Mobility datasets could help India exit the lockdown

During India’s fifth week of lockdown, measures have slowed COVID-19’s spread, despite testing and reporting concerns. A staggered lifting of restrictions is planned, balancing disease control with economic needs. Utilizing mobility data can guide reopening strategies, but privacy concerns must be addressed, especially in less dense areas.

This article was first published in The Mint. You can read the original at this link.

As India moves into the fifth week of its lockdown, it appears that we have managed to slow down the spread of the disease somewhat. Despite allegations that we are not doing enough testing and that deaths are being under-reported, the extraordinary measures that the country took to keep people from moving about has resulted in a lower number of fatalities than would otherwise have been the case.

But while these measures have been successful, it is clear that we cannot keep the country locked down forever. We have to open up some time, and when that happens it will likely be a staggered relaxation. Lockdown restrictions will gradually be rolled back in areas where the spread of the epidemic is trending downwards, while disease hotspots will remain tightly sealed. But no matter when people go back to work, it is likely that the disease will still be active in some shape or form.

This is why we need to implement any relaxation thoughtfully. Doing anything else runs the risk of giving the disease a new lease of life, allowing it to spread again as our workplaces and streets get crowded. This is what happened in Singapore and Japan, where, even though they had brought the disease under control, it resurfaced and has begun to surge again.

Unlike with previous epidemics, we have a wide range of technology tools at our disposal today. In an earlier column, I had written about how the use of smartphone-based contact tracing solutions could slow down the community spread of covid-19. But that is only useful when we know which patients have tested positive. To evaluate how and when to relax lockdown measures, we need an altogether different technology.

For a while now, epidemiologists and public health officials have used aggregated mobility datasets to fight disease outbreaks. Given the ubiquity of smartphones in the general population, our ability to log the movement of these devices gives us new ways in which to model how people move about in society, letting researchers develop deep insights into how diseases spread.

These mobility datasets have been used to assess where people might go after they leave a disease hotspot, giving public health administrators advance warning of where new disease clusters might pop up. In Pakistan, Telenor analysed aggregate mobility information gathered from the call data records of millions of customers to understand how dengue spreads from hotspots to yet-unaffected areas, giving local health administrators early warnings of how the disease was developing. Similar technology was used in West Africa in the relation to the Ebola epidemic. While covid-19 is different from those diseases, we should be able to apply those principles to the current epidemic.

For instance, it should be possible to evaluate the efficiency of various lockdown restrictions in achieving social distancing by comparing current mobility patterns with baseline levels from before the curbs came into force. By using data collected from its users who have opted to keep their location history active, Google has already generated national-level reports that provide that information. According to the India report, it seems that our restrictions have worked well, with an 80% reduction of mobility in retail and recreation areas, 55% drop in grocery and pharmacy areas, 64% fall in workplaces, and 69% reduction in transit areas. The report also shows an entirely explainable increase in mobility in residential areas, by 30% over the baseline.

This sort of data will be just as useful in measuring the effectiveness of specific relaxations of lockdown restrictions. For instance, if we can assess the mobility impact of allowing 50% of the workforce to return to work, particularly as it applies to transit areas, restaurants and other locations where people tend to gather, we will know how we need to further calibrate our administrative decisions. If it turns out that a 50% increase in workforce still results in a 90% increase in peak hour mobility in transit areas, that might suggest the need to implement additional staggered work hours.

There are many sources of aggregate mobility data. While the big global social media companies have already begun releasing this data publicly and under data-use agreements to research institutions, in the Indian context, the most accurate source of mobility data will likely come from telecom companies. With mobile phone penetration at over 95% in India, there is no greater proxy for mobility in the country. It is only a question of time before this data will be put to use for this purpose.

When that happens, the greatest risk is to personal privacy. We must ensure that telecom datasets are properly anonymized, so that it is impossible to identify individuals within these aggregated datasets. While this might be easy to do in urban areas, where the population density is so high that every last trace of individual information is hidden in large crowds, this is not so easy to do in less dense rural areas. Care must be taken to dynamically size the cells from which mobility data is derived to ensure that, regardless of density, personal data remains anonymous at all times. In addition, where new fields of information are layered over existing datasets, incremental information should not result in breaches of privacy.

As useful as mobility data might be, we must do our best to preserve personal privacy—even in the extreme circumstances brought about by covid-19.