Deep Medicine: How Artificial Intelligence Can Make Healthcare Human Again

Image of Deep Medicine: How Artificial Intelligence Can Make Healthcare Human Again
Release Date: 
March 12, 2019
Basic Books
Reviewed by: 

“if the United States wants to cut health care costs, improve outcomes, and help patients take more ownership of their health, smart, AI-enabled medicine will need to be a top priority.”

Anyone who experiences that U.S. healthcare system will find a lot to agree with in Dr. Eric Topol’s Deep Medicine. Topol, a cardiologist, medical researcher, and author, argues that medicine as it is practiced today is “shallow” rather than deep, and that accounts for many of its problems. By deep, Topol is referring to more meaningful encounters between clinician and patient, rather than the seven minutes a doctor spends on average with a repeat patient.

To get to a world of “deep medicine,” Topol argues that artificial intelligence (AI) will play a key role. He writes that “the promise of artificial intelligence in medicine is to provide composite, panoramic views of individual’s medical data; to improve decision making, to avoid errors such as misdiagnose and unnecessary procedures, to help in ordering and interpretation of appropriate tests; and to recommend treatment.”

All, to be sure, tall orders; but ones in dire need of happening. If the reader should take one lesson away from this book, it is don’t place complete trust in your doctor, for most are prone to make avoidable errors. Topol not only describes many cases he has seen in his clinical practice where clinicians made errors, he reviews the medical literature.

For example, 30% to 50% of CT scans in the United States are unnecessary. One study of doctors found that of the ones who were 100 percent certain of the accuracy of their diagnosis before a patient’s death, 40% were wrong.

A Mayo clinic study found that only 12 percent of second opinions from the clinic agreed with the original diagnosis. Some of these mistakes are because clinicians are overworked and have limited time to really study particular cases. But some of these mistakes stem from the fact that health care is so complicated. As Topol notes, there are over 10,000 different diseases and not even the best doctor “who could recall a fraction of them.”

This is where AI could come in. It is possible that AI tools could understand all diseases, including symptoms and cutting-edge treatments. But while Topol is optimistic about the promise of AI, he is not Pollyannaish, and in fact argues that for the most part AI is not ready for broad applications like diagnosis of diseases and recommendations of treatment.

What AI can do extremely well are narrow applications, such as reading chest X-rays and eye scans and with a high degree of accuracy. This is because todays AI is often very good at pattern recognition. But this does not mean that AI systems could not improve to be able to be better than the best doctors in making diagnoses.

In fact, AI is getting better and its use is broadening. For example, a startup company in California was able to use AI to understand the relationship between patients’ ECG tests to the amount of potassium in the blood in order to be able to identify patients with dangerously high potassium levels.

AI is also being used to help develop new drugs. In 2018 there were more than 60 startups and 16 pharmaceutical companies in the United States using AI for drug discovery. Given that the number of possible molecular compounds is virtually infinite, using AI to narrow areas for human inquiry and testing is likely to be an important tool.

One of the most interesting applications of AI is to personal medical assistants. With wearables like the Apple watch and Fitbit getting better and now specialty devices coming on the market that use AI, individuals are increasingly able to collect real time health information. Combining that with medical records, including lab results, can mean that AI-enabled personal assistants can give advice, including being virtual real time health coaches. Topol is right to argue that individuals should have a right to all their health data in machine readable forms.

Perhaps the most disappointing aspect of Deep Medicine is that Topol completely ignores the role of government in bringing about this change. In example after example, Topol stresses the importance of having robust data sets to train AI algorithms on. Yet this is something federal policy has a clear role in, particularly how the Health Insurance Portability and Accountability Act (HIPAA)—those annoying forms you fill out when you have to go to a doctor) make data sharing extremely difficult.

Finally, while Topol is optimistic about AI in improving health care, he argues that its greatest benefit is to enable deeper and more sustained relationships between patient and doctor. But such a change would appear to cost money. If AI does much of the routine work for doctors, why would society not use those savings to reduce the cost of health care instead of increasing the time doctors spend time with patients? Having more time with patients is certainly nice; its why there has been a big increase in concierge medicine for upper income individuals where they pay significant annual fees in order to experience “deep medicine.”

But society as a whole cannot afford deep medicine, especially as more and more of the population becomes aged. However, what is clear is that if the United States wants to cut health care costs, improve outcomes, and help patients take more ownership of their health, smart, AI-enabled medicine will need to be a top priority.