Machine learning and algorithms can support the human touch
DOC TALK by Stewart Foxman
If you think Siri and Alexa offer great virtual support now, imagine when people have this. Bob’s wearable sensor alerts him that his heart rate and BP have been up for days. An automated voice comes on; it’s Bob’s virtual medical assistant (VMI). It asks Bob to pull out his smartphone retina imaging app. Bob takes a picture; signs are good. The VMI notes Bob’s risk profile for heart disease, and runs through symptoms. Bob mentions one peculiar sensation. Could be angina. The VMI books a doctor appointment.
Or this — Karen tells her own VMI that she has been having discomfort in her belly. The VMI asks how long and on which side, what she ate last and when, and if Karen has other symptoms like nausea. Then the VMI has her get her smartphone ultrasound probe and place it on her belly. She does, and the VMI tells her to move the probe up and to the right. The VMI says it looks like gallstones, which goes with a family history and her genomic risk score, and gets Karen’s doctor on the phone.
These scenarios are just one possible outcome of the AI revolution. Dr. Eric Topol writes about it in his new book Deep Medicine: How Artificial Intelligence Can Make Healthcare Human Again. He’s a California-based cardiologist and researcher, and has written previous books on the future of medicine. Dr. Topol talked to Dialogue about AI and the often overlooked effect on doctor-patient communications.
Many experts view AI as a cure-all. Machine learning might enable us to diagnose, interpret and classify as never before — better and faster than any medical professional ever could. AI might eliminate workflow inefficiencies. Or predict the future. Yet AI isn’t just about technology, says Dr. Topol; it’s about what it will enable humans to do.
Freed from other roles, doctors can get back to basics. “What’s at the core is the human bond, and the ability to experience compassion. That’s what we’ve lost, and we can get it back,” Dr. Topol told Dialogue.
Why was it lost? The demands in medicine keep rising, infringing on the doctor-patient relationship. AI’s capabilities can return something invaluable to doctors, says Dr. Topol: the gift of time.
The promise of the new tools
In Deep Medicine, Dr. Topol writes that AI’s promise is to provide composite, panoramic views of patients’ medical data; improve decision-making; avoid errors (like misdiagnosis and unnecessary procedures); assist in ordering and deducing tests; and recommend treatment.
Advances in new tools — big data, image recognition, modelling, wearables and the like — are accelerating. AI isn’t the future; it’s already happening.
Consider the Peter Munk Cardiac Centre, part of Toronto’s University Health Network. A team there is creating a “data lake.” It’s a computer platform filled with an enormous pool of patient information: data from labs, pathology, radiology, echocardiography, EMRs and more. The data lake already has 1.6 billion points from 2 million patients, and is adding 315,000 new pieces of information per day.
AI analyzes the data to find patterns. Drawing on that, trends in subpopulations become clear. Doctors can turn the data into diagnosis models and tailored treatment plans.
In health care, we may have a data advantage. “The large amounts of data collected by Canada’s public health systems are ideal for enabling machine learning insights and solutions,” wrote Garth Gibson, President of the Vector Institute, in a recent opinion piece for the Globe & Mail.
The Vector Institute is a not-for-profit research body dedicated to advancing AI. It’s funded by the Government of Ontario and the Government of Canada (through the Pan-Canadian AI Strategy). Vector is backing a series projects to implement AI-assisted technologies in the health sector. As Gibson wrote, “Improving patient care while reducing costs is the Holy Grail.”
The future isn’t machine medicine. It’s humane medicine, deep medicine, enabled in part by AI.
Dr. Eric Topol
What Dr. Topol calls deep medicine requires three components. AI can improve two abilities and, in so doing, facilitate the third. The first is deeply defining each individual; in essence, Dr. Topol says, that means “digitizing the medical essence of a human being.” The second element is deep learning, which touches on everything from pattern recognition to VMIs.
All of this supports the third and most important component, says Dr. Topol: deep empathy, and the connection between patients and clinicians.
“What’s wrong in health care today is that it’s missing care,” he writes in Deep Medicine. “We don’t get to really care for patients enough, and patients don’t feel they are cared for. The greatest opportunity offered by AI is not reducing workloads or even curing cancer. It is the opportunity to restore the connection between patients and doctors.”
If all three elements — deep definitions, deep learning and deep empathy — create deep medicine, then what’s shallow medicine?
Dr. Topol says it’s a 15-minute office visit where 13 minutes is spent looking for information and two minutes talking to the patient, instead of the other way around. It’s incomplete or inaccurate information, and a missed diagnosis. It’s more time looking at a keyboard or screen than at the patient, and feeling like a data clerk.
“Patients exist in a world of insufficient data, insufficient time, insufficient context and insufficient presence. Or, as I say, in a world of shallow medicine,” says Dr. Topol.
Two outgrowths of shallow medicine are waste (in costs and time) and harm (in useless procedures, fear and anxiety).
With AI and algorithms as “work partners”, Dr. Topol says doctors can have the information and time they need to be present.
For some doctors, patient profiles might change too. AI might help give patients the information needed to practice self-care, and avoid seeing the doctor for minor issues. Dr. Topol told Dialogue that will give patients more autonomy and empower them — longstanding terms in health care that he says will now have more relevance. Doctors might be left seeing the people who need their care the most, and spending that time on what’s needed most.
Can we eliminate bias?
There are some caveats. Some people assume that AI systems might remove diagnostic biases. Not necessarily, says Dr. Topol. Bias can enter AI by virtue of the fact that humans design it. For example, there are biases in medical research because patients enrolled in studies aren’t always reflective of the population as a whole. Using such data as inputs into AI algorithms, and then applying them for prediction or treatment of all patients, could be problematic.
Another potential issue: AI might be able to forecast the probability and timing of diseases with greater precision than humans, yet we don’t always understand how. Will we need to explain the decisions of automated systems? Doctors, hospitals and health systems will be accountable for decisions that machines might make. A human mistake can affect one patient; an AI system error could affect hundreds or thousands.
Any critiques of bias creeping into AI must recognize that humans make key decisions that aren’t always based solely on reason and logic. Doctors are human too, and can have a range of biases, as Dr. Topol notes: overconfidence; rule-based thinking (the patient is too young to have heart disease); confirmation bias (favouring information that matches beliefs); and a desire to find simple conclusions that explain everything.
“All these humanistic interactions are difficult to digitize, which highlights why doctors are irreplaceable by machines”
Still, the very humanity of doctors vs. machines is an unmatched advantage in the art of medicine. As effective as they might be, algorithms are cold and inhumane tools, says Dr. Topol. They’ll never know a flesh-and-blood person.
Evolving along with the machines
According to one study cited in Deep Medicine, the impact of AI might free up time for patient care by, on average, 25% across various types of clinicians. Time isn’t enough. What do you do with it?
With the gains of AI, the need for doctors to integrate and explain will become more pronounced. It’s how doctors interact with patients that will make medicine truly deep, Dr. Topol says.
As machines get smarter, doctors need to evolve too, he says. It’s not a matter of making a radical shift but of honouring the fundamentals: listening, understanding, being there, providing comfort and promoting a sense of healing.
“All these humanistic interactions are difficult to digitize, which highlights why doctors are irreplaceable by machines,” writes Dr. Topol.
Still, the machines will lay the groundwork for bringing the human touch to health care even more profoundly.
AI can outperform people in many ways, and the pace and breadth is broadening. Do doctors grasp AI’s potential and impact? “At the moment, the state of the field is early, and the understanding of AI is poor,” Dr. Topol told Dialogue.
He suggests doctors don’t know if the reality will live up to the hype. Some solutions may be years or even decades away. Yet that’s what doctors should embrace, he says.
“Most people who went into medicine did so because they’re caring. Time with patients is precious. The trust, the empathy — that’s attainable,” Dr. Topol told Dialogue. Even more so, he says, with the technology on the horizon.
The future isn’t machine medicine, he says. It’s humane medicine, deep medicine, enabled in part by AI. That’s the power. “It’s our chance,” says Dr. Topol, “to bring back real medicine.”
Using AI in the COVID-19 Fight
The potential of AI to support major health care emergencies was illustrated in the COVID-19 response.
Consider the work of BlueDot. The Toronto-based social enterprise combines public health and medical expertise with advanced data analytics. On Dec. 31, 2019, 10 days before the World Health Organization notified the public of an outbreak in China, BlueDot sent clients a notice about a novel coronavirus spreading throughout Wuhan. The company then used its big data platform to forecast the next 11 cities where the virus would spread.
BlueDot’s tech can continuously crunch data like animal disease reservoirs, disease surveillance, global air travel, population densities and more. The goal is to build solutions that track, contextualize and anticipate infectious disease risks.
The company was founded by Dr. Kamran Khan, an infectious disease physician and scientist at St. Michael’s Hospital, and a Professor of Medicine at the University of Toronto. In March, the federal government announced that the Public Health Agency of Canada would use BlueDot’s platform to support modelling and monitoring of the spread of COVID-19, and inform government decision-making.
In Quebec, meanwhile, Mila (the province’s AI institute) leveraged its machine learning expertise to help develop a tracing app. The ideal is to be able to predict a person’s probability of having COVID-19 based on their contact history and medical information.
Another Mila effort involved using data analysis tools to provide a mechanistic understanding of the COVID-19 disease progression. The goal is to assess risk of given medical/patient profiles, and help identify binding targets for antiviral agents and potential vaccines.
The outbreak also saw the creation of AI Against COVID-19 Canada. This special task force aims to map and coordinate AI projects across the country that are related to the pandemic. The task force is led by a community of researchers from the Vector Institute, Mila, CIFAR (which supports research collaboration) and Amii (the Alberta Machine Intelligence Institute).
In April, Markham, Ont. became the first Canadian city to use the virtual agent IBM Watson Assistant for Citizens to offer 24-hour customer service for residents seeking COVID-19 information. The use of the AI-driven tool allowed Markham to deliver real-time information to residents on various communication channels. The information related to symptoms, testing, guidance on gatherings, closures and more.
The content was adapted to Markham’s needs, with information collected from the city, the Public Health Agency of Canada, the Ontario Ministry of Health and York Region Public Health.
Disseminating critical information during a pandemic is yet another dimension of AI’s utility — lessons that will prove helpful to help address future outbreaks.
