How Facebook mobility data could help predict COVID clusters early

Anonymous “mobility data” you either didn’t know or don’t care that Facebook collects could have played a role in helping predicting the growth of some of Australia’s most worrying coronavirus clusters ahead of time.

The anonymised data, provided by Facebook to select research partners as part of the social media and advertising giant’s “Data For Good” program has been made available to a group of Australian researchers, who used it to analyse coronavirus outbreaks at the Cedar Meats processing facility in Victoria and the Crossroads Hotel in NSW.

RELATED: Follow our live coronavirus coverage

RELATED: Bizarre new COVID-19 conspiracy theory

The Cedar cluster was first officially recognised on April 29 with four cases confirmed by the Victorian Department of Health and Human Services.

When it was first mentioned during a daily update on May 2, the number of cases had doubled to eight. 43 cases were confirmed over the next week. More than 100 infections were eventually linked back to the site.

Similarly, in NSW, the Crossroads Hotel cluster — which genomic sequencing linked to Victoria — quickly swelled in size.

Both clusters were localised outbreaks during a time where there was otherwise very low or undetected community transmission.

Researchers from The University of Melbourne, The University of Adelaide, Monash University, The University of New South Wales, as well as the Victorian government, looked at Facebook mobility data for the week before the clusters were detected, using the data to assess the risk of transmission by looking at the proportion of travellers who entered the location where the outbreak took place.

RELATED: Facebook facing yet another lawsuit

RELATED: ‘I want to just die’: 40 women sue Pornhub

There are a few limits in using the data this way as the researchers point out: One is that the anonymised data comes only from people who have Facebook installed on their phone.

Another is that Facebook requires there to be more than 10 unique users detected in an area over an 8 hour period in order to make the data anonymous.

The researchers said that latter stipulation means their methods might not work as well in regional areas.

A third case study took a different approach, looking at Melbourne’s disastrous second wave that led to months of arduous lockdowns for residents.

The aim was to look at whether the areas that were seeing higher levels of community transmission in late June and July could have been predicted using mobility data along with the active case numbers in early June.

“Our examination of the second wave of community transmission in Victoria showed that several weeks before it was recognised, the spatial distribution of a small number of active cases was indicative of the outbreak distribution more than 30 days later when interventions were introduced,” the researchers discuss in their paper, published in the peer-reviewed Journal of the Royal Society Interface.

The results suggest that even when cases numbers were small, low-level community transmission may have been taking place throughout metropolitan Melbourne, and that “earlier selective lockdown measures including extending the borders of regions in which cases had been identified, may have been more effective at containing transmission”.

The data did not seem as effective in tracking community transmissions as it was for clusters, particularly ones like the Meatworks cluster where the context provides a high-risk transmission environment.

The research did however suggest there was “an optimal usage window that opens when case counts are high enough for aggregate mobility patterns to shed light on transmission patterns, and closes when these transmission patterns begin to determine the distribution of active cases which then predict their own future”, which could meet “a need to anticipate which populations and locations are at heightened risk of exposure”.

The study also notes that once cases start to rise and intervention measures like social distancing, restrictions on movement and stay-at-home orders are brought in, the mobility data becomes less useful.

Leave a Reply

Your email address will not be published. Required fields are marked *