Thank you, Madam Chair, for the invitation to appear.
I represent the Partnership for Women's Health Research Canada, which unites the Women's Health Research Institute in B.C., Women and Children’s Health Research Institute in Alberta, Women's College Hospital in Ontario, the Research Institute of the McGill University Health Centre in Quebec and IWK Health in Nova Scotia. Together we are the largest national network of women's health researchers, with 74 Canada research chairs and more than 1,200 members.
The health of girls, women and members of the 2SLGBTQI community is central to their employment, economic advancement and productivity, the topic of this particular study. In fact, McKinsey estimates that closing the women's health gap would contribute $37 billion annually to Canada's GDP. Ensuring that women and gender-diverse individuals are not left behind in the AI health care revolution is essential for closing that gap.
AI systems analyze massive datasets, interpret complex medical images and synthesize research evidence in seconds. This creates an opportunity to accelerate learning in health care in ways that simply weren't possible before, but AI is only as good as the data it trains on. Currently, AI makes existing biases and discrimination in health care worse. This is true for race, ethnicity, sex, gender and sexual orientation.
Let's take women's heart disease as an example. It's the leading cause of premature death among Canadian women, but important historical research studies included men only. Men are still more likely to be diagnosed and appropriately treated than women. AI exaggerates these biases. It's less likely to diagnose women's heart conditions than men's heart conditions, and it's more likely to attribute women's symptoms to mental health issues.
But if we build it and train it properly, AI can improve accurate diagnosis for women.
An example is hypertrophic cardiomyopathy. It's an inherited genetic condition that makes the heart muscle abnormally thick. It's equally common in men and women, fifty-fifty, but two-thirds of the people who are diagnosed currently are men, because the diagnostic criteria, like so many, are based on the male body. Researchers developed an AI tool to analyze MRI heart scans, what AI can do better and quicker than humans can, but they also built it to adjust the diagnostic criteria for age, sex and body size. The number of women diagnosed increased dramatically, to 44%, better aligned with the actual occurrence in the population.
AI can improve women's health and health care rather than compromise it further, but to do so, it needs to be appropriately trained. There's world-class Canadian women's health research happening. It emphasizes that we need sex-specific normal ranges, diagnosis and treatment options. When we do this in our specialized women's heart health centres, women are receiving women-specific care and experiencing better outcomes. This is the information we need to train AI on.
To change the current situation and ensure that AI is applied ethically and effectively, we recommend that the federal government fund the proposed Canada strong for women's health research and data platform, making the research and data from women's hospitals and research institutes from coast to coast available for AI applications. This requires data from patients who are not hospitalized, as 80% of health care is delivered in outpatient settings, and should be led by women's health experts.
The government should implement existing equity, diversity, inclusion and SGBA commitments that have already been made by the federal government and most provinces and territories, such as implementing standards for disaggregated sex and gender identity; going further by improving standardized data collected on race and ethnicity to understand access to and appropriateness of care; and having Health Canada require that AI providers test their systems for bias against specific populations before an AI tool can be licensed for sale.
The federal government should work with provinces to ban the use of commercial AI models that use health care spending or past clinical utilization as a proxy for disease severity. These metrics under-represent women and 2SLGBTQI people and others who face systemic barriers to accessing care.
Finally, as others have said, mandate postdeployment audits—
