Toronto Tech Institute Tracks Long COVID with AI and Social Media
The Toronto Tech Institute uses artificial intelligence and social media to track and identify the most common long-term symptoms of COVID-19.
Vector Institute, an artificial intelligence organization based at the MaRS tech hub in Toronto, has teamed up with telecom company Telus Corp., consulting firm Deloitte and diagnostics and pharmaceutical company Roche Canada to help healthcare professionals learn more about the symptoms people with long COVID experience.
They built an AI framework that used machine learning to search and process 460,000 Twitter posts from people with long COVID, defined by the Government of Canada as people who continue to show COVID-19 symptoms for weeks or months after their initial recovery.
Research has shown that social media is often used to express experiences of illness. Users rely on social networks for information as well as emotional and social support. This makes them a rich resource for researchers, but only if they can reliably filter the flood of daily posts and identify those that are relevant to the task.
The platform analyzed the posts to determine which were first-person accounts of long COVID, and then counted the symptoms described. Fatigue, pain, brain fog, anxiety and headaches were found to be the most common symptoms, and many people with long-term illness experienced several symptoms at the same time.
The hope is that these patterns can reveal clues from everyday life outside the hospital about when and how often symptoms occur and where clusters of the condition appear. The findings can be shared with clinicians and researchers to refine questions, identify trends early or inform treatment strategies.
“This research is a clear example of how AI can help improve lives. It provides the medical community with early signals and firsthand information about long-term COVID symptoms that can help with further investigation and treatment. It also shows how AI specialists, the corporate sector and the healthcare sector can work together and share their expertise to help address some of the most concerning consequences of COVID-19”
After a series of experiments, preliminary results showed that patterns related to symptom frequency, co-occurrence and distribution over time could be successfully detected and visualized using this framework.
Replicating this research without AI would require an enormous number of working hours and staff who would have to manually find hundreds of thousands of social media posts from people, filter out those without long-term COVID symptoms in first-person accounts, and count the symptoms.
“AI is very good at working with large sets of big data to find patterns. This is for things that are too big for any person to hold in their head”
The framework speeds up research into a virus that is evolving rapidly and still surrounded by many unknowns.
So far, long COVID-19 is not fully understood. There is no single way to diagnose it or a single treatment to ease or cure it. Information is key to improving patient outcomes and ensuring hospitals are not overwhelmed in the coming years.
“Long COVID can manifest in different ways. We are still studying the heterogeneity of this condition and how to treat its symptoms, and there is no one-size-fits-all approach. The findings of the study conducted by Vector using social media help us better understand long COVID, especially its more distressing symptoms. This proof-of-concept study can be applied in many areas of medicine and can be used to inform healthcare providers in various ways, from identifying early signals of a pandemic to early signals of side effects from new drug therapies”
A survey conducted in May 2021 among 1,048 Canadians with long COVID, also known as post-COVID syndrome, identified more than 100 symptoms or difficulties in daily activities.
About 80% of adults surveyed by the Canadian Long COVID Patient Support Group and the Canadian charity Neurological Health Charities reported one or more symptoms in the period 4 to 12 weeks after their first infection.
60% reported one or more symptoms in the long term. The symptoms were so severe that about 10% are unable to return to work in the long term.
Researchers and those behind the technology hope it will quickly contribute to the global fight against lingering COVID-19, but they are already envisioning ways to further advance this framework or apply it to other situations.
"This is a new kind of tool. I don't know of anyone else doing this, and so I think it could really be very useful in health research," said Dr. Angela Chung, senior scientist at the University Health Network, which is conducting two major long COVID studies.
Researchers reported that preliminary use of the scheme shows it can help identify patterns related to symptom frequency, co-occurrence and distribution over time.
Indeed, the value of this machine learning system does not begin and end with COVID-19. Researchers now have a tool and process that could make social media a key resource for understanding other population-level health events, such as emerging infectious diseases, rare diseases or the impact of booster vaccines on infection.
Vector Institute is an independent non-profit corporation dedicated to advancing artificial intelligence, with expertise in machine and deep learning. Its vision is to achieve excellence and leadership in AI knowledge, creation and use in Canada to drive economic growth and improve the lives of Canadians. The Vector Institute is funded by the Province of Ontario, the Government of Canada through the Pan-Canadian Artificial Intelligence Strategy delivered by CIFAR, and industry sponsors across Canada.
