Chronic diseases such as diabetes and high blood pressure currently affect six in 10 Americans and are responsible for 1.7 million deaths each year from heart attacks, strokes, cancer and other complications.

These deaths are directly related to a lack of prevention and effective disease management. Currently, high blood pressure is the leading cause of stroke, yet only 55% of cases are properly controlled. Diabetes, the leading cause of kidney failure and the leading cause of cardiovascular disease, is even less well controlled. We know that with best practices we can achieve control rates of 90% or more, but current approaches just can’t.

According to the CDC, 30% to 50% of life-threatening complications from chronic diseases could be avoided with effective management. Teaching medical students how to use generative AI to provide continuous rather than intermittent monitoring would dramatically improve the health of patients and the nation as a whole.

Doctors today have access to wearable monitors that can measure blood pressure, blood glucose levels, etc. By integrating these tools with GenAI, patients’ health data can be robustly analyzed and medical advice can be provided based on the expectations set by the clinician.

This combination takes the guesswork out of whether a doctor’s appointment is necessary – the patient knows – and their expertise allows doctors to intervene sooner when a problem occurs, while reducing unnecessary visits when chronic conditions are well managed.

According to CDC data, successful training of the next generation of physicians to effectively monitor and manage chronic diseases is estimated to save 510,000 to 850,000 lives each year and reduce healthcare costs by $163 billion to $272 billion annually.

2. Diagnosis: From confirmation bias to continuous second opinions

In the classroom and during clinical training, medical students are still taught to rely on memory to confirm diagnoses and recommend optimal treatments.

Often in the hustle and bustle of clinical practice, doctors fall prey to human cognitive biases and make frequent unintentional mistakes. Misdiagnosis leads to 400,000 American deaths each year.

American doctors are smart, skilled, and dedicated to their patients. But mistakes happen. GenAI gives doctors the opportunity to reexamine their assumptions and reduce the risk of mistakes — at no extra cost to them.

AI can analyze vast amounts of patient data, including symptoms, medical history, and diagnostic test results. And computer applications are free of cognitive errors such as confirmation, overconfidence, and proximity bias. While GenAI is not perfect, the technology serves as a valuable tool to complement human analysis. And because it can compare patient data to a large, comprehensive database of known diseases and conditions, it can identify diagnoses that doctors may miss.

Already, AI has shown great potential to reduce misdiagnosis in emergency rooms. A recent study evaluated an AI’s ability to triage patients and found that it performed as well as doctors and nurses in accurately identifying high-risk patients. A second study evaluated the diagnostic accuracy of an AI based on patient symptoms and test results, and found that the AI ​​was consistently more accurate than doctors at correctly identifying likely diagnoses.

And the technology is only getting better: Experts predict that by the time medical students entering today complete their fellowship programs in 10 years, generative AI will be 1,000 times more powerful.

3. Research: From human hypotheses to data mining

Clinical research is the foundation behind medical advances, providing doctors with the information they need to improve health and save lives. In school, doctors learn how to design research studies, analyze data, and write papers.

When researchers today create a study, they pose a clinical question, extract information from medical records, and perform statistical analysis. GenAI provides an opportunity to reverse engineer this traditional approach.

GenAI now allows physicians to analyze massive amounts of data generated by bedside monitors, surgical robots, and other digital sources. Currently, U.S. hospitals create 50 petabytes of data each year, 97% of which goes unused. That’s the equivalent of all works published by humanity in any language since the beginning of recorded history. All of this data is currently excluded from clinical research because the sheer volume exceeds researchers’ ability to analyze patterns embedded in the data, making it impossible to separate the signal from the noise.

By analyzing this data with GenAI models, doctors can advance medical knowledge much faster than they can today. The technology can help clinicians accurately predict which conditions in hospitalized patients will worsen in the next 24 hours and take preventative measures. Oncologists can determine the optimal chemotherapy dosage with fewer complications. Surgeons can identify the best surgical technique to remove cancer. Leveraging GenAI to analyze data can quickly answer questions that would take years to answer.

Imagine if researchers from dozens of academic institutions agreed to pool and share monitoring and patient data. With this wealth of information, dozens of researchers could simultaneously access and study this data. Instead of asking narrow, specific questions and digging through clinical information to answer them, scientists could address bigger questions and advance clinical practice in a fraction of the time.

This approach of pre-loading vast amounts of data and letting the technology organize and analyze it reflects the way GenAI tools are designed, and would be a radical departure from traditional research efforts.

But before that can happen, physician-researchers need to be trained in new data analysis methods and equipped with the interpersonal tools to enhance collaboration and cooperation. These skills, often taught in business schools, are easily transferable to medical students.

The next era of healthcare is upon us, and the need to act is clear. Academic medical centers must not only incorporate generative AI into their core curriculum, but also teach the next generation of clinicians how to use this technology to fundamentally improve chronic disease management, diagnostic accuracy, clinical research, and many other outdated healthcare processes. Classes are scheduled to begin this fall, so the time to act is now.