key: cord-0891137-d61l4o7u
authors: Husain, Z.; Das, A. K.; Ghosh, S.
title: Did the National lockdown lock COVID-19 down in India, and reduce pressure on health infrastructure?
date: 2020-05-29
journal: nan
DOI: 10.1101/2020.05.27.20115329
sha: 129bfd98d41ed28f64fb99a88a680d25efe157e2
doc_id: 891137
cord_uid: d61l4o7u

Background & objectives: The spread of COVID19 in India has posed a major challenge for policy makers. Policy response in form of imposition of a prolonged national lockdown has imposed substantial costs on the entire population. But the extent to which it has contained the spread of the epidemic needs to be assessed. Methods: We use an Interrupted Time Series model to assess the success of lockdowns in containing COVID-19. In the second step, we use four variants of the SIR models to develop a counterfactual- what would have happened without the lockdown. These results are compared with actual data. The analysis is undertaken for India, and Maharashtra, Gujarat, Delhi, and Tamil Nadu. Results: Lockdown has reduced the number of COVID-19 cases by 23.65 to 337.73 lakh in Class I cities and towns, where COVID has mainly spread. It has averted about 0.01 to 0.10 lakh deaths. At the regional level, however, lockdown has averted a health crisis as existing ICU and ventilator facilities for critically ill patients would have been inadequate. Interpretation & conclusions: Overall, the results for three of the four models reveal that lockdown has a modest impact on spread of COVID-19; the health infrastructure at the national level is not over strained, even at the peak. At the regional level, on the other hand, lockdowns may have been justified. However, given that identification of new cases is limited by levels of daily testing that are low even by Asian standards, analysis based upon official data may have limitations and result in flawed decisions.

This study estimates the impact of the lockdown on the Indian economy, specifically those residing in Class I cities, to contain the spread of 2019-nCov-popularly referred to as COVID-19. Specifically, we have estimated the effect of the lockdown in terms of reducing number of COVID-19 cases, deaths, and pressure on health infrastructure in India.

COVID-19 is a group of viruses affecting human beings through zoonotic transmission. COVID-19 was manifested in the Province of Hubei, China in December 2019. By 13 th March, the outbreak had spread to 114 countries with more than 118,000 cases and 4,291 deaths, leading WHO to declare it a pandemic 1 . The major reason for concern with COVID-19 is its the global scale of transmission, significant number of deaths, infection and mortality of care providers and healthcare workers (HCWs), and higher risk of death in vulnerable or susceptible groups 2 . In such circumstances, particularly in the absence of a licensed vaccine or effective therapeutics for COVID-19, slowing or breaking the transmission dynamics through quarantining and social distancing has been adopted as a strategy 3 . Quite a few countries, including India (vide MHA Order No. 40-3/2020-D dated 24/3/2020), have also adopted a lockdown of the economy.

Lockdown imposes a substantial economic and humanitarian burden on societies, particularly in developing countries, that persists even after the epidemic dies out. Such a cost is justified on the grounds that it decreases number of cases, and delays peak, and consequently reduces pressure on health infrastructure. In a press statement made on 23 rd May, 2020, the Niti Ayyog claimed that without lockdown there would have been 36-70 lakh cases, and 37-71 thousand deaths 4 

There have been several studies attempting to estimate the impact of travel bans, social distancing, containment and lockdown on flattening the transmission curve using exponential growth models and the SIR model 5 , and its variants. Such attempts compare actual cases with projected cases estimated using SIR models, and its variants. The limitation of such exercises is that parametric assumptions are either data driven (so that noises in the data, particularly due to . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 29, 2020. . https://doi.org/10.1101/2020.05.27.20115329 doi: medRxiv preprint low testing rates, distort parametric assumptions), or are based upon parameters from studies in other countries (which may not reflect the socio-economic realities of India).

This study estimates a SIR model based on actual number of cases reported for India in the John Hopkins and MoHFW websites as counterfactual. Alternative parametric values are used to estimate the trajectory of COVID-19 without lockdown. While two projections uses parametric values estimated from data, we also present another set using data on daily number of contacts before lockdown obtained from an online survey, along with assumptions based on different studies. We also use an Interrupted Time Series model, commonly used to study the impact on interventions on population level data, to study the impact of the first three lockdowns upon actual spread of cases 6 

There have been several studies attempting to estimate the effect of policies like social distancing, quarantining, and lockdown on restricting the spread of COVID-19. Such studies have produced different estimates of the efficacy of lockdowns in containing transmission of COVID-19 cases in India. A study using estimates based on exponential and SIR models reports that symptomatic quarantine can achieve meaningful reductions in peak prevalence, and spread out the outbreak over a longer period even with high R 0 6 . Mandal and Mandal observes that lockdown had reduced number of cases by 78 percent 8 . A study by IIPS reports that strict implementation of the third lockdown will ensure that confirmed cases at the peak will be only three million; if, however, the R 0 prevailing during the second lockdown continues, this figure . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 29, 2020. . https://doi.org/10.1101/2020.05.27.20115329 doi: medRxiv preprint will be 20 million 9 . A study for Kochi, Bengaluru, Mumbai, and Pune using the eSIR model 10 reports that:

"without enforcing the interventions, the predicted counts are going to exceed the estimated capacity of hospital beds in India (estimated at 0.7 per 1000). … We also note … for example the predicted number of cases by May 15 will be at 161 per 100,000 without any intervention (2.2 million total cases nationwide) and will reduce to1 per 100,000 (13,800 total cases nationwide) with the most severe form of intervention" 11 (p.

Comparing projected increase in confirmed cases in absence of lockdown, a study based on an exponential growth model reported that the projected number of confirmed cases would have been 86,373 on 23 rd April, compared to actual count of 23,039 12 . The estimated number of COVID-19 cases in government hospitals would be about three per hospital, but this figure has been significantly reduced to 0.82 through Government policies 12 . Another study reported that lockdown decreased R 0 from 1.862 to 1.2; duration of the epidemic increases from 2 to 6 months, but would die out by November. Only 0.23 percent of the population would be affected 9 .

A study using Bayesian analysis, however found that a three-week lockdown would be insufficient to prevent resurgence; a more sustained lockdown with periodic relaxation was suggested 13 .

In any impact evaluation exercise, identifying the counterfactual-what would have happened in the absence of the intervention-is the crucial problem. While the SIR model and its variants have been utilized to estimate the counterfactual scenario, a major problem with this approach is that there are no reliable estimates for the parametric values embedded in the model 14 . For instance, current models in India either do not account for the contact distribution, or end up assuming that this distribution can be extrapolated from Western countries, which are markedly different in their social and living arrangements 15 . Exponential models, for instance, provide a poor fit to the growth in confirmed cases, both in the initial period and subsequently.

The impact of the lockdown on reducing COVID-19 cases in India is estimated in this study by comparing between:

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (b) Estimated progress of COVID in India in absence of lockdown: This is "missing". We have, therefore, used the SIR model to estimate the probable progress of COVID in the absence of lock downpossible number of COVID-19 cases without the lock down.

Interrupted time series (ITS) model is a tool suitable for evaluating the effectiveness of a largescale public health interventions at a population level, in the absence or presence of suitable counterfactuals 6 . It has been widely used in public health interventions such as vaccination, evaluation of health impacts of unplanned global financial crisis, effect of smoking ban in public places on hospital admissions for acute coronary events etc. 16, 17 . The ITS model can be presented as:

where Y t : the aggregated outcome at time t, in our case the number of new cases of COVID19 T: the time since the start of the study X t : a dummy variable indicating pre-intervention period (coded as 0) or post intervention period (coded as 1)

 0 represents the intercept or the initial value of the outcome at T=0,  1 is the slope or trajectory represents the change in outcome associated with an increase in the time unit (representing the underlying pre-intervention change),  2 represents the level change following the intervention,  3 indicates the slope change following the intervention (using the interaction between time and intervention), and  t is the random error term, following a first order autoregressive [AR (1)] process.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 29, 2020. . https://doi.org/10.1101/2020.05.27.20115329 doi: medRxiv preprint It is estimated in STATA Version 15, using the ITSA module 18 , followed by testing for appropriate lag 19 , after adding two sets of extra terms for lockdowns 2 and 3.

In this study we use the SIR model to project the spread of COVID-19 in the absence of lockdowns. The SIR model explains the dynamics of transmission of infectious diseases in a fixed population (of size N). The population is divided into three groups: S (Susceptible), I (Infected), and R (Recovered). Other components of susceptible population are excluded, such as (a) Exposed: The model aims to project the epidemic without any intervention; so it is assumed that entire urban population is exposed and will get infected over time; (b) Quarantined: In the absence of any contact tracing no quarantine measures will be adopted (c) Immune: Herd immunity will increase as population recovers over time. Moreover, there is no vaccine available to assume that certain population can be immune through public health immunization program.

Also, the chances of re-infection are not considered to be the major drivers of COVID-19 at the time of writing this article 20, 21 .

The model is expressed in terms of three simultaneous non-linear differential equations tracing the time path of the number of susceptible, infected and recovered persons from their starting values (S 0 , I 0 and R 0 ):

The two parameters of the model,  and , represent transmission rate of disease and recovery rate, respectively. The transmission rate is the product of the number of daily contacts of an infected person (c) and the probability that the disease will be transmitted from an infected person to a contact (). The recovery rate is the reciprocal of average number of days for recovery.

These parameters define the basic reproduction rate of the disease (R 0 ), given by /. If R 0 is greater than unity, the infection will spread in a self-sustaining manner to become an epidemic;

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 29, 2020. . https://doi.org/10.1101/2020.05.27.20115329 doi: medRxiv preprint on the other hand, for values of R 0 lower than unity, the infection will die out in the absence of fresh cases entering the society from outside.

There are three parameters in the SIR model-daily contacts (c), probability of infection from contact with an infected person (), and recovery period (). Their combination gives R 0 (=c/).

Although studies on daily contact rates exist for developed countries 22, 23 , there is no similar study on daily contacts before lockdown for Indian societies. An online survey conducted by us during the lockdown found mean contact rate before lockdown to be 32. 

The SIR model is run from 6 th March to 17 th May (the last day of the third lockdown) using alternative values for  and . Results generate predicted number of cases in the absence of lockdown. We have four sets of predictions-OWNL, OWNH, EG and ML, corresponding to each model. These give us the counterfactual, and are used to predict the number of COVID cases without lockdown for alternative scenarios. The difference between actual cases and deaths, and estimates using the SIR model (OWNL, OWNH, EG and ML) is an estimate of COVID cases and deaths averted by lockdown; it is used to calculate reduced mortality and pressure on health care system. (Table A1 ). The estimates for required hospital beds, ICU beds, oxygen, and ventilator support are estimated for the peak. State-wise information on health infrastructure is also used in this study 36 .

Results of the ITS model is given in Fig. 1 (see also Appendix Table A2 ). The correlation between predicted and actual values is 0.9869, indicating a very good fit. It may be seen that . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 29, 2020. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 29, 2020. . https://doi.org/10.1101/2020.05. 27.20115329 doi: medRxiv preprint The reduction in cases, deaths and pressure on health infrastructure due to lockdowns is summarized in Table 1 . While the OWNL and OWNH models indicate a substantial impact of lockdowns, the results from the EG and ML models are quite modest. Another point to note is that the hospital infrastructure does not seem to be strained-only two predictions (OWNL and ML) indicate that there may be a deficit in requirement of ICU beds.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 29, 2020. . https://doi.org/10.1101/2020.05.27.20115329 doi: medRxiv preprint 

The national picture hides considerable regional variations. Given the high incidence of cases in some states, it is necessary to also explore the regional picture.

The result of the ITS model (Fig. 2) for Maharashtra indicates failure of the three lockdowns to contain COVID-19; in Gujarat, there is an increase in level after the first lockdown, followed by an increase in slope after the second lockdown. Only in Delhi and Tamil Nadu, lockdown seems to have achieved limited success-a decrease in level after the Lockdowns 1 and 2, along with a decrease in slope after the third lockdown (Delhi), and a decrease in level after Lockdown 2 (Tamil Nadu).

Source: Estimated by authors . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 29, 2020. . https://doi.org/10.1101/2020.05.27.20115329 doi: medRxiv preprint infrastructure for these four states assuming that no containment policy had been enforced ( Table   2 ). The estimates have been made based on the SIR model using the parametric assumptions for OWNH and ML models. Results indicate that COVID-19 cases and mortality would have been increased substantially in these states in the absence of policies. A comparison with actual figures, however, indicates that lockdown did not succeed in reducing incidence and mortality substantially (Appendix Table A3 ). Results also indicate, while hospital beds are sufficient, a major shortage of ICU and ventilator would have occurred in these states.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 29, 2020. . https://doi.org/10.1101/2020.05.27.20115329 doi: medRxiv preprint 1. Hospitalisation cases estimated at peak as maximum of total of 14 days cases.

2. Death, ICU cases and ventilator support cases estimated as in Table 1 [35] .

3. Utilisation figures is obtained by using information on health infrastructure [36] 

The exercise, however, has a major limitation. The SIR model is based upon initial values of the epidemic, and assumes an exponential growth. In India, however, the number of new cases seems to follow a linear fit, with periodic changes in intercept and slope. While this may reflect the impact of containment policies, another possibility is that the trend in cases may reflect, not the spread of disease, but the increase in daily testing. Regression results (Table 3) 

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 29, 2020. . https://doi.org/10.1101/2020.05. 27.20115329 doi: medRxiv preprint It is widely accepted that lockdown is a drastic public health measure 37,38 with far-reaching consequences and adverse health outcomes 39,40 . The adoption of such an extreme step requires careful analysis. Our study indicates that the lockdown has reduced the spread of COVID-19 cases in India. The gains at the national level, however, seem to be modest (except for OWNH model projections); it is also unlikely that the health infrastructure would have been overstrained. Given that COVID-19 in India has varied widely across states, regional analysis is also

important. An analysis of the situation in four states with about 68 percent of cases reveals that lockdown may have averted a major health crisis as the health infrastructure (ICU and ventilator support) required for treatment of critically ill patients available in these states would have been inadequate.

In conclusion, we would also like to point out that lockdown is primarily a measure to buy time and create the health infrastructure required to fight COVID-19. This has been done to a limited extent. Consequently, when the lockdown is ultimately lifted, a significant spike can occur at the national level, with the emergence of new hotspots where migrant workers return.

Secondly, estimates-though not conclusions-vary widely depending upon parameters used.

The sensitivity to parametric assumptions is a major limitation of existing studies. It calls for more sophisticated calibration of models using parameters relevant for the Indian context.

Finally, our analysis also indicates that the available data may reflect the increase in testing, rather than only the spread of the disease. It implies that the actual number of cases may remain hidden due to limited testing. So, unless we increase daily testing to at least levels comparable . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 29, 2020. . https://doi.org/10.1101/2020.05.27.20115329 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 29, 2020. . https://doi.org/10.1101/2020.05.27.20115329 doi: medRxiv preprint Source: Estimated by authors . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 29, 2020. . https://doi.org/10.1101/2020.05.27.20115329 doi: medRxiv preprint

World Health Organization. WHO announces COVID-19 outbreak a pandemic

The 2019 novel coronavirus disease (COVID-19) pandemic: A review of the current evidence

Advisory on social distancing measure in view of spread of COVID-19 disease

Lockdown has averted 14-29 lakh infections

Contributions to the mathematical theory of epidemics--I. 1927

Segmented regression analysis of interrupted time series studies in medication use research

Prudent public health intervention strategies to control the coronavirus disease 2019 transmission in India: A mathematical model-based approach

COVID-19 pandemic scenario in India compared to China and rest of the world: a data driven and model analysis. medRxiv 2020

Projecting the future trajectory of COVID-19 infections in India using the susceptible-infected-recovered (SIR) model -An analytical paper for policymakers

Predictions, role of interventions and effects of a historic national lockdown in India's response to the COVID-19 pandemic: data science call to arms

Assessing the impact of complete lockdown on COVID-19 infections in India and its burden on public health facilities -A situational analysis paper for policy makers

Age-structured impact of social distancing on the COVID-19 epidemic in India

Tracking the impact of interventions against COVID-19 in absence of extensive testing: Active surveillance for SARI is urgently needed

India's response to coronavirus can't be based on existing epidemiological models. The Print

Interrupted time series regression for the evaluation of public health interventions: A tutorial

How do you know which health care effectiveness research you can trust? A guide to study design for the perplexed

COVID-19 and postinfection immunity: Limited evidence, many remaining questions

Risk of reactivation or reinfection of novel coronavirus (COVID-19)

The effect of public health measures on the 1918 influenza pandemic in U.S. cities

Social contacts and mixing patterns relevant to the spread of infectious diseases

COVID-19 outbreak on the Diamond Princess cruise ship: estimating the epidemic potential and effectiveness of public health countermeasures

Multiple estimates of transmissibility for the 2009 influenza pandemic based on influenza-like-illness data from small US military populations

Early transmission dynamics in Wuhan, China, of novel coronavirus-infected pneumonia

Serial interval of COVID-19 among publicly reported confirmed cases

Report of the WHO-China Joint Mission on Coronavirus Disease 2019 (COVID-19): Geneva: WHO; 6-14

How generation intervals shape the relationship between growth rates and reproductive numbers

Estimation of the Reproductive Number and the Serial Interval in Early Phase of the 2009 Influenza A/H1N1 Pandemic in the USA

The R0 package: a toolbox to estimate reproduction numbers for epidemic outbreaks

A new framework and software to estimate time-varying reproduction numbers during epidemics

Coronavirus COVID-19 Global Cases

Impact of Non-pharmaceutical Interventions (NPIs) to Reduce COVID-19 Mortality and Healthcare Demand

India confirms over 1 lakh Covid-19 cases: How many ICU beds and ventilators does the country have? Huffpost

COVID-19 in India: State-wise estimates of current hospital beds, intensive care unit (ICU) beds and ventilators