COVID-19 surveillance program in Pune, India
Pune city-located in western India around 150 km east of Mumbai (Fig. 1a)—launched a COVID-19 surveillance program during the early stages of the pandemic (January 2020). Pune Municipal Corporation (PMC) collaborated with multiple public and private health facilities to establish SARS-CoV-2 diagnostics, quarantine facilities for asymptomatic persons, and hospital/critical care beds for moderate to severely ill patients diagnosed with COVID-19. In addition, community-based workers were mobilized to conduct contact tracing activities. A publicly accessible dashboard was established to report the cumulative COVID-19 caseload in the PMC’s 41 prabhags (also known as electoral wards). The number tested and individual-level data, such as age, sex, residential address, COVID-19 test results, and COVID-19 outcomes, were centrally compiled on a regular (almost daily) basis.22.
Nationwide and Pune regional COVID-19 pandemic management
India’s initial response to the pandemic comprised travel advisories on international travel and suspension of visas from mid-January through mid-March. During this period, COVID-19 testing was administered to travelers who were returning from China and other foreign countries and had fever, cough or other viral respiratory symptomstwenty. Those testing positive were hospitalized for quarantine, and their close contacts were traced and underwent COVID-19 testing. The first nationwide lockdown was implemented from March 25th to April 14th, 2020 (Lockdown 1). Nearly all services and factories were suspended with reports of arrests for lockdown violations. During this time, Pune city expanded COVID-19 testing capacity, making testing available to persons with viral symptoms or within 14 days of COVID-19 exposure. The nationwide lockdown was extended from April 15th to May 3rd (Lockdown 2). Agricultural activities and essential services were allowed to function from April 20th, and Pune city areas were classified into red, orange, and green zones based on infection clusters. Red zones were defined by the central government based on case counts, doubling rate, and testing/surveillance findings. Initially, the central government defined the red zone as a particular area/district with more than 15 active cases. The area with <15 cases with no recent surges were defined as the orange zone. The area with zero COVID cases were green zones. However later as the cases emerged in the country, the central government allowed the states to categorize the zones. Notably, interstate transport was allowed for stranded individuals, and during the month of May alone, approximately one million migrants traveled via roads or trains to their home states, mostly from Maharashtra state. The lockdown was extended again from May 4th to May 17th (Lockdown 3), but with more relaxations in green zones where lower infection rates were reported. The final extension spanned May 18th to May 31st (Lockdown 4). States were given more authority to demarcate infection zones, and red zones were further divided into containment zones, which maintained stricter enforcement of lockdown norms than other zones.
The unlocking (resumption) of economic activities began in June 2020. During the first phase (Unlock 1, June 1st to June 30th), interstate travel was allowed with few state-specific restrictions while containment zones continued to follow lockdown norms. Phased unlocking continued in July (Unlock 2) when the authority to impose lockdowns was further decentralized to local governments. Pune city and the adjoining areas implemented a regional lockdown from July 14th to July 23rd in response to a sharp rise in COVID-19 patients. City and state authorities enforced a strict lockdown during the first week—a complete shutdown of all essential services, except emergency healthcare. This resulted in minimal movement in Pune’s public spaces. Slight relaxations in the supply of essential goods and services followed during the second week and Unlock 2 summarized in Pune on July 24th. August 1st to August 31st (Unlock 3) witnessed further relaxations in interstate travel and an end to nationwide curfews. Pune shopping malls and market complexes could remain open until evening, and cab services could operate with a restricted passenger load. However, lockdown restrictions continued in containment zones. During September (Unlock 4), gatherings of up to 50 persons were permitted while containment zones continued to follow lockdown norms. Early in September, Pune experienced a sharp rise in COVID-19 patients and became a top national COVID-19 hotspot (The lockdown events are summarized in the supplemental Fig. 1).
The area within PMC limits is divided into 15 administrative units, called ward offices (Fig. 1b), which are further divided into 41 electoral wards with similar populations, called prabhags. Individual-level data were included for the time period spanning February 1st to September 15th, 2020. According to daily press reports released by PMC, a total of 542,946 samples were collected for COVID-19 testing during the study period, and of these, 313,373 records were available. These data were curated to remove records with missing data. The remaining records were assigned to a prabhag using a machine learning based geocoder that was developed in house. The geocoding methodology is described in the supplementary material 1. Records with a confidence score below 0.5 out of 1.0 (provided by the ML geocoder) and records for persons residing outside PMC limits were removed. The final dataset used for this analysis comprises 241,629 records.
This analysis was done retrospectively on programmatic data without personal identifiers, hence individual patient consent was not obtained as infeasible. The Ethics Committee of the Indian Institute of Science Education and Research, Pune, India approved the analysis of COVID-19 programmatic data and has waived the need for obtaining the consent. The analysis and reporting were performed in accordance with the relevant guidelines and regulations.
Statistical analysis and mathematical modeling
The primary endpoint was weekly change in incident COVID-19 patients. The secondary endpoint was weekly infection rate; infection rate was calculated as the number of positive SARS-CoV-2 results divided by the total number of tests per 1000 population. Other endpoints included risk of COVID-19, defined as an incident COVID-19 case. Primary and secondary endpoints were assessed pre-lockdown, during lockdown and post-lockdown in the overall dataset and by population characteristics, namely sex, age group, and ward office-specific subcategories (population density and proportion residing in slum areas). Population density was calculated as number of people per 1 square kilometer and has been reported for all 15 PMC ward offices. For this analysis, population density was binarized as high (above the 3rd quartile of PMC ward office density, n = 6) or low-average (below the 3rd quartile of PMC ward office density, n = 9) (Fig. 1b). Since differences in infection rates existed among ward offices, the effect of lockdown on the primary endpoint was assessed using a multilevel Poisson regression model with random effects for ward office and test week. Change in the weekly infection rate over the study period was estimated using quasi-Poisson regression analysis. Logistic regression was used to assess the effect of risk factors on mortality. Epidemic curves for trends of incident patients over time were plotted using nonparametric locally weighted regression for the overall population and by sex, age group, and ward-specific subcategories.
We modeled the trajectory of the natural epidemic to estimate the delay of the peak of the pandemic. For this, we used a 9-compartmental model INDSCI-SIM that enables robust predictions taking into account the effects of various non-pharmaceutical measures (Supplementary appendix)23.24. There are a wide range of estimates for the value of R0; for example, Hilton and Keeling estimated R0 between 2 and 325 While India specific study by Sinha found out the value to be around 1.8. In order to avoid overestimation of total patients, we also considered R0 =1.826. Although there is no unique way to estimate actual number of patients, we assume infection on the first day (taken to be 1st April 2020) of the simulation to be three times reported patients. We note here that the choice of R0 and initial values may affect the final outcome, but our choices are conservative and more accurate estimation may make the results worse than reported here. We assessed the geospatial spread of COVID-19 patients over time and the visualizations were generated using the Python library geopandas (version 0.7.0, https://pypi.org/project/geopandas/0.7.0/). (Supplementary appendix). Data were analyzed in Stata Version 14·2.