How Artificial Intelligence Can Make Doctors More Human
Technology has helped cardiologist Eric Topol save lives. While on an airplane several years ago, a flight attendant asked if there was a doctor on board—a man was suffering from chest pain at 30,000 feet. Topol was able to obtain an electrocardiogram from the man by using a heart activity–reading gadget that attached to his smartphone, made by the medical device company AliveCor.
“It turned out to be a big anterior heart attack I could see right on my smartphone,” says Topol, director and founder of the Scripps Research Translational Institute in San Diego. “I had to tell the folks to land the plane. He wound up doing pretty well.”
Although Topol used his smartphone to conduct the ECG, it wasn’t an algorithm that led to the man’s diagnosis; it was Topol’s years of knowledge as a cardiologist. The technology could complete the test, but without Topol to interpret the results, it would have been useless. “It was me, the human algorithm,” he says.
But in late 2017 — just six years after the in-flight medical emergency — the U.S. Food and Drug Administration (FDA) approved an algorithm developed by AliveCor for the Apple Watch that no human can match. Thanks to the algorithm embedded in the watch’s band, the watch can continuously monitor your heartbeat for signs of atrial fibrillation, a common disorder that carries a high risk of stroke. It learns your heart rate at rest and when you’re active; if it detects an apparent abnormality, the watch alerts you to put your thumb on the watchband to record an ECG. (Apple has since developed a similar algorithm that is a feature in the most recent version of its watch, Apple Watch 4.)
The technological transition of the ECG from a test that requires a doctor’s interpretation to one that can be done on a wearer’s wrist is just one example Topol cites in his new book, Deep Medicine: How Artificial Intelligence Can Make Healthcare Human Again. Topol spoke to OneZero about Deep Medicine and the many ways A.I. offers an opportunity to fix a broken health care system and how it could change the medical experience for better or for worse.
The interview has been condensed and edited for clarity.
Eric Topol: We now have the capability to know the medical essence of a person. We never had the ability to do that until now. We didn’t have all these layers of data, whether it’s from your wearable sensors for your physiology, high-resolution scans, your genome, your gut microbiome, all your data from your traditional medical records. But it’s too overwhelming to make sense of it. We need help. The whole idea now is that with artificial intelligence, we can take all this data and do so much with it. This is a remarkably exciting time for health care.
The problem is that we have very little time with patients. Looking at it both from the doctor and the patient side, the whole relationship has been eroding steadily throughout the 40 years since I graduated from medical school. It’s not just the time, though that’s a big part of it. It’s also all the mistakes. It’s the inability to have all of a person’s data, the inability to keep up with all the medical literature.
Another dimension is the burnout. It’s estimated that 50 percent of clinicians, not just doctors, are in this state of burnout because medicine has lost its way. This sets up a vicious cycle, because people who burn out have double or more risk of making medical errors. But I believe the biggest problem here is the combination of a lack of deep understanding of each human being and a lack of the deep empathy that is necessary to build a trusting and caring patient-doctor relationship. Those are the two fundamental issues. If we can fix those — which I think we can with deep learning and A.I. — then we can get back on track.
Well, that’s certainly the whole paradox. Up until now, technology has taken us south in terms of our humanity. Electronic health records have broken the backs of clinicians and made them into data clerks. So why would anyone in their right mind think that we could have a rescue through technology? I went into this whole multiyear book project without this notion, but I became acutely aware that this is our best shot ever. That’s because we’ve never had a technology that could actually give us the gift of time. A.I. would markedly augment human performance, doctor performance, clinician performance, and at the same time empower consumers and patients to take charge of their health.
There’s one big caveat, and that is it can make things worse. The administrators, the bean counters, say, “See more patients, read more scans, read more slides.” That could get worse if you’re more productive with machine support. “Oh, you read 50 scans a day? Well, good, now you can read 400.” We can’t let that happen, and that’s going to take unprecedented activism in the medical community. It’s vital that this time we get it right, because we’ll likely never have another opportunity like this, and we’ve got to seize it.
The only sure-fire way we can get A.I. broadly implemented is through rigorous research that provides unequivocal proof of benefit for patients. It’s early in the A.I. medical era, so naturally there is a real paucity of peer-reviewed, prospective trials with independent replication of outcome improvement. In fact, there isn’t even one that fulfills all that now. Most of what we have so far is retrospective “in silico” work from large data sets that has yet to be taken to real-world medical practice. There can’t be any exceptionalism for a strong evidence basis required for A.I. to change health care. Many will argue that the bar should even be higher because of the potential harm for algorithms to be rapidly applied to patients at scale.
It’s small, but we live in a world with pervasive hacking. Beyond that, we’re all familiar with the occurrence of software glitches. We need to be mindful of this possibility, albeit small or remote. Time will tell, but it’s certainly one of the things I’m worried about.
Well, this is a silly notion that we’re not going to need doctors, that people are just going to get treated by algorithms. That’s absurd. Machines don’t have judgment; they don’t have context. They have an insatiable hunger for data, whereas people have early satiety with data. So we want to basically offload this analytical side. But you don’t want to entrust a machine to make a major decision about a diagnosis, a treatment, or a surgery without human oversight.
We can have algorithms that go off the track. They get hacked. They have glitches in them that are unforeseen. They don’t perform like they’re supposed to. These algorithms that are going to be central as we go forward, they’ve got to have a backup. So I strongly refute the notion that we’re going to have a lack of need for clinicians.
In cardiology, there’s so much room for this. The first thing is keyboard liberation: no screen and no keyboard in the patient encounter. It’s actually a real human bonding experience, intimate, with trust and presence. That’s step one. Second, tests like the cardiogram or echocardiogram would be machine-read — and not like the pathetically error-laden machine-read electrocardiograms we’ve had since the 1980s. But with deep learning, it’s not just about fixing up the error problem. We’re going to learn all these things that we didn’t know before.
With deep learning, the machines can see what we could never see, because you feed in a million Mayo Clinic electrocardiograms linked to the “ground truths,” the actual details of what the patients had, and all of a sudden it’s ridiculously intelligent.
The other big thing in cardiology is patients generating their own data. Rather than send people for a lab test or a one-off blood pressure reading during an office visit, we just tell them to use their smartwatch, or we send them a sensor bandage to monitor their heart rate or manage their glucose. So people are much more engaged and have data that’s being processed through algorithms in their real world, and they work with it.
We’ve had these food pyramids and national guidelines on what you should eat, never based on any real science but on the simplistic notion that the same diet is right for all people. That was really stupid. The food diaries and the way people have done these studies all these years are so inaccurate.
Now, with A.I., we’re starting to figure out the individualized diet. Machine learning, for instance, can help predict a person’s unique glycemic response to their diet. Or it can integrate what a person eats with data about the time they eat, their level of physical activity, and their medications, lab tests, genome sequence, and gut microbiome. It’s just beginning — we don’t have all the data in. But the nutrition science world will be upended, because it was not really science—it was kind of a stab in the dark.
When I ask a group of people, “Have you been roughed up by health care in the U.S. today?” hardly anybody says no. We’ve got to stop that roughing up. We’ve got to use the future to bring back the past and get to the point where we have time and reflection and the human bond. The deterioration of the doctor-patient relationship is so fundamental to what ails us. I’m trying to be optimistic. I know full well this is going to take a while. This is as powerful a technology that we’ll likely see for fixing what ails health care in our lifetime.
How Artificial Intelligence Can Make Doctors More Human
Research & References of How Artificial Intelligence Can Make Doctors More Human|A&C Accounting And Tax Services