Population-Based Serosurvey for Severe Acute Respiratory Syndrome Coronavirus 2 Transmission, Chennai, India

We conducted a cross-sectional survey to estimate the seroprevalence of IgG against severe acute respiratory syndrome coronavirus 2 in Chennai, India. Among 12,405 serum samples tested, weighted seroprevalence was 18.4% (95% CI 14.8%–22.6%). These findings indicate most of the population of Chennai is still susceptible to this virus.

O n August 15, 2020, India had the third highest number of coronavirus disease (COVID-19) cases globally (1). The Indian state of Tamil Nadu reported 332,105 cases and 5,641 deaths on August 15, and ≈35% cases were from the state capital, Chennai (2). Administratively, Greater Chennai Corporation (GCC) is divided into 15 zones that are further divided into 200 wards with populations ranging from 4,400-104,558 (3). The total population of GCC is 7.1 million and 31% of the population resides in slums.
As a part of nationwide containment strategy, Chennai was under lockdown beginning March 25, 2020; beginning May 4, the lockdown was relaxed in a phased manner. Wearing facemasks in public has been mandatory since April 13. However, the number of COVID-19 cases has been increasing in Chennai since May.
Serologic surveys can provide a comprehensive picture of community spread of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), the causative agent of COVID-19 (4). During the first week of May, the unweighted seroprevalence in Chennai was 2% (5). We conducted a communitybased serosurvey in July 2020, to estimate the seroprevalence of SARS-CoV-2 in GCC.

The Study
We conducted a household-based cross-sectional survey among usual residents >10 years of age in GCC. To estimate a seroprevalence of 2%, with 20% relative precision, design effect of 2.5, and 95% CI, we needed a sample size of 11,710 persons, which we rounded to 12,000. We used a multistage cluster sampling method to select the survey participants. In the first stage, we selected 51 wards by using probability proportion to population size method. In the second stage, we randomly selected 6 streets from each ward from which to recruit participants. The survey team selected a random starting point in each street and visited contiguous households to enroll >40 consenting persons >10 years of age. When no one was home or household members were unavailable, the team proceeded to the next house and completed the survey until >40 persons were enrolled from each street. We included all eligible persons in the household who consented.
After obtaining written consent from persons >18 years of age, and assent and parental or guardian approval from persons <18 years of age, we interviewed participants to collect information. We used the Open Data Kit application (https://opendatakit.org) to collect sociodemographic details, and information on exposure to laboratory-confirmed COVID-19 case, history of COVID-19 symptoms in the past 3 months, and COVID-19 testing status.
We analyzed the data to estimate weighted seroprevalence of SARS-CoV-2 and 95% CI by using appropriate sampling weights. We further adjusted the seroprevalence for assay characteristics (6). We estimated the total number of SARS-CoV-2 infections among persons >10 years of age and infection-to-case ratio (ICR) (Appendix).
The survey teams visited 7,234 households from 321 streets across 15 zones. Of the 18,040 residents >10 years of age in the visited households, 14,839 (82.3%) were available at the time of survey, among whom 12,405 (83.6%) consented to participate (Appendix Table 1). The mean age of survey participants was 41.1 years (SD 17.3 years); 52.7% were female and 47.3% were male. Among 496 (4%) persons who reported prior reverse transcription-PCR (RT-PCR) testing for CO-VID-19, 119 (24%) reported testing positive (Table 1).

Conclusions
Our community-based survey indicated that ≈1/5 persons in Chennai was exposed to SARS-CoV-2 by July 2020. We noted a wide variation in the extent of infection across wards and seroprevalence ranged from 2%-50% (Appendix Table 3).
Seroprevalence was higher in northern Chennai and adjoining wards of central Chennai than in southern Chennai (Figure). Chennai witnessed a surge in COVID-19 cases in last week of April 2020 and >65% of cases were in northern Chennai (7). The number of cases showed a declining trend after the first week of July. Northern Chennai has a higher population density (55,000/km 2 ) than Chennai (27,000/ km 2 ) and has several slum areas (7). High population density and persons living in close proximity might have contributed to the higher seroprevalence observed in northern Chennai.
Seroprevalence was lower among male participants. Laboratory surveillance data in India showed a higher proportion of laboratory-confirmed COVID-19 among male than female patients (8). Comparable seroprevalence between children and adults suggests exposure within and outside of the household settings. Lower prevalence among persons >60 years of age could be due to lower exposure to infected persons or stricter adherence to nonpharmaceutical interventions. Serosurveys conducted in Santa Clara County, California, USA reported lower seropositivity among persons >60 years of age (E. Bendavid,   (9) and in Greece, seroprevalence was higher among persons >60 years of age (10). Most seropositive participants in our survey did not report any symptoms nor had any known contact with COVID-19 patient. IgG developed among most (107/119; 90%) recovered COVID-19 patients in our survey. Among 105 participants for whom >15 days had passed between RT-PCR confirmation of COV-ID-19 and blood sample collection for our serosurvey, 99 (94.2%) had seroconverted. Even after accounting for a 2-week delay for development of antibodies (11), ≈6% of COVID-19 patients were seronegative. Discordance between RT-PCR test results and presence of IgG might be due to poor B cell response or antibodies waning over time (12).
The ICR ranged from 19-21 and was lower than the ICR of 82-130 reported during the nationwide seroprevalence survey in India conducted in May 2020 (5). Lower ICR reflects a high level of case detection, resulting from extensive COVID-19 testing in the city. By July 15, 2020, Chennai had conducted 14,270 tests/ million population.
Our study had 2 limitations. First, ≈1/3 persons from the visited households did not participate in the survey. Among them, 17.7% were not available at the time of visit and 13.5% refused to participate. Due to time constraints, we did not revisit households where persons were not available. The proportion of female participants and children 10-19 years of age was higher among persons who did not participate in the survey (Appendix Table 2), which might have influenced the seroprevalence estimates in either direction. Second, we might have underestimated the seroprevalence because antibodies to  nucleocapsid protein have been shown to decline after infection (13).
In conclusion, ≈80% of the population in Chennai is still susceptible to SARS-CoV-2 infection. Transmission is expected to continue in wards with lower seroprevalence. Maintaining high testing rates and monitoring adherence to nonpharmacological interventions in GCC should be continued. In addition, periodic serosurveys would help monitor the trend of infection and assess the effects of varying containment measures in the city.
This study was funded by Greater Chennai Corporation public health department (PHDC no. 2797/20 dated July 9, 2020).
We used a random effects logistic regression model to address the clustering effect of estimates at all levels of hierarchical structure identified in the design. The hierarchical structure used in the analysis was ward, street, and household levels.
We modeled overall seroprevalence by using random intercept model for inclusion of each of the levels with design weights. We used the Akaike Information Criterion to select the final model. We also used this model to estimate seroprevalence for other factors, such as age and sex.
Seroprevalence estimates were obtained by exponentiating the log odds values obtained from the model and converting into probability to calculate corresponding 95% Wald confidence interval. We used the lme4 package from R software (R Foundation for Statistical Computing, https://www.r-project.org) to perform analysis.
We compared the weighted seroprevalence by selected demographic characteristics, history of respiratory symptoms, contact with laboratory-confirmed case of coronavirus disease, and coronavirus disease testing. We considered p<0.05 statistically significant.
We adjusted the weighted seroprevalence for test characteristics by using the following formula (3):