key: cord-0226059-2bm7mofj authors: Harsha, Prahladh; Juneja, Sandeep; Mittal, Daksh; Saptharishi, Ramprasad title: COVID-19 Epidemic in Mumbai: Projections, full economic opening, and containment zones versus contact tracing and testing: An Update date: 2020-10-29 journal: nan DOI: nan sha: 1e46850b306a557e8a7447675a8a2e08391ce262 doc_id: 226059 cord_uid: 2bm7mofj Mumbai, amongst the most densely populated cities in the world, has witnessed the fourth largest number of cases and the largest number of deaths among all the cities in India (as of 28th October 2020). Along with the rest of India, lockdowns (of varying degrees) have been in effect in Mumbai since March 25, 2020. Given the large economic toll on the country from the lockdown and the related restrictions on mobility of people and goods, swift opening of the economy especially in a financial hub such as Mumbai becomes critical. In this report, we use the IISc-TIFR agent based simulator to develop long term projections for Mumbai under realistic scenarios related to Mumbai's opening of the workplaces, or equivalently, the economy, and the associated public transportation through local trains and buses. These projections were developed taking into account a possible second wave if the economy and the local trains are fully opened either on November 1, 2020 or on January 1, 2021. The impact on infection spread in Mumbai if the schools and colleges open on January first week 2021 is also considered. We also try to account for the increased intermingling amongst the population during the Ganeshotsav festival as well as around the Navratri/Dussehra and Diwali festival. Our conclusion, based on our simulations, is that the impact of fully opening up the economy on November 1 is manageable provided reasonable medical infrastructure is in place. Further, schools and colleges opening in January do not lead to excessive increase in infections. The report also explores the relative effectiveness of contact tracing vs containment zones, and also includes very rudimentary results of the effect of vaccinating the elderly population in February 2021. • We use the data from [5] for age distribution in high-density areas and low-density areas to capture the fact that the high-density areas have a younger population (see Figure 1 ) . This results in a lower fatality rate per infection within the high-density areas, and also an increase in spread within the high-density areas since there are more individuals in the working age-group. • The population of Mumbai is taken to be 12.8 million as opposed to 12.4 million. This in consonance with mid-year estimated population 2019 data with MCGM that updates the 2011 census data [6] , [7] . • The intervention modelling includes lower compliance and higher community level interactions during the festive seasons. The details of this are explained in subsection II-A. The net result of these changes is that in our model slums see high infections early on largely during April to June. Infections thereafter are primarily from non-slums. This matches the Mumbai scenario much better. Later in this report we also conduct some rudimentary experiments to ball park gauge the benefits and the costs of introducing perfect vaccines amongst the older population of Mumbai. Our simulations further suggest that by mid-January 2021, the prevalence (fraction of the population infected) can be seen to be stabilising close to 80% in slums and 55% in nonslums. This stabilisation and high prevalence indicates that Mumbai city may have more or less reached "herd immunity" by then. By this we mean that the new infections and related medical indicators in the city will be substantially reduced compared to their peak values in mid-May and June 2020. In these simulations we also conduct counterfactual experiments where the containment efforts as well as the contact tracing and testing efforts are varied and their impact on the health indicators is measured. Our simulations suggest that containment efforts do a better job of slowing infection spread compared to increasing the contact tracing and testing efforts. Increase in the latter leads to only marginal improvement in slowing the infection. Below, we list some policy recommendations for opening up the workplaces and schools and colleges in Mumbai that incorporates above considerations. • We had in the earlier version of this report [4] , recommended gradual opening of the workplaces so that the increase in infections that may result from increased occupancy in local trains and other public transport is manageable. We continue to recommend 4 similar gradual opening up of the city and that the economy may be fully opened up by November 1 or soon thereafter, again with the opening up process carefully based on observed infections. • Schools and colleges may be opened by first week of January 2021. As mentioned earlier, our simulations suggest that the resulting second wave from this opening up is minimal. • Social distancing in public transport, staggering of office times, use of shifts to the extent feasible is recommended. • Prevailing hygiene measures such as mandatory use of masks/face-covers, encouragement of regular hand-hygiene, regular disinfection of "high-touch surfaces" in trains and workplaces [8] etc. should continue as before. • Our analysis of containment zones vis-a-vis contact tracing and testing suggests that wherever feasible and when the economic costs are not prohibitive, containment in regions where infection is seen to be present is a desirable option to slow the infection spread. In addition to the various modeling assumptions listed in our previous reports [3] , [9] , our recommendations rely on two important assumptions. 1) Our first assumption is that the population, by and large, will continue to observe social distancing precautions including wearing of masks. This could change as public perception of risk changes over time. This may lead to increase in infections not accounted for in our projections. 2) Our second assumption is that the reinfection probability is sufficiently small for that population of Mumbai that it can be ignored in opening up of the city. We note that cases of reinfection have recently been reported. However, the number of such reports continues to be very few. If this changes and reinfection happens to a non-negligible proportion of the population, then our projections become less valid. Vaccines: It is believed that a COVID-19 vaccine will become available sometime in 2021. Important decisions related to prioritising people to vaccinate and to manage the storage and distribution of vaccines would require careful analysis. This analysis would also include the cost and time taken in vaccine production, the effectiveness of the vaccines, the frequency of administering them to each person, the time that a vaccine takes to provide immunity, and the duration for which the immunity is provided. These are complex issues that require detailed analysis. Later in this report we conduct some very rudimentary simulation experiments 5 related to administering vaccines under simplified assumptions to get a ball park idea of the potential costs and benefits, in terms of number of fatalities and the load on medical facilities, of administering vaccines to relatively older population in Mumbai. Our broad conclusions are that Greater Mumbai has an estimated 13.1 lakh people aged 60 and older. If these are all vaccinated by February 1 with a vaccine that provides instantaneous and perfect immunity, then the number of fatalities post February 1 will reduce by estimated 53% from around 950 to 450 in the next six months. The hospitalisations (including critical cases) will reduce by estimated 40% from around 8840 to 5340 in the next six months. Similarly, if estimated 29.3 lakh 50 years and older Greater Mumbai residents are vaccinated on February 1, then the number of fatalities post February 1 will reduce by estimated 64% from around 950 to 340 in the next six months. The hospitalisations (including critical cases) will reduce by estimated 67% from around 8840 to 2910 in the next six months. There have been varying news-reports on the differences among the population in how the disease spreads with respect to age, especially among the younger population. In particular, the role of the younger population in the transmission of the disease is not well-understood. In our earlier report [4] , we took a cautious view and recommended opening of schools and colleges only as late as January 1, 2021, several months after the opening of the economy. Recent studies in the Indian states of Andhra Pradesh and Tamil Nadu [10] suggest that children in the age group of 0-14 pose a significant transmission risk, corroborating our recommendation. Caveats: We emphasize that our simulator is intended primarily as a tool for comparing the effectiveness of different non-medical interventions to assist decision making. In particular, the simulator, due to the inherent model uncertainty, is not intended as a tool for predicting absolute numerical values of COVID-19 cases. In our informal view (which is difficult to validate scientifically), a confidence interval of ± 20% to 30% around the projected numbers may be reasonable to capture the model uncertainty. This number may be larger when estimates with small values are considered. On the other hand, the statistical error due to the random noise in the simulations is much smaller and is easily controlled. We also recognize that many of the non-pharmaceutical interventions considered in our study, especially when they remain implemented over a long duration, may lead to important social and economic concerns and consequences, beyond their effect on the evolution of the epidemic. Prolonged restrictions, besides economic concerns, may also lead to disruption of many essential supply chains and medical services. The World Health Organisation Pulse 6 report [11] , based on responses from over 100 countries, show significant reductions in routine vaccinations, diagnosis and treatment of noncommunicable diseases, antenatal care, cancer diagnosis and treatment, and many others. These are important factors that must be taken into consideration when implementing any prolonged restrictions. However, we do not know how to quantitively project many of the socio-economic effects of a non-pharmaceutical intervention. The scope of our simulator is to project purely COVID-19 related stresses on the medical infrastructure. The modelling of such socio-economic effects, though important, remains beyond the scope of our simulator. Similar to our previous report, we emphasize that this report has been prepared to help researchers and public health officials understand the effectiveness of social distancing interventions related to COVID-19 in terms of the stresses on the medical infrastructure. The report should not be used for medical diagnostic, prognostic or treatment purposes or for guidance on personal travel plans. Greater Mumbai (consisting of Mumbai and Suburban Mumbai) has a population of about 1.28 crores (12.8 million) and a population density of roughly 21,000 per km 2 making it one of the densest cities in the world 1 . Further, about 53% [12] of Mumbai lives in cramped dwellings with shared sanitation facilities where the population density may be 5 to 10 times larger than other parts of the city. In addition, crowded suburban trains are the lifeline of the city where the suburban railway system serves more than 80 lakh (8 million) passengers on a weekday, in normal times [13] . It is generally believed that the infection spreads faster in denser areas, due to increased contacts in these areas. Given these factors, the public health threat in Mumbai is particularly acute. The importance of modelling the effect of infection spread arising from the gradual opening and relaxation of lockdown measures, for a city like Mumbai, cannot then be over-emphasized. We model the spread of infection in the city using our IISc-TIFR agent-based city simulator [3] . For completeness, we briefly review it below. Agent-based city simulator (ABCS): As described in detail in [3] , our agent-based simulator creates a synthetic model of about 1.28 crore (12.8 million) residents of Mumbai that matches the city population ward-wise, and matches the numbers employed, numbers in schools, commute distances, etc. This is done by suitably populating households, schools, and workplaces with people. Several interaction spaces including households, local communities, 1 Some of the discussion in the Introduction first appeared in [9] progression parameters are based on [14] . In this work we report the following scenarios: • Long term forecasts: We develop long term forecasts till March 15, 2021 under the following six scenarios: -Containment effort set at 75% and at 60%. Exact modelling of containment effort relies on the modelling feature 'neighbourhood containment zones' introduced in [3] and is discussed later in Section III. -Train infection levels are set at normal β T = 0.19 × β H (see [9] for detailed calculations to arrive at this number; that report also discusses the household transmission parameter β H and the rationale that relates it to β T . The value of β H used is calibrated primarily to fatality data, and is given in Figure 2 .) as well as to the larger value of β T = 0.30 × β H to account for the additional infections that may occur in trains through passengers coming to Greater Mumbai through the neighbouring areas in the Mumbai Metropolitan Region (MMR). We further consider a more pessimistic higher value of β T = 0.40 × β H to account for the inherent uncertainty in estimating β T without availability of relevant data on infection spread through trains. The workplace attendance is a good measure of economic activity. It is set in our model as follows: Lockdown till May 17. Mobility to workplaces set at 5% from May 18 to May 31. In June this is set at 15% and it increases to 25% in July. It is set at level 33% in August. For September and October it is set at 50%. And thereafter it fully opens (100% attendance) from November onwards. 8 Festivals in Mumbai are a time for increased intermingling amongst the population, and during this time compliance on wearing masks, maintaining social distance, etc. is likely to be less. We account for this in our model as follows: -To account for increased intermingling due to the Ganpati festival, from August 20 to September 1, we increase β for community by 2/3, and we reduce compliance from 60% in non-slums and 40% in slums to 40% in non-slums and 20% in slums. With this adjustment we observe that the fatality data from the model is fairly close to the actual observed fatalities in months of September and October. -We do a similar adjustments for Navaratri and Dussehra for the period October 19 to October 25. For Diwali we conduct similar adjustments for the period November 8 to November 14. The developed model is validated by comparing the model projections with the observed health data, that is, observed number of fatalities, hospitalisations and critical cases. As in [9] , in all our simulations we continue to assume that outside the festival times, 60% of households are compliant in residential, relatively low density areas (non-slums), while 40% of households are compliant in slums or high density areas. • Fully operational economy: We consider the following three scenarios: 1) The workplaces fully operational on November 1 and school/colleges open on January 1, 2) workplaces fully operational on November 1 and school/colleges remain closed, 3) workplaces and school/colleges fully operational from January 1. Fully opening workplaces or the economy implies that the trains are back to the capacity as in normal pre-covid times. To err on the side of caution, we keep the train beta at a high risk level, that is, β T = 0.4 × β H , in these scenarios. • Containment zones and contact tracing and testing are regarded as two important policy tools available to decision makers in slowing the epidemic spread. Our small network framework (introduced in [3] ) through the neighbourhood cells allows us to model neighbourhood containment efforts with reasonable accuracy. Further, the small network framework with the introduction of the community of friends and neighbourhood community, allows us to plausibly model the contact tracing and testing efforts. The methodology to aid in measuring contract tracing and testing in our model is introduced in [3] . This modelling feature is further discussed in Section IV. Through our model we evaluate the performance of containment efforts by measuring the health indicators as a 9 Age Group Slum age Distribution Non Slum Age Distribution Figure 1 : Age Distribution for Greater Mumbai. Age distribution for slums and non-slums for 12 years and above is obtained from [5] . We combine this with [6] and [7] , assuming that both slums and non-slums have the same percentage of population younger than 12 years, to arrive at the overall age distribution for Greater Mumbai. function of varying containment efforts. We similarly evaluate the the impact of varying level of contact tracing and testing efforts on the health indicators. While containment zones are relatively easier to administer compared to contact tracing and testing, our simulations suggest that the former may also be more effective in slowing the spread of the infection in the city. A caveat to keep in mind is that containment zones lead to restricting movements of relatively large number of people, and so may come at a significant economic and social cost. • Impact of Vaccination: We consider the following simplistic scenarios: Population above the age of 60 is vaccinated on February 1. The vaccine comes into effect immediately and is 100% effective so that the vaccinated population instantly becomes immune. We also consider the scenario where population above the age of 50 is vaccinated on As detailed in our previous report, each of the interaction spaces is further broken down in subnetworks that corresponds to most of the interactions that an agent has within that 10 As in the earlier report [3] , the contact rates are calibrated to match the observed growth of fatalities, and to have roughly equal contribution of infections from the household, community and workplace networks (including the subnetworks) in the "no-intervention" scenario. While in [9] the containment zones were modelled at the ward level for computational ease, in the current implementation we aim for more accuracy through a finer and more accurate model of containment. In particular, we model containment at a neighbourhood cell level. Recall that our synthetic city is divided into a grid of square cells where length of each cell is 178 meters. Containment effort is modeled as an increasing adaptive function of the active hospitalisations observed in the neighbourhood cell. Number of hospitalisations is taken as the decision variable since it is easily observable as compared to tracking the number of positive cases in a cell, which may be harder to estimate accurately without extensive testing, and the two are highly correlated. The cell is incrementally closed as more number of hospitalisations are observed in it. 11 Specifically, suppose that the containment effectiveness (CE) =75% and there are 3,000 residents in a neighbourhood cell. Then, first hospitalisation leads to movement restriction of 25% internally amongst the residents as well as to and fro from the cell; second hospitalisation leads 50% movement restriction, and third hospitalisation onwards leads to 75% movement restriction. If, on the other hand, the neighbours in the cell are less than a thousand, then as long as there exists a hospitalised person in the cell, movement of every resident within the cell, as well as movements into and out of the cell are restricted by 75%. More precisely, If containment effectiveness is set to a fraction y, and the neighbours in a cell equal n thousand, then every hospitalisation leads to y/n restriction in movement, and total movement restriction is capped at y. Thus, percentage of activity restriction (internally as well as in entering and leaving the cell), or containment effectiveness, is our control and we set it to min (hy/n, y) , where h denotes the number of people in the neighbourhood cell that are hospitalised 2 . As in [9] , in all our simulations, outside of festival times, we set the compliance levels to 60% in residential or non-slum areas, and at 40% in high density or slum areas. This level of compliance with additional measures such a mask usage, case-isolation and home quarantine post lock-down, restriction on those above 65 to stay home, closed school and colleges match reasonably well the observed data on fatalities. Further, accounting for the results of the Mumbai SeroSurvey [2] , and in deviation from our earlier analysis in [9] , we reduce the proportion of symptomatic population amongst those exposed to the Covid-19 disease to 40% from the earlier 66.67%. In addition, in our earlier simulation runs for Mumbai in [3] , [9] , [4] β values for homes and communities in slums were kept at two times the values for non-slums. In the current simulations this factor is increased to 3. As mentioned earlier, this better matches the observed prevalance estimates for Mumbai. The model is recalibrated to data using these adjustments. We first discuss the long term projections. As mentioned earlier, we consider these under the workplace attendance scenario where after lockdown till May 17, there is 5% attendance 12 from May 18 to May 31st. This increases to 15% attendance in June, 25% in July, 33% in August, 50% in September and October and fully opens November onwards. In addition, as discussed earlier, we adjust for increase intermingling in population during the three festival times. These projections are developed till March 15, 2021 under the following six scenarios. • The containment effectiveness kept at 75% as well as 60%. • The infection rate from trains kept at base risk level β T = 0.19 × β H (recall that β H corresponds to the household transmission parameter). This was derived as a plausible rate of infection in trains in [9] . To account for the uncertainty in such calculations and population from outside Greater Mumbai using Mumbai locals, we also consider the more pessimistic medium risk setting of β T = 0.30 × β H and high risk setting of As in Reports [3] and [9] , in our simulations, a synthetic city is created that match the aggregate Mumbai demographic data. For this city we run 5 independent simulations. The reported results are the average of these five runs. In Figures 3a and 3b , we show the daily as well as the cumulative number of infections under the six scenarios. These results suggest that the growth of infections in Mumbai started to slow down from June, with slight increase every time the economy opened further or due to the festival season. From January 1 onwards the number of new infections becomes very small indicating that by and large herd immunity has been reached by the city. Furthermore, this suggests that city will stabilize with close to 8-9 million residents infected. In Figure 4 , we map the prevalence for all of Mumbai suggested by the model when CE is set to 0.60 and β T is set at 0.4 × β H . We also separately plot the prevalence for slums and non-slums. The salient observations are that herd immunity is reached at different level of prevalence in slums and non-slums. In slums this is attained at around 85%, while in non-slums the number is closer to 60%. The Mumbai SeroSurvey [2] suggested around 55% prevalence in slums and 15% prevalence in non-slums of the three wards of Mumbai that were sampled around the first two weeks of July. Due to a typical gap of 1-2 weeks between the time the infection happens and it can be detected by a serological test, we compare this with our model output on July 1. Our respective numbers on July 1 of 55% and 15% are more or less identical to theirs. Figure 5a shows the daily number of the hospitalised patients as per our model under the six scenarios. In Figure 5b , we compare our hospitalisation numbers to the hospitalisation numbers reported by BMC in their Dashboard [15] . The BMC dashboard reports Dedicated Covid Hospital (DCH) as well as Dedicated Covid Health Centre (DCHC) aggregated together 13 and these are reported as hospitalisations under DCH and DCHC. We make the following adjustments to the data series from our model as well as from the BMC Dashboard to make the comparison between them more apples to apples. as Nasik. These are roughly estimated to equal 30% (based on feedback from a BMC official). To account for these, we increase our hospital patient projections by 30%, with the understanding that this increase is reasonable for comparison with observed data beyond mid-July. ICU. In our model we report patients under ICU (critical patients) separately. Thus, we remove the ICU patients in the Dashboard data from the DCH and DCHC data. 3) Further, based on the snapshot data provided by BMC on August 1 and August 20, we inferred that about 13% of patients in DCH and DCHC are asymptomatic. The disease progression data in our model is taken from [16] and [14] , and here hospitalised patients correspond to those with serious symptoms. Thus, to compare our projections with the observed data, we further reduce the DCH and DCHC reported numbers by 13%. It can be seen from Figure 5b that with the above corrections, post mid-July, our projections are reasonably close to the adjusted DCH and DCHC numbers. In Figure 6a , the projected daily number of the critical cases under the six scenarios are shown. Again, to compare with the Mumbai ICU numbers as per the BMC Dashboard, we scale our numbers by 41% in Figure 6b . These again our based on a snapshot input provided by BMC where 41% of the critical beds used by the Mumbai population were in use by population from outside the city. Here too, the match between the adjusted model and data after mid-July appears reasonable. From May 27 to June 16 the occupancy of reported ICUs was seen to be above 98% as per the BMC Dashboard (except on May 29, when it was 97%). This may at least partially explain why the model numbers for critical cases in this period are much higher than the actual numbers. Figure 7a shows the projected fatalities based on our model and compares them to the fatality data as reported by BMC through their Dashboard. Figure 7b shows cumulative number of fatalities as a function of time. As pointed out in earlier reports [3] and [9] , in 14 our model the key transmission rates are calibrated to primarily match the initial observed fatality data in Mumbai as well as in the rest of India. Further, the compliance parameters used in the model are fine tuned so that the model fatalities are close to the reported fatality data (as reported by BMC through their Dashboard [15] ). As evident from the two figures, the match between the model generated fatality and the reported fatality data appears to be quite good. Few points are in order. • BMC in mid-June had updated the reported deaths data. Figure 7b shows both the original and the updated reported fatality data. Observe, that our model (the data series corresponding to CE 0.60 and base or medium train risk β T level) slightly underestimates this series from mid-May to mid-June. One reason for this may be that while in our model there is no limit on ICUs for critical patients, the city of Mumbai did observe this shortage around that period. • Our model reports higher number of deaths (under CE 60%, and high train risk β T value) compared to the reported from July onwards. This may be partially explained by the fact that around mid-June BMC stopped testing dead bodies for covid [17] . A broad conclusion suggested by our model is that the fatalities in the city will stabilize within 13,000 to 14,000 by March 2021. In Figures 8a and 8b , we report the observed daily and the cumulative cases in Mumbai. In Figure 8a we also report the daily tests conducted to illustrate the correlation between these and the observed cases. This correlation is especially strong from September first week to mid-October. Since these numbers are a function of the testing strategy followed, they are difficult to estimate. Later, in Section VI, we discuss our contact testing and tracing strategy that was tailored to the case data from mid-May onwards until the last ten days of August. However, since then the amount of testing in Mumbai has significantly increased and we do not have an accurate model to estimate the number of positive cases thereafter. In Figure 9 we report the time series of fatalities in slums and non-slums as per our model slums is attained at around 80%, while in non-slums it is attained at prevalence close to 55%. 17 (a) Simulated daily hospitalised patients in the city under the workplace opening schedule 5% attendance, May 18 to May 31st, 15% attendance in June, 25% in July, 33% in August, 50% till October and 100% from Nov. Includes Ganpati, Navratri/Dussehra and Diwali relaxations. (b) Comparison of simulated daily hospitalised patients in the city with the DCH and DCHC numbers reported by BMC. The simulated numbers are increased by 30% to account for estimated patients coming from the other MMR areas. These cases came to Mumbai mainly from around mid-July. The ICU numbers are removed from the reported DCH and DCHC numbers. These numbers are further reduced by 13% to remove the estimated asymptomatic patients in DCH and DCHC to facilitate comparison. For the comparison between the two strategies, containment zones are implemented as explained in Section III, while contact tracing and testing strategy designed to reflect the testing strategy and data from mid-May to mid-August is explained below. The contact tracing machinery can be briefly described as follows: 1) An individual that the simulator deems as a hospitalised case undergoes a COVID-19 test with some probability (specified by the protocol given in Figure 13 ). If the test turns out to be positive, this agent is deemed as a hospitalised index case. This hospitalised case is typically tested with probability 1, although later in Section VI when we evaluate medical statistics under different testing protocols, we allow this probability to take lower values of 0.66 and 0.8 as well. 2) When an index case is identified, a fraction of agents from their subnetworks (specified by the protocol) are quarantined and marked as primary contacts. 3) Each primary contact is tested with some probability (specified by the protocol). If the test is positive, then such agents are marked as positive index cases. The newly discovered positive index cases would additionally initiate contact trace around this agent like in the hospitalised index cases. In the current implementation, 0.5% of the neighbourhood cell (which is 5 agents on average) and 100% of all other subnetworks are deemed as the agent's primary contacts. Furthermore, the testing probabilities in our current implementation are set to match a rough test positivity rate of 30% to 40%, which is the observed test positivity rate during the months of June and July in Mumbai [15] . The conclusion is that while contact tracing and testing does help in slowing the spread of the infection, the amount of reduction appears much less compared to that achieved through containment efforts, particularly since the latter appears to be cheaper and easier to implement. The comparison between benefits of containment vis-a-vis contact tracing and testing is crystallised in Figure 18 where we plot the peak of moving ten day average of daily hospitalised patients as a function of these efforts. The contact tracing and testing effort measured on the x axis of the right hand figure corresponds to the hospital index probability for values 0.66, 0.80 and 1.0. We also consider the two higher contact tracing and testing cases where the hospital index probability is kept fixed at 1 but the remaining testing probabilities in the protocol are increased by 25% in one case and 50% in the other. These cases correspond to x-axis values of 1.25 and 1.5 in Figure 18 . The demographics of Greater Mumbai as estimated from Figure 1 suggests that approximately 13.1 lac Mumbai residents are aged 60 years or above. According to our simulations, out of these 13.1 lac, approximately 5.6 lac will be susceptible on Feb 1. Similarly, the total number of residents aged 50 and above is 29.4 lac whereas around 10.3 lac will be susceptible on Feb 1. Since, we may not be able to, or it may not even be desirable to, separate the susceptibles from the recovered population from vaccination viewpoint, we may have to vaccinate everyone in the respective age group to provide them immunity. Easing of restrictions and phase-wise opening of lockdown. (Mission Begin Again) Seroprevalence of SARS-CoV-2 in slums and non-slums of Mumbai, india, during City-scale agent-based simulators for the study of non-pharmaceutical interventions in the context of the covid-19 epidemic Covid-19 epidemic in mumbai: Long term projections, full economic opening, and containment zones versus contact tracing and testing Seroprevalence of SARS-CoV-2 in slums and non-slums of Mumbai District Census Handbook 2011 -Mumbai District Census Handbook 2011 -Suburban Mumbai Q&A: Considerations for the cleaning and disinfection of environmental surfaces in the context of covid-19 in non-health care settings Covid-19 epidemic study ii: Phased emergence from the lockdown in mumbai Epidemiology and transmission dynamics of COVID-19 in two Indian states Pulse survey on continuity of essential health services during the COVID-19 pandemic: interim report Municipal Corporation of Greater Mumbai Mumbai Sub-urban Rail Passenger Surveys and Analysis Estimates of the severity of coronavirus disease 2019: a model-based analysis BMC COVID-19 Response War Room Dashboard Impact of non-pharmaceutical interventions (NPIs) to reduce COVID19 mortality and healthcare demand Maharashtra stop testing dead bodies to ease painful delays for kin We thank our colleague Piyush Srivastava for many useful suggestions that helped our analysis. We thank Piyush as well as our IISc collaborators R. Sundaresan, P. Patil, N.Rathod, A. Sarath, S. Sriram, and N. Vaidhiyan for their tireless efforts in developing the IISc-TIFR Simulation model [3] and their key role in our earlier report on Mumbai [9] .We thank Mrs. Ashwini Bhide, AMC, MCGM for her insights and for her crucial data inputs. We also thank Bhaskaran Raman for his critical inputs on the socio-economic effects of prolonged non-pharmaceutical interventions. We thank Siddarth Raman, a volunteer at BMC, for his help with data.Authors acknowledge support of the Department of Atomic Energy, Government of India, to TIFR under project no. 12-R&D-TFR-5.01-0500. RS is also supported by the Ramanujan Fellowship of DST. We also acknowledge the support of A.T.E. Chandra Foundation for this research.