Relatives mourn a man who died due to COVID-19, at a crematorium in New Delhi, June 2020. Photo: Reuters/Adnan Abidi
- Official COVID-19 deaths greatly underestimate the horror of the pandemic: excess deaths are between 7- and 13-times higher.
- According to a new analysis, excess deaths rise and fall with the waves of disease, reflecting the pandemic and not other longer-term underlying trends.
- This toll is also broadly what we’d expect given India’s age structure, the extent of spread of the virus and COVID-19 fatality rates from global data.
The Economist recently estimated that there have been around 16 million pandemic excess deaths worldwide. These are 16 million deaths during the COVID-19 pandemic over and above what we would expect in normal times. By this measure, the global toll of the pandemic is more than three-times the official toll of COVID-19 deaths that were officially reported.
A number of research groups have also now put out estimates of India’s pandemic excess deaths. They range from a little under 3 million deaths to around 5 million deaths. India could thus account for around a quarter of global pandemic deaths.
To put India’s numbers in context: during normal times, around 9 million people die each year. An extra 3-5 million deaths is around one third to a half of normal annual deaths. By this measure – i.e. pandemic excess deaths as a fraction of normal yearly deaths – India stands among the hardest hit countries in the world.
India’s excess deaths and its COVID-19 deaths
In a recent preprint, we presented our own estimates of excess deaths. These align well with the other estimates. Our central estimate for the period from April 2020 to June 2021 is 3.8 million deaths – but we also present optimistic and pessimistic estimates ranging from 2.8 million to 5.2 million.
Much of the variation reflects different scenarios for how the pandemic might have affected levels of death registration. We’ll discuss this further below, but even the lowest estimate means that India has experienced a mortality crisis not seen since its independence.
One thing is very clear: official COVID-19 deaths greatly underestimate the horror of the pandemic. Excess deaths are between 7- and 13-times higher.
We may never know precisely how many of these excess deaths were from COVID-19. But here are two things we found:
- The excess deaths rise and fall with the waves of disease. Whether the deaths are all from COVID-19 or not, they clearly reflect the pandemic, and not other longer-term underlying trends.
- The toll is broadly what we would expect given India’s age structure, the extent of spread of the virus, and COVID-19 fatality rates from international data. This makes it plausible that most of the excess deaths were indeed from COVID-19.
Estimating excess deaths using civil registration data
Many of the estimates, including our own, are based on death registration data from the Civil Registration System (CRS). In order to estimate excess deaths, we need first to estimate how many deaths to expect. In countries where all deaths are registered and there is good historical data, estimating expected deaths is fairly straightforward. In India, it is more complicated.
For a start, we cannot be very certain how many deaths occur during normal times. When we wrote earlier that around 9 million deaths occur every year, it was just one estimate. The government’s Sample Registration System (SRS) survey estimated around 6.2 deaths per 1,000 population per year in 2018, implying a total of around 8.3 million deaths that year. The UN, on the other hand, put the death rate around 16% higher, giving around 9.6 million deaths per year.
While the absolute numbers differ, both government and UN estimates agree on one thing: yearly deaths are quite stable, remaining flat or rising gradually in line with population growth.
Registered and unregistered deaths
In order to use CRS data, we need to estimate levels of death registration. Here again, there are uncertainties. The most optimistic estimates imply that around 8% of deaths went unregistered in 2019, but a variety of other calculations, including from the UN, suggest that the true figure could be closer to 20%.
Whatever the actual level of registration, registration coverage has been improving. Between 2015 and 2019, total death registrations went up on average 5% per year, and we know that the majority of this change reflects improved registration rather than increasing deaths.
When we look at individual states, however, we see rather unsteady improvement. Government data puts the fraction of deaths registered in Uttar Pradesh during 2015-2019 at 44.6%, 40.6%, 38.7%, 61.6% and 63.3%, respectively. In Bihar, the corresponding numbers were 29.2%, 25.7%, 38.7%, 31.2% and 51.6%.
Based on these volatile percentages, it would be hard to predict the level of death registration to expect in either state during 2020. Formal calculations of trends may miss crucial reasons why registration levels changed from one year to the next, and underestimate how vulnerable improvements are to being reversed.
In particular, it is hard to gauge the scale of disruption to registration during the pandemic. The early period saw major drops in registered deaths. For example, the 12 states whose data we use reported around 10% fewer registrations during March-April 2020 than during the same period in 2019. Registration may have recovered after this, but we can’t be sure because the pandemic surge in mortality subsequently obscured the picture.
The picture is almost certainly very mixed. In each state or territory, registration was disrupted to various extents at different points during the pandemic, and may have returned to 2019 levels or above at others. Interestingly, where such data is available, we find drops in birth registrations during 2020. To unravel this story will require looking at each region carefully, and may need further data from surveys.
Are there alternative explanations for the data?
Could the pandemic mortality surge be a mirage? The Ministry of Health and Family Welfare has implied this, calling some of the estimates speculative and erroneous. It has also claimed that “while some cases could go undetected … missing out on deaths is completely unlikely”; and that the “Civil Registration System (CRS) ensures all births and deaths in the country get registered”.
The first claim appears to be telling us to trust the official COVID-19 toll. The second appears to be telling us to trust CRS data. But as we have seen, official COVID-19 deaths and CRS data tell very different stories.
Some commentators have implied that the surge in registered deaths reflects natural increases in mortality and/or an increasing fraction of deaths being registered during the pandemic. Our analysis finds such claims to be highly implausible. Yearly deaths were steady or slowly rising prior to the pandemic. Registration was improving gradually, before the pandemic disruption. Our most optimistic estimates factor in natural increases in mortality, and significant improvements in registration. And even so, we still find close to 3 million excess deaths.
Also, we have to consider the fact that the excess deaths surged and declined precisely as COVID-19 surged and declined. This is not the pattern we would expect from long-term trends in mortality. In fact, the rise and fall is clearest in states with strong and timely civil registration.
Punjab and Andhra Pradesh are two such states. According to their data, Punjab’s pandemic excess mortality was relatively low, while Andhra Pradesh’s was very high. But in both cases the time-pattern is consistent: registered deaths rose during the first wave, returned close to baseline levels in early 2021, and then surged to new heights during the second wave.
Given that interpreting mortality data can be complex, the consistency between the estimates from different groups is remarkable. Yes, there are gaps in the data. For example, much of it is from online systems and to some degree incomplete. How we factor in the uncertainties changes the details; but the broad story of a huge mortality surge remains the same.
Some of the uncertainties could quickly be resolved if state governments simply released all the civil registration data they hold. But, given low levels of death registration in some states and fluctuations during the pandemic, survey data will also be needed to fill in some of the gaps.
Survey data could also help understand fatality rates in different age groups, and might clarify the extent to which excess deaths were directly from COVID-19. The data could help answer crucial questions about how the pandemic has interacted with existing inequalities.
For example, a recent small-scale survey of mortality in Bihar 1 found higher levels of mortality during the second COVID-19 wave than expected from CRS data. It is likely that marginalised social and economic groups have been most severely affected by the pandemic, but the data to confirm this is largely lacking.
It is important to understand that India’s pandemic mortality is roughly as expected. The absolute numbers are shocking, but we know from seroprevalence surveys that the disease spread very widely. International data tells us approximately how many infections will result in death once we account for India’s relatively young population. Taking these factors into account, the excess mortality estimates are neither unexpectedly low, nor unexpectedly high.
What is startlingly high, and varies very much between states, is the disparity between official COVID-19 deaths and excess deaths. It is important to understand why surveillance of pandemic deaths has been so poor overall, and particularly bad in some parts of India. This is key to understanding the Indian epidemic, and also to gaining insights into death surveillance in India more generally.
The data and analysis that inform this piece are available in our preprint paper on India’s pandemic excess mortality on medRxiv. Detailed state-level factsheets on mortality and available death registration data are available here. There are further estimates of India’s pandemic excess mortality from other groups here, here and here.
Murad Banaji is a mathematician at Middlesex University London. Aashish Gupta is a demographer at Harvard University.
The authors here also co-authored this study↩