India’s national COVID death totals remain undetermined. Using an independent nationally representative survey of 0.14 million (M) adults, we compared COVID mortality during the 2020 and 2021 viral waves to expected all-cause mortality. COVID constituted 29% (95% confidence interval, 28 to 31%) of deaths from June 2020 to July 2021, corresponding to 3.2 M (3.1 to 3.4) deaths, of which 2.7 M (2.6 to 2.9) occurred in April to July 2021 (when COVID doubled all-cause mortality). A subsurvey of 57,000 adults showed similar temporal increases in mortality, with COVID and non-COVID deaths peaking similarly. Two government data sources found that, when compared to prepandemic periods, all-cause mortality was 27% (23 to 32%) higher in 0.2 M health facilities and 26% (21 to 31%) higher in civil registration deaths in 10 states; both increases occurred mostly in 2021. The analyses find that India’s cumulative COVID deaths by September 2021 were six to seven times higher than reported officially.
As of 1 January 2022 and prior to the current surge driven by the Omicron variant, India reported over 35 million cases of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), second only to the United States (1). India’s official cumulative COVID death count of 0.48 million implies a COVID death rate of ~345 per million population, about one-seventh of the US death rate (2). India’s reported COVID death totals are widely believed to be underreports because of incomplete certification of COVID deaths and misattribution to chronic diseases and because most deaths occur in rural areas, often without medical attention (3, 4). Of India’s 10 million deaths estimated by the United Nations Population Division (UNPD) in 2020, over 3 million were not registered and over 8 million did not undergo medical certification (fig. S1 and table S1).
Model-based estimates of cumulative COVID deaths through June 2021 in India range from a few hundred thousand to more than 4 million, with most suggesting a substantial official undercount (5–12) (table S2). However, models start with official reports and apply varying assumptions, leading to wide or implausible estimates. In the absence of near universal and timely death registration and the lack of release of data from India’s Sample Registration System (SRS), which tracks deaths in a random sample of about 1% of Indian homes (13), alternative approaches are needed to estimate COVID deaths. Recorded increases in all-cause mortality during peak pandemic transmission are likely nearly all caused by COVID infection (14). The World Health Organization (WHO) has recognized such counts as a crude but useful method to track the pandemic (15). Reports by journalists and nongovernmental organizations using civil registration system (CRS) data have documented a large increase in deaths from all causes compared with previous years (16). Unfortunately, CRS data are reliably available only in states that cover about half of the estimated total deaths in India and may be affected by changes in the level of registration. Given the marked heterogeneity in the temporal patterns of confirmed COVID mortality cases and deaths across states (17), and the variable background of mortality rates from chronic diseases affected by COVID infection (3), extrapolating from selected states has its limitations.
To fill the gaps in national-level estimates, we quantified COVID mortality in India using one independent and two government data sources. The first study is mortality reported in a nationally representative telephone survey conducted by CVoter, an established, independent, private polling agency, which launched the survey on a nonprofit basis to help track the pandemic [see materials and methods, p. 2 (18)]. The COVID Tracker survey covers 0.14 million adults (including a substudy of 57,000 people in 13,500 households with more exact reporting of COVID and non-COVID deaths in immediate family members) (18, 19). In addition, we studied the Government of India’s administrative data on national facility-based deaths and CRS deaths in 10 states (fig. S2).
The CVoter Tracker survey is a nationally representative, random probability-based computer-assisted telephone interview survey carried out daily to track governance, media, and other socioeconomic indicators (19). In March 2020, it began to capture COVID symptoms among adults aged 18 years or older, covering ~2100 randomly selected respondents weekly, drawn from ~4000 local electoral areas in the whole of the country, providing a rolling 7-day average of COVID symptoms and deaths. The survey covers >98% of Indian population by geography, with interviews in 11 languages. The response rate was 55%; 137,289 respondents in all states and union territories were interviewed from March 2020 to July 2021.
Our numerator was defined as the average weekly percentages of surveyed households reporting a COVID death (defined by the household, as medical certification remains uncommon in India; fig. S1). We excluded the 16% of reported COVID deaths that were below age 35 years (confirmed COVID deaths below this age are infrequent; fig. S3) and subtracted a fixed percentage of 0.59%, which was an assumed value for reported deaths that did not occur among immediate family members. The assumed value drew on observed background rates during February–March 2021, when few COVID cases or deaths were reported in the official government data (see materials and methods, p. 3). Results using survey weights or raw proportions were similar, so we used the latter. We compared these survey-reported COVID deaths to a denominator defined as the expected weekly percentage for all-cause deaths, based on 2020 death totals from the UNPD’s comprehensive demographic estimates that combine censuses, survey data, and models (20) (Fig. 1). India had about 296 million households in 2020, with an average household size of 4.6 (21). Dividing this into the 10.16 million deaths estimated by the UNPD in India in 2020 yields ~3.4% of households expected to report a death from any cause in that year (with nearly identical results for 2021). To this expected all-cause proportion, we applied the weekly variation observed in the Million Death Study, a large and representative mortality study conducted within the SRS (3).
