CIFAR is a Canadian-based global organization that convenes extraordinary minds to address the most important questions facing science and humanity. CIFAR is taking immediate action in the global response to COVID-19, by working with governments to provide advice on artificial intelligence (AI) applied to health research and development related to COVID-19.
On March 20, 2020, CIFAR held a conference call open to all members of its Learning in Machines & Brains program. The group discussed projects and initiatives that Fellows and Advisors are engaged in and that connect AI and machine learning applications to COVID-19. CIFAR also launched a call for CIFAR AI and COVID-19 Catalyst Grants which is open until April 3, 2020.
On March 23, CIFAR convened an international virtual roundtable of over 70 experts in AI, infectious disease, public health, and policy to explore how AI might accelerate key approaches to ending this pandemic faster. On March 24, CIFAR held a virtual briefing for Canadian and international policy-makers on the key insights from the international roundtable.
A summary of the briefing is provided here as well as notes from the discussion that followed.
Roundtable report overview
CIFAR President and CEO Alan Bernstein began by providing an overview of the key themes in the report and the importance of high quality, high dimensional data.
Public Health/Mathematical Modelling:
Lack of access to high-quality, high-dimensional data is a major research limitation
Need access to epidemiological, clinical, and non-traditional data
Biology, Drug, and Vaccine Development:
Vaccines are being studied at an accelerated pace, but deployment will take 12-18 months
AI can help model vaccine production needs and vaccine impact
Need for high-quality, high-dimensional data
Testing of COVID-19 cases needs to be more comprehensive and scaled up in order to develop the size and quality of the data sets needed to further this research.
There is a need to look beyond traditional public health data in central repositories to Big Data from apps. Platforms like Facebook and Google have high-resolution data on user location and movement, but there are current legal and privacy hurdles that prevent the sharing of such data.
Some platforms such as Google TakeOut allow users to voluntarily data share with researchers.
There are questions about surveillance and how to keep tracking the disease when we are not able to test everyone. This data is important for effective distancing measures and policies.
It is estimated that less than 20% of symptomatic cases are currently being reported, raising the need for a way to track mild community cases.
Elissa Strome (Interim VP Research and Executive Director, Pan-Canadian AI Strategy) presented the potential for AI within each of the four themes.
1. Public health and mathematical modelling
With better data sets, AI can be applied to:
Disease surveillance: tracking disease activity to reduce ICU burden
Case prediction: determining the likelihood infection before test results are available to help with prioritizing patient beds
Mortality risk: deciding patient prioritization for ventilators
Assessing clinical displacement: analyzing the long-term implications for patients who are not receiving clinical care for other illnesses
Tracking how patients can be moved within the health system based on different demands and system stress
Peer-to-peer AI apps may be used for contact tracing and even encourage behavioural change to reduce risk
2. Biology, drug, and vaccine development
For vaccine development, AI can help with:
Prediction models to help manufacturers accurately assess how many vaccines to manufacture to ensure equitable access
Unravelling large and complicated data sets that could indicate correlates of protection or biomarkers of harm
Modelling vaccine impact to make decisions on coverage and deployment
For drug development and testing, AI is being used for:
Modelling chemical structures
Interrogating genetic interactions
Correlating clinical and cellular data to study side effects
Teasing apart confounding factors and effect heterogeneity in clinical trials
Synthetic chemistry capacity is a major research limitation at this time, as many university labs are closed. Universities and funders need to examine the possibility of reopening labs and mobilizing research.
3. Clinical trials design
Several large drug screening and clinical trial efforts are ongoing where AI can be applied. WHO announced a global trial, SOLIDARITY, to find out whether any promising existing medications (including antivirals, interferon beta, and chloroquine) work against COVID-19. AI can be used to better understand the data from clinical trials in the Global South, where populations tend to be younger, which may have an impact on disease severity.
4. Health system capacity and resilience
AI could be used to:
Optimize patient intake, including the queuing and circuit/flow within hospitals to avoid contamination of non-COVID-19 patients
Perform high-resolution modelling for resource planning by hospitals. This is not just for physical resources such as PPEs, but also for scheduling doctors and nurses to avoid burnout.
Analyze global supply of testing: sourcing of reagents, transportation, prediction of surges in demand, etc.
When looking at regions that seem to have made better progress in containing COVID-19 (e.g. South Korea, Singapore, Taiwan), it is necessary to disentangle relative contribution or effectiveness of different public health strategies. AI could be helpful in teasing these apart, for example, by analyzing between-region variability.
Key points of discussion and potential next steps
There were a number of questions on the points raised in the roundtable report. The following three points - data access, app development, international and national policy briefings - engaged a number of participants who indicated an interest in connecting further. For all three topics, a summary of the discussion is provided as well as some of the outstanding questions from participants. CIFAR proposes to hold virtual meetings on these topics through the spring. Details to come.
1. Data Access
A bottleneck for researchers hoping to tackle the pandemic with AI technology is the lack of access to data sets. AI researchers are equipped with the right tools but require access to robust and real-time data.
There are currently a number of gaps in the data available to researchers and modellers. Consistent, real-time data on positive and negative test results, and the numbers of hospitalized patients, patients in ICUs, and patient deaths is required. These data also need to be more demographically granular, including infromation such as age, sex, and ethnicity, to better understand how the disease is behaving.
There also needs to be clearer data on how many people are infected but asymptomatic, crucial for effective policies to enforce distancing and control spread.
There is also a need for data on whether patients are immune after recovery, which is important for vaccine development and understanding the immune response to the virus. Some of these data can become available with the roll-out of serological tests, but more testing still needs to be done to fill this gap.
Finally, more genomic data on the virus will be important for tracking mutations as it continues to spread.
Another key rate-limiting step in the application of AI to health, previously identified by the CIFAR AI4Health task force, is a lack of coordination and integration across the country. There are already some initiatives aimed at improving this coordination. The Canadian Institute of Health Informatics (CIHI) has been working in Kingston, ON and across Alberta on improved data sharing to get near-real-time surveillance; the SPOR Canadian Data Platform, led by Kim McGrail at the University of British Columbia, has been developing a single portal for health data across jurisdictions.
In the Canadian context, there is a question about how data access and governance systems should be coordinated at the federal or provincial level. Both levels are required, as health is an area of provincial responsibility, but cross-province cooperation is also needed. Each province also has different privacy legislation. Moreover, inconsistent COVID-19 responses and policies across provinces, particularly with regards to social distancing, can lead to public confusion and doubt. One option noted was that the federal government could possibly play a role in coordination, with provincial input on the system’s design.
Questions on this topic included:
Proposed Next Steps
CIFAR proposes to coordinate a follow-on virtual meeting on health data to organize a data access working group. This could be an opportunity to work with jurisdictions to share data across the country. Also discussed at the meeting, representatives from British Columbia, Ontario, and Alberta to coordinate cross-provincially to share best practices and explore models for sharing health data with researchers. CIHI will provide a future update on linking administrative data to other datasets to obtain real-time surveillance data.
2. App Development
Using phone applications to track users’ movements and geolocations was discussed as an effective method for tracing COVID-19 and there was interest in exploring this option further. An app developed by MIT researchers was identified to streamline the process for tracking contact with the virus. Deployment of the app in Canada could be beneficial, but software development would be required. Companies exploring crowd-sourced apps to monitor population transmission dynamics and reduce individual risk may be another effective app.
However, the legal and ethical challenges of securing real-time data, even anonymized, for such location tracking applications was discussed. It was noted that public engagement and buy-in would be critical for successful deployment. Consistent messaging across jurisdictions for how to use the apps would also be important for the app’s success.
Questions on this topic included:
To what extent can anonymized data from Google, Facebook, etc be shared with governments to get a better grip on whether populations are complying with social distancing recommendations?
How do we gain social licence for AI-based contact-tracing apps? Are there any learnings or best practices from other jurisdictions, such as South Korea?
How can we engage and motivate students, either through micro grants or other mechanisms, to develop new AI applications?
A number of companies are approaching governments with ideas for crowdsourced apps to monitor population transmission dynamics and reduce individual risk. Given that some existing or newly-developed apps (such as the ones from Singapore and MIT) are already gaining traction, what should be the advice to these inquirers?
Proposed Next Steps
CIFAR proposes to hold a follow-on virtual briefing on opportunities for app development. In addition, it was discussed that collaborators from Vector, Mila, McGill University and MIT are exploring how to bring a phone application to Canada that can use geolocations of user movements to track the virus’s spread.
3. National and international AI and COVID-19 Virtual Policy Meeting
Several participants requested that a second virtual briefing session be set for participants to share insights from across jurisdictions across Canada and internationally. Participants were interested to hear about jurisdictions, technology companies and international organizations, such as the WHO, deploying AI technologies across the four themes identified in the report and models for data consolidation, sharing and access.
Questions on this topic included:
Are there examples from other jurisdictions where AI tools are being used for health system planning and resilience, e.g., regarding clinical flow design in hospitals that accomodate COVID and non-COVID patients, or in planning for resources like PPE?
Proposed Next Steps
CIFAR proposes to hold a follow-on virtual briefing and requests that individuals let us know if they would like to speak to specific initiatives underway or have a topic of particular interest. Based on the level of interest and timing available, CIFAR will determine the topics to be discussed.
Resources identified by CIFAR and participants
CIFAR Action on COVID-19
Private Kit: Safe Paths, open-source, contact tracing app with privacy protections
App for capturing patient-generated data, track their symptoms and communicate with their healthcare provider
BC guidelines for appropriate use of virtual care
Health Research Data Network - SPOR Canadian Data Platform
COVID-19 resources from Canadian Institute of Health Information
Some activities underway in Canada and abroad
US OSTP has released CORD-19 dataset; 29k full text articles; Kaggle challenge for machine reading.
For this data set, Element AI will provide full-text access within a (beta) semantic search engine front-end, to allow researchers to identify similarities across articles or individual research results. Their goal is to then progressively integrate additional unstructured and structured datasets to this engine, and enable asking natural language questions against them.
BlueDot is working with various governments (California, Canada) to use mobile devices to track people moving between health systems.
Several large drug screening and clinical trial efforts are ongoing
Covid-19 Therapeutics Accelerator, funded by Gates, Wellcome Trust and Mastercard;
SOLIDARITY trial coordinated by WHO
CIFAR is a Canadian-based global charitable organization that convenes extraordinary minds to address the most important questions facing science and humanity.
By supporting long-term interdisciplinary collaboration, CIFAR provides researchers with an unparalleled environment of trust, transparency and knowledge sharing. Our time-tested model inspires new directions of inquiry, accelerates discovery and yields breakthroughs across borders and academic disciplines. Through knowledge mobilization, we are catalysts for change in industry, government and society. CIFAR’s community of fellows includes 20 Nobel laureates and more than 400 researchers from 22 countries. In 2017, the Government of Canada appointed CIFAR to develop and lead the Pan-Canadian Artificial Intelligence Strategy, the world's first national AI strategy. CIFAR is supported by the governments of Canada, British Columbia, Alberta and Quebec, Canadian and international partners, as well as individuals, foundations and corporations.