Historic 21-day lockdown, predictions for lockdown effects and the role of data in this crisis of virus in India

This article is the second one of a three part series on the novel corona virus outbreak in India. You can read the first part (pre-lockdown) here and the final part (unlocking the lockdown) here. You can also access a related pre-print here. All data source, code and interactive plots are available at covind19.org.

India, a democracy of 1.34 billion people has been aggressive in adOn the evening of March 24, India’s Prime Minister Narendra Modi announced a three-week nationwide central lockdown — starting Wednesday, March 25 at midnight, to slow the spread of the novel coronavirus. This is the biggest lockdown in the history of the world happening in a democracy of 1.34 billion people. At that time India had reported 536 COVID-19 cases and 11 fatalities. The people of India received essentially 3 hours’ notice to prepare for a three-week long lockdown. This is a wrenching decision for any nation, particularly for India with 22% of the population living below the poverty line, 90+% of its workforce employed in the informal sector, and poor housing conditions for one in six urban dwellers. As debates rage around the world regarding India’s central lockdown, and while policy experts are analyzing the timing, management and implementation of such a monumental decision, we present our thoughts on lockdown-related issues through a data-centric lens. As a follow-up to our article on Medium.com on Saturday, March 21, 2020, we want to clarify the role of epidemiological models, projections and predictions in informing critical policy decisions. In addition, we introduce a new interactive visualization platform for COVID-19 data in India for public use with open source software code.

1. Was a central lockdown necessary? With the 21-day lockdown period in place, how many cases can we expect in India? How many fatalities can we expect in the next few weeks?

A central lockdown, if rightly done, can be a highly effective prevention and control strategy for arresting the spread of the virus. Epidemiologic and mathematical models and the examples of other countries provide support for the effectiveness of a lockdown. However, decision-makers cannot act from a pure epidemiologic perspective as there are huge economic, ethical, moral and social implications associated with this choice. Policymakers face unchartered territory here with economic and public health interests competing with each other. It will take us years to know what the optimal decision for any nation was from a rigorous risk-benefit analysis framework. Governments and policymakers across the world are faced with a truly daunting scenario with massive consequences on either side. Any decision will lead to different types of misery, suffering, anxiety and pain for the public. As stated in a recent article in the Harvard Business Review, it is important to quickly learn from successes and failures of an intervention implementation in its early phase and nimbly turn the course if needed to effectively manage the COVID-19 crisis. Flexibility, data-adaptability and elasticity in decision-making are critically needed in this rapidly evolving landscape.

The COVID-19 crisis is a complex problem requiring a multi-pronged, long-term, systems-based approach instead of discrete tactical moves. The mathematical forecasting models cannot capture the devastating consequences a lockdown may have on the lives of people. An all-encompassing and timely cost-benefit analysis of the situation that identifies the optimal course of action is extremely difficult to formulate right now, if not impossible. For example, the goal of any intervention like a lockdown is to reduce mobility and transmission probability. With millions of daily wage earners and migrant workers crowding in buses, or walking miles to return to the hinterland and congregating in shelters and soup kitchens [NYTimes, CNN] the lockdown may even increase the spread initially before the spread (hopefully) settles down. Management and execution of a lockdown for a country of this size has to balance between the collateral damage due to the lockdown and the efficacy of the lockdown itself to stop the virus.

From a public health perspective, if a lockdown had to be imposed at some point of time during the course of this pandemic, it is best to do it early. A lockdown with a few hundred cases may appear to be an overreaction early in the disease course, but in reality, it can save many lives from COVID-19. Getting through the virus crisis will also lead to eventual economic recovery, thus a centralized national lockdown is one of the recommended choices by experts [Bill Gates on Anderson Cooper. “Stop the virus before you can do anything about economics”].

Effect of lockdown on the number of cases: From a purely epidemiologic/mathematical modeling point of view we created some projections using the eSIR model [Wang et al. 2020] under the following hypothetical scenarios of interventions so that the readers can get a sense of the predicted number of COVID-19 cases prevented by an intervention on a relative scale:

• Natural course of the epidemic is followed till April 30 with no intervention
• Social distancing and travel ban continue from March 12 to April 30 without a central lockdown, and
• Social distancing and travel ban till March 25 followed by a 21-day lockdown till April 14, followed by a gradual return to normal (pre-intervention) activities till April 30

We try to understand where we will stand on April 30 under these various speculative scenarios. Figure 1 shows that the growth in the number of cases can appreciably slow down with an effective lockdown, potentially reducing the cumulative number of cases from 35,796 (which is the prediction under no lockdown but with just social distancing and travel ban) to as low as 3,951 (with the lockdown) on April 30. As with many forecasting models with limited data, these estimates should not be over-interpreted as they have large uncertainty and assumes lockdown is strictly followed (upper credible interval or CI 192,985 and 31,979, respectively for these projections). Thus, the 21-day lockdown will likely have a strong effect on reducing the predicted number of cases in the short term. If the implementation of lockdown measures and public adherence takes longer (a scenario we present in our Shiny app), the reduction in the number of cases will be less. However, regardless of the speed with which adherence to the lockdown takes place, the lockdown, if implemented correctly in the end, has a high chance to reduce the number of COVID-19 cases in the short term and buy India invaluable time to prepare its healthcare and disease monitoring system.

Figure 1. Daily growth in cumulative case counts in India up to April 30. Observed data are shown for days up to April 1. Predicted future case counts for April 2 and onwards based on observed data until April 1 using the eSIR model. Three hypothetical forecasting scenarios are plotted: no intervention (red), social distancing and travel ban without lockdown (yellow), and lockdown with moderate release to normal activities after three weeks (blue). The dashed blue line represents the upper credible limit for the lockdown with moderate return scenario. Daily updates to the plot available at covind19.org. Please note the log-base 10 scale on the y-axis.

Fatalities: It is hard to predict fatalities this early for India as we have 53 fatalities so far (April 2; bing tracker). The cumulative number of cases in India is also low compared to other countries (Figure 2). Estimation of fatality rate in general is laced with the problem of selection bias in case ascertainment. We can only speculate by looking at data from other countries. South Korea has relatively fewer COVID-19 related fatalities — 167 out of 9887 cases (1.7%) — compared to several other countries with significant outbreaks, where the case fatality rate often exceeds 4% [CEBM Report]. South Korea achieved this low fatality rate without a central lockdown because they could arrest the spread of the epidemic earlier in its course with social distancing guidelines and rapid widespread testing (free for suspected cases and cluster testing of communities where cases belong), plus expansive contact tracing using credit card activity and mobile phone tracking [Can South Korea be a model for virus-hit countries?]. On the other hand, Hubei province of China arrested the growth rate of deaths by a combination of strategies: lockdown, centralized quarantine of confirmed and suspected cases, their close contacts (asymptomatic) and those with COVID-19 symptoms (separating them from the rest of the population), and rapid expansion of healthcare infrastructure [Wang et al. 2020].

Figure 2. Cumulative number of cases in India compared to other COVID-19 affected countries. Daily updates to the plot available at covind19.org.

India has acted early in the disease course (see Table 1 for comparison with other countries), and so we are hoping that India’s death rate remains closer to or less than what South Korea has seen, leading to approximately 63 deaths (upper CI 511) by April 30. If the fatality rate is more like 4%, the expected number of deaths by April 30 is 158 (upper CI 1279).

Table 1.

2. Will a lockdown get rid of the coronavirus?

No! — we have to remember that the lockdown is not a “cure all” solution. The virus is not going to disappear. The lockdown just delays the epidemic and buys us additional time so that we can prepare our army to go on this war against COVID-19. The war is still on and we should use this time to operate as if we are on war-footing. During this lockdown period, we need to expand and ramp up our testing capacity, set up a nationwide task force for community inspection, surveillance and contact tracing, set up temporary COVID-19 care centers and quarantine facilities, produce and acquire protective equipment for healthcare workers (not just doctors and nurses but also other workers such as hospital cleaners or medical technicians at the frontline of this epidemic). An estimated 5–10% of the cases will require critical care and ventilator support. India has only 35–58 thousand ICU beds and at most 1 ventilator per 2 ICU beds [Does India have enough ventilators, hospital beds?]. India needs to acquire such medical supplies soon in this 21-day period and work on optimal strategies for resource allocation based on estimated influx of patients. Without these strategies, the lockdown will go to waste. Lockdown is an action that buys time, flattens the curve in terms of healthcare demand that a pandemic of this scale entails, and gives us a second window of opportunity for mobilization of resources. The health care system in India, like most countries in the world, is used to individual patient-centric care instead of the community-based care needed during an epidemic, and it has to scale up.

In the intervening time, we also need to make sure there is economic and social immunity and subsidy for the people of India who are most vulnerable during the lockdown. We need a long-term strategy to conquer this epidemic and it requires partnership between the public, the government, the health care sector, the private sector and the scientific experts. This applies to every country affected by the virus. We are all connected in this process and in this collective war. As Dr. Anthony Fauci, Director of the National Institute of Allgery and Infectious Diseases, USA, stated, “We do not make the timeline. The virus makes the timeline.”

3. Italy has been under lockdown since March 9. It has not seen an appreciable reduction in the number of deaths yet or number of cases yet. When can we expect to see a reduction in India?

If we look at the data from Hubei, there is approximately a period of 14–18 days when the number of new cases starts going down after introduction of lockdown, centralized treatment and isolation strategies (Figure 3a). This is because the median incubation period of the coronavirus is 5.2 days (95% CI: 4.4, 6.0), and 97.5% of those who develop symptoms will do so within 10.5 days (95% CI: 7.3, 15.3) of infection [Lauer et al. 2020].

Figure 3a. Case profile in Hubei, China.

South Korea, on the other hand, did not impose a complete lockdown but took several aggressive measures (including testing more and more people, fining the quarantine violators harshly), social distancing and partial closings, starting mid-February. Within 2 weeks, South Korea started seeing a decrease in growth rate of new cases (Figure 3b).

Figure 3b. Case profile in South Korea

Compared to Hubei and South Korea, Italy imposed lockdown/stringent measures later into the epidemic and thus reduction in number of cases in Italy will take a longer time (Figure 3c). As for death, the numbers typically go down at a longer lag after aggressive intervention since “it takes around three weeks on average for someone to die” [Why COVID-19 deaths will keep rising even as the coronavirus outbreak wanes].

Figure 3c. Case profile in Italy

To explain the concept of lag in simple terms, suppose someone contracted the virus on March 23 while the lockdown started on March 24. Let us assume that this person is not aware of their exposure and remains asymptomatic in the beginning and starts showing symptoms on March 30 and gets confirmed as a case on April 1 and finally dies on April 10. Thus, this case will still be counted as confirmed and fatal after the lockdown though the infection was contracted prior to lockdown. Every day we are essentially seeing a snapshot of what transmission occurred about two weeks ago. Because of this longer time to progress to severe infection and fatality [Yang et al. 2020], the number of deaths typically go down even later after the lockdown ends, possibly about three weeks. For comparison with Hubei, South Korea and Italy, we provide a similar figure for India through March 31 (Figure 3d). We can expect to start seeing significant changes in case counts by April 8–15 in India if the lockdown is effective.

Figure 3d. Case profile in India

While the lag issue is easier to explain, we have to remember that testing conventions are changing rapidly in every nation. An increase in testing availability and a broadening of testing criteria will increase the case count, which is not necessarily a bad thing. The ability to test is completely confounded with the case-counts. The number of “true” cases is a composite of the number of unreported/untested infected people and the ones that are confirmed. We will never know the “truth” unless we test more. The rise in the reported numbers in the coming days could be due to more testing becoming available and our predictions will also change. This will eventually lead to knowledge of almost all true cases and their contacts and help us enable post-lockdown prevention strategies. We will have to watch the next few days carefully for the case-counts in India as well as for the rest of the world as more tests are carried out.

4. What would success under a lockdown look like?

Hopefully, the ideal situation (subject to many assumptions) on April 14 when the lockdown in India is lifted, the number of new cases have been contained, testing and contact tracing task forces have been set up, healthcare facilities and mobile labs for testing and treatment are better prepared, other public spaces have been repurposed into temporary care and quarantine facilities, we have more ventilators, and we all are adherent to the hygiene recommendations. Each small step taken by each citizen can make a big difference for the country at this time. The success of the lockdown depends on how people adhere to it and take the recommendations seriously. Success also hinges on how the government utilizes the time to strengthen the healthcare infrastructure. At the same time, the disruption caused to public life will need to be minimized ensuring continuation of supply of essentials, food chain supply, access to critical services and economic protection for the poor and daily wage earners.

5. What can we expect after the lockdown is lifted — will the number of cases go back up? Will there be more deaths?

Our analysis shows we need to have some measures of suppression in place after the lockdown for the best outcome. We created statistical models [Wang et al. 2020] to study the effect of the lockdown under various hypothetical scenarios after it is lifted on April 14. We assume that the initial reproduction number R0 (the average number of people who will catch the virus from one infected person) of 2.0 came down to 1.5 via social distancing and travel ban, and further to 0.8 because of the lockdown. Now we consider four scenarios:

1. Perpetual social distancing and travel ban with no lockdown (“soc. dist. + travel ban”) R0 stays at 1.5 from March 12 in perpetuity without a lockdown.
2. Increased contact probabilities (due to reconnecting after lockdown and the festival of Baishakhi heralding summer in many parts of India) with R0 going up from 0.8 to 2.0 (“normal (pre-intervention)”)
3. Contact probabilities will go back up to where it was on March 25 (due to reconnecting after lockdown) with R0 rising from 0.8 to 1.5 (“moderate return”)
4. Slightly decreased contact probabilities (more caution and fear of the virus) with R0 increasing from 0.8 to 1.2 (“cautious return”)

We estimate that 15 (upper credible limit: 99) and 70 (upper credible limit: 1095) cases per 100,000 are avoided by May 15 and June 15, respectively, by instituting a 21-day lockdown with a cautious release compared to perpetual social distancing and travel ban (without a lockdown period) (Figure 4a). This boils down to preventing roughly 196 thousand (upper credible limit: 1.3 million) COVID-19 cases nationwide by May 15 and 943 thousand (upper credible limit: 14.7 million) COVID-19 cases by June 15 by an effective lockdown. Drawing an imaginary horizontal line across Figure 4a gives us a sense of how much time is bought under the different scenarios. Thus, we essentially buy ourselves time through this 21-day pause and flatten the curve (Figure 4b).

Figure 4a Predicted cumulative number of cases under different hypothetical social intermingling scenarios after the lockdown ends on April 14, from April 30 to August 31. Daily updates to the plot available at covind19.org.

As for deaths, we do not expect to see a reduction in the death rate from COVID-19 immediately after April 14 due to the longer lag but this will go down as case-counts go down. However, we have to be mindful that deaths due to other causes such as hunger in low-income families and lack of access to hospital care may increase during this period. There have been reports of deaths from mental health disorders, sense of isolation and loneliness [How Loneliness from Coronavirus Isolation Takes Its Own Toll]. We have to be mindful of the possibility of excess death or casualty due to the indirect effects of the lockdown and be proactive with instituting the right preventive measures. Devoting all our attention and resources towards controlling the number of COVID-19 infected cases is not an ethically viable strategy.

Figure 4b Predicted number of new cases under different hypothetical social intermingling scenarios after the lockdown ends on April 14, from April 30-August 31. Daily updates to the plot available at covind19.org.

6. Is a 21-day lockdown enough? Would a longer lockdown be more effective?

Several recent papers have argued for a longer period of lockdown [Ferguson et al. 2020, Singh & Adhikari 2020] from a modeling perspective. However, we refrain from commenting on this choice at this point. From a purely epidemiologic point of view, doing one long lockdown (2–3 months) in a strict manner may be effective enough to reduce the R0 further below 1.0 and arrest the pandemic. The other choice is to come out of lockdown in a modulated way with some measures of suppression in place and be prepared to go into another lockdown if more cases appear. However, science and knowledge are evolving every day and since such recommendations have major implication on people’s lives, and particularly for the lives of poor and the underprivileged, we decide to revisit this question after more data come in. While through the devastation of the pandemic it may be easy to reflect on how things could have been handled differently, we must work together and focus on what can be done now in the short term . Any credible long term projections with our models will need more data.

7. How vulnerable are India’s healthcare/frontline workers given that resources are overstretched?

This is a very serious concern not just in India but also in high-income countries. In the US, many fatalities are happening among frontline healthcare workers [KPEL965, USA Today, MedScape]. These heroic individuals are taking care of COVID-19 patients and getting sustained exposure. In China, around 3300 healthcare workers were infected and at least 22 died by the end of February [The Lancet March 21, 2020]. Within a month of COVID-19 hitting Italy, 20% of the responding healthcare professionals were infected thereby straining the healthcare system further [Remuzzi & Remuzzi 2020]. India has only 0.8 doctors (2.1 nurses and midwives) per 1000 people compared to 4.1 (5.9) in Italy, 2.4 (7.0) in South Korea, 1.8 (2.3) in China, 2.8 (8.3) in the UK, 2.6 (8.6) in the USA, and 1.1 (1.9) in Iran. It is extremely important to protect these frontline workers. China and other countries who are on the mend have donated protective gears to other countries. Large scale production of these equipment should start now. Strategies like full protection, stringent safety protocol and training of healthcare personnel are known to work since almost none of the 42,000 external workers who went to help Wuhan, China got infection [Professor Xihong Lin’s Lecture Slides]. Industrial sectors need to pitch in. India has brilliant scientists, inventors and academics. Everyone should think about increasing production capacity, as they currently are. Philanthropists are coming forward with their contributions (e.g., Ratan Tata, individual big-ticket donors from the film industry). Statistical modelers can play an extensive role to predict county and state level health care and hospitalization resource needs so that optimal distribution of resources can be made when the supply is limited [Emanuel et al. 2020, ESRI Impact Planning]. We hope more data driven strategies are adopted in this process of resource allocation. Moreover, from a collective social conscience, COVID-19 patients and caregivers should not be stigmatized or sacrificed [Stigma: the other enemy India’s overworked doctors face in the battle against COVID-19].

8. What can the Indian government do during and after the lockdown?

The Indian government has taken multiple proactive measures including a quick shut down of India’s borders, suspending visas, swiftly promoting quarantine and social distancing guidelines, a central lockdown and a $23 billion stimulus package. Many of the state governments have been remarkably agile and strategic in this fight. Some key points to keep in mind:

• Test, test, test: Testing and contact tracing, cluster testing for contacts of exposed cases cannot stop and needs to expand. Make it easy to get tested with prompt return of results. As of March 27, India has performed nearly approximately 20 tests per million compared to South Korea with roughly 7500 tests per million. Scalable rapid testing is becoming available, for example, a 15-minute COVID-19 test recently received FDA approval, produced by Abbot Laboratories. Self-swab has been shown to be equally effective so dropping off or mailing samples to labs will remove the need for additional personnel [Self-Service Diagnosis of COVID-19]. Creative design and distribution of test kits are needed to track the pulse of the epidemic.
• Testing random subjects and setting up community surveillance is extremely important. Iceland has set up a random testing framework early in the disease course and seen success. Israeli scientists have suggested pooling of biosamples for fast and cost-effective testing. Monitoring admission due to respiratory illness in hospitals, healthcare facilities, mining database for insurance claims in metros could provide alerts for new outbreaks. We cannot fight this war blindfolded, we need transparent data.
• Many daily wage earners are returning to the hinterland from metros and coming in close contact with others through walking miles together and seeking food and shelter collectively. This could further increase transmission. It is important to set up fever clinics and surveillance system in rural areas leveraging community health workers to monitor this exodus from the metros and arrange for free testing and treatment in villages.
• Enable efficient and humanitarian implementation of the lockdown and continue to provide a punctuated, modulated level of infrastructure support with smooth transition. Distribute food, hand soap, thermometers and make sure that the essential supply chain is not broken down.
• Clear communication and transparency are key here. Government advisories are often unclear to common people. In absence of clear and consistent messaging, the public turn to less reliable social media sources for information. Partnering with communication experts to individualize communication strategies for people with different education, literacy level and socioeconomic background is critical at this time.
• Send strong messages not to stigmatize COVID-19 patients, for public to adhere to quarantine rules, and provide guidance on hand hygiene through all possible channels.
• Use pragmatic real-time use of data for optimally deploying surveillance, community inspection and health care resources. This is key with limited resources.
• Arrange steady supply of protective gear (for health care workers), ventilators, treatment supplies. Expand and authorize manufacturing of these equipment.
• Set up temporary hospitals for critically ill COVID-19 patients and quarantine centers for isolating exposed contacts and asymptomatic cases.
• Reduce non-essential medical care but ensure critical care patients are not afraid to seek care. Create COVID-19 and non-COVID-19 care streams in healthcare facilities. Avoid excess deaths from other causes and ensure appropriate supply chain of medications during lockdown.
• As our simulations suggest (Figure 4a), post lockdown, there should be some measures of suppression and modulated return should be ensured. Without some form of suppression, the epidemic can grow again.
• Long term surveillance and management of COVID-19 crisis is needed with not just public health in mind but also to take care of the economic, social, and psychological trauma that it will leave on the people. Reviving the economy will be critical in the coming months.
• Leading with empathy and efficiency and being prepared to scale up interventions that are working and shutting down ones that are not is key for flexible policymaking now.

9. What can the people of India do during and after the lockdown?

• Stay at home during lockdown. Follow guidelines and co-operate with the government, officials and in particular the Ministry of Health. This includes individually adhering to social distancing rules, hand hygiene and cough/sneeze etiquette, and collectively co-operating with public health, and public safety officials enforcing these rules. More cases are prevented and more lives are saved if adherence to lockdown is quick and consistent.
• Be proactive in getting tested if exposed to the virus or showing symptoms, and not be afraid of the COVID-19 diagnosis. You can save others by isolating yourself and seek proper care for yourself with an early diagnosis.
• The general public can provide support and guidance to daily wage earners in their own households and social circle. This can be in the form of paid leave for short-distance domestic helpers and group D staff, or excused rent and sustenance support for long-distance daily wage earners (e.g. construction and industrial labors).
• Industrialists can support and accelerate manufacturing initiatives for COVID-19 testing and treatment equipment, and protective gears. Grassroots level organizations (e.g. student unions, non-profit organizations), universities can mobilize initiatives for raising public awareness and understanding of the pandemic. Much of this is already happening in India which is encouraging.
• Corporations and individual donors can contribute to the COVID-19 cause with philanthropic funds to support healthcare capacity building, as well as reduce suffering for the large impoverished segment of the population affected by the lockdown.
• People should refrain from hoarding food, medicines and essential supplies and be mindful of the needs of other fellow humans. Hoarding causes unnecessary panic and increases food price, thereby alienating the poor further.
• Stop stigmatizing and isolating healthcare professionals. We need to stand beside healthcare workforce in this heroic battle against the pandemic.
• Individuals should take care of personal health and the health of those around them. It is important to pay attention to both physical and mental health at a time of isolation and distress like this. Networking, connecting with each other through virtual platforms, organizing local initiatives to bring food, medicines for senior citizens are examples of ways people can stay socially connected and maintain a sense of community during the lockdown. The physical distancing does not have to imply complete social isolation.
• All of us should be prepared for potential longer social distancing measures and a gradual and cautious return to normal life after the lockdown. It is important to remember that the virus will simply not disappear after the lockdown and large mass gatherings may still be risky.
• Travel restrictions in some form will possibly prevail across states and across the world for a period of time. We should mentally prepare for a more static and constrained life in the near future.

10. Why are there so many models and so many projections and why do they vary so much? How much should we rely on these projections?

There are many epidemiological models to predict the course of an infectious disease [Mandal et al. 2020, CDDEP, Singh & Adhikari 2020]. Some use age-structure, contact patterns, spatial information to finesse their prediction. Some consider the possibility of a latent number of true cases, only a fraction of which are ascertained and observed [Wang et al. 2020]. The model we used here is an extension of a standard SIR model, called eSIR model [Wang et al. 2020], where we can create hypothetical intervention scenarios in a time dependent manner. The goal of any intervention is to reduce the chance that an infected person meets a susceptible person. We create models for declines/drops in contact probabilities when an intervention is rolled out. Thus, there is some intrinsic ad-hocery to our assumptions. Any statistical model is wrinkled with such assumptions. Similarly, the predictions themselves have large uncertainty (as reflected by the upper credible limits). As we interpret the numbers from any model, let us use caution in not over-interpreting them. A rigorous quantitative treatment often allows us to analyze a problem with clarity and objectivity, but we recommend focusing more on the qualitative takeaway messages from this exercise rather than concentrating on the exact numerical projections or quoting them with certainty.

We did explore some alternative assumptions and conducted thorough sensitivity analysis before settling on the model presented here. In one example, we assumed that there are actually ten times the number of reported cases to date to reflect potential underreporting of cases due to lack of testing. In another scenario, we assumed these cases occurred in metropolitan areas to reflect a potential intensification of case clustering. In yet a third scenario, we hypothesized that a single infected individual would infect 2.5 susceptible individuals, on average, instead of 2. These scenarios did not appreciably change our conclusions in broad qualitative terms, though the exact quantitative projections are quite sensitive to such choices. Across these scenarios, the projected total number of infected cases by the first phase of the pandemic varied between 1–10% of the population, again showing the significant variability in these numbers. The estimates we present here may appear conservative and are at best underestimates, and, in all cases, our confidence in these projections decreases markedly the farther into the future we try to forecast. It is extremely important to update these models as new data arise.

In our strong commitment to reproducibility and dissemination of our research, we have made the code for our predictions available at GitHub and created an interactive and dynamic R Shiny app to visualize the observed data and create predictions under hypothetical scenarios with quantification of uncertainties. These forecasts will get updated daily as data come in. We hope these products will remain our contribution and service as data scientists during this tragic global catastrophe, and the model and methods will be used to analyze data from other countries.

Finally, our message to the public is to proceed with prudence and caution, and not panic or drown in despair. We should draw hope from the success of South Korea and China and the initial promising containment in India. We need to support the community around us and help the government of India to manage the crisis with the best strategies, resources and science. The lockdown has given us time to prepare and act, let us make the best use of it. We are still in a state of national and global emergency and it will take a considerable time for humanity to recover from this global pandemic and return to normalcy. In the mean time we root for public health, for innovation and science, for home testing kits [there is none yet], for FDA approved drugs [Solidarity Trial], and for a vaccine [3 phase I clinical trials ongoing]. In these frightening times we find inspiration in the power of the common people and the magic of human kindness.

The COV-IND-19 Study Group is comprised of an interdisciplinary group of scholars and data scientists. The following members contributed to this piece:

• Maxwell Salvatore — Departments of Biostatistics and Epidemiology and Center for Precision Health Data Science, University of Michigan
• Lili Wang — Department of Biostatistics, University of Michigan
• Rupam Bhattacharyya — Department of Biostatistics, University of Michigan
• Soumik Purkayastha — Department of Biostatistics, University of Michigan
• Shariq Mohammed — Departments of Biostatistics and Computational Medicine and Bioinformatics, University of Michigan
• Aritra Halder — Department of Statistics, University of Connecticut
• Daniel Barker — Department of Biostatistics, University of Michigan
• Michael Kleinsasser — Departments of Biostatistics and Health Behavior and Health Education, University of Michigan
• Alexander Rix — Department of Biostatistics and Center for Precision Health Data Science, University of Michigan
• Mousumi Banerjee — Department of Biostatistics and Institute for Health Policy and Innovation, University of Michigan
• Veera Baladandayuthapani — Department of Biostatistics, University of Michigan
• Debashree Ray — Departments of Epidemiology and Biostatistics, Johns Hopkins University
• Bhramar Mukherjee — Departments of Biostatistics and Epidemiology and Center for Precision Health Data Science, University of Michigan

Contact Bhramar Mukherjee (bhramar@umich.edu) with questions and inquiries.