In the final article in our series on healthcare, we speak to the renowned cardiologist and author Eric Topol about his new book, Deep Medicine.
8 minute read
Eric Topol works at the very forefront of medicine. A world-renowned cardiologist and geneticist, he founded the Scripps Research Translational Institute, which aims to individualise medicine by drawing on the latest advances in human genomics and digital technologies. In 2017 he was commissioned by the UK Health Secretary to conduct a review into the use of technology in the National Health Service; his findings were published in The Topol Review in February 2019.
Topol’s latest book Deep Medicine argues that AI and other new technologies can make healthcare “human again” by freeing up medical professionals to focus on the doctor-patient relationship. He spoke to AIQ about the developing role of big tech in healthcare, the relative merits of the US and UK health systems, and the future of medical training in an era of AI.
In your book you outline a distinction between ‘shallow’ medicine and ‘deep’ medicine. Could you start by defining shallow medicine?
Shallow is what we have today, unfortunately. What it involves is very limited time between a patient and their doctor. The information is shallow, because no human can really assimilate all a patient’s data, and the context is minimal, because even within that brief window there are distractions and interruptions. The biggest problem is that doctors feel like they can’t execute their charge, which is caring for patients. That is what has led to disillusionment and a high rate of burnout, involving at least half of doctors: it’s a global epidemic. That’s along with a high rate of clinical depression, affecting one in five doctors, and even suicide. All these aspects of disenchantment are at peak levels in the medical profession.
What is deep medicine and how would it improve on the current situation?
Deep medicine is about giving time back. The gift of time is derived from the efficiency and productivity of AI. It’s about leaning on machines to assimilate a person’s data, because that takes a lot of time and data is often dispersed between many sources. It’s also about putting patients in charge of their own data and giving them algorithms to interpret it, especially when it comes to routine, non-serious illnesses: treatment of those could become autonomous. There are many ways to ensure that when doctors and patients come together, there is the chance for real communication, trust, compassion, empathy. All the things that should be the essence of medicine have been largely lost over recent decades.
You mention that, compared with other industries, medicine has been slow to take up new technologies associated with the ‘fourth industrial revolution’. Why is this?
There is tremendous conservativism around technology. In the US system, unless something is going to enhance business and reimbursement, there is awfully slow adoption. And you are talking about doctors losing control, which is not exactly an alluring concept, as doctors have been used to calling the shots for two millennia. There is also a fair amount of education and retraining that will be necessary, and that’s difficult to achieve for doctors that are already out in practice.
What are the best examples of the use of AI in healthcare today, based on impact? What are the examples that you find most exciting?
There’s one example that really transmits the power of AI, and it’s the story of the retina picture. When you show a picture of a retina to international experts on the retina, and you ask whether the eye belongs to a man or a woman, the chance of the experts getting the answer correct is 50 per cent. If you put it through a machine-learning algorithm you wind up with 97 per cent accuracy.
Now that example isn’t clinically useful, but it shows that when you take hundreds of thousands of images you can train machines to see things humans can’t.
In radiology, studies show that scans can be interpreted extremely well by machines, reducing missing diagnoses; humans tend to miss around 30 per cent of findings and we could get that down to the single digits. And the same is true in pathology: you can train machines not just to make a diagnosis of cancer and to tell whether it has spread, but to find other mutations and viruses – the machine eyes can see things that we could never have conceived was possible. And with machine learning there is an insatiable hunger for data, whereas humans have early satiety; we can’t handle too much information. So we can train machines to help doctors and clinicians perform better.
Companies such as Google and Amazon are devoting resources to healthcare. Will this trend develop further, and how do healthcare professionals view the involvement of Big Tech?
The technology in medicine today is pathetic. In the average hospital room, an alarm goes off 135 times a day, and 99 per cent of those are false alarms. The fact we haven’t been able to fix that over the years tells you something. So we really do need enhancements, we need every type of innovative and ingenious technology, as long as it is relatively inexpensive. We need help.
Google, Amazon, Apple and Microsoft are the four that have put in the most dedicated efforts of the tech titans in healthcare. I think they are going to have an unstoppable impact, as they have made it a priority. These companies now have lots of doctors and clinical staff and they are building up because they recognise this is a largely undeveloped opportunity, although it doesn’t mean there won’t be lots of start-ups and other challengers to the incumbents in medicine.
Do we need to be cautious about the involvement of big tech firms given sensitivities around the commercialisation of medical data?
We need regulation – in fact we already have some, with respect to commercialisation of algorithms. But there are lots of dimensions to unpack. One is that these are autodidactic algorithms and currently the regulators approving them do not take that into account, which misses the idea that more data will make them more powerful. We also have to fix privacy and security. We have seen breaches and hacks and cyber-thievery, so we must do much better. There has been some progress with GDPR standards in Europe but in the US nothing has been done yet to take that on.
Some commentators suggest China has an innate advantage in healthcare AI, given the centralised bureaucracy and the fact there are fewer protections over data sharing. Will China lead the way in this area?
That’s their goal and they are putting in big resources. I have become quite familiar with what is happening with AI in healthcare in China and there are some surprises. One is that they have the same problem as the US with respect to hospital electronic records, which are not uniform; they may work in a particular city, but they have no interoperability. That is holding them back, as with the US.
On the other hand, what’s fascinating is that China claims not to have had any breaches of data, whereas in the US 60 per cent of people have had their medical data hacked. It’s hard to verify that, as there’s not much transparency, but it would be impressive if true. The biggest advantage China has is the mass of data: Their hospitals go to 20,000 patients compared with 2,000 in most of the rest of the world, so everything is scaled up 10x in data. They are implementing AI in many different ways and they have a head-start.
You write about the huge inefficiencies, particularly in US healthcare – could AI help improve efficiency and cut costs at the same time as improving outcomes?
I think that’s possible. I mentioned the radiology story – you could have the scans screened by deep learning algorithms before the radiologist started to look at them, to make that process more accurate and faster. Pathologists with slides, dermatologists with skin lesions: anything that involves recognising patterns will benefit from AI. Getting rid of keyboards and switching to speech-recognition technology in doctors’ offices also brings tremendous economic value, because speech is much faster and we want to get rid of data-clerk activities for clinicians. These things would be ideal for lowering costs because you are extracting so much more efficiency.
The other limb of this is getting patients more autonomy. For example, in the UK, you can now get a urine tract infection diagnosed using an AI kit from a pharmacy, and that is just the beginning of where this is heading. Children’s ear infections, skin rashes; there is a long list of things that are not serious that could go ‘doctorless’ to decompress the workload among physicians and nurses.
Given the focus among Democratic presidential candidates on healthcare, do you see a possibility that the US might introduce significant reforms and perhaps even a universal system?
I hope so. Having tens of millions of people with no or limited access to healthcare is a serious problem and it accounts for why our outcomes are inferior. That’s showing more and more in comparison with other OECD countries. Life expectancy has been falling for three years and rates of infant and maternal mortality are the highest in the OECD.
It doesn’t have to be Medicare for all, but it has to be healthcare as a human right. A country as well off as the US has zero excuse for not providing healthcare for everyone. The problem we have is that there are all these incumbent entities – lobbying, interested parties with power and influence. It’s probably not going to happen in the short term, even with a new president from the Democratic side. We saw during the Obama administration how difficult it was to even get the Affordable Care Act, which was only in some ways a baby step in this direction, because of the special interests and perverse incentives in the system.
You were commissioned by the UK government to write an in-depth report on the potential for new technologies the NHS. What were your key findings?
There is a big wing of NHS Health Education England responsible for training and education of clinical professionals, which is important – that’s why the UK is the capital of the world in genomics. That training is ongoing and will cover AI and sensors and digital medicine, all the different technologies we reviewed in the report. In many respects the UK is well positioned: the government has been planning the technology uptake and the training of the workforce, and very much embraced this gift of time to improve the patient-doctor relationship. When an emergency department in Leeds decided it was getting rid of keyboards and switching to speech recognition, they just did it and that project is holding up well. They can now get that going through the entire NHS infrastructure.
You write movingly in the book about your father-in-law and the palliative care he needed at the end of his life. What can AI do to support healthcare systems that are increasingly focusing resources on patients’ final months?
For people who need palliative care, there are better ways to predict outcomes using AI. Most importantly, when a patient is in that state, AI could enable them to stay at home, with monitoring; they don’t need to necessarily be in special facilities, as we will have ways to keep people in their home environment with their loved ones. There is the technology to do that and it should be relatively inexpensive, relative to having to use special facilities and personnel. But this area is not as well developed as the ability to process images and patterns for diagnosis.
You write about the potential for personalised medicine based on mobile technology and virtual assistants. Imagine we’re in the year 2029: Do you see us speaking to holographic doctors on our smartphones and treating ourselves based on their advice?
Yes. It won’t work for everyone, of course, but avatars that have your data and could give you the feedback you need to prevent an illness or better manage an illness or condition will exist. It’s exciting, because if you have the continuous processing of your data – your genome, your gut microbiome, all of your prior data and all of the medical literature up to the moment – if you have all that and it’s continuously learning about you and giving you counsel and advice, this virtual medical coach is exciting. We are already seeing it for specific conditions, such as diabetes. Some people will find it creepy; some will find it too much. But for most people the idea of having this assistant in your pocket or on your smart speaker would be an attractive prospect.
In your final chapter you speak about the need to ‘rewire the minds of medical students’. How will medical education and training need to change with the advent of AI?
The problem we have today is that we largely select future doctors based on their test scores and their college grades, and those are not necessarily reflective of where we need to go. We need the highest emotional intelligence, the best communicators, the people with the most compassion and empathy for others. We need to adjust, as the ‘brainiac’ premium is going to be reduced when you have all this data-processing capability and AI that knows the literature and achieves a higher accuracy of diagnoses with fewer errors. It’s nice to have great minds, but we need to put an even higher priority on the humane elements, the essence of medicine, which is a connection between patients and their doctors, nurses, genetic counsellors, physical therapists. That’s going to mean different priorities of selection and nurturing in medical school.
Finally, would you say you are optimistic about the future of AI in healthcare? What’s the best-case scenario for our healthcare systems over the coming decades?
We need a path that can get us out of the multi-dimensional trouble we’re in – with all the errors, all the waste, the lack of humanity, patients getting roughed up, doctors’ burnout – it’s an awful mess. AI could alleviate these conditions and get us to a whole new plane. It’s almost like a Back to the Future story – we need to go back three or four decades to when the intimacy and humanity in medicine was still there. I am optimistic, but it’s going to take time; I don’t want to give anyone the impression this is going to be a quick switch, but it takes the mindset, it takes what the NHS is doing in embracing this potential and not making it worse. Because if you take the power of AI and you ask doctors to do more of the same, that will make everything worse. It takes a committed, dedicated effort to use these things in the right way – not just for better economics, but for better people.