How data-driven models map Covid-19 risk prone states, aid policy response

Standard epidemiological and statistical models by researchers across the world are showing the warning signs, and the way in which policy response should be directed

India’s Covid-19 ordeal is not ending any time soon, as it approaches the two-million mark in confirmed cases, with more than 50,000 new infections per day at the national level—more than any other country. 

At the local level, however, the problems are different. Cities such as Hyderabad or Pune as well as densely-populated districts such as Nalanda in Bihar are grappling with inadequate health infrastructure to deal with the pandemic. 

The Centre, states, district administrations and local governments are struggling to have a coordinated effort due to this. One fact that has emerged out of this is that the speed of the spread, the fatality rates, the extent of contact tracing and the availability of hospital beds has varied across states, districts and localities. 

To make sense of this seemingly complex and multi-layered information, data scientists and researchers across the world have attempted to develop insightful models for monitoring the spread of Covid-19 in India. 

The good part is that a few states are taking inputs from these models to design policy responses, effectively making way for data-driven, evidence-based policy making. These models track real-time data and offer the best insights available from it, more than just the headline number of confirmed cases and deaths. 

The Adaptive Control model 

One of these data-driven models goes by the name Adaptive Control, and is developed by researchers led by the University of Chicago. It dynamically calculates the “reproductive rate” of Covid-19 infections in India by using the standard statistical model used for epidemiological analysis, the SIR, or the Susceptible, Infected and Recovered model. The addition of S, I and R is equal to the population of the region under study. Mumbai-based think tank IDFC Institute is a collaborator in this model. 

Emergence of new hotspots, though a common phenomenon now, is what is making things worse for all administrations. By its virtue, the Adaptive Control model attempts to reduce the element of surprise by giving out district-wise reproductive rate of infections, that direct policymakers to the regions that will soon climb the risk ladder. . 

“Using the SIR model, we update the reproductive rate across Indian districts using varied static and dynamic variables. We used Census data for mapping long-term migrations, Google data for current mobility, and by correlating historical data with current data, we get insights on people’s movement that is crucial to the spread of Covid-19,” Anup Malani, professor at the University of Chicago Law School, and one of the lead investigators in the group told Business Standard.

The model calculates the dynamic reproductive rate or Rt, which shows how the R0—number of new people infected per infected person—is changing over time.

The Adaptive Control model shows that Rt is rising in India since July 15, about the day when confirmed cases in India crossed one million. This suggested earlier in the day that the two-million mark would be crossed earlier than expected. 

 

It also points to the states where the relatively stronger policy response is needed. Though Maharashtra and Tamil Nadu lead in the total number of cases, the dynamic reproductive rate in Andhra Pradesh at 1.46 is much higher than 1.1 for the two top infected states. Note that the data pertains to July 27.

Delhi, on the other hand, was recording the highest daily cases in India at one point of time in June. Now, the Adaptive Control model shows that the dynamic reproductive rate for Delhi has consistently been below 1.0.  

Separately, data analysed by researchers from the University of Michigan points to similar trends in reproductive rates (see map), even though certain parts of the methodology and the variables used are different to some extent.

Malani at UChicago argues that this is the most that can be done using available data. For example, the data on the segregation os symptomatic and asymptomatic cases is not available, and thus, it is difficult to model if symptomatic cases are spreading Covid-19 faster than the asymptomatic carriers.

“Though some states have begun providing granular data in some areas, most data, if incomplete, is akin to make-believe,” Malani added. 

Luis Bettencourt, the architect behind the Adaptive Control model likens the data exercise to a detective story.

Bettencourt, who is the director at Mansueto Institute of Urban Innovation at the University of Chicago, said that his model suggests that even at this evolved level of Covid-19 spread, contact tracing is more important than mobility restrictions. 

“It is important to disentangle efficient tracing from mobility. Better data can help to get a better measure of contacts. This is what our model points to,” he told Business Standard. 

He also said that a small Rt does not necessarily mean that the situation is better. If the base number of confirmed cases is very high, as is the case in Maharashtra, Gujarat and Tamil Nadu, a smaller Rt can also be very serious. 

In terms of policy implementation, he said that dealing with clusters or smaller areas—such as stratified localities in cities—is crucial in dealing with Covid-19. 

The University of Michigan has, on the other hand, designed a public health model, and not an operational model, said Rupam Bhattacharyya, a lead researcher at the School of Public Health. 

“While our model gives the forecast of the growth in cases across a few time horizons in absolute terms, the relative scaling or the distribution in between different states, and the regional variabilities are the key outcome,” Bhattacharyya told Business Standard. 

“Our model suggests that the national lockdown was neither a complete failure, nor a complete success,” he added. 

The website of their model named COV-IND-19 shows that in a “moderate return” scenario, which loosely translates to moderate efforts to contain, India would likely see close to 6 million cases by September 3. In a “cautious” scenario, confirmed cases in India could touch 7.6 million, it shows. 

Another data exercise, by the US-based firm Infinite Analytics used ad-tech data to map public mobility and innovate contact tracing. Cities in Maharashtra and some other states have used these analytics for improving tracing and preventing spread to some extent. 

States that Business Standard spoke to refused to comment on how they are involving researchers and the terms on which they are using specific statistical models to bolster their policy response, but admitted that they are using all the data-driven insights that they are commissioning from researchers and companies around the world.