COVID –19: Effects of 3 weeks-long lockdown on lives and livelihoods in India – a data perspective (Part-2)

Photo of a young boy receiving the Schick Test from a doctor. A boy is accompanied by a mother and younger sibling (1915). (source credit: unsplash)

How can we bend the curve for India?

In this blog, we will further explain that point. To identify the right strategy for testing we wanted to use the latest techniques from the reinforcement learning (RL) domain [1]. Our colleagues from have applied RL for building investment strategies in the financial markets. We collaborated to develop the RL-based models that drive the experiments and Covid19 testing capabilities explained in this blog. To see more detail on the richness of the modelling efforts, see their blog “Optimal Resource Utilization for Covid19 using RL”.

Need for Increased Testing

As data suggests, China and South Korea could bend the curve of the infected cases after 25 days and 7 days respectively, post-detection of the first 100 cases. For India, it took almost 7 weeks (January 30th to 15th of March) to reach the first 100 infected cases. Since then, the number of infected cases has been doubling every 4 days. The government is taking measures like the closure of borders, complete lockdown of the country, testing, and hospital capacity increase, etc. Despite all of these steps, we see an upward trend in the daily number of infected cases in India. In our previous post “COVID –19: Effects of 3 weeks-long lockdown on lives and livelihoods in India – a data perspective (Part-1)” we have shown that 3-weeks long lockdown in India has just delayed the saturation point (total infected people > total number of ICU beds with ventilators available) by a couple of weeks. Hence the biggest question remains, what should we do now to bend the curve? How should we act as a country to contain the virus spread and save valuable lives and livelihoods in India?

If we look into the pattern of spread and containment of COVID-19 for other countries like South Korea and China, we can observe that the main reason for the disease containment is mass-testing. Performing mass testing on a population of 1.3 Bn Indians has different challenges than for countries like South Korea with only 50 Mn people. The number of testing kits available as well test execution capacity pose significant limitations.We have examined this problem along with associated real-world constraints, and propose a selective testing solution that yields maximize Covid19 case detection.

Smart, Pre-emptive Testing for Better Results

For government agencies like municipal corporations or city councils, one of the major pain points is to understand how to distribute their limited testing capacity across wards or counties (including all public and private testing facilities). Hence we aim to build a decision support system that can help government officials/responders to determine test kit distribution strategy that can detect the maximum number of infected people with an optimal number of tests.

We consider the population demographics and profile of each specific ward/county along with the current Covid19 infection status. The contact potential between neighboring people is considered. In addition, people will move across ward/county boundaries and we consider such movement dynamics as well. Covid19 disease’s characteristics like transmission rate (R0) and incubation period are also included in the analysis. The number of test kits available and time is taken to obtain test results complete the picture for our system to provide optimal recommendations.

Our system models asymptomatic transmission, thus enabling a government official to deploy test kits for pre-emptive testing in areas where infection cases are not necessarily visible. This will significantly help in flattening the curve since COVID-19’s spread will be contained early in the cycle.

In our experimental setup, we could efficiently predict not only the current hotspots but also future hotspots based on their probability of human-to-human transmission patterns. Our models could learn the best action strategy (i.e. test kit deployment strategy) by analyzing corresponding rewards (successfully finding out the infected population with the optimal level of testing).

Infection spread model for geographic clusters

The output of our system comes to life in the form of suggested actions (deploying the right amount of testing kits as shown in the screenshots below) that can help decision-makers such as health officers or a ward-officers to understand and execute optimal mass-testing steps at the ward/district/city level.

Screenshot of decision support system that helps crisis responders

Testing Scalability

Determining the testing strategy is one part of the overall efforts to contain the spread of the disease. Another important aspect is to understand how testing is currently conducted on the ground and how would we could scale it going forward. Currently, the suspected patients often travel several Kms to the nearest lab to get the testing done. The possibility of transmission is likely maximum during travel. Hence, we need to look for other strategies for the execution of tests, for example

-Mobile testing van (similar to telephone booth style testing kiosks in South Korea[2]) can be deployed to localities having a higher chance of infection and transmission

-We can use an existing mass task execution framework (such as the polio vaccination network [3]) to perform such testing at scale.





This site uses cookies to offer you a better browsing experience. By browsing this website, you agree to our use of cookies.