On November 19, Trillium Health Partner’s (THP) Institute for Better Health (IBH) hosted its fifth annual Hazel McCallion Lecture. This year’s theme, AI for Better Health, focused on the potential for artificial intelligence to advance health care and foster healthier communities. The event featured a thought-provoking keynote and discussions, which underscored the importance of approaching AI as a people project that augments human capabilities, rather than seeing AI as solely a new technology or tool that automates or replicates human tasks.
The evening included a keynote address from Dr. Sendhil Mullainathan (Peter de Florez Professor in Economics and Computer Science at MIT; Co-PI at the MIT Algorithmic Artifacts Lab; and Co-Founder of AI health initiatives Dandelion and Nightingale) followed by a panel discussion with health care, community and industry leaders, including:
- (Moderator) Dr. Laura Rosella, Chief Scientist and Stephen Family Research Chair in Community Health, IBH
- Aderemi (‘Remi) Ejiwunmi, Vice President, future Shah Family Hospital for Women and Children, THP
- Tomi Poutanen, CEO, Signal 1; Co-Founder, Vector Institute for Artificial Intelligence
- Tess Romain, President and CEO, Partners Community Health
Humans at the centre
Dr. Mullainathan’s compelling keynote address, “Medicine in the age of algorithms,” introduced the concept of “bicycles for the mind,” a perspective that shifts the focus of AI from automation to a tool that, when used effectively, can enhance human judgment and uncover deeper insights. To get here, he explained the three views of AI – generative expressions of existing information, such as large language models (LLMs); supervised learning and pattern recognition; and “bicycles for the mind”, which challenged us to view AI as a way to see things and ourselves differently.
Dr. Mullainathan discussed certain impressive capabilities of LLMs but also emphasized how currently they can be unreliable for certain tasks. As Mullainathan explained, humans need to be in the “loop” to verify and evaluate information produced by LLMs to maximize their potential, especially in critical fields like health care.
Where do we start?
Human-centred AI solutions are “centred around the patient,” said Ejiwunmi. “Ensuring we [point-of-care teams] are using tools to allow us to make care more accessible to patients, see patients as a whole person, and help make our job easier as [health care] providers in a challenging environment.”
To effectively implement human-centred AI solutions in health care, three key areas of focus emerged from the keynote and panel discussion:
1. Co-design AI tools to meet real-world needs
The successful implementation and adoption of AI solutions starts with a collaborative approach to identify the real-world challenges faced by health care and community providers. Ejiwunmi emphasized the importance of engaging point-of-care staff in the design process to co-create AI solutions that address the most important areas to improve hospital systems and patient outcomes while lessening the burden on the health care system and its people.
Romain shared an example of a pilot project that addresses adequate nutrition, a real problem faced by long-term care homes. Wellbrook Place – a long-term care residence – partnered with RxFood, PointClickCare, DIGITAL and IBH on a project to use clinically proven AI nutritional assessment technology to assess their residents’ food intake.
With one photo, the AI tool analyzes a resident’s food tray and estimates the amount of each food item remaining, how much was consumed, and the nutritional value. Traditionally, health care staff use notetaking to estimate the percentage of food consumed. This innovation delivers a detailed and accurate snapshot of a resident’s nutritional health, enabling clinicians to make more informed patient care decisions, streamline food ordering and reduce food waste.
If you think about this technology beyond replacing an existing task, we move toward using this information to learn more about how best to optimize nutrition in this population.
2. Leverage AI to further strengthen human decision-making
Dr. Mullainathan emphasized how AI tools can transform the way humans perceive themselves and their diagnoses, offering a new way to challenge existing perceptions and boundaries, which can move us toward a more equitable system.
For example, rather than using AI to summarize patient records, an area often challenging to evaluate without manually repeating the task, Dr. Mullainathan explained that AI could instead be used to review a human-developed patient summary, flag human oversights and uncover unintentional biases. This approach leverages AI’s strong analytical capabilities while keeping humans in the loop to augment and learn from the AI outputs.
Strengthening human decision-making with AI is not about replacing or automating decisions but bringing in new information and perspectives that are going unnoticed or coming in later. One compelling example uncovered how applying AI to knee X-ray images can help remove human bias in assessing pain severity. As a result, the AI algorithm can better explain disparities in treatment for knee pain among Black patients, leading to increased osteoarthritis diagnoses in a population previously underrepresented in clinical studies.
3. Build diverse data sets for human augmentation
For AI tools to be effective, they require diverse data sets that reflect the populations they serve. Without diverse data, AI solutions risk developing biased algorithms that exacerbate existing biases in health care and fail to address the needs of all patients. The panel discussion highlighted THP’s diverse patient population, reinforcing its unique capability to create and share these data sets to unlock more responsible and equitable AI.
In the Peel Region, Dr. Rosella is co-leading an AI for Diabetes Prediction and Prevention (AI4DPP) Solutions Network to partner with the community to responsibly deploy machine learning models to inform diabetes prevention and management. This involves building a comprehensive diabetes dashboard for the Peel Region, one of Canada’s most demographically diverse communities, which will account for the complex demographic and social factors that are essential to consider in addressing diabetes in populations.
Trillium Health Partners is rising to the challenge
“The power of the 905 is amazing” said Simone Harrington, Vice President, IBH. “The diverse population that we serve – the data, the capabilities, and the incredible mix of health care spaces across [THP’s] 1.7 million patient encounters, our potential is enormous.”
THP is leading the way in people-centred AI solutions to build a healthier Mississauga and Ontario. Recent initiatives, like the partnership with Canadian health AI startup, Signal 1, demonstrate THP’s commitment to lead in AI-powered health care.
With an AI governance model for ethical and responsible adoption, a capacity building program to empower staff, the implementation of Signal 1’s monitoring solution and a community of practice focused on responsibly driving a new kind of health care for a healthier community already underway, THP is setting the standard for creating best practices to leverage human-centred AI for healthier communities.