Supplied
Think about the last time you felt let down by the health care system. You probably don’t have to go back far. In wealthy countries around the world, medical systems that were once robust are now crumbling. Doctors and nurses, tasked with an ever-expanding range of responsibilities, are busier than ever, which means they have less and less time for patients. In the United States, the average doctor’s appointment lasts seven minutes. In South Korea, it’s only two.
Without sufficient time and attention, patients are suffering. There are 12 million significant misdiagnoses in the U.S. every year, and 800,000 of those result in death or disability. (While the same kind of data aren’t available in Canada, similar trends are almost certainly happening here as well.)
Eric Topol is an American cardiologist, leading medical researcher, and bestselling author. In his book “Deep Medicine: How Artificial Intelligence Can Make Healthcare Human Again” Topol writes about how actual face time between doctors and patients has been reducing over the years, due to administrative tasks such as record keeping, which have to be squeezed into the same measly time slots.
Just what the doctor ordered: How AI will change medicine in the 2020s
In a recent episode of Machines Like Us, Topol joins host Taylor Owen to discuss how artificial intelligence can not only bring back the doctor-patient bond, but also predict and prevent serious diseases in a manner simply not possible before.
How do we distinguish between productivity tools that just make doctors busier versus tools that actually enable time for increased patient interaction?
Let’s use an example of an AI-powered synthetic note taker, that transcribes conversations between the physician and their patient. It makes notes that are far better than any notes that exist today, but it’s beyond that.
It also has audio links for the patient to go back and listen to, because people often forget or don’t really understand what’s discussed. So they have that transcript and audio links to listen to.
That’s just the beginning though. This virtual note taker goes even further and schedules follow-up appointments, does pre-authorization for insurance companies, and books tests and procedures that need to be set up. It can also nudge patients about the things that were discussed during the visit, such as medication and dietary reminders.
Increased efficiency may prompt medical administrators to say “oh, now [that AI tools are freeing up your time] you should see more patients, read more scans or slides.” That’s where we have to stand up against overburdening doctors even further. This has been compromised to a terrible extent, with electronic health records, which were purported to facilitate better care, ending up just being tools to improve billing efficiency.
What are some other ways in which AI is being introduced to improve health care?
My biggest aspiration is to refocus on the patient-doctor relationship. That is overarching. But the next application of this AI is alleviation of medical errors in diagnosis.
In the U.S. there’s over 12 million serious diagnostic errors made each year. According to Johns Hopkins, over 800,000 Americans either become disabled or die each year because of serious diagnostic errors. These misdiagnoses can be reduced if we employ large language models that use AI.
In the interpretations of scans:
When a radiologist is looking at a chest X-ray for signs of pneumonia in the lungs, they would not be thinking about making the diagnosis of pancreatic cancer, that’s not on the requisition form. The radiologists’ workload is so extraordinary, that they are unlikely to be looking for things that weren’t why the scans were ordered in the first place. But the AI tool may see something is wrong with the pancreas.
An AI tool that’s been trained to be very highly accurate is going through a patient’s records. It has the ability to sift through them very quickly and come up with a summary of key issues, the diagnoses, abnormalities, lab tests and the uncertainties. In fact, the machine has superhuman eyes that can spot things in scans that we can’t see.
Further, it can also review medical literature relevant to that patient’s specific illness. So, things that would take us hours to do can be done in the flash of seconds. The tool can process information, to help lead to accuracy, so things don’t get missed while making a diagnosis.
What is medical forecasting, and how could it transform advanced diagnosis of serious illnesses?
Advanced AI techniques have led to an extraordinary tool for weather forecasting called graph cast. The accuracy of the weather forecast is going to improve greatly because the tool has been trained with years worth of [meteorological] data at any locale.
AI is being harnessed in medicine in a similar manner.
There’s layers to human beings, not just what’s in an electronic health record, but their genome, their gut microbiome, their environment, their physiology through sensors, their historic lab results, etc., all these are layers of data.
If we use Alzheimer’s as an example, there’s a biomarker for Alzheimer’s that picks up the accumulation of amyloid in the brain very early. There are genes that show you are at risk for Alzheimer’s, so you could tell 20 years before any symptoms manifest that someone is at very high risk. Once tracked, you could use those 20 years to try to prevent the illness from ever happening.
We’re starting to have ways to do that now that we have never had before. For serious illnesses like cancer, cardiovascular (diseases,) Alzheimer’s, Parkinson’s, etc., we can now potentially use multimodal AI for medical forecasting, defining high risk people, tracking them, and determining the time of intervention. We’re in very early stages of using this, but this will reboot our whole approach to preventing illnesses.
Are people hesitant to learn about future risk if preventative tools or potential cures aren’t as certain?
Yes. If it isn’t actionable, there’s no good.
Nobody wants us to say, “oh, you’re doomed, and I can’t do anything about it.” They only want to know if they can prevent that serious condition. It’s not just about the forecast. It’s also about the viability of taking action to prevent that illness.
Where do you see humans being substantially better than AI. Where’s our comparative advantage?
We want to bring back the human bond in medicine. That’s the one thing that’s essential, and AI will never be able to do that.
However, the big surprise was how well AI could promote empathy and better communication.
So if we go back to that synthetic note and say, “critique the doctor” the AI might say, why did you interrupt Mrs. Jones after nine seconds? She never got to express her concern. Why didn’t you ask about this?
So AI can coach the doctor to be a better, more empathetic communicator. This has been shown now through several studies.
AI doesn’t know what empathy is, but it’s promoting it by knowing good language. So here again, we want to be as great a communicator and empathic as possible. And for the patient to know that this doctor has your back.
Will this mean we’ll need to rethink what it actually means to be a doctor?
Yes, absolutely. Right now, the criteria for medical school admission is that you need very high grades and you have a very good score on your medical college admission test, but there’s nothing about “do you have good interpersonal skills,” “do you care about people?”
So in the future, we don’t need brainiacs as much. We need people that obviously, have good judgment, are curious and want to read and keep up with their field and have good reasoning power. But knowing all what’s in the medical literature and all the things that we used to memorize, we don’t need that any more. That’s where we can lean on machines, and AI.
These new capacities could make things far better or maybe even worse. Are you concerned about what could come from this if we’re not careful?
I am worried that the efficiency that you mention could actually make things worse.
In the U.S., most health systems are run by administrators, not by clinicians, and they are charged with bringing in more revenue. So this is going to be a major tension going forward.
For instance, right now, at Emory Health System in Atlanta, the physician that’s in charge of the information technology is getting love letters from his doctors about how much time they’re saving, hours a day because of these synthetic notes. As this gets more pervasive, the administrators may say, “Oh wow, they have all these extra hours a day. I’m going to assign them more patients to see per day.”
We can’t let that happen. This is a turning point in medicine. So if we don’t stand up and say to these overlords, no, the gig’s up. This is about the care of patients.
We’ve never stood up as a physician community. We’ve just let this happen. And this time that’s not going to work because we may not get another chance like this.
Your new book is about the adoption of AI and how it can change what it means to be a doctor. These are pretty foundational critiques. How has the medical community reacted to this?
The AI book has been interesting because, for the first time, there is a large segment of the medical community that is saying, maybe this can really help us because things are getting pretty desperate. We still are in a wait-and-see period, and there’s many pieces that are not developed as they should. Hopefully we’re going to get there.
There has been a doctor-patient dynamic where doctors felt the need to control information in order to protect patients. Will the adoption of AI flip this dynamic, giving patients more power and better access to their data?
It is the patients’ data, and they should own it. The medical community is just not letting go of that data, and that’s a serious problem.
I believe until patients are granted access to their data, they can’t have the charge. For example, we rely a lot on medical portals. They only give patients the de minimis about your record, and don’t provide the raw images or a lot of what’s in there.
The problem is all of that should be the patients’. And if you have AI models that can process your data but you have inadequate, incomplete inputs, it’s only going to do a job that’s inadequate. So the unwillingness to let go is a real issue.
For patients to access all their medical data should be a civil right. We should be doing this, especially in preparation for the AI world that we’re going to be living in. That is inevitable.
You write that the current medical system is focused on curing not healing. What does it mean to rebuild a medical system that is focused on healing?
I think healing comes from the human bond between the patient and their doctor.
If doctors start interacting with patients with increased empathy, in a more human way, those with illnesses will be reassured that “the doctor really cares about me, and they are doing everything they can to get me to be better. I may not be cured, but I’m going to get over this.
“I’m going to have a much better chance of surviving without suffering.”
So we’ve got to do much better. True cures in medicines are very rare. But healing, that’s what should be the norm. It isn’t something that happens when you give somebody a pill. The healing happens between people. It’s something that’s often subconscious, you know, subliminal. But it’s essential because that’s something AI will never do, that we can do.
This interview has been edited for clarity and length.