Interview with Dr. Eric Topol, digital health leader, physician-scientist, founder and director of the Scripps Research Translational Institute, and author of the book “Deep Medicine. How Artificial Intelligence Can Make Healthcare Human Again”.
Dr. Topol, you said that a visit to the doctor these days is mechanical and robotic, that medicine is broken. However, is digital healthcare itself an answer to such healthcare challenges as low-quality care, long waiting times, rising costs, aging populations, or the rising burden of non-communicable diseases?
To be clear, I’m not suggesting that we dehumanize medicine any further, by making it virtual. But what I am suggesting is there are many things that we can off-load from the current standard and tradition, so that when a doctor and patient get together, it is for something urgent.
For example, concerns like a urinary tract infection, an ear infection in a child, a skin rash, or many other things that are not serious – these could be handled by the people concerned, or by the parents if it is a child, without seeing a doctor at all, just using a virtual connection. This mechanism is one way to remove some of the overwhelming workloads from doctors. Overwhelming due to the mismatches, for all the reasons you described, such as the aging population.
However, this is only one part of the solution, there are many others. Doctors are still heavy keyboard users, even though we thought that AI would record the voice and transcribe it directly into notes. Liberation from the keyboard! We also see doctors toiling over charts to gather enough information about the patient, which could be done for them using suitable algorithms. These are some ways we can outsource this burden to machines, so that every visit to the doctor takes less time and is less prone to error, thus improving the whole relationship. In this way, we could use digital tools to restore humanity in medicine.
Will AI give doctors the gift of time? Right now, many doctors are frustrated as they have even more work in this era of computers, as they spend time clicking instead of talking to the patient.
The gift of time can be illusory because it can make things worse. Improving productivity and efficiency can also encourage administrators to think: “Oh, good! Now you can see more patients and read more scans.” They then squeeze doctors more, making things worse. Already more than half of the doctors face burnout because they are unable to achieve their mission. This burning out of doctors is accompanied by a doubling of the medical error rates, and today about 20% of doctors suffer from clinical depression. This is a problem because administrators have already squeezed doctors to an unacceptable extent. It could go the other way, where doctors stand up and say: “No, you cannot do this to us anymore, or to our patients. We went into medicine to spend time with our patients, to provide care, to have a presence, to use our hands for the examination, to be able to think, to be able to care for patients with empathy, to have time to discuss things with them.” All this has been lost to a large degree, because of the squeeze.
So the gift of time is to outsource enough to machines to help improve the error rates and the bureaucracy, and to give patients the ability to work with their own data; to make their own diagnosis for non-serious matters, and to manage their personal data for things that can be validated and proven to be helpful. Many types of digital AI capabilities could get us back to where we were, decades ago, before medicine became big business.
The subtitle of your latest book is “How artificial intelligence can make healthcare human again.” Why do you believe in the power of AI in medicine? What threats do you see?
The biggest threat is that this highly significant opportunity to restore the patient-doctor relationship could be lost. And it is conceivable that it could get worse. Today there is an ever-increasing number of threats: the risk of malware hacking, insecure privacy, and the breaching of algorithms and data; the danger of bias embedded in the algorithms, whether it is gender, racial or some other bias. There are many ethical issues, concerns about the scale of AI use or whether an AI algorithm could hurt many people quickly if it lacked surveillance, or that there is a lack of validation trials of the post-spectrum to prove the value of AI. There are many things, and all of them are, I believe, solvable over time. My biggest concern is the vulnerability of the medical community to using the power of AI to make things worse.
A few digital trends have dominated the health market, for example, wearables, mobile apps, digital therapeutics. Which of these seem overhyped in your opinion?
I think that most overhyped has been the prediction side. Classification and interpretation, the strengths that AI brings, has been especially noteworthy for images, whether they are medical scans, pathology slides, or even skin lesions, anywhere there is a pattern. This is deep learning. The sweet spot. The one where it has been much better at predicting outcomes, like short-time survival, hospital re-admission, length of stay in hospital and predicting conditions, such as Alzheimer’s. These studies are done using retrospective data sets, not perceptively.
Moreover, they have all sorts of issues concerning the interpretation of the data, so these algorithms are derived from the data sets. I consider there to be much noise, it is not nearly as refined as the image inside. There is so much hype around being able to predict all these things, even though it is not that clear yet.
There are many issues that digitalization has to face: interoperability, cyber security, ethics, and transparency of algorithms. Which of them would you – as an enthusiast of newly emerging technologies – consider to be the most important in healthcare?
My biggest worry is the lack of data completeness for any individual. The problem here is not just a lack of interoperable data, but that a patient will see many different doctors in various health systems and hospitals, and yet no one can access all their data, from when they were in the womb to the present time. There is also a lack of inputs about any individual, considering all their data, not just traditional medical data but also their genomic data or environmental data. For deep-learning algorithms, we need as much comprehensive data about a person as possible. We have a compromised situation where we can only access limited pieces of data. This is not ideal because deep-learning algorithms are about inputs and outputs. If we have a compromised, incomplete input, then we cannot actualize the potential of AI.
In the book “The Patient Will See You Now,” you focused on the empowerment of the patient, driven by digitalization, and the democratization of healthcare. However, data from smart devices is still useless and is usually not included in the electronic patient records. Patients that want to take control of their health are seen as problematic. How can healthcare be made patient-oriented in an era of disruptive technologies?
One of the most significant barriers to democratization is the fact that most doctors are paternalistic – and they have not yet ceded control to patients. Neither do they accord respect to patients who can generate their own data and want more responsibility. Therefore, most doctors still refuse to allow their patients to have access to their visit notes, even though it is the patient’s body, and in the patient’s most vested interest. In healthcare in the USA, the patient has paid for it.
We have a real problem with paternalism, and we need to make it go away. It is essential to realize that data is eminently portable. Electronic data needs to be shared and, ideally, the patient needs to own their own data and decide with whom and when they want to share their data, whether this means their doctor or a medical research study. We still have a long way to go to abolish the dependency of patients on doctors. It has to stop, doctors need to let go and say: “You know, if you can generate your own data and if you can have a validated algorithm to help interpret the data, then I support that.” Compare that with the situation today where the patient arrives with their own data and says: “Here is all my data,” and the doctor replies: “I don’t want to look at that. It’s junk.” We need to get over this inability to share. The doctor’s notes should also be the patient’s notes. There is this sense of compartmentalization and the wrong model of ownership. All these things have to improve.
Your latest publication, “Deep Medicine” highlights, how deep-learning algorithms, as applied to wearable sensors, genomic information, and medical data, can create a bespoke treatment plan.
Well, we can already see this. A virtual code, a virtual medical code, is already being used for people with diabetes to help them self-manage, to learn what causes their glucose to go out-of-range. This could be coaching them into more physical activity, avoiding specific foods, or sleeping more. We are seeing the rights of virtual codes under particular conditions, and this will eventually mature to the point where all our data can be continuous if we are interested. I mean it’s not for everyone, but if anyone wants to use a virtual medical code, it will take all their data, everything that is available. It will absorb all the medical literature, and use it to coach them into preventing illnesses they might otherwise have. It could prevent asthma or heart failure; it could prevent many of the conditions that we experience today by giving us actionable data, algorithmic support that continuously and seamlessly accumulates our data and gives us feedback. And that data, that output, could then be shared with the doctor when we need to do so. The doctor may only see us once a year, or maybe less frequently, while we experience our health from moment to moment. And so the virtual code aligns with that principle so that our healthcare is not a one-off but a continuous story. The only reason we have not got there yet is that we had no way to deal with the data. And now we do.
So far, the healthcare system remains stuck in old structures. A conservative healthcare ecosystem with many different stakeholders is not only complex but very hard to change. The consequences are clear: the adoption of a digital solution is very slow, too slow in comparison with our expectations. What has to happen to speed up the digitalization of healthcare?
There are many reasons why it is very slow-moving because the medical community is resistant to change. It refuses to accept change, whether we are talking about reduced reimbursement, loss of control, paternalism, authority, ceding of that authority, and giving power to patients. Resolving much of this involves training and education. There are so many factors here, within a slow-moving profession in terms of change, but what we see now are other forces that are slowing the change even further. All these things, I believe, will come about. We are witnessing progressively more embracement of digital technologies, although the pace of adoption is slower than it could be.
You describe today’s medicine as “shallow medicine” and AI-driven medicine as “deep medicine.” How will things change for patients in the future, when “deep medicine” becomes widespread?
Well, I think the most significant thing is time, the gift of time, where the time spent between the patient and the doctor can be significantly increased from the single digit minutes that exist today. With an AI presence and no need for a keyboard, the doctor does not have to spend time trawling through all the different electronic or paper pages or charts, because everything is already organized. So there is time to listen instead of interrupting the patient, as happens today within seconds of starting. It is time to hear the patient’s story, as their life story can never be digitized, so it will still require a human-to-human bond. Therefore shallow medicine is medicine with errors. There are little time and no context, giving a small presence. This is what we have today. It compromises the essence of what medicine is all about, which is a human bond. Here is the place for deep medicine, with a greater understanding of everyone through machine-supported analytics. There is again the ability to listen, and to query, to have the required care, compassion, and empathy. All the things that we see too little of today.
Thank you for your time!
I have a small favour to ask…
This content is free of charge. This website is free of commercials. Please support aboutDigitalHealth.com (€1+). It only takes a minute. Thank you!