Open this photo in gallery:
Think about the last time your health care system let you down. It probably wasn't that long ago. In wealthy countries around the world, once-robust health care systems are now crumbling. Doctors and nurses are taking on broader responsibilities, busier than ever, and have less and less time to devote to patients. In the United States, the average doctor visit lasts seven minutes. In South Korea, it's just two minutes.
Without enough time and attention, patients suffer: There are 12 million serious misdiagnoses each year in the United States, 800,000 of which result in death or disability. (Similar data is not available for Canada, but similar trends are almost certainly occurring there.)
Eric Topol is an American cardiologist, leading medical researcher, and best-selling author. In his book, Deep Medicine: How Artificial Intelligence Is Rehumanizing Healthcare, Topol writes about how face-to-face time between doctors and patients has decreased over the years as administrative tasks like record-keeping have to be crammed into the same tiny time slot.
Just what the doctor ordered: How AI will transform healthcare in the 2020s
In a recent episode of “Machines Like Us,” Topol joined host Taylor Owen to discuss how artificial intelligence can not only restore the doctor-patient connection, but also predict and prevent serious illnesses in ways never before possible.
How do we distinguish between productivity tools that only make physicians busier and those that actually free up time for more patient interactions?
Take the example of an AI-powered synthetic note-taker that transcribes doctor-patient conversations: This will produce better notes than anything that exists today, but it will go beyond that.
We also have an audio link for patients to go back and listen to because often times people forget or don't quite understand what was discussed, so patients can listen with the transcript and the audio link.
But that's just the beginning: the virtual note-taking tool goes further to schedule follow-up appointments, obtain pre-authorizations from insurance companies, and schedule any necessary tests or procedures. It can also remind patients about medications or dietary precautions that were discussed during the appointment.
With increased efficiency, healthcare administrators can now say, “Oh, now [that AI tools are freeing up your time] “See more patients, read more scans and slides.” We must confront the increasing overburdening of physicians, which has been so badly compromised that the electronic medical record, which was supposed to facilitate better care, has become nothing more than a tool to improve billing efficiency.
What are some other ways AI is being deployed to improve healthcare?
My biggest goal is to refocus the patient-doctor relationship, which is the overarching goal, but the next application of this AI is going to be reducing medical errors in diagnostics.
Over 12 million serious diagnostic errors occur each year in the U.S. According to Johns Hopkins University, over 800,000 Americans are disabled or killed each year due to serious diagnostic errors. Deploying large-scale language models using AI can help reduce these misdiagnoses.
In interpreting the scan:
When a radiologist is looking at a chest X-ray looking for signs of pneumonia, they probably won't think about making a diagnosis of pancreatic cancer because that's not listed on the job description. Radiologists have such a huge workload that they're unlikely to be looking for something that's not why they ordered the scan in the first place. But an AI tool might notice that there's something abnormal with the pancreas.
AI tools trained to be highly accurate are examining patient records. They have the ability to sift through records very quickly and create an overview of key issues, diagnoses, abnormalities, lab tests, and uncertainties. In fact, these machines have superhuman eyes and can spot things in scans that humans cannot see.
It can also look up medical literature related to a patient's specific illness, so what would take us hours to do can now be completed in a matter of seconds. The tool processes the information to improve accuracy and ensure nothing falls through the cracks when it comes to diagnosis.
What is medical predictive medicine and how can it transform the advanced diagnosis of serious diseases?
Advanced AI techniques have given birth to an amazing tool for weather forecasting called GraphCast, which is trained on years of data and will greatly improve the accuracy of weather forecasts. [meteorological] Data for any locale.
AI is being used in the medical field as well.
Humans have layers, and not just the contents of our electronic health records, but our genomes, our gut microbiomes, our environment, our physiology from sensors, our past test results, everything is a layer of data.
Take Alzheimer's disease for example. We have biomarkers for Alzheimer's disease that detect amyloid buildup in the brain at a very early stage. We have genes that indicate risk for Alzheimer's disease, so we know that person is at very high risk 20 years before symptoms appear. If we can track them, we can use those 20 years to prevent the onset of the disease.
This is now possible in ways that have never been possible before. For serious illnesses such as cancer, cardiovascular disease, Alzheimer's, and Parkinson's, multimodal AI could be used to make medical predictions, identify at-risk individuals, track them, and determine when to intervene. While this is still in its early stages of use, it could revolutionize our entire approach to disease prevention.
Are people hesitant to learn about future risks when there is no certainty about preventative measures or potential treatments?
Yes, if it's not actionable it's pointless.
Nobody wants to be told, “It's too late, there's nothing we can do.” All they want to know is whether or not they can prevent that serious condition. It's not just about prediction, it's also about whether it's feasible to take action to prevent the disease.
In what ways do you think humans are significantly better than AI? Where do we have a comparative advantage?
We want to bring the human connection back to healthcare, that's what's most important, and AI can never do that.
But the big surprise was how much AI can foster empathy and better communication.
So if you go back to that composite note and say, “Criticize the doctor,” the AI ​​might say, “Why did you interrupt Mrs. Jones after nine seconds? She didn't have a chance to express her concerns. Why didn't you ask about this?”
In other words, AI can coach doctors to communicate better and more empathetically, as several studies have demonstrated.
AI doesn't know what empathy is, but it fosters empathy by knowing the right language. Again, we want to be as communicative and empathetic as possible, and let the patient know that this doctor is on their side.
Does this mean we need to rethink what it really means to be a doctor?
Yes, of course. Currently, the standard for getting into medical school is very high grades and doing very well on the medical entrance exam, but it doesn't take into account “good interpersonal skills” or “caring about others.”
So in the future, we won't need as many brilliant people. We will need people who, of course, have good judgment, are curious, are knowledgeable in their field, up-to-date, and have good reasoning skills. But we won't need to know the contents of the medical literature or all the things that we used to memorize. That's where we can turn to machines and AI.
These new abilities could significantly improve the situation or make it worse. Are you worried about what could happen from this if you're not careful?
I worry that the efficiencies you speak of will make things worse.
In the United States, most health systems are run by administrators, not physicians, and they are responsible for bringing in more revenue, so this is going to be a big tension going forward.
For example, right now at Emory Health System in Atlanta, our information technology physicians are getting love letters from doctors saying that these composite notes are saving them hours a day. If this gets more widespread, administrators might say, “Wow, we have so many more hours in our day. Let's allocate more patients to see in a day.”
We can't allow that to happen. This is a tipping point in medicine. And it's going to end unless we stand up and say to these rulers, “No, we're done. This is a patient care issue.”
We have never stood up as a physician community, we have just let this happen, and this time, it's not going to work because we may never get a chance like this again.
Your new book is about the introduction of AI and how it will change what it means to be a doctor. These are some pretty basic criticisms. How has the medical community responded to this?
The AI ​​book is interesting because for the first time we have a big part of the medical community saying this could really help because the situation is pretty dire. We're still in the wait-and-see phase and there are a lot of pieces that aren't developed as they should be. Hopefully we'll get there.
There has been a power dynamic between doctors and patients where doctors have felt the need to control information to protect their patients. Will the introduction of AI reverse this dynamic, giving patients more power and access to data?
It's patient data and it should be owned by the patient. The medical community is reluctant to give up that data, and this is a real problem.
I believe that until patients have access to their own data, they cannot be charged for it. For example, we rely heavily on medical portals that give patients minimal information about their records, and do not provide raw images or much of what is recorded there.
The problem is that all of that should be the patient's responsibility, and even if you have an AI model that can process the data, if the inputs are poor and incomplete, it can only do a poor job. So the reluctance to let go is a real issue.
It should be a civil right for patients to have access to all of their medical data. We should do this, especially as we prepare for the AI ​​world we will be living in. It's inevitable.
You write that our current health care system is focused on treatment, not cure. What does it mean to restructure our health care system to focus on healing?
I believe healing comes from the human connection between patient and doctor.
If doctors treated their patients with more human empathy, sick patients would feel more reassured, thinking, “My doctor really cares about me, and he's doing everything he can to help me get better. He may not be cured, but he can get through this.”
“You'll have a much better chance of surviving without suffering.”
So we have to try harder. It's very rare for a drug to actually cure something. But healing should be the norm. You don't just give someone a drug and they're cured. Healing happens between people. It's often unconscious, subliminal. But we can do that because AI can never do that.
This interview has been edited for clarity and length.
