Amitabh Sinha: How high do you think the third wave peak can go, and how long do you think this wave will continue?
For Mumbai, the third wave should peak somewhere around the middle of this month. Same with Delhi. According to our current calculation, which is preliminary, because we don’t have enough data for entire India, we expect the wave to peak somewhere in the beginning of next month. The height of the peak is not being properly captured currently because the parameter values are changing rapidly. As of now, as an estimate, we predict a wide range between four and eight lakh cases a day.
The Delhi and Mumbai curves are likely to come down as rapidly as they have gone up. The all-India curve has just started to rise. It should take another month’s time to peak and come down. By the middle of March, the third wave of the pandemic is likely to be more or less over in India.
Amitabh Sinha: How reliable are predictions of a computer model considering that there is a great degree of randomness in the spread of a pandemic?
It is true that pandemics by nature are extremely random phenomena, but there are some basic principles. Infection gets transferred when an infected person and an uninfected person come in contact. It’s quite a simple analysis that the more infected persons there will be, the more new infections will get created, because the more transfers can happen. The more uninfected people there will be, the more infected people will get created. Based on this, one creates a model.
The basic model was created about 100 years ago. It is called the SIR model and has been very useful in predicting the trajectory of several pandemics. We have applied some tweaks to this model to account for some local ground realities. In our model, we have allowed the parameters to learn their values from input data itself. All we need is the daily time series of reported new cases. From that time series, we are able to estimate the parameter values required for our model.
This also means that the parameter values should not be changing when we are doing an estimate. If they are changing, then our estimates will go wrong. The model requires some time for the parameters to stabilise. Every time the parameters change, we have to re-compute. The good thing is that apart from input data, the model doesn’t need any other calculation to compute the parameter values; it picks up from the data itself. That’s where we have been successful in predicting or capturing trajectories, when many other models could not.
Amitabh Sinha: How good is the quality of data that the state governments are putting out?
We have found that the quality of Indian data is superior to that of many other countries, including a few very advanced countries. Sometimes we don’t give ourselves enough credit, but this is at least one time when I think our machinery can justifiably claim the credit.
Anil Sasi: The initial assumption was that the India trajectory in third wave would perhaps mirror South Africa but the Indian trajectory clearly seems much more steep. What has gone wrong?
Initially, since there was no Indian data, we thought to run our model on South African data because it is closest to India in terms of population, age-profile, as well as level of natural immunity. We thought India would have a similar trajectory. It didn’t happen. The reason is something that virologists and biologists should be able to explain. In India, the loss of immunity, especially loss of natural immunity, seems a lot more as compared to South Africa. I don’t know why.
Maulshree Seth: How significant would the loss of immunity be?
Having learned our lesson from the second wave, when our predictions went wrong, this time, we are a little more cautious. We assumed complete loss of immunity, that’s the worst case. However, there is one additional data point that needs to be factored in. It has been observed everywhere that when a vaccinated or some immune person gets infected, then the intensity of infection is a lot less. So, we have factored the worst-case assumption, that everyone has lost immunity, and our current projections are based on that. It is possible that the third-wave peak goes beyond our upper limit of eight lakh cases in a day, but I don’t think it would be too high, possibly around 10 lakh. When the current phase of rapid growth stabilises, we will be able to make a more accurate prediction.
Asad Rehman: How effective are lockdowns?
In the first wave, the very strict lockdown cut down the spread rate by a factor of two. During the second wave, different states adopted different strategies. The states which properly imposed a mild or medium lockdown were also able to cut down the spread. So it helped.
A strict lockdown always helps more but then it has to be traded off with the downside, which is the complete loss of livelihood for a lot of people. We always talk about Covid-induced deaths, but we should also sometimes talk about deaths induced by this loss of livelihood.
For cities, where we are expecting the peak to be in the middle of January, there is absolutely no need for a lockdown. For states that are still in the growth stage, such as Tamil Nadu, which has imposed a lockdown, it is a little premature, because this time the hospitalisation cases are not that many.
It could be worthwhile strategy to let it grow if your hospitalisation or medical system can handle it. Let it grow and be done with it quickly. That will minimise the time of discomfort for the entire population. Of course, your medical system should be ready to deal with that kind of short, but intense pressure.
Ritika Chopra: The SUTRA model has not always given accurate predictions. Is the team working to improve it?
We are keen to improve the model. The drawback of the model is that when the parameter values are changing, it has no way of predicting what will be their eventual value. When there may be a possibility of that kind of prediction with a more advanced analysis, we would certainly love to take anybody’s help to do that.
Amitabh Sinha: One common criticism has been that all that you are doing is retrospective curve-fitting. Another is that there isn’t any biology in your model.
That charge is absolutely correct. There is no biology in our model. But I don’t think a modeller should look at whether or not there is biology or philosophy or anything else in a model. The objective simply is, are you able to predict accurately? Our model hasn’t been 100% correct. But I would submit that we have been better than any other models in existence. And I would love to be proven otherwise.
Shubhajit Roy: You’ve been conducting these mathematical models for the last year and a half. Has there been any conversation with the government on this?
We have been asked to make presentations by the Central government, as well as several state governments. Since it’s the Central government, the Covid Task Force had taken input from us on several occasions. Several state governments, including UP, Delhi, Maharashtra and others, have occasionally asked to share our findings with them.
Sohini Ghosh: How are epidemiological factors accounted for in a mathematical model?
The great advantage of a mean field model like ours is it just averages out everything. The reproduction rate, for example, is going to be different in a very dense region compared to a very sparse region but when you take a mean-field approach, you just kind of take the average out of the entire race. Delta, Omicron, and maybe some other variants are also floating around. So, the mean-field model will just compute one single average value of the reproduction rate.
Sohini Ghosh: Would that not then affect the accuracy of the models? Other epidemiological factors like immune response and immune escape are also getting evened out.
At the individual level, there is a lot of randomness such as individual immune responses. How a person responds to the pandemic could be very different. There is this famous principle of statistics, the law of large numbers, which is that whenever there are large numbers, the individual or local random variations smoothen out. Just with a few parameters you can capture the entire phenomenon.
Kaushik Dasgupta: Why do you expect the cases to drop as quickly as they are rising?
It is the nature of any pandemic essentially that if you don’t apply too many external controls, then the curve on the rise will mirror the fall also.
Sandeep Singh: Do the small numbers in UP come as a surprise to you, because maybe it isn’t being reported properly?
It takes time. UP is a large state, and in any large state, this particular variant will take time to move about. So, it’s not a surprise. We always start with larger cities and then move on to smaller cities and to villages. I expect cases in UP to go through the roof over the next couple of weeks.
Amitabh Sinha: Several studies have put the actual number of dead in India between 4 and 5 million. Do you think that such large numbers can go completely missing from records?
It’s very unlikely. We are not living in the stone age where such a large number of deaths will go completely unreported. There have been reports from many states about graveyards being full, long queues outside but those have happened in a very small time period—a week or 10 days—when the second wave was at its peak. When you average over the entire timeline of the pandemic, the impact of such excess deaths will not be that high. So, I find it very difficult to believe that the number of deaths would be 10x of what has been reported. Yes, 2x or 3x is certainly possible.
Amitabh Sinha: A lot of states report death data with a lag. Given that, on a day-to-day basis, there would certainly be some number of deaths that would be missed, what do you think would be the best estimate for the number of deaths?
I would say somewhere between 2x and 3x. How do I arrive at this estimate? Partly by my own studies, but also there was a very interesting paper that came out in The Lancet, which looked at the excess deaths in Chennai. They found that there were around 25,000 excess deaths during the pandemic period. If you look at the reported deaths in July, they were around 9,000. This means that it is slightly less than 3x to the actual deaths in Chennai.
I have tried to look at every single study that has come about which claims some factor of deaths. In every ‘10x’ type of claim that I’ve seen, the methodology used is seriously flawed.
I will give you one example, which I believe is clearly a mischievous methodology. They look at the CRS data from July 2020 to June 2021 and compare it with July 2019 to June 2020 data and look at the deaths in these two periods. This is a somewhat unusual time period to choose. Why chose July-starting and June-ending specially, when we know that April, May and June 2020 was a very strict lockdown period in the entire country, literally everything was shut down. If there are deaths happening in that three-month period, it stands to reason that they will all get reported or recorded after July.
Without there being any excess deaths, if you just compare the deaths in these two time periods, you will find the second time period reporting a much larger number of deaths because some deaths are missing from the first period and those deaths get added to the second.
Amitabh Sinha: One of the big points of discussion on this death debate has been the fxloating bodies that were seen in UP and that has been used to make a claim that the state had a massive underreporting of deaths. You actually praised the handling of the Covid pandemic by UP government. Why do you suggest that UP has handled the pandemic better than it is given credit for?
What many people do is that they look at a state’s performance vis-a-vis some other state’s performance without factoring in the health infrastructure of the state. There is no doubt that many deaths happened in UP. I come from UP, from Prayagraj. During my childhood, we would find bodies floating in the Ganga. It’s not a new phenomenon. But it’s still quite possible that this time the numbers are much more. UP has many poor people and clearly they cannot find resources to cremate a body. This is why they resorted to putting bodies in the Ganga. One has to evaluate the performance of UP with respect to the health infrastructure that it has, which is in a pathetic situation. If you start with a pathetic health infrastructure, you are expected to encounter utter disaster. The fact that that did not happen is at least, to a significant part, thanks to the way it was handled.
Ritika Chopra: Given that we don’t have correct figures for Covid deaths for the second wave, do you think it’s a little premature to actually label a state’s handling of the pandemic as great or otherwise?
Since the death data was not available, we did not analyse the handling from that aspect. But data was available to analyse so many other aspects. For example, the handling of the migrants, loss of livelihood, distribution of food. Every single village was lucky to have these committees, which were created three years ago to handle Japanese encephalitis. So these committees were tasked to monitor every household in the village. They were given basic preventive medicine, paracetamol and such stuff. It worked really well.
There may still have been many deaths. But what if none of this was done? The number of deaths would have gone through the roof.
Harikishan Sharma: Did you factor the fact that if the panchayat elections that happened in UP had been held in a Covid-appropriate manner, we could have saved thousands of lives? What initiatives of the UP government do you think were effective?
The state government made a very important decision to provide transport to all the workers returning in the second wave. Apart from gaining some political mileage, they could control when and where the migrants arrived.
So, they set up centres at each of those points. Every migrant who came down there recorded full details, including the expertise of that migrant. They were quarantined, symptomatically treated at those places, provided with some food and then sent to their homes. Along with that, in their district, the DMs were given a list of these migrants that contained information about their expertise, with the suggestion that wherever there is a project going on, to see if they can be employed. A lot of them found employment through that mechanism.
About the elections, here is something we have been sitting on for a while. We did an analysis of 16 states during the second wave. Five of them had elections—West Bengal, Kerala, Tamil Nadu, Assam and Puducherry—and 11 states did not. We ran proper statistical experiments to study the surge in these states.
And we found statistically there is no difference between the two groups. Which means or suggests that elections did not play a major role in the spread of the pandemic in the five states. This came as a surprise, but it is true.
It’s not that election rallies do not cause a spread, of course they do. But there are many other things which are causing the spread, and this is just one of them. So, if you just remove one and retain the others, it does not seem to be making much difference.