TRACKING THE SPREAD OF COVID-19
Data from millions of cell phones show how spring break was a catalyst for the pandemic in Quebec
May 7, 2020
As Canadian authorities are thinking about using geolocalized data from cell phones to track the spread of the coronavirus, Radio-Canada got access to the data from 7.9 million cell phones across the country. Our analysis shows how Quebecers multiplied their movements at the worst possible moment: during spring break and at a time when health authorities were still underestimating the presence of COVID-19 in the province.
In mid-March, Quebec Premier François Legault announced that the province would be put "on pause." Many companies were left with little to do. This included Drako Media, a company that specializes in mobile marketing. "Our activities came grinding to a halt," explained the company's co-founder, Laurent Elkaim. "So we asked ourselves if we could share our data to better understand what is going on." This included sharing 6.7 billion location points from 7.9 million cell phone users in Canada.
The data was provided for free. It had been anonymized and aggregated before being shared with us, out of respect for each individual's privacy. The precise locations and the phone identification numbers were never shared with us. From this data, we created models to observe the movements across provinces, regions in Quebec and boroughs in Montreal. We conducted our analysis independently. Our methodology can be found at the end of this article.
Movements between provinces
Options for the number of cases
From Feb. 2 to Feb. 23, weeks before the beginning of the pandemic in Canada, 133,000 Canadians travelled from one province to another each day. There was a significant uptick in movements between the country's most populated provinces, Quebec and Ontario. These provinces will also become the most affected by the virus.
At the end of February, movements continued to increase and hit their peak on March 6, right in the middle of Quebec's spring break. At this point in time, the virus had already travelled half way around the world, from China to Canada. There were already 49 confirmed cases of COVID-19 in Canada. But the situation was already far worse: there were eight times more Canadians infected than reported.
Epidemiological reports published weeks later by Statistics Canada show that 395 Canadians were already infected at that time. Of that number, 142 were community cases: they had not travelled or had been in contact with anyone who had travelled recently. In other words, the virus was already spreading from person to person in Canada.
"This data shows how significant spring break was on the number of people infected in Quebec."
- Benoît Barbeau, professor in biological sciences and virology expert from UQAM
"People weren't really aware that they were carriers," he explains. "So as they travelled between provinces and regions, they came into contact with other people and they likely transmitted the virus."
On top of all of the movements within Canada, there are on average 165,000 travellers that enter the country from abroad every day. Epidemiological reports show that hundreds of them brought the virus home. International borders would only be shut two weeks later.
Should provincial borders have also been closed? "I believe that provinces that were the most affected had a responsibility to limit the propagation of the virus and stop it from spreading to other provinces," adds Benoît Barbeau.
On March 13, confinement measures started to be announced and the amount of movements dramatically dropped. However, the number of people infected continued to rise.
Travel between Quebec's regions
Options for the number of cases
Because of their proximity and small area, Laval and Montreal are considered as a single entity.
During the first three weeks of February, 343,000 people travelled from one region of Quebec to another each day. As spring break began, Quebecers started travelling even more. On March 6, the amount of movements between regions doubled. The vast majority of movements were people coming in and out of the greater Montreal area, which would become the epicenter of the pandemic in Quebec.
On this same day, Quebec had only tested 306 people for the coronavirus, most of them travellers coming back from abroad. Yet, on average, 23,000 travellers come back into the province every day. The virus had already been detected in over 94 countries. And community spread had already started occuring, as documented by La Presse on April 4.
"We saw a significant increase in cases two weeks after spring break. Two weeks is roughly the time before an infected person develops symptoms."
- Nimâ Machouf, epidemiologist and lecturer at the School of Public Health at the University of Montreal
Today, one in two cases in Quebec are Montrealers, even though they only represent one out of every five Quebecers. "When I saw the concentration of cases at the beginning of April, I told myself we have to close Montreal, we have to protect the regions," says Nimâ Machouf. Public health authorities, who have access to even more information, should have realized this earlier, she believes. "It's an island. It would have been easy to close off and people would have understood."
The government preferred to restrict travel between certain regions that were considered more vulnerable, for example ones with higher concentrations of seniors. The gradual reopening of those same areas started on May 4.
Movements between Montreal boroughs
Options for the number of cases
Note that the number of cases per borough and demerged cities was not available before March 29, 2020.
In Montreal, the peak of movements between boroughs and demerged cities can been observed on March 12, the day before Premier Legault announced one of the first confinement measures. Montrealers seemed to be conscious of the dangers of spreading the virus and started to reduce their movements within city limits.
At this point in time, Montreal only had three confirmed cases. But each day, more than 600 planes were landing or taking off from the airport. Half were travelling within Canada, the other half were international flights. "This was our biggest mistake," says Nimâ Machouf. "This is where the virus came from. We should have told travellers to go into quarantine." It was only on March 16 that air travel ground to a halt.
On March 25, Montreal authorities revealed for the first time the number of confirmed cases in each borough. The extent of the problem could be seen: the propagation of the virus was widespread in the city. At the time of publication, businesses were allowed to reopen in all areas, except in the greater Montreal region.
What about privacy issues?
"This data and your analysis demonstrate the gaps and the failures in some of the measures implemented," says Benoît Barbeau. "This type of data is even more important because we are anticipating new waves of infected people in the near future."
Radio-Canada and Drako Media have anonymized all the data to make it impossible to retrace a cell phone user's identity. However, a dozen institutions and companies have offered their expertise to provincial and federal governments to create applications that would trace contacts between infected people and other people.
For this expert, the source codes for the applications should be shared and verified. The effectiveness of the application should also be demonstrated. He also believes that collecting geolocalized data would be a mistake. He says following a person's movement is extremely invasive and this data is only accurate from five to 10 meters above the ground.
"Bluetooth is a technology that is less risky," believes Sébastien Gambs. This type of signal can, for example, connect wireless headphones to a phone. It works at short distances. It could be used to detect cell phones users that are near an infected person's cell phone. But in order to respect privacy, such a system should be on a voluntary-basis and the anonymity of participants should be guaranteed. "The goal is to know if you might have been exposed, therefore increasing your risk of contracting the virus. It is not to find someone to blame."
In any case, Canadians need to think twice before accepting applications that can trace an individual's movements. "Normally, it is a temporary solution," says Sébastien Gambs. But we are now talking about second and third waves. So if such a system is put in place, it is very likely that it will stay and that it could lead to more surveillance in the future." He gives the example of anti-terrorism laws that increased the government's surveillance powers in the early 2000s, but then stayed in place for several years. The difference this time, is that the threat is viral.
Methodology
The data was provided by Drako Media, with the help of their analyst Gabriel Mongeau. This company receives the latitude and longitude of the phones through applications installed on cell phones. Their sample contains data from 7,889,466 devices in Canada (21% of the population), including 1,652,389 in Quebec (19% of the population) and 702,665 in Montreal (34% of the population).
Individuals must explicitly agree to share their location with an application before their data is collected. The accuracy is the same as a GPS, with a margin of error of five to 15 metres.
In order to divide the country into different zones, we used Canada Post's forward sortation areas. These areas correspond to the first three characters of postal codes. There are 1,620 in Canada.
We defined movement by detecting a cellular device in two different regions on the same day.
To calculate movements between provinces, we compiled the movements between forward sorting areas located in two different provinces. We applied the same logic for travel between health regions and between cities and boroughs related to Montreal.
Only areas with a minimum of 100 incoming or outgoing movements were considered in our analysis.
To estimate the number of Canadians who travelled, we multiplied the number of trips in our data in proportion to the population of Canada, Quebec and Montreal.
In some cases, the location data is incorrect and the devices all return the same latitude and longitude. We have seen this type of error in Alberta, in particular. To avoid including anomalies in our analysis, we filtered out the sorting areas whose traffic suddenly doubled in 24 hours. These regions represented 4% of the total and their exclusion did not change the general trend in Quebec and Montreal.
Naël Shiab data journalist, Francis Lamontagne designer, Melanie Julien editor. With the contribution of reporter Mélanie Meloche-Holubowski for the translation.
