For readers who aren’t aware, what are some of the key ways that AI can improve healthcare?
Artificial intelligence is all about using computers to perform tasks that normally require human intelligence. For organizations of all types, AI presents three interrelated opportunities: information processing, prediction and optimization. When you apply that lens to healthcare, there are numerous ways to be smarter and more targeted about the kinds of healthcare interventions we undertake and to improve the efficiency of our operations.
One key way to apply AI to healthcare is by automating menial tasks that are labour-intensive but don’t require sophisticated human intelligence. By doing this we can relieve the burden on the workforce, which is already over-stretched. Simple examples include scheduling staff and inventory management. When we take these tasks away from workers, they can focus on other things that add more value for people.
A second set of applications is more aspirational in nature, as it relates to improving clinical decision-making. Put simply, AI can help ensure patients receive the precise treatment that is right for them at a specific moment in time. Achieving this will involve using powerful computational methods to predict which patients are at risk for which outcomes by mining data about a specific patient, cross-referencing this with the vast body of scientific evidence, and then choosing the appropriate treatment based on that. Overall, this is really about personalizing care, which some people call precision medicine. AI is not a silver bullet, but it could be an important part of redesigning the Canadian healthcare system to address the critical challenges of volume pressures and wait times.
Having said all of that, how would you describe the current state of AI adoption in Canadian healthcare?
There are examples across the country where AI solutions are being used, but by and large there is very little AI in clinical care today. However, I do think we will see this change in the next three to five years and that AI will enter the system broadly in a few different ways.
The first is through electronic medical record systems, which are increasingly being adopted across the system. A lot of these come with built-in AI tools for predicting risks for patients and suggesting treatments, and I think we’ll see a lot more of that. Another way AI will be embraced quickly is with more personal digital tools that people can use to maintain their own health, whether that be wearable devices or mobile health apps. We are already seeing AI in this space with things like chatbots to help people with mental health issues or to answer simple clinical questions.
We will see a proliferation of these technologies in the coming years, and it’s critical that these tools are refined by interdisciplinary user communities as they move through development. End users like patients and clinicians must be engaged from the very beginning, and the process must also include IT experts, engineers, data experts as well as people from social sciences and humanities to ensure new technologies are being used ethically and wisely.
Can you provide a couple of examples of how AI will improve outcomes?
Many of the worst outcomes in hospital patients are things that might be preventable and addressable if only physicians and nurses were able to identify them. One example is sepsis, which is the body’s extreme response to a severe infection. Sepsis has a very high mortality rate, but if nurses and doctors detect and treat these infections early, we can sometimes prevent them from progressing to sepsis, which can save lives. At St. Michael’s Hospital, we have an early warning system called CHARTwatch that detects patients who are at risk of deteriorating from problems like sepsis, to help clinicians intervene earlier and prevent harm.
You have said that our health data infrastructure needs to be redesigned before systemic solutions can be developed. What are the key issues with the current system?
Right now our data systems are largely fragmented around individual healthcare providers and organizations. But that’s not the case everywhere. Some provinces, like Alberta and Nova Scotia, have made investments in provincial-scale solutions that are more uniform and combat fragmentation. But generally, our health data is siloed and is not user-friendly for those seeking to develop or test AI solutions.
It’s important to figure out, first, how to create an environment where data can be shared and used by the scientists and technologists who are developing innovative solutions; and second, what protections we need to put in place to ensure that the data is used ethically and responsibly.
Above all else, we need a broader infrastructure. I sometimes use the analogy of a highway system: Right now we have some impressive major freeways in a few areas; but in the majority of the country, people are not connected to it yet. If we want AI to help everyone — and not to widen disparities that already exist, for example, between healthcare in rural communities and what’s delivered in our cities — we need to make sure that the health data infrastructure is invested in and developed right across the country.
We also need to consider computing power, of course, and importantly, who can access that computing power, whether it be public or private. We have to think about — having enough support in place — professional, skilled personnel to support the productive use of that infrastructure and ensure its accessibility.
Are electronic medical records (EMRs) a good news story in Canada?
That is one of the few bright spots. Nearly 90 per cent of physicians are using EMRs for at least one core aspect of their practice. But the good news comes with a caveat: many core elements of healthcare are not included in EMRs. For instance, most do not have a function for prescribing or renewing drugs or for consulting with other physician colleagues. And no single EMR accounts for more than 17 per cent of the market share in Canada. If our data systems aren’t talking to each other, we’re going to face real challenges trying to learn from data. So there is definitely much progress to be made.
AI solutions in healthcare will ultimately be constrained by the data we have access to. What data would you love to be able to collect?
We have to dream big about the data we want to have access to in the future, and build those solutions today. We need to ask questions like, Who is represented in this particular data set? Whose perspectives are being amplified? And what does that mean for the solutions we develop? We need a system of robust governance around health data to ensure the privacy of our patients and to ensure data is used ethically; but at the same time, we also need data to be shared openly, so individual stories are honoured and can be used for public benefit. This will require a fine balance.
In terms of what is missing today, first and foremost, in Canada we don’t do a good job of collecting information about people’s social and economic backgrounds, and that is critically important to understanding fairness in healthcare delivery. It’s obviously very sensitive to collect information about things like race, income or education, so we need to collect this data thoughtfully, in partnership with disadvantaged communities who have been harmed by this kind of data collection in the past. It is crucial that we do this, and the sooner the better, because AI-based solutions do have the potential to widen the inequities in our system.
The second thing I’d love to see is more data around the kinds of experiences patients are having. Right now we have very little understanding of the patient experience. Did the patient feel they were treated with dignity? Did they have to wait a long time for care? Did they feel physically better after the healthcare interaction? Given that health is ultimately all about how people feel, if we can’t measure subjective experiences and well-being, it is harder to focus resources on improving those things. And arguably that’s really what our healthcare system is trying to do.
Building data infrastructure across the country faces a key hurdle: Canada has to literally double its computing resources in order to reach the average of G7 nations. What will it take to bridge this gap?
That statistic comes from a needs assessment report produced by the Digital Research Alliance of Canada, which is a large federal body that was given a substantial amount of funding to try to modernize Canada’s digital research infrastructure. Since that alliance was created, it’s become clear that Canada is very much lagging. Only one thing can change that: partnerships between the federal and provincial levels of government to prioritize these major investments.
These are not the sort of investments that small groups can make. Nor can they be siloed within specific jurisdictions. Computing resources should be available everywhere in Canada to serve people in a digital and virtual environment. One data centre in BC or a supercomputer in Alberta could serve people across the country. These investments need to be made with a pan-Canadian lens, and even an international lens, because investments in massive super-computing will actually require some global collaboration. Our governments need to work together and partner with industry to make it happen. A lot of the resources, talent and infrastructure to support massive computational power will come in partnership with the private sector.
Tell us about your work with the GEMINI initiative.
In 2014, my Unity Health Toronto colleague Dr. Fahad Razak and I set out to bridge the gaps that exist in hospital data collection. This has grown from a pilot project anchored at a few Toronto- area hospitals to the largest network of its kind in Canada.
We are currently collecting data from 30 hospitals representing 60 per cent of Ontario’s inpatients and two million medical and ICU hospitalizations.
These data now sit on a state-of-the-art computing environment called HPC4Health that is powered for machine learning. We go to great lengths to protect patient privacy by removing personal identifiers from the data and ensuring high ethical and security standards are met. This data can be accessed by both scientists and students, and by the end of 2022, they had published more than 100 studies using it.
This initiative is powerful, to be sure, but we are just scratching the surface of the potential in this space, in terms of both scientific discovery and impact on patients. We have now partnered with the Vector Institute to continue making progress on AI in healthcare and are working with them on numerous projects — including predicting clinical outcomes and ensuring algorithms are fair.
Tell us a bit about the project on delirium, and how AI will help.
Delirium is an acute confusional state that is experienced by 20 to 30 per cent of all adult inpatients in hospitals. We know that when patients develop it, they have worse outcomes: They are twice as likely to die in hospital; they typically stay eight days longer than the average patient; they are 2.4 times more likely to be placed in a nursing home after their stay; and they cost hospitals close to $11,000 more than the average patient. And yet until now, there has been no way to reliably measure delirium.
This is a huge issue, because up to 40 per cent of these cases could be prevented through interventions, including cognitive stimulation, nutrition and hydration, exercise and sleep enhancement. This is a prime example of a case where AI can help, because routinely collected data points capture only 25 per cent of cases. We’re working on machine-learning models to predict the occurrence of delirium using routinely available data such as demographics and lab, radiology and pharmacy information. We have found that AI models create threefold better detection than routine hospital data alone. We’re now working on a Delirium Identification Tool for widespread use, with the goal of using the tool to measure and prevent delirium in hospitals.
Apart from the work being done at GEMINI, are there AI initiatives around the world that you wish we had in Canada?
There are definitely pockets of excellent innovation, but no one country has put all the pieces together yet. The UK has done a great job of connecting its digital health records and making them available for research and innovation. Healthcare providers in the U.S. like the Kaiser Permanente network and centres of excellence at Duke University and NYU have been able to implement real-time AI solutions because they have access to connected digital information systems. And the Veterans Affairs Network in the U.S. has figured out how to cheaply run clinical trials using connected electronic medical records. Scandinavian countries, like Denmark, are also leaders in running large clinical trials that use routinely collected health data. Much of our real-world evidence about COVID-19 comes from Israel, because their large insurance providers have integrated information systems.
We can learn from all of these projects, but ultimately, progress in Canada will come down to three things: connected digital infrastructure; communities of innovators that responsibly use data and infrastructure; and putting incentives in place around the first two items, so the technology that is developed gets to market quickly and can benefit as many people as possible.
You have been on the front lines of fighting COVID-19 over the past three years. As it begins to recede, what important lessons have you learned?
The biggest thing I take away from my experience on COVID-19 wards is that, in a state of emergency there is a lot of chaos, and the ability to respond effectively is very much driven by good leadership. There is nothing like the crucible of a public health emergency to shine a light on an organization’s (or a system’s) resilience and point out the cracks that exist. When I think about the organizations that had smoother responses than others, it was really about the effectiveness of their leaders as team builders, communicators and decision-makers in the early days of the pandemic. As we think ahead about preparing our systems for the future, ensuring we have excellent leaders in place is crucial.
The second key takeaway for me has been the importance of trust in our systems. Our public response in Canada — for example, to the initial vaccination campaigns — was generally very positive and contrasted with many other countries around the world. Research suggests that we had some of the highest early vaccination rates and very good adherence to public health measures. And a lot of that came down to trust in our systems of governance. At the same time, as the months passed, we saw those systems and structures fray, and some trust was definitely lost. In terms of our resilience for future challenges, paying attention to levels of social trust and cohesion is crucially important.
My final key takeaway from the pandemic is the importance of the health of our healthcare workers. Right now, we’re seeing the consequence of several years of incredible strain on these people, who are suffering extraordinary rates of burnout, mental illness and exhaustion. As a consequence, the workforce is depleted and the service patients receive is getting worse.
As we think about incorporating new technologies into our system, we must attend to the impact they will have on the human workforce and ensure that we use these technologies to enhance their well-being. Because at the end of the day, every system is only as good as the people who work within in it. Looking ahead, I remain hopeful. I firmly believe that it is my generation’s challenge to integrate data, analytics and advanced computing in medicine to improve both the quality and the humanism of healthcare in Canada.
Dr. Amol Verma is an Assistant Professor in General Internal Medicine at St. Michael’s Hospital and the University of Toronto and the 2023 Temerty Professor of AI Research and Education in Medicine at the University. He co-leads GEMINI, a data platform that partners with hospitals across Ontario; is an inaugural Provincial Clinical Lead for Quality Improvement in General Medicine with Ontario Health; and co-leads the Ontario General Medicine Quality Improvement Network and the COVID-19 Hospital Analytics Laboratory. Recipient of the 2022 CIHR-IPPH Trailblazer Award, he is also leading the development of a machine learning tool to predict and prevent death and critical illness at St. Michael’s Hospital.
Share this article: