A healthcare worker checks the temperature of a resident during a medical campaign for COVID-19 at a slum area in Mumbai, July 2020. Photo: Reuters/Francis Mascarenhas/Files.
After Delhi, Ahmedabad and Mumbai, Pune is the fourth major Indian city to report the results of a seroprevalence survey aimed at detecting the presence of immunoglobulin G antibodies to SARS-CoV-2, the virus responsible for COVID-19.
Pune’s survey reported a very high prevalence of the antibodies, higher than any of the other surveys. Although smaller than the surveys in the other three cities – and somewhat hard to interpret because of choices in the survey design – Pune’s survey nevertheless provides valuable insights into the city’s COVID-19 epidemic.
The sero-survey was carried out in five of the city’s 41 wards between July 20 and August 5. A total of 1,664 individuals were tested, of whom 51.5% tested positive. Adjusting for the test’s sensitivity and specificity – 84.7% and 100%, respectively, according to the survey report – yields an even higher prevalence in the sample of 60.8%.
The participants were aged 18 and over and were chosen from wards where the infection was reportedly prevalent. Four of the five wards are also clustered together, geographically. So it is not clear whether the survey’s results can be considered to be representative of the city as a whole.
Effects of housing poverty
As in Mumbai, the seropositivity was found to vary considerably with the nature of housing – but the differences were perhaps less stark. Raw seroprevalence ranged from 62% in “hutments” down to 33% in “apartments”. Shared toilets were a good predictor of high seropositivity, although this does not of course imply causation.
Correcting for the test’s sensitivity, seropositivity among those with shared toilets was an astonishingly high 73.5% against 53.5% amongst those with independent toilets. Housing poverty thus appears strongly as a factor in the rapid spread of COVID-19 in the city. However, the rapid spread even in non-slum areas is a notable difference between Pune’s and Mumbai’s results.
An estimated 40% of people in Pune live in slums. The survey data, if extrapolated to the whole city, yields a corrected prevalence of 61.5% at the time of the survey. Note, however, that the wards were chosen precisely because they were high-incidence wards, so it seems likely that the results overestimate prevalence in the entire city. (Pune’s mayor has called Lohiya Nagar, the ward with the highest incidence, “the Dharavi of Pune”.)
We also know from Mumbai that the number of cases does not necessarily reflect prevalence in different parts of the city, and that the disease can spread through different geographical areas unevenly over time. So we need to be cautious about extrapolating the results.
Estimated infection fatality rate
Despite all these reservations, it is interesting to explore the implications of assuming the prevalence inferred from the survey could be generalised to all of Pune. What would this tell us about the infection fatality rate (IFR) of COVID-19 in the city?
(According to the WHO, the infection fatality rate “estimates the proportion of deaths among all infected individuals” whereas the case fatality rate “estimates the proportion of deaths among identified confirmed cases.”)
If indeed the city had a COVID-19 seroprevalence of 61.5% at the time of the survey, there would have been about 2.1 million infected people there by mid-July. But there had been 1,744 COVID-19 deaths in Pune city by August 9, so we get a rough and naïve IFR of 0.08% for the city.
Note that this IFR figure ignores possible fatality undercounting and relies on the assumption that the prevalence among slum and non-slum areas in the whole city is captured in the wards that were surveyed.
Has there been COVID-19 fatality undercounting in Pune? The data available for Pune district shows a very marked and steady fall in the cumulative case fatality rate (CFR), alongside a fairly steady rise in the cumulative test positivity rate through the duration of the epidemic. Pune city accounts for about two-thirds of cases and deaths in the district; we may assume these trends hold in the city as well.
A falling CFR without any evidence of better detection of infections can be a sign that death undercounting is becoming more common. We can use modelling to explore this further and suggest IFR values consistent with the whole trajectory of Pune’s epidemic. To do this, we need to make various approximations about prevalence in Pune district.
Let’s assume based on the case and fatality data that Pune city accounts for about two-thirds of all infections in the district. So the prevalence in the district as such would have been about 3.1 million shortly before the survey.
When we model COVID-19 across time, we obtain IFR values that are considerably higher than the naïve one. The methods, assumptions and limitations of the modelling approach have been previously described in a similar analysis for Mumbai. The results of two example simulations (below) can be seen as the two extreme possibilities obtained from modelling. (More technical details are available here.)
We find that values of IFR consistent with the approximate seroprevalence of 3.1 million in the district by late July range between 0.32% and 0.45%.
These estimates in turn correspond to a massive 70-80% of COVID-19 fatalities going ‘missing’ by the end of July.
This can be interpreted either as a very significant drop in IFR in the district – or increasing death undercounting over time. The simulations also suggest that, which ever way we interpret the data, much of the drop happened in the month of May, with the change later being more gradual.
So something rather dramatic seems to have happened in Pune district vis-à-vis COVID-19 fatalities.
A combination of effects?
COVID-19 has spread quickly and widely through Pune. Using the data for Pune district, it is currently unclear whether the epidemic has peaked. While the number of new cases every day appears to be dropping slightly, the same is not true of daily deaths. One possibility is that the epidemic has peaked in the city but not yet in the rest of the district. But Pune city is still recording high numbers of cases and deaths as well.
As in Mumbai, it seems that the speed of the spread was higher in areas with poorer living conditions. While the naïve IFR in the city is just 0.08%, the survey – having been conducted in wards with higher prevalence – likely exaggerated the prevalence estimates and thus lowered the IFR estimate.
In contrast to the naïve IFR estimate, simulations of the whole epidemic in Pune district suggest IFR values of 30% or more. The district, and likely the city, has experienced what appears to be a very dramatic drop in its COVID-19 IFR during the course of the epidemic.
We saw the same phenomenon play out in Mumbai, and indeed several other Indian cities. This could be interpreted as a genuine drop in fatality rate, possibly resulting from disease spreading in a younger population or from improved treatment. Alternatively, there has been very significant and growing fatality undercounting. A combination of the two effects is likely. Without excess mortality data, it is hard to know where the truth lies.
Murad Banaji is a mathematician with an interest in disease modelling.