How reliable are COVID-19 burden estimates for India?

With nearly 31 million reported COVID-19 cases and 410 000 deaths,¹ India is one of the countries with the heaviest burden of COVID-19 cases and deaths. There is near-universal consensus that the country's reported morbidity and mortality data are substantial underestimates. The majority of the morbidity and mortality in India are a consequence of the second wave, which started in March, 2021,¹ and which is attributable largely to the delta SARS-CoV-2 variant. There is some suggestion that India was largely spared from the COVID-19 disease burden in the first wave of the pandemic that began in June, 2020.¹ In the absence of good vital registration data and electronic health records that are available in more well resourced countries, good-quality surveillance data are relied upon to estimate disease burden. In this context, the study by Ramanan Laxminarayan and colleagues² in The Lancet Infectious Diseases makes a valuable contribution by reporting results from a large-scale active SARS-CoV-2 surveillance programme in Madurai, Tamil Nadu, during the first wave of the pandemic. In this study, prospective testing through RT-PCR was done from May 20, 2020, to Oct 31, 2020, for individuals with fever or acute respiratory symptoms as well as selected groups of individuals at high risk of COVID-19, including returning travellers, frontline workers, contacts of laboratory-confirmed COVID-19 cases, residents of containment zones, and patients having medical procedures. The authors also report data from a cross-sectional serosurvey done from Oct 19, 2020, to Nov 5, 2020.

On the basis of this surveillance, Laxminarayan and colleagues² report that the proportion of individuals who tested positive after RT-PCR was 3·6% overall (5·4% among symptomatic individuals and 2·5% among asymptomatic individuals). Although the number of males and females who received RT-PCR tests was broadly similar among symptomatic individuals, more males were tested than females among asymptomatic individuals. Adjusted odds of symptomatic SARS-CoV-2 infection were 21% higher among males than females, although this difference was reversed for asymptomatic infection. The case-fatality ratio among RT-PCR-confirmed cases was 2·4%. Although these findings are important for understanding the risk profile of symptomatic and asymptomatic infections at the population level, interpretation of these findings should be made in the context of several potential biases. First, the surveillance case definition for symptomatic individuals, which only considered fever and respiratory symptoms, was not comprehensive; other common COVID-19 symptoms, such as loss of smell and taste,³ were not considered, which could lead to both selection and misclassification biases. Second, testing among asymptomatic individuals was not random but targeted, and thus could lead to selection bias. Third, misclassification bias could also occur through comorbidity status (given that this was self-reported), which could bias the association between SARS-CoV-2 infection and risk factors towards null. Fourth, although the surveillance system in Madurai allowed 13·5 diagnostic tests per 100 inhabitants to be done, almost twice the national average for this period, the number of tests done per day was not uniform across the study period, which probably contributed to ascertainment bias. Lastly, socioeconomic deprivation and occupation were not considered in the analysis, which could confound the aforementioned association.

The Article from Laxminarayanan and colleagues² appears to suggest that cases and deaths were substantially underestimated. The authors report an overall weighted seroprevalence of 40·1% in their study population, whereas of the 440 253 RT-PCR tests that were done, only 15 781 were positive, thus indicating an infection-case ratio (ICR) of about 67. This is substantially higher than previously reported seroprevalence estimates of 18·4% (with an ICR of about 20) by Selvaraju and colleagues⁴ for Chennai, Tamil Nadu, in July, 2020. Another nationwide serosurvey done between August, 2020, and September, 2020, by Murhekar and colleagues⁵ reported a weighted nationwide seroprevalence of 6·6% (with an ICR of about 30), with an unweighted seroprevalence of 33·5% in Chennai. Laxminarayanan and colleagues² also report an overall infection-fatality ratio (IFR) of 0·043% from the serosurvey, a figure that is broadly similar to the nationwide study by Murhekar and colleagues⁵ in India (0·1%) but substantially lower than nationwide studies in England⁶ (0·9%), France⁷ (0·8%), and Spain⁸ (0·8%) during the first wave of the pandemic. The authors claim that only one SARS-CoV-2 death was reported for every 9·1 deaths expected to occur, which is a cause for concern; this estimate should be interpreted with caution—the expected deaths were crudely estimated by applying the external age-specific IFR estimates from a meta-analysis⁹ of studies from Spain, Geneva, New York City, England, Italy, Kenya, Portugal, and Sweden. Laxminarayan and colleagues² have made a courageous assumption that the difference in IFRs between Madurai, India, and other countries is entirely attributable to underreporting, although it is probable that several factors other than age could have contributed to variations in IFR, such as comorbidity,¹⁰ socioeconomic status, and occupation.¹¹ Nevertheless, this study serves as a call to action for substantial investments in developing good data systems to gather accurate data on COVID-19 morbidity and mortality, to inform policy decisions both for India specifically and low-income and middle-income countries in general. In the absence of good vital registration data, this type of data system would require strengthening existing demographic and health-surveillance systems with the addition of high-quality mortality surveillance, collection of nasal swabs twice per week even in the absence of symptoms and repeated serosurveys, and linking these results to clinic and hospital records. Such data systems would also support national burden estimates for other respiratory infections such as influenza, and close the data gaps for future pandemics.

© 2021 Flickr - Gwydion Williams

YL reports grants from WHO and the Wellcome Trust, outside the submitted work. HN reports grants from the Innovative Medicines Initiative, WHO, and the National Institute for Health Research, personal fees from the Bill & Melinda Gates Foundation, Janssen, ReViral, and AbbVie, and grants and personal fees from Sanofi and the Foundation for Influenza Epidemiology, outside the submitted work.