Delhi’s huge third surge in Covid-19 cases – and deaths – reminds us that there are few certainties when it comes to epidemics caused by the coronavirus. With cases rising steeply eight months after the disease reached the city, it serves as a warning that a long harsh epidemic is no guarantee of light at the end of the tunnel. Moreover, fatality is not lower this time round – the data suggests those infected are increasingly at risk of dying, as we will see.

To make sense of it all, residents of the city would probably like to know the answers to a number of questions. How many have had Covid-19 in the city? How many of those infected have died? What has been driving surge after surge, and what could break this cycle? How has the disease impacted different areas, different age-groups and people living in different kinds of housing?

Inadequate data makes Delhi’s story hard to decode

Unfortunately, the city’s daily bulletins with their minimal detail do not shed much light on these questions. They compare poorly with Mumbai’s highly detailed dashboards that give a breakdown of cases and deaths by age and provide information on geographical spread, ICU beds, contact tracing, and much more.

This data, along with Mumbai’s carefully designed serosurveys, made it possible to disentangle the slum and non-slum epidemics and trace the origins of the city’s second wave.

Interpreting Delhi’s data has been further complicated by major changes in testing. Increases in testing are welcome, whether they involve RT-PCR tests or less sensitive rapid antigen tests. But the ups, downs and shifts mean that every time we see a rise in cases people naturally ask: are infections accelerating or is it all about changes in testing? To be clear: both second and third waves have been real – daily infections have indeed been rising.

Fatalities can provide the most valuable insights into an epidemic. But early systematic fatality underreporting and delays followed by “reconciliation” created a confused picture in Delhi. (This is also the case in Mumbai.) Delhi’s fatality data has no doubt become more reliable over time. But understanding early trends – particularly the surprisingly low early fatality rate – might have helped make sense of the current surge and increasing fatality rate.

A market in Delhi on October 19. Credit:Anushree Fadnavis/Reuters

Making sense of Delhi’s four serosurveys

Could the city’s four serosurveys fill in some of the gaps? The aim of these surveys was to estimate how many people had been exposed to Covid-19 based on the presence of antibodies to the disease in the blood.

Such surveys provide some important clues about the current surge, but their value is diminished once again by poor transparency. There has been no technical documentation describing the surveys or their results. The Delhi government released neither any analysis of its own, nor the data which would allow others to do such analysis.

There was little consistency in the surveys. From a report in the Hindustan Times, it emerged that surveying methodology, and even test-kits, changed between surveys. Ultimately, we don’t know if the population was sensibly sampled, making it hard to use the results to speculate about disease spreading through different communities at different points in time.

The best that can be done in the circumstances is to try and factor in the multiple uncertainties and hope to still emerge with some conclusions. The process I use is referred to as “Monte Carlo simulation”: uncertainties are estimated or guessed, and we generate lots and lots of estimates for prevalence, fatality rate, detection of infections, and so forth, over time, to see if any useful patterns emerge. Basically, we get lots of stories consistent with the data, and see if some features of these stories are actually quite general.

The four serosurveys spanning late June to mid-October reported seroprevalence of around 23%, 29%, 25% and 25% – these were the fractions of those surveyed who had measurable antibodies to SARS-CoV-2, the virus responsible for Covid-19. As the pandemic raged throughout this period, the data only makes sense if we assume “waning antibodies”: namely, that older infections are less likely to be detected by antibody tests. Since this phenomenon seems to depend at least in part on the antibody tests used, we can refer to it as waning test sensitivity.

Waning test sensitivity was found to be key to interpreting Mumbai’s second serosurvey data. Although the antibody tests used in the two metros were different, Delhi’s data suggests that, as in Mumbai, the waning can be quite rapid. This is also consistent with news reports that antibodies could not be measured in over 40% of Delhi-ites who had previously tested positive for Covid-19.

Some results: despite high exposure, the surges are real and fatality is increasing

Given waning sensitivity, changing test-kits, and uncertain sampling, do any credible messages emerge from the simulations?

We find, with high confidence, that by late September, somewhere between 43% and 59% of Delhi’s population had been exposed to the disease. The magnitude of the third surge suggests that the lower estimate could be closer to the truth; but however we look at it, it is extremely likely that at least half of the city has been exposed to Covid-19 by mid-November. We do not yet know enough about reinfection to be certain of how much this half are now protected from the disease.

The simulations indicate that the second surge was real. But they also show a very clear and consistent improvement in the detection of infections. The peak in September was magnified by this improvement – June’s surge was, in fact, far greater in terms of daily infections. This mirrors the pattern seen in Mumbai: an early surge with very poor detection was followed by a second surge that appeared larger because of improved detection.

What about Delhi’s third, ongoing, surge? Although the serosurveys so far only give us information upto early October, we can use what they tell us about fatality rates and case detection to conclude that there have been a very large number of new infections in the past month. Although still probably not on the scale of the June surge, the disease may now be hitting a population more vulnerable to hospitalisation and death.

This brings us to one consistent and worrying observation from the simulations. They show that the naive infection fatality rate – recorded deaths as a fraction of estimated infections – has been increasing. This could be partly a consequence of improved transparency and surveillance around death recording. But it could also be that Covid-19 has affected a more elderly population over time as occurred in Mumbai. Although Delhi’s recorded Covid-19 mortality remains below Mumbai’s, it may catch up if current trends continue.

A municipal official fines a man walking in Delhi's Connought Place without a mask. Credit: Prakash Singh/AFP

How could we decode the current surge?

Delhi’s dramatic third surge in cases reflects an acceleration of the epidemic. With an increasing fatality rate and hospitals running short of ICU beds, the priority must be to trace why it is occurring.

Several reports have speculated that the festive season, dropping temperatures and rising pollution could play a part. Experience from Mumbai suggests that a likely additional ingredient is uneven spread of infection. Sub-populations that have seen relatively little infection can drive new spread, even if the city as a whole has seen high infection. The general principle that “a lot of people have had it, so disease can’t spread fast” falls apart when those who have had Covid-19 are not distributed evenly through the population.

Running with the hypothesis that there are communities/localities where the majority are still susceptible to the disease, the question must be who and where. Delhi has tested very widely – much more than Mumbai. This data must have generated insights into where and how spread is occurring, but these insights have not been shared. Alongside proper reports on the serosurveys, the Delhi government would do well to release data on cases and deaths from different areas over time, ideally broken down by age.

Perhaps the main lesson – for Delhi, but other city administrations too – is about transparency: do not be afraid to share the data early and share it widely. Therein lies the best hope of understanding and controlling the epidemic.

A detailed and referenced technical document on which this piece is based is available here.

Murad Banaji is a mathematician with an interest in disease modelling.