does c.s. research, builds robots, codes, does stand-up, has a podcast & writes.

You are reading something Soham Sankaran wrote. Return to the writing index.
Feel free to find him on twitter @sohamsankaran or email him at soham@(this website).

More With Less: Reopening India with limited COVID-19 testing resources

A strategy for reopening India piece-by-piece from the coronanvirus lockdown based on the counterintuitive insight that our limited testing capacity is best used to determine where COVID-19 is not, rather than to confirm where it is

Published April 15, 2020

Note: This article was co-authored with Sahaj Sankaran, and was also published as an op-ed by CNBC India on April 15, 2020.

There is only one certainty in the once-in-a-lifetime crisis we find ourselves living through — India's total national lockdown must end, and soon. The economic costs of continuing total shutdown are vast all over the world, yes, but while the developed world trades off money for lives, in India, the calculus of economics has always been denominated in human bodies — we are not sacrificing our economy for lives, but rather lives for other lives. The longer total lockdown is sustained, the more likely it becomes that the sheer humanitarian cost of the long-term economic devastation to follow matches even that of an unfettered coronavirus epidemic.

One way or the other, the country must begin to reopen, and this must start to happen before the non-isolated case count hits zero. The government seems to agree — the Prime Minister announced this week that they will begin reviewing districts for a relaxation of the lockdown starting on the 20th of April. The question, then, is how to do this without undoing the herculean efforts of every Indian to slow the spread of this virus. Short of the introduction of a vaccine or highly effective therapeutic intervention, the best way to reopen safely would be to conduct frequent tests at population scale. Ideally, you would test every Indian every day for the next 3 months, immediately isolating those who test positive. This would completely solve the problem, but you would need more than 100 billion tests to accomplish this, and the likelihood of that capacity emerging quickly is low, though it’s probably worth a try. Even testing a non-trivial fraction of the population is likely out of reach. We would need 500,000 tests a day to match the lauded testing rates of South Korea, which has only tested 1% of the population despite being a much smaller country with a much better-developed healthcare infrastructure — ICMR’s stated goal is to eventually reach 100,000 tests per day from our current daily volume of about 18,000.

We need a strategy to safely reopen parts of the country over the next few months that takes into account our limited testing capacity. We believe that the best candidate for this is a method that combines geolocated self-reported symptom data with limited randomized testing to gradually reopen areas that can be demonstrated to be virus free. This method takes advantage of a key piece of infrastructure we do have — near-universal cell phone access.

In this strategy, the entire country must be divided into a large number of small subdivisions, about the size of individual neigbourhoods, which will be designated ‘Red’, ‘Yellow’, or ‘Green’, representing different levels of movement restrictions. Red areas are in total lockdown, as the entire country is today. Yellow areas would allow limited movement within themselves, but their borders would remain sealed. Green areas would allow movement internally and between other neighbouring Green areas. Each day, every resident in every subdivision would be told to self-report any symptoms via app, text message, or voice call — this would only take 30 seconds. This data is then processed into a risk score for each subdivision using statistical estimators that have proven effective in past flu epidemics.

The risk scores will be closely linked to coronavirus rates, but the exact correlation would need to be determined empirically. Therefore, we will select a small number subdivisions that are in aggregate representative of the whole distribution of risk scores and carry out extensive random testing within them to estimate actual coronavirus prevalence. This information would allow us to correlate the risk score to actual coronavirus rates, and in particular would let us designate a risk score safety threshold — for instance, we might find that if the risk score is below 30, there is little to no coronavirus present in that subdivision.

Subdivisions with risk scores below our risk score threshold are designated Yellow, and we allow limited internal movement — though practicing social distancing — within them. Yellow zones are monitored carefully for at least 14 days. If they remain below the threshold for those 14 days, they are subjected to randomized testing of non-isolated individuals. If the virus does not appear to be present, they are designated Green and full internal movement is allowed; further, movement between any neighboring Green zones is opened up.

Green and Yellow subdivisions whose risk scores start to rise are immediately designated Red and locked down completely, with people only allowed to leave their homes for food and essentials (at the moment, the entire country is effectively under Red regulations).

This strategy is optimal for a number of reasons. For one, the daily symptom data and risk score calculation allows us to monitor improving or worsening situations in real time. Testing data is useful, but lags behind coronavirus spread since a coronavirus-positive person could spread the virus to any number of people between symptoms appearing and being tested. Further, the strategy is optimized for safety: any Yellow or Green zones that worsen are immediately locked down again, but Red and Yellow zones are never fully opened up without testing and a monitoring period.

The key insight is the somewhat counterintuitive idea that, given our limited testing infrastructure, we need to use it not to determine where coronavirus is but rather where it is not. This follows because Red zones don’t actually need extensive testing; we already know or strongly suspect that there are developing coronavirus clusters present. Instead, we should concentrate medical resources - doctors, PPEs, quarantine beds — in the Red zones and save our limited testing capabilities to confirm the safety of zones that we can then open.

This strategy will work particularly well in India because, thankfully, things aren’t as far gone here as they are in America or Italy; we have strong reason to believe coronavirus is relatively concentrated, and that most of the country, particularly rural areas, would quickly register as Green or Yellow zones, allowing us to open large parts of the country’s economy in short order. As isolated Red zones are identified, isolated, and quarantined, entire states with low case counts such as Goa, Chattisgarh, and Tripura can open up travel internally and to each other, reopening transport and commerce networks bit by bit. What we will likely be left with are virulent patches of Red in and around the major sites of spread - places like Mumbai, Delhi, Bangalore, and Indore. The full weight of the country’s medical resources can then be thrown at these Red clusters. By this point, most of the country is likely Green, meaning testing infrastructure can be repurposed and deployed to comprehensively test the residents of these zones. Subdivision by subdivision, district by district, cases will be identified, quarantined, and treated. Step by step, inch by inch, we will contain and push back coronavirus even as the Green zones nurse our economy back to health.

The strategy is simple, feasible, effective, and safe, but we are running out of time. Every day that we wait brings untold long term consequences for the economy. Small businesses may never recover, and daily wage labourers may never get back on their feet. We cannot simply sit back and hope that the lockdown will clear the country of this malign specter — it will not, and if it is extended again and again, hundreds of millions will find their lives broken beyond repair.

. . .

Soham Sankaran is the CEO of Pashi, a Y Combinator-backed startup building software for manufacturing. He was previously a doctoral researcher at Cornell in Computer Science, and a researcher at the Yale Institute for Network Science working with Professor Nicholas Christakis on human social networks. He has a BS in Computer Science from Yale University.

Sahaj Sankaran researches the history of economic planning and public policy in India at Yale University.

Soham can be contacted at (his first name) [at]

You can read more of his writing at, follow him on twitter @sohamsankaran, and get new writing via email by subscribing below.