How can artificial intelligence and data analytics improve patient care while maintaining public trust? In this episode, Dr. Amol Verma explores how health providers can learn from each other to improve clinical decision-making and system efficiency by sharing data across different institutions and using digital tools in a way that’s transparent and privacy protective. -- Comment l'intelligence artificielle et l'analyse de données peuvent-elles améliorer les soins aux patients tout en préservant la confiance du public? Dans cet épisode, le Dr Amol Verma examine comment les fournisseurs de soins de santé peuvent apprendre les uns des autres afin d'améliorer la prise de décision clinique et l'efficacité du système, en partageant des données entre différentes institutions et en utilisant des outils numériques de manière transparente et respectueuse de la vie privée.
Dr. Amol Verma is a physician and scientist in General Internal Medicine at St. Michael’s Hospital and the Temerty Professor of AI Research and Education in Medicine at the University of Toronto. Dr. Verma co-founded and co-leads GEMINI, Canada’s largest hospital clinical data research network.
Resources:
Info Matters is a podcast about people, privacy, and access to information hosted by Patricia Kosseim, Information and Privacy Commissioner of Ontario. We dive into conversations with people from all walks of life and hear stories about the access and privacy issues that matter most to them.
If you enjoyed the podcast, leave us a rating or a review.
Have an access to information or privacy topic you want to learn more about? Interested in being a guest on the show? Comment on our posts on BlueSky and LinkedIn or email your ideas to podcast@ipc.on.ca.
Patricia Kosseim:
Hello, I'm Patricia Kosseim, Ontario's Information and Privacy Commissioner and you're listening to Info Matters, a podcast about people, privacy, and access to information. We dive into conversations with people from all walks of life about privacy and access to information issues that matter to them. Welcome to Info Matters. Today we're talking about artificial intelligence and data analytics in healthcare, exploring how research and AI can ultimately improve patient services. Digital innovation promises more connected responsive care, but alongside that promise are important questions about privacy and patient trust. Joining me today is Dr. Amol Verma, a physician and researcher at St. Michael's Hospital at the University of Toronto. He co-leads the GEMINI Project, which is Canada's largest research network of hospital clinical data. Dr. Verma, welcome to the show.
Dr. Amol Verma:
Thanks so much for having me, Commissioner. It's such a privilege to be able to be part of this conversation and I think in particular for this conversation to be led from your office, Commissioner, I think it's so important. You flagged the importance of trust and it's central to all of this work. People hold up privacy as something of a barrier and I'm really excited to be able to have a conversation with you to highlight that we can do innovation while still respecting and maybe even enhancing privacy as we go.
Patricia Kosseim:
Well, that sounds like a line right out of my book. Absolutely. I agree. Privacy and innovation can go hand in hand. Just to begin, I'm interested in learning a bit more about your background, how you came to specialize in this area of medicine, AI, and data analytics.
Dr. Amol Verma:
I was always really fascinated with human biology and went through medical school, I became a specialist in internal medicine and it became increasingly clear to me, particularly as our healthcare system digitized over this last decade or more, how much information is locked away in siloed information systems of individual hospitals and how important it was to try to unlock that and share that information, not only so it can be used for clinical care, which we know is really important, but so it can be used to understand how we can improve the way healthcare is delivered. So as a doctor, one of my colleagues and I at St. Michael's Hospital had this question of just, can I be doing better from grade school to university, to medical school? You're used to constantly being evaluated and graded. But once you get out into clinical practice, you get so little feedback about the care that we're providing.
And so we actually reached out to colleagues in the greater Toronto area and said, "Do you have any information about the quality of care you're providing?" And it turned out we all felt this gap, that we really needed more information about the care we were providing. And that was really the genesis of our hospital data sharing and research network, which we call GEMINI, which is a not-for-profit research program hosted at Unity Health Toronto, but which is really a collaboration of many hospitals. And we just started with that question, what is the quality of care we're providing? And then we started working with hospitals to say, "Can we get data that informs that quality of care?" And so basically hospitals started extracting information from their electronic medical records and sharing that information with us so that we could start measuring the quality of care, we could start doing research about ways to improve care. And so that's kind of how I came to this conversation.
Patricia Kosseim:
So as doctors, this is certainly more robust data than Rate My Doctor.
Dr. Amol Verma:
There's a really important piece of information. You know, what feedback might a patient have for me, right? But something like Rate My Doctor, for example, it's not systematic at all, right? It would be a very selected set of opinions someone may choose to put on the internet. And so we need a way of being able to systematically collect data about things like, how am I ordering treatments? Am I ordering the same kinds of treatments that my colleagues are for the same types of patients? Am I ordering more or less tests than my colleagues? How are my patients doing? Are they staying in hospital the same amount of time as my colleagues?
All of that systematic information that you really need to collect by measuring the entirety of care an individual, a clinical team or a hospital in my case or a health system provides. Right. You need to be able to capture information about all of the patients and try to provide some really good estimates so people can use that as a starting point for understanding where we can improve. Everyone can improve in some direction, right? And so this is really about giving individual care providers and individual healthcare organizations data that they can use to improve.
Patricia Kosseim:
We've been doing research with healthcare data for decades. What's new about this now? What makes GEMINI a different research project than any other research project that uses health data to improve patient services?
Dr. Amol Verma:
There are some things that are genuinely different at this moment in time. The first is the digitization of our healthcare system. So historically, we were used to being able to measure and study healthcare using a very narrow slice of information generated by the healthcare system, which is the information that comes from administering healthcare, things like claims that are submitted for insurance to cover a certain drug prescribing or something like that. Right. Those kind of financial transactions provide a very thin slice of information about healthcare. What has since changed is that everything that was written in a paper chart is now digital. And so you have this enormous depth of information about every healthcare interaction. That includes the nurses and physicians' notes about their clinical interaction, it includes the images, medications being prescribed, vital sign measurements like blood pressure, heart rate. All of that really rich information is now accessible in a way that can be pooled and aggregated and shared across a system to generate enormous insights about the quality of care that's currently being delivered.
So that's the first wave, this digitization of healthcare. The second wave is the rise of computing technologies that allow us to analyze that volume of data. And that's the rise of artificial intelligence and other kinds of advanced statistical methods where we can process these large quantities of data and we can generate insights from that data that would not be obvious to the naked eye. I think we can all agree we want a healthcare system that's continuously improving at the system level. People call that the learning health system, right? We hear about all the challenges that we're facing daily, emergency room waiting times or quality of care across the system. We need to be able to use all this information that's locked away to help us find efficiencies, find ways to make healthcare safer, higher quality. And I think that's the pretty unique and extraordinary opportunity that we have now, is taking advantage of all that information and these new analytics to do healthcare and deliver healthcare more intelligently.
Patricia Kosseim:
So two waves, digitization and new technologies like artificial intelligence. But from a very practical perspective, have you solved for the infrastructure challenge of getting access to these data across institutions? How do you solve for interoperability issues, the cultural barriers, legal barriers, interprovincial barriers that allows you to do a project of this size and scale?
Dr. Amol Verma:
It's a technical challenge and it's a social challenge. And in the word social, I would include regulatory and cultural components to encouraging sharing. GEMINI started more than 10 years ago now and started with seven hospitals in the Greater Toronto area. One of the challenges is each organization will store the data in its own format with different meanings. The same lab test name might have all different kinds of abbreviations at every site. All of that is a challenge, that the systems don't talk to each other. So we worked with each hospital to develop mechanisms to extract that information out of each hospital's system and then put it in a format that could be shared. And then our team works really carefully to standardize the data across sites so that the same lab test name can mean the same thing across every organization. So we grew from seven hospitals to about 40, 45 hospitals that care for about 60 to 70% of the patients in Ontario.
So it's such an exciting story I think for people who use healthcare, the idea that our healthcare organizations are sharing data and they're learning from each other and they're trying to do better. And so the solution on the interoperability is partially a technical one where we have a team that works together to standardize data and extract all of that data out of all of the different EMR systems, more than five different companies within Ontario that we've worked with their different electronic medical record systems. But then it's also a question of the culture of, do you want to share, right? And then how do you share in a way that's privacy compliant? And I think the culture has really shifted. When we started this work in 2015, there was a lot of fear about data sharing and people were worried what they might see in their data, people were worried about losing motivation or clinicians feeling punished or overwhelmed.
But over the last 10 years, I think the world has become so much more data oriented. We're used to seeing data in all different walks of life. We're used to our Fitbit or our Apple Watch telling us how healthy we are. All of these things have changed and partially, I think our work of working closely with clinicians, care teams, hospitals to socialize the use of data for improvement. And now I think people are really eager to receive data, to learn from each other, find ways to improve. And so we have found ways to bridge the social cultural part of it. Of course, the whole project governed by research ethics board, governed by data sharing agreements that enshrine privacy by design principles into them and then a robust system of cybersecurity around the whole system.
So it is a complicated challenge. We're now starting to work with different provinces. How can we find ways to connect data infrastructures in Ontario, Alberta, and Quebec while still respecting provincial authority over health data without necessarily moving data across provincial boundaries, but it does require a lot of harmonizing the data across different sites, harmonizing our approaches and leveraging new technologies, something called federated analytics. But I think that what we've seen over the last 10, 15 years is one, there can be extraordinary benefits from sharing data and two, an increasing willingness for us to collaborate not only within provinces, but across provinces because we recognize that delivering healthcare in a way that's sustainable and high quality is a challenge we need to tackle together.
Patricia Kosseim:
So you talked about data being pooled from different institutions within Ontario and the next phase will be to look at compiling or pooling data from three or several provinces. Are these static dumps of data or are they regular feeds of data? How does that work?
Dr. Amol Verma:
The way our platform works in Ontario today is what we would call data dumps. So each hospital that participates extracts a large batch of data for, let's say, every patient they saw over the last year or over the last three months and they share that large batch of data with us. One of the benefits of that kind of approach is you're able to control a little bit more tightly the data that are shared after patients are all discharged from hospital. And so that has historically been the way that this kind of work has been done. The downside, of course, is that if you're only sharing data every three months or every 12 months, there are long lags to receiving that data and that can be a major limitation for having a health system that can respond rapidly to emerging crises or just to be able to improve.
If you introduce a new change in your healthcare organization, your hospital, you don't want to wait a year to find out if that worked, right? You want to be able to rapidly identify, oh, that did work or it didn't work and we can refine and have these rapid cycles of improvement. And it's important for our system to move to that kind of capability. And so that's where we're working now. We're trying to work on a few different pilots where we can see how quickly can we extract that data. So that's where I think we need to go.
Patricia Kosseim:
You mentioned that you're trying to reduce the lag between those regular data dumps in order to have access to more real-time accurate data in a dynamic environment. And my question of course is whether this data is identifiable or is it de-identified? What form is this data when it gets dumped into the research pool that you're analyzing at GEMINI?
Dr. Amol Verma:
When we think about privacy risk, we try to think about layers of privacy protection. So part of it involves thinking about how we can minimize risk by reducing identifiable information within the dataset. And your office has been really helpful in leading and putting out some advice and guidance on how we can think about this. So we follow a framework called the Five Safes framework. The first safe is to make sure that your data are as safe as possible. And so the way we do that is we receive data from hospitals that do contain patient identifiers, things like health insurance number. The reason those data contain health insurance number as a starting point is one of the core functions of our platform is to check the quality of data that's extracted. And so that requires a re-identification step at the beginning where what we do is we have an individual at each hospital pull up the electronic medical record and manually collect a bit of information and we make sure that the data we've extracted from computer systems matches what you would see if you looked in the medical record.
And that quality step is so important. We've caught dozens of errors in the extraction of data from computer systems that would have really critical implications for the analysis that we do downstream, things like completely wrong dates and times on transfusion data, for example, that would totally change the insights that you were getting. And so we need to have some identifiers to allow us to link the manually collected data to the extracted data. So the first step is to remove all those direct identifiers when we receive it. We remove them, we store them separately from the dataset. We then code the data, so pseudonymize where we add an encryption layer to a health insurance number so that it's sort of a one-way coding. Each hospital admission then has an encrypted version of a patient's health insurance number attached to it, which itself is not directly identifiable.
What it allows us to do is identify when the same person has visited multiple hospitals within the network so you can trace someone's information over time, which is really important for understanding trajectories of disease and things like that. So we pseudonymize the data in that fashion and then we're left with a dataset that no longer contains the direct identifiers, but we recognize actually you have to go one step farther than that with today's data, which is that there's a lot of what we would call unstructured data in the dataset. For example, a physician's note or a nurse's note. In the physician's note, they might write your name, or they might write your brother's name, or your cousin's name, right? And so we need to run text processing algorithms over that data to identify potential personal health information and remove it and then it actually masks it. So it replaces it with fake information so that if you were to have a malicious actor break into a system, for example, a human wouldn't be able to distinguish real identifiable data from fake masked identifiable data.
So we go through sort of a detailed step at trying to minimize risk, reduce the identifiability risk in the dataset by taking these steps on structured and unstructured data. But we're cognizant that when you have a big database with a lot of information, there is the opportunity for identifiability to happen by linking multiple pieces of information. We know from other large databases this kind of thing is possible. So we add other layers of protection. We ensure that every researcher that gains access to the platform has a track record of working with sensitive health data. They have an affiliation with a academic institution. We ensure that that's the sort of Safe People part of it. We ensure that every project has ethics oversight and has been improved by a data access committee. So that's the Safe Project layer of the Five Safes framework. We ensure that our research environment, so no data, no record level data or patient level data ever leaves the research environment. It stays within a secure research environment.
And then we have regular threat risk assessments, cybersecurity investigations on that environment, what they call penetration testing and threat risk assessments to ensure that that environment is secure. Invariably, the assessment identifies gaps that we can remediate and we improve and we're proactive about managing those things. Our team just did a tabletop exercise for an incident response plan to make sure we have a good plan in case there's a privacy breach, those kinds of things. So we take those layers really seriously and we think about risk minimization and I think that's got to be the way we think about privacy risk in this context.
Patricia Kosseim:
I know you've been in discussions with staff in my office, really carefully thinking through the process that you described and understanding that it's an evolving situation that you can never take for granted that you're dealing with de-identified data. You always have to test and reevaluate as the context changes as more and more information is made available and the risks increase. You mentioned federated learning, and so maybe you can explain what that is in layperson's terms and how you use federated learning in your GEMINI project.
Dr. Amol Verma:
So federated analytics is the idea that you don't need to put data together to analyze it together. Let's just take a simple example of two datasets. If you have two datasets, historically the only way to really analyze them together in a rich way was to pull them together and then run mathematical equations, our statistical models across all that data. And extraordinary advance has been the insight that you can keep those data separately and you can run your mathematical models sequentially where the statistical model starts in one dataset, it starts to learn but it doesn't complete its learning cycle and it sends a partially learned equation over to the other dataset. And then that other dataset, it continues learning and then it sends an updated version back to the first dataset in cycles. And by cycling that way, you actually can converge on a statistical model that is very similar to if the data were pooled, but you never had to pool the data.
And what's exciting about that is you can keep the data in distributed environments, whether they're at the level of an individual hospital or healthcare provider or at the level of a province and you can still gain really extraordinary insights at the similar level of depth and statistical robustness that you would if you were pooling the data. And so this is the approach that we are taking for the multi-provincial initiative that builds on GEMINI, which we're calling the vital platform, where starting with Ontario, Alberta, and Quebec, we're looking to build this network across 160 hospitals across the three provinces and use that federated model so that data are pooled within each province, but then to analyze the data across each province, we use this federated method, so we don't share data physically across provinces. If we can get everyone rowing in the same direction, we can achieve this federation that allows us to understand everyone's data and learn from our population on a national scale, right?
It's really important for, for example, studying rare diseases where no single province has enough population for us to study it. Right. It's really important for public health and understanding how diseases emerge. It's important for running large-scale clinical trials. So it's really important for all these reasons. Where federated platforms have sometimes fallen down in the past is they only work as far as each node is able to maintain its data standards, all of these steps around privacy protection, everything at the same level. So you could imagine if you tried to federate 160 hospitals in Ontario, our hospitals are just not resourced to do that, especially the smaller hospitals. So you need to find the right scale at which to federate. That's where we're thinking pooling within each province, have a centralized infrastructure in each province that takes on the management, the expertise required to do all that work, but then federate across the provinces would allow us to both meet the needs of maintaining this infrastructure but also be flexible and federated across provinces.
Patricia Kosseim:
You mentioned that 10 years ago physicians would've been hesitant to share data, to have their practices examined or evaluated by others and it's taken a while for them culturally to begin to see the benefits of the kinds of projects like yours that can give them back important feedback and actually help improve their practice and make them better doctors essentially. I was wondering if you can give us a couple of concrete examples of what the benefits are of a project like this.
Dr. Amol Verma:
So let me give you a very specific example. GEMINI supports the General Medicine Quality Improvement Network run by Ontario Health. Ontario Health convenes every month webinars where from each participating hospital physician leader, operational administrative leader, and there's a vice president or executive level leader included in the network and a data leader. They come together in webinars to review data that they've received to identify learning opportunities. And so one of the initial parts of our quality reporting back to hospitals was reporting back how long patients stay in hospital, the hospital length of stay for their general medical wards. We also report back on things like the readmission rate, so how often people end up back in hospital unexpectedly and hospital mortality. What we found through an evaluation of our first report delivery between 2022 and 2024 is that clinical teams and hospital administrators seeing some of this data presented to them in an actionable way for the first time were sending teams to the wards to understand where efficiencies might be gained if they were an outlier, for example, and sending teams to other organizations.
And the upshot, the benefit of all of that work was we saw that in the 18 months after hospitals received their first report through safe efficiency gains, they were able to reduce hospital length of stay to a magnitude of across the network, 48,000 bed days saved. So that means 48,000 days where a patient was not in the hospital because of the improvements from this network, which translates to $51 million of avoided costs in the healthcare system, really dramatic benefits for the healthcare system, and that's all through audit and feedback. So that's a really tangible example. It's also a tangible example of why you need to share data across organizations. You need to identify who to learn from. You need to see yourself as an outlier or not, right? And individual organizations don't have that window if they just do it themselves.
So a second example, one of the most leading causes of complications in hospitals is delirium. Delirium is a state of acute confusion. It affects about one in four adults when they're admitted to hospital with a medical or surgical problem. People become confused because of injury that's happening to their brain during that other illness. And that brain injury can be really severe. It leads to a twofold increase in the risk of someone dying in hospital, but even if people survive, they don't always recover their brain function. And so it leads to a two and a half times increased risk of ending up in a nursing home after you survive your hospital admission. 40% of cases could be prevented by a bundle of healthcare interventions that hospitals could deliver that include helping people sleep better in hospital, helping people eat better in hospital, helping them move and mobilize in hospital, ensuring that they are cognitively stimulated, ensuring that people have their glasses and hearing aids. Most of those things get seriously disrupted in hospital, right?
So our research showed a couple of really important things. The first thing we found is that delirium is very poorly recognized in health data. So routine health data capture only 25% of delirium cases. So when a hospital looks to say, "What is the delirium rate in my hospital and should I improve or not?" They miss 75% of their cases and they can't make delirium improvement a real corporate priority without good data about them. What's amazing is the data we get through GEMINI, all this rich data can be used through artificial intelligence algorithms to improve the recognition of delirium to a 90% accuracy that allows hospitals now to track their delirium rate much more reliably and identify where they need to devote resources. And hospitals are using this now as part of their routine corporate quality measurement.
The second tool that we've developed, which is really exciting, is an AI tool that can predict which patients will develop delirium at the time they're admitted to hospital. And you can then target those high-risk patients with this intensive bundle of interventions and try to reduce risk even with our current staffing levels on the units. And so we're now rolling out that tool in a clinical trial across 13 hospitals funded by the government in Ontario to see if we can safely reduce the risk of delirium with our current staffing levels and really leveraging these AI technologies to improve care in our hospitals. I really love that it's using AI to make healthcare more human, like to make healthcare more comfortable and help people recover better and provide better nursing care. Right. It's not about making care more or less personal. This AI tool is allowing us to really support those who need it the most and I think that's really exciting.
Patricia Kosseim:
And now, of course, the most important part that we haven't talked about is the patients. Basic, basic question, how do patients feel about this? What are they saying? Do they even have a say in GEMINI?
Dr. Amol Verma:
All of our projects ensure that we include patient and caregiver advisors or representatives just to really have that perspective, whether it's on a steering committee, for example, for the General Medicine QI Network, whether it's a project specific steering committee. The delirium project, for example, is being co-designed with patient and family advisors from every participating hospital. And I think that's so important because it helps us address really important questions, like how do you talk to a patient and their family about whether their loved one is at high risk for delirium and how do we make sure that they're engaged in a core part of any solution without feeling overburdened? So part of the approach is to ensure that every project has that perspective built into its governance or leadership structure, and we certainly do that very purposefully, encouraging our patients to be engaged, to ask questions, to be critical on the whole, a very positive thing.
Patricia Kosseim:
How do you deliver on the transparency that's needed or the notice that's needed for individuals to even know that a project like GEMINI is happening behind the scenes, what it does, who's involved, what data is being used?
Dr. Amol Verma:
So the first thing is we have a public facing website. We have a frequently asked questions section that tries to explain a lot of these in layperson's language to make sure that people could read and understand what's happening to their data. The second part of that is hospitals have to publish a notice how they're using their data and things like that. A third part of it is when we do a specific project like let's say the delirium project, there will be informational posters on the units. The clinical staff will be trained to interact with the patients and talk about the project to them so that there is that level of transparency, but also thoughtfulness in the way you present that information. You don't want to just walk into a patient's room and say, "The computer thinks you're high risk for delirium." You have to have some sensitivity in the way that these things are brought forward in the context of someone's healthcare. So those are all the steps that we take.
Patricia Kosseim:
So I think that's really important, that transparency piece. So maybe we'll end on a more philosophical note and ask you as AI becomes more and more embedded in the system, what do you think we need to get right now to avoid problems later on?
Dr. Amol Verma:
I think we should avoid making bad decisions, which is obviously an easy thing to say, but what I mean is when new technologies come forward, we should evaluate them rigorously and make sure that they pass a level of rigor that we're ready to really deploy and scale them. And relatively few AI technologies meet that bar today. And then when we deploy new technologies, we should evaluate to see what impact they're having locally. Just because it worked somewhere else in a different system, it may not work as well here. And so there's a really important aspect of local testing and local evaluation that I would say we just have to bake that in. We have to get better now at taking advantage of the right technologies. And so how do we get ready, right? How do we build a system that is ready to incorporate these new technologies?
And I'd just point to three areas. The first is we need to get our people strategy right, which means we need to have people with the right level of skill to understand these technologies in our healthcare organizations and in our health system. And we need to have the right mix of people talking to each other where it's clinicians as well as the computer scientists, as well as the lawyers, as well as the privacy experts like, in a room together and helping us make decisions because these technologies are evolving rapidly. Underpinning that people's strategy is you need networks so that this expertise, this skillset and the deployment doesn't just serve a small, narrow number of large health organizations like our academic health centers, but we make decisions that will benefit the far reaches of our province like everyone. So that's where we need to have our health system in networks where we're learning from each other, where the problems in Northern Ontario are prioritized as much as the problems in downtown Toronto when it comes to looking for AI solutions.
So getting the people and network strategy is really important. Second thing, getting the data right. We need to have our data organized in a fashion that we know what's happening when we use AI with our data. The AI outputs will only be as good as the data we give those systems. So we need to really get our data structures right. And one critical part of that is sovereignty around our data. One of the things that's happening increasingly today is our health system is leaking data to American companies and I'm really worried about that. Some of our electronic medical records providers, most of them are American companies and we're just giving away our data to American companies. We need to think about protecting that data and making it a resource for Canadians, a public good for Canadians. So I think getting the people, the networks, and the data right, that'll set us up for really being ready to take advantage of this technology as it changes dramatically.
The last part is that computing infrastructure piece, but that part is a really big moving target. How much computing you need to run AI today is different than what it was before. So we do need to get that strategy right, but I think if we get the people, the networks, and the data right, we'll be able to figure out the infrastructure computing part of it as well.
Patricia Kosseim:
So for projects like GEMINI to be successful, the one single ingredient I think you need above all else is patient trust. So what's the one thing you think they need to know in order to secure their trust, that projects like GEMINI and AI enabled research really are intended for their benefit and will improve the care that they receive without having to pay a hefty privacy or ethical price?
Dr. Amol Verma:
Yeah. I'm so glad this conversation kind of started with trust and we're ending with trust. It's so important that one, we not paint with too broad a brush. Every AI technology is not the same. And even within a class of AI technologies, take AI Scribes, every vendor or every solution is not the same. We don't ask people to just trust all medications. We ask them to trust a specific medication for a specific use for a specific person in a specific time, right? We have to come to the same approach. When we have a new AI technology, we have to explain to people how it's been developed, how we know that it's working safely in this context for this patient for this time and be able to express that efficiently so that it's routinely part of care, right? And that's a process. The doctor explains to the patient why they're receiving this new medication. They have the information to be able to do that, right?
In the same way within AI technology, the clinical operator, whether it's a nurse or a physician or whoever is taking the outputs of that AI tool and using it to inform their care has to be able to... They don't need to know how the machine learning model is working just like I don't know the technology that's inside of a medication, right, but I need to be able to explain why it's the right tool for this patient at this time and what the side effects might be and that kind of thing. Right. That's the standard we should get to. And until we can do that, I don't think we can look a patient in the eye and ask them to just trust a system. And so from a policymaker perspective, from a system perspective, the question is, how do we get to the end stage where our health professionals know enough about AI intervention that they can have that conversation and where the AI intervention has been tested rigorously to provide that information. That's the system that we need to set up.
Patricia Kosseim:
I'm so happy we ended the conversation as we began, on the theme of trust. And I just want to thank you so much, Dr. Verma, for a fascinating conversation. I've learned so much.
Dr. Amol Verma:
Thanks so much for having me and thanks for your leadership in this space.
Patricia Kosseim:
As we've heard today, AI and data are already playing an important role in how healthcare is being delivered, but how these tools continue to evolve and how they're used will depend on the choices we make today around privacy, transparency, ethics, and trust. These are conversations that are still unfolding and will continue to shape the future of healthcare. If you'd like to explore this topic further, you can watch the video from our recent privacy day event, Trustworthy AI in Health: The Promise, Perils and Protections, where Dr. Verma was a panelist. It's available on IPC's YouTube channel and there's a link to it in the show notes.
Well, that's it for today folks. Thank you for listening and until next time. I'm Patricia Kosseim Ontario's Information and Privacy Commissioner and this has been the Info Matters Podcast. If you enjoyed the show, leave us a rating or a review. If there's an access to information or privacy topic you'd like us to explore on a future episode, we'd love to hear from you. You can comment on our posts on BlueSky and LinkedIn or email your ideas to podcast@ipc.on.ca. Thanks for listening and please join us again for more conversations about people, privacy, and access to information. If it matters to you, it matters to me.