Public health interventions slowed but did not halt the spread of COVID‐19 in India

Abstract The government of India implemented social distancing interventions to contain the COVID‐19 epidemic. However, effects of these interventions on epidemic dynamics are yet to be understood. Rates of laboratory‐confirmed COVID‐19 infections per day and effective reproduction number (Rt ) were estimated for 7 periods (Pre‐lockdown, Lockdown Phases 1 to 4 and Unlock 1–2) according to nationally implemented interventions with phased relaxation. Adoption of these interventions was estimated using Google mobility data. Estimates at the national level and for 12 Indian states most affected by COVID‐19 are presented. Daily case rates ranged from 0.03 to 285.60/10 million people across 7 discrete periods in India. From 18 May to 31 July 2020, the NCT of Delhi had the highest case rate (999/10 million people/day), whereas Madhya Pradesh had the lowest (49/10 million/day). Average Rt was 1.99 (95% CI 1.93–2.06) and 1.39 (95% CI 1.38–1.40) for the entirety of India during the period from 22 March 2020 to 17 May 2020 and from 18 May 2020 to 31 July 2020, respectively. Median mobility in India decreased in all contact domains during the period from 22 March 2020 to 17 May 2020, with the lowest being 21% in retail/recreation, except home which increased to 129% compared to the 100% baseline value. Median mobility in the ‘Grocery and Pharmacy’ returned to levels observed before 22 March 2020 in Unlock 1 and 2, and the enhanced mobility in the Pharmacy sector needs to be investigated. The Indian government imposed strict contact mitigation, followed by a phased relaxation, which slowed the spread of COVID‐19 epidemic progression in India. The identified daily COVID‐19 case rates and Rt will aid national and state governments in formulating ongoing COVID‐19 containment plans. Furthermore, these findings may inform COVID‐19 public health policy in developing countries with similar settings to India.

deaths have been reported worldwide (WHO, 2020b). As data collection efforts and testing protocols continue to evolve, these values likely represent a considerable underestimation of total cases and deaths (Richterich, 2020).
The successful impact of public health interventions on COVID-19 in Wuhan, China, which included social distancing, isolating infected individuals and quarantine supported implementation of similar measures in many other countries (Anderson et al., 2020).
For instance, mass-gathering events have been reported to pose a considerable public health risk and therefore have been avoided in most COVID-19-infected countries. Modelling studies suggest that, whereas highly effective contact tracing coupled with case isolation has potential to contain COVID-19 outbreaks, proper containment likely requires additional measures, for example contact mitigation (Hellewell et al., 2020). However, with recent declines in COVID-19 case volume, many countries are turning to contact mitigation relaxation plans.
The federal government in India responded swiftly to COVID-19 by implementing staged lockdown periods across the country; strict, intense contact mitigation was implemented initially, followed by phased relaxation as deemed appropriate. As India is among the world's most populated countries, COVID-19 has considerable potential for widespread morbidity and mortality and containment has global implications. Evaluating impacts of COVID-19 public health interventions is important to inform future effective public health and social interventions. The effective reproduction number (R t ), which captures time-dependent variations in the transmission potential of an infectious disease in a given population, is an important parameter for evaluating effectiveness of public health interventions (Pan et al., 2020). Although epidemiological characteristics of COVID-19, including rates of confirmed cases and R t , were investigated in Wuhan, China, across various periods of social distancing interventions (Pan et al., 2020), the impact of government-imposed mitigation strategies on transmission dynamics in India remains unknown. In this study, we estimated the laboratory-confirmed daily case rate and R t of the COVID-19 epidemic in key lockdown periods in India. Estimates presented here will provide key information for ongoing COVID-19 prevention and control in India.

| Time series data
Indian COVID-19 time series incidence and fatality data were extracted on 15 August 2020 using the WHO coronavirus disease 2019 (COVID-19) situation reports 10-193 (WHO, 2020a). We selected data from states/union territories where > 200 COVID-19 cases (positive/recovered/deceased) had been recorded before 22 May 2020, which consisted of data from 17 states/union territories. Time series data on daily counts were not available for 5 states (Gujarat, Rajasthan, Uttar Pradesh, Jammu and Kashmir, and Bihar); therefore, these were excluded from the study. Time series data for the remaining 12 states were extracted from official state Health and Family Welfare department websites (Appendix S1).

| Census data
The official human population 2011 Indian census data were used (COI, 2011a). In 2014, the state of Andhra Pradesh was bifurcated into two states, Telangana and residuary Andhra Pradesh. As separate census data for these states were not available, combined results for Andhra Pradesh (Telangana and residuary Andhra Pradesh) are presented.

| Case definitions
Cases were defined based on laboratory confirmation using throat/ nasal swab real-time reverse transcriptase-polymerase chain reaction (RT-PCR) assay, the diagnostic test recommended by the Indian Council for Medical Research (ICMR) for the diagnosis of COVID-19 (ICMR, 2020b). Only laboratory-confirmed cases were included in the analyses. The ICMR strategy for COVID-19 testing included all symptomatic individuals who had undertaken international travel in the last 14 days, all symptomatic contacts of laboratory conformed cases, all symptomatic health care workers, all patients with severe acute respiratory illness (fever and cough and/or shortness of breath), and asymptomatic direct and high-risk contacts of a confirmed cases (ICMR, 2020b). For hotspots/clusters, large migration gatherings and evacuation centres, all persons having fever, cough, sore throat or runny nose were recommended to be tested (ICMR, 2020b).

| Classification of 7 time periods
The study period comprised an initial unmitigated social contact period when temperature screening was imposed at international checkpoints. This was followed by strict contact mitigation and then phased relaxation. Dynamics of the COVID-19 epidemic were evaluated in discrete periods, according to changes in government policies ( Figure 1). Based on these government-imposed interventions, we classified 7 periods for analysis, including the following:

| Mobility index
A domain-specific mobility index was constructed using India's mobility report (Google Inc., Mountain View, CA, USA) (Google, 2020).
These data are publicly available from Google and represent the per cent change from baseline mobility within various domains (retail and recreation mobility, grocery and pharmacy mobility, parks mobility, transit stations mobility, workplace mobility and residential mobility) according to cell phone-user geolocation data. As data were available as per cent mobility change from the baseline value, we considered 100 as the baseline value; therefore, 100 was added to each value to transform the raw Google data to domain-specific mobility per day. The domain-specific mobility index was constructed for the country and 12 Indian states.

| Outcomes
Daily rate of laboratory-confirmed cases per 10 million people per day was estimated across periods for the 12 Indian states and for the country. We used number of cases in each period, divided by the number of days in each period (52,24,19,14,14,30, and 31d, respectively) and the total population of the selected region as per the 2011 census (COI, 2011a). The R t was calculated to determine the transmission of SARS-CoV-2 in each of the 6 periods. The R t was calculated based on the method developed by Cori et al. (Cori et al., 2013) that is able to detect changes in the R t following public health interventions.

| Statistical analyses
All statistical analyses were conducted using R version 3.6.3 (R Development Core Team, http://www.r-proje ct.org). Epidemic curves were plotted, based on laboratory diagnosis date and intervention periods described. Choropleth maps describing geographical distributions of COVID-19 case rates for the country and 12 Indian states through the 7 intervention periods were generated.
The R t for each of 12 Indian states, as well as for the entire country, were calculated as per the method developed by Cori et al. (Cori et al., 2013). This method estimates R t from the incidence time series and incorporates uncertainty in the distribution of the serial interval. We used the daily number of reported COVID-19 cases from the above-mentioned official data sources. The serial interval required to estimate R t (mean = 7.5 d, SD = 3.4 d) was derived from previous studies in Wuhan, China Pan et al., 2020). The serial interval was considered constant across all periods. A 5-day moving average was used to estimate R t and its 95% credible interval on each day.
The first locally transmitted COVID-19 case is thought to have occurred in India on 5 March 2020 and the Indian government closed international borders and air travel on 22 March 2020; therefore, March 22 was taken as our first day of R t estimation. This allows for a complete serial interval from the first locally transmitted case, and enables presumption of a closed population and ensures an appropriate total case count (Cori et al., 2013). Therefore, the R t was estimated from 1) March 22 to May 24; and 2) May 18 to August 7; and results of the initial burn-in period (when both imported and locally transmitted cases were reported; Figure 1a) are not presented. For the 12 states, R t was estimated for the whole period, but was presented from the day when 50 cumulative cases were reported (or from March 22 onwards, whichever is later) given the limited diagnostic cases and capacity in the preliminary period. Although country and state populations were assumed to be closed from March 22 onwards, the possibility of imported cases being reported cannot be ruled out at the state level.
Descriptive analyses to report changes in mobility index were conducted according to key intervention periods.
As of 31 July 2020, the state of Maharashtra had the highest number of cumulative cases (n = 411,798) followed by Tamil Nadu (n = 239,978) and NCT of Delhi (n = 134,403) (Figure 2b).
The R t was > 1 for all states and the entirety of India during the period from 18 May 2020 to 31 July 2020 except NCT of Delhi where R t was 0.79 (95% CI 0.77-0.81) in Unlock 2 (Figure 5b; Appendix 7-53).

| Mobility index
Median mobility was 21% for retail and recreation, 53% at the grocery and pharmacy, 42% at parks, 34% at transit stations, 38% at workplaces and 129% at residential places from 22 March 2020 to 17 May 2020 (Table 2). Mobility at grocery and pharmacy (median 37%), transit stations (median 29%) and workplaces (median 33%) was lowest from 22 March 2020 to 14 April 2020, during Lockdown Phase 1 (Figure 4a). Interestingly, mobility for retail and recreation (median 15%) and parks (median 38%) was lowest from 15 April 2020 to 3 May 2020, during Lockdown Phase 2 (representing relaxation). Note that these values were derived in comparison to a 100% baseline mobility in the country/state value when no such interventions were imposed.
Median mobility subsequently increased in all the domains except residential during the period from 18 May 2020 to 31 July 2020 as compared to the period from 22 March 2020 to 17 May 2020 (Table 3, Figures 4b and 6b). Median mobility in the 'Grocery and Pharmacy' returned to levels observed before 22 March 2020 in Unlock 1 and 2. Median mobility was 41% for retail and recreation, 95% at the grocery and pharmacy, 49% at parks, 60% at transit stations, 67% at workplaces and 115% at residential places from 18 May 2020 to 31 July 2020.

| D ISCUSS I ON
TA B L E 2 Domain-specific mobility in the country and 12 Indian states during key intervention periods. Note that a 100% baseline mobility was considered when no interventions were imposed bans on mass-gathering events, quarantining of positive cases and their contacts, and improved medical care. In Wuhan, China, the R t decreased from 3.0 in 26 January 2020 to 0.3 in 1 March 2020, within 40 days after multifaceted public health interventions were implemented (Pan et al., 2020). For the United States, the R t reduced from 4.02 to 1.51 between 17 March and 1 April 2020 (Gunzler & Sehgal, 2020). As of 9 March 2020, the R t was reported to be 3.10 for Italy, 6.56 for France, 4.43 for Germany and 3.95 for Spain (Yuan et al., 2020). The reduction pattern in R t in India was similar to other areas of the world after implementing intense mitigation strategies.
Although the R t remained > 1 in most states of India during all 7 periods, the R t in India was lower than in many European countries as   Previous studies reported a varied basic reproduction number (R 0 ) 1.03 -4.18 in India (Mandal & Mandal, 2020;Rai et al., 2020;Senapati et al., 2020) due to differences in time periods, data sources or methods employed. The time periods considered in these studies varied between: 4 March 2020 to 3 April 2020, with an estimated R 0 of 2.56 (Rai et al., 2020), using initial epidemic growth phase data and reporting R 0 to be 4.18 (Senapati et al., 2020); and before 9 April 2020 and reporting an R 0 of 1.03 (Mandal & Mandal, 2020). We used official data from 22 March 2020 to 17 May 2020 and 18 May 2020 to 31 July 2020 to derive R t estimates, so these estimates cannot be compared directly.
The mobility index highlighted the adoption level of the public health contact mitigation interventions imposed by the Indian government; clearly, they were highly effective in substantially reducing social contact in the country. Mobility in the country was < 50% in all the domains except grocery and pharmacy, where mobility was 53% unexamined. Therefore, current estimates should be carefully interpreted. Notwithstanding, in our opinion the current study provides much-needed information for further control and prevention of COVID-19 in India. Availability of additional data, hopefully in the near future, is expected to further improve these efforts.
The Indian government imposed strict contact mitigation followed by phased relaxation, which slowed the spread of COVID-19 epidemic progression in India, as evidenced by the decreasing R t demonstrated in this study. These findings will also inform policy development for the control of COVID-19 epidemic in other regions and countries.

ACK N OWLED G EM ENTS
The authors acknowledge India's National and State Health departments for collecting the daily COVID-19 epidemic data and releasing it in the public domain.

CO N FLI C T O F I NTE R E S T
The authors declare no conflicts of interest.

E TH I C A L S TATEM ENT
Informed consent for collection of epidemiological data was not required, as these data were already coded and available in the public domain. No identifiable personal information was used in this study.

DATA AVA I L A B I L I T Y S TAT E M E N T
The analysed data are available along with the manuscript. Sources of the raw data used in the analysis have been cited.