CESSI IISER KOLKATA OUTREACH INITIATIVE ON COVID19

The coronavirus pandemic is unparalleled in modern human history. Its impact has spread far and wide across the world and through all sections of the society. India has not been spared and our daily numbers of CoViD19 infections are still rising. Our country is in the midst of an unprecedented lockdown which has had disproportionate consequences on people with varied economic backgrounds. New information, myths, hoaxes and claims bombard our collective consciousness every day. In these challenging times, research geared towards understanding different aspects of the novel coronavirus virus SARS-CoV-2, creation of scientific awareness about the disease, and effective science communication that may inform public behaviour and guide policy, are crucial towards mitigating the adverse impacts of this pandemic.

The Center of Excellence in Space Sciences India (CESSI), IISER Kolkata have utilized their in-house modelling and data analytics capabilities to create resources intended for spreading scientific awareness about the pandemic among the general public and guiding future policies relating to the same. The resources available here are based on the CESSI-nCoV-SEIRD model which has been optimized for the Indian context at IISER Kolkata, data analysis of India specific and some global data on the progress of the pandemic, and informational graphics and social media messages created by the Indian Scientists’ Response to CoViD19 (ISRC) group – to which IISER Kolkata scientists have contributed.

Further details on the epidemiology model developed at CESSI can be found in the "Model" section. India-specific information on the disease progression and critical parameters characterizing the progression of the pandemic can be found in the "Data Analytics" section. Socio-scientific awareness materials can be found in the "Public Outreach" section.

We have also analyzed the COVID19 progression for different Indian states and cities which can be found below by selecting the name of the state or the city.

Select a state:


Important Note: All model outputs depend on certain starting assumptions and governing parameters that have to be reasonably constrained by observations. For the coronavirus pandemic these are still early days and most observational data are under-sampled. Our simulation set-up, modelling assumptions and governing parameters would be updated as and when more reliable constraints are available and a better physical understanding is achieved. Therefore, we advise that any policy actions based on our research should lay more emphasis on the qualitative trends implied in our simulations rather than actual numbers.

Spotlight India

In this section, we highlight some issues that are very relevant to the coronavirus pandemic in the Indian context. We tackle some outstanding questions and provide model-based solutions that can guide public policy and catalyze socio-scientific awareness. These answers are backed by our model predictions and data analysis of the observed trends in India.


1. Is the Indian national lockdown necessary; what would have happened if there were no lockdown?


The above simulation based on our epidemiological model shows the progression of the coronavirus pandemic in India under a no-containment scenario. The results of this simulation indicates that the number of infected and deceased individuals would have risen rapidly and could have been unacceptably large. Eventually, almost all the individuals susceptible to the disease (in our simulation assumed to be the population of India) could have become infected. Note that we do not include herd immunity in our model; however, even if herd immunity was included, the number of infected individuals would likely have been 60% of the whole Indian population! We could have had the highest number of coronavirus deaths in the world. Our health care infrastructure and hospitals would not have been able to deal with this free-flowing pandemic in the no-containment scenario. A containment slows down the growth rate of the disease leading to more recoveries and less deaths in the long run. A containment also flattens the curve of infection implying a lesser number of infected individuals at any point in time; this allows our healthcare facilities to cope with the pandemic and buys crucial time for formulating strategic plans to deal with the disease. This simulation clearly demonstrates the catastrophic scenario that the Indian national lockdown aims to avoid.

2. How efficient is the Indian national lockdown?


While the containment has been undoubtedly effective in lowering the growth rate of the disease, it has not been ideal. In the above simulation we implement a "close to ideal" containment with the epidemic growth rate reduced to 10% of the originally assumed unrestrained growth rate. For this nearly ideal containment scenario, the simulation indicates that the active infected individuals would have been between 5000 and 15000 with the pandemic peaking in late March. However, this does not agree with the number of observed active infections which has already surpassed this containment scenario by a significant margin and is still increasing. The total number of deaths in an ideal containment scenario would be anywhere between 1500 and 5500, taking reasonable uncertainties into account. Nonetheless, the rapidly increasing mismatch between the observed and simulated active infections indicates that the Indian national containment is not absolutely perfect. An imperfect containment is to be expected when one takes into account human behavioural traits and random events which disproportionately contribute to “super-spreading” the disease.

3. What is the simulated most-likely-scenario of novel coronavirus progression in India; what does this India-specific simulation tell us about the national containment efficiency and eventual numbers of affected individuals?


Over and beyond the intrinsic growth rates and reproduction numbers, the efficiency of the national lockdown and human behaviour controls the growth rate of this pandemic. Based on model fits to the observed growth of the pandemic, our simulation indicates that an increasingly efficient containment (which is implemented via a declining step function across different phases of the Indian lockdown) is able to reasonably match the early available observations. This best-case scenario simulation (black curve) forecast performed by us in late April indicated a peak active infection of about 60,000 around 17 May. Clearly, it appears that the current observation (blue-dashed curve with current number indicated by a dot) has surpassed that indicating a more inefficient containment than assumed. We perform multiple simulations with increasing inefficiency in lock down (i.e., equivalent to more relaxations in the containment) to assess possible future progression of the pandemic. We find that with increasing relaxations in the containment, the simulated active infections rise more rapidly, as expected. The current active infection growth curve is progressing at a rate suggestive of a relaxation in the containment to 30%-40% level (i.e., working at a 70%-60% efficiency).

4. Would it have been better if India imposed a complete national lockdown even earlier in February 2020?

Our model simulations with a lockdown imposed on 26 February 2020, that is a month earlier compared to the actual date of imposition of the Indian lockdown (25 March). This indicates that implementation of an early and strict containment is necessary to control the growth rate of the pandemic. This is because the initial numbers of exposed and infected individuals were lower, and an early imposition of the containment would have contained the disease much more effectively and quickly. However, one must add the caveat that this would still have left a large number of susceptible individuals in the population and only an extended containment until mid-May - even if imposed earlier - would still have been necessary.

5. What do empirical fits to the data tell us about the progression of the disease?

We have binned the data available in public domain for total infections into week-wise distributions after lockdown was declared - i.e. from March 25 (week 1). For each week, we fitted an exponential fit of the nature y = A(t)e-τ t, and noted the values of A(t) and τ. Next, we fit τ as a function of weeks from lockdown, and observed a clear decreasing exponential fit to that as well. Then, we performed another extrapolation to find out the projected value of the time constant for future weeks. The extrapolation of the optimized fit shows that around May 25 (week 9 of lockdown) the value of the exponential time constant would reach around 0.036 - at which point the total number of cases will saturate, i.e. we shall not observe new cases occurring. However, if the time constant for the next week were to be higher than the extrapolated trend line, the time taken to saturate will be longer, and we shall also observe a higher number of total cases. For example, the time constant obtained after the end of Week 6 of lockdown lies above the optimized fit-line, and if stays the same over the next few weeks, we shall have a total of 2.35 lakh cases at the end of May 31. We have also performed an exponential fit biased towards the data obtained on May 5, and observe that just doing that extends the saturation date to June 9. Thus, it would be interesting to observe the value of the time constant over the next few weeks in order to understand whether the time constants would lie on the optimized fit (which would indicate a slow-down), or on the biased fit or even higher, which would indicate a considerable extension of the saturation date and correspondingly, an increase in the total number of cases.

To find the total number of projected infections, we also found out how the amplitude A(t) would evolve with time, for which we noted the fitted values of A(t) and τ from the exponential fits to the lockdown data of different weeks. We proceeded to determine their inter-dependence, which also turned out to be an exponential. Thus, we observe that with the biased fit - the saturation would occur at around the end of the 11th week of lockdown, with the saturation value of cases to be around 91000 ± 6000 (for the normal fit, the same value is around 64000 ± 4500, with the saturation reaching around May 25). Interestingly, we also find that the predictions from the analysis from the biased fit matches closely the number obtained by fitting a sigmoidal function to the entire data between March 1 to May 5. However, it is important to note that these predictions are entirely assuming that conditions remain absolutely unaltered from present existing ones. Also, the predicted value obtained by combining the decay of the exponential time constant and the projected amplitude value should be updated continuously by correcting the time constant and amplitude with respect to actual data released.

Update after Week 12 of lockdown: Our predictions for this week have matched actual numbers pretty well - where we predicted the total number of cases as 3,53,020 with 77,021 new cases - the actual number is 3,54,148, with 78,149 new cases. Thus, we are within 1.5% of the actual number for new cases, which indicates that conditions remain more or less stable in the country with respect to the spread of the virus. Naturally, our prediction for the time constant of the exponential has also been very close to what has actually been observed - with the new value of time constant being almost right on top of the trend line we plotted last week with the fit biased with the data point for May 26 (black dashed line in the figure displaying evolution of the time constant). Again, the best fit (red solid line) has also approached the biased fit a little more this week. After updating the amplitude dependence of the time constant, we now predict the number of new cases to be 4,38,439 ± 20,000 for the coming week (June 17-23). This means around 84,291 new cases with an error of around 20,000 (or 25%). The saturation is now around July 28, with a total number of expected cases of 6.45 ± 0.6 lakhs in India.The sigmoidal fit now gives around 7.5 ± 0.15 lakh cases, which means that the values from the two methods now agree at the 1σ level. The flattening of cases by this approach is around the fourth week of September, around 2 months later than the prediction from the decay of exponential time constant model. Updated on: June 17 2020.

6. Does empirical model fitting to the observed data on the novel coronavirus pandemic progression in India motivate the extension of the Indian national lockdown to 17 May 2020; how reliable are such empirical model fitting based extrapolations?


It is possible that the extension was in parts motivated by empirical understanding of the progression of the coronavirus pandemic in India. Our empirical fit using an appropriate function to the Indian data as on 1 May 2020 indicates that the growth of cumulative infected individuals and total number of deaths would saturate just before 17 May 2020. This empirical model fitting implies the pandemic would be over by 17 May. However, this empirical extrapolation to the future assumes that the fit parameters remain constant and would not change over time. By breaking up the available India data on disease progression into different time slabs we perform additional analysis to test this assumption. We find that the empirically derived fit parameters change over different time windows. This indicates that a simple, empirical curve fitting of past data to predict the future course of the pandemic may provide misleading information. For example, the overall empirical fit which fits the past data well also predicts that by now the number of daily new infections should be saturating. This appears to be contradicted by the current observational trends of significant growth in the number of daily infections. We shall have a better idea about the applicability of this empirical model extrapolation by 7 May and will update this comment accordingly. We note that our dynamical epidemiological model indicates the lockdown may need to be extended even beyond 17 May, in contradiction to the indications from the simplified empirical fitting.

7. What is the best strategy for continuing a national lockdown or isolated regional lockdowns in various parts of India?

We are currently performing research with our epidemiological model - optimized for India - to figure out a desirable containment strategy that can inform public policy. While we are exploring various strategies through predictive modeling, our early experience is that there is no one magic solution. This also indicates the difficulties that confront public policy makers at central and state levels who must rely on such scientific inputs. As and when we are able to identify some plausible strategic interventions, we shall make them available to the nation.

MODELLING AND DATA ANALYTICS TEAM

The team involved in this work consists of faculty, graduate and undergraduate students at the Center of Excellence in Space Sciences India (CESSI) and the Department of Physical Sciences at IISER Kolkata.

Modelling: Shaonwita Pal, Soumyaranjan Dash and Dibyendu Nandi
Data Analytics: Agnibha Banerjee, Vishal Singh, Dibyendu Nandi, Rajesh Nayak and Ayan Banerjee.
Webpage: Soumyaranjan Dash, Agnibha Banerjee and Prosenjit Lahiri

CONTACT

Prof. Dibyendu Nandi
dnandi at iiserkol dot ac dot in

ACKNOWLEDGEMENTS

CESSI is funded by the Ministry of Human Resource Development, Government of India under the Frontier Areas of Science and Technology (FAST) Scheme. The research and analysis documented herein utilized the CESSI computational facilities. Soumyaranjan Dash is funded by the INSPIRE program of the Department of Science and Technology, Government of India. Shaonwita Pal is funded by the University Grants Commission, Government of India. Agnibha Banerjee and Vishal Singh are funded by KIshore Vaigyanik Protsahan Yojana of the Department of Science and Technology, Government of India.

counter free counter free