Written By: Isaiah Hazelwood, Science Staff Writer
On March 11, 2020, the World Health Organization declared COVID-19 a pandemic, and our lives changed dramatically as we moved to remote work and online classes. Governments declared lockdowns and mandated masks, and the virus continued to spread despite our best efforts. One year later, the pandemic is as large as ever and we continue in protective isolation, though the rollout of new vaccines promises a slow return to safety and normalcy. To mark this date, the Trinity Times presents interviews with University of Toronto’s experts on COVID-19. Dr. Dionne Aleman is an Associate Professor in U of T’s Department of Mechanical & Industrial Engineering, holds an appointment in U of T’s Institute for Pandemics, and leads the Medical Operations and Research Lab. Her work is currently focused on creating computer models of disease spread for the province of Newfoundland and Labrador. Our interview with her has been edited for clarity and length.
Prior to COVID-19, what were your general areas of research? How did they change in this last year?
My research is several applications of advanced analytics to medical areas and healthcare decisions. Pandemic planning and predictions were one of those areas before COVID-19, though it’s a central focus now.
Besides pandemic modelling, I also research collaborative operating room scheduling, which is a method of scheduling surgeries by placing everyone into a queue where they are helped by any available surgeon instead of our current system where your family doctor refers you to one specific surgeon. It’s much more efficient than our current system. You can do what is currently 12 months of surgeries in 9 or 10 months, and that’s with our models accounting for disruptions like the lack of downstream beds, some patients demanding a specific surgeon, or other irregularities. With the pandemic putting elective surgeries on hold, having a method to very quickly and efficiently schedule many surgeries is even more pressing in the coming year.
I’ve done some work in radiation therapy, particularly in using mathematical optimization to design radiation treatments which send the right dose of radiation to the tumour and avoid sending radiation elsewhere. Once the treatments are designed, I use machine learning to automatically generate quality assurance for those treatments.
I also perform some modelling in organizing bone marrow and kidney transplants. There are many donors and a lot of patients needing transplants, so machine learning and mathematical optimization can try to best match them up and predict transplant outcomes.
Can you provide an explanation of how your pandemic models work? How are they different from other models that epidemiologists use?
A lot of models we see from governments, in Canada and globally, are compartmental models. Those include the SIR – Susceptible, Infectious, and Removed – model, where there isn’t much differentiation between individuals in the population because it models proportions of the population moving from healthy to infected to recovered. Compartmental models are easy to make and don’t require much data. Once you have the viral reproductive number, the number of contacts each person has, the recovery rate, and the death rate, you can make differential equations which model the virus’s spread decently.
However, compartmental models aren’t good at modelling detailed situations. What happens if you close schools in one specific area or target vaccinations on one group of people? My models are agent-based simulations which consider each individual as a unique agent with an age, a home, a workplace, household companions, a group of contacts, time spent on public transit, chance of becoming hospitalized after getting sick, and more. The model simulates everybody living their lives and interacting with one another, and each time a person interacts with an infected person their exposure increases. At the end of the day, people with more exposure are more likely to become infected, infected people have a chance to recover, then we simulate another day with those new infected people. Agent-based simulations let you track the effects of targeted policies on specific groups of people, which is great for testing health policy. They do have some downsides though, as they require a lot of data on the area and the people living in it. When that’s not available, we have to make estimates. They also need a lot of computational power; I run my model on a supercomputer and not every group gets access to that.
In recent positive news, we have new vaccines getting approved and rolled out. In the past you published some research on vaccine modelling. Are you doing any of that now?
I’m actually just now ramping up vaccine strategization research. I use graph theory-based approaches to look at community contact networks and try to calculate who should be vaccinated to most reduce the spread of disease. It’s not always who you think should be vaccinated. The people with the most contacts might be in a heavily connected part of the network, so vaccinating them doesn’t do as much as you expect. Optimization algorithms and graph theory metrics can point to who to target to best limit spread. Then, I use machine learning to perform rule mining, which gives us implementable rules that we can turn into public policy.
We have lots of public health advice to limit the spread of COVID-19, from washing hands to distancing to mask-wearing. In addition to that, what do you think we should do to slow and stop the pandemic?
Those three you mentioned are the big ones; physical distancing and mask wearing in particular are very good preventative measures better than almost everything we have. Unfortunately, a lot of people either don’t want to or can’t engage in those measures. Some people still have to come into work despite the stay-at-home mandate, and allowing businesses to decide who is essential opens opportunities for potential abuse by forcing people into an enclosed workspace when they don’t need to be. Even when you’re wearing masks and staying apart, you’re still not 100% safe, and more contact means a higher chance of infection. It’s a systemic problem, and health policy alone won’t eliminate the unnecessary risks. A good way to address them is by implementing policies like paid sick leave or financial support for businesses, so workers and employers can abide by health recommendations without feeling the need to stay open and risk the health of them and their customers.
In the future, what would you recommend so we can prepare for the next pandemic?
One thing COVID-19 has shown is, at least in Canada, there’s a finite window of time where you can expect people to completely comply with protective measures like closing businesses, staying home, doing school online, and all those disruptions from pre-pandemic life. If those measures drag on, people start wanting to reopen, and they ask their politicians to reopen, which can pressure the government to open even if it isn’t safe. For future pandemics, we need to better utilize that first six to eight months to dramatically close, get everything under control, and be very reactive and fast to respond to new cases. After the long initial lockdown, we will be in a much better situation to reopen and go back to normal, though we will definitely need two-week circuit breaker lockdowns if cases start to reappear.
What advice would you give to students hoping to pursue a career like yours?
The most important thing is to be very comfortable with math and computers. Operations research and modelling uses lots of optimization techniques, algorithms, and math, so you’ll need to be able to design and write computer programs to execute your ideas.