A lab worker in PPE removes vials of AstraZeneca’s Covishield vaccine candidate from an inspection machine at the Serum Institute of India, Pune, November 30, 2020. Photo: Reuters/Francis Mascarenhas/File Photo.
In mid-September, India’s COVID-19 cases and deaths began a steady decline, which then slowed and became more uncertain in late October.
The data appears to tell a familiar story. There is a peak, the worst seems over, but then comes a roadblock: the number of cases refuses to fall any further and sometimes starts to rise again.
What could lie behind this story of peak, decline and “getting stuck”? And what is likely to happen next? Answering these questions is hard – but spelling out the difficulties could help.
A sum of many epidemics
First, the complex national picture is a sum of very diverse local pictures. Even within individual states, there is huge variation. A seroprevalence survey in Haryana in October detected antibodies to SARS-CoV-2 in over 30% of those surveyed in Faridabad, a predominantly urban district in the National Capital Region, but only 3% of those surveyed in Bhiwani, a rural district further west. Even omitting urban hotspots, such studies tend to reveal widely varying levels of infection.
Epidemics may also be connected. It is not so surprising that Faridabad has seen high levels of infection given its close relationship and proximity to Delhi. Indeed, Delhi’s recent surge seemed to trigger surges in Gurgaon and Faridabad.
We see this pattern frequently. In Maharashtra, Ahmednagar district’s cases closely track neighbouring Pune’s, although with a difference of scale and a delay.
But we can’t assume that nearby areas will necessarily see closely linked epidemics. Mumbai’s first seroprevalence survey found very different levels of infection in slum and non-slum areas, even though these are geographically entangled. It’s likely that lockdown and mitigation played important roles in decoupling the slum and non-slum epidemics.
Variable surveillance
So the national COVID-19 story is the sum of varied and more or less interconnected local stories. Local patterns of development seem to strongly influence how the disease spreads geographically.
But there is another factor making it hard to interpret case and death data: differences in disease surveillance. In some areas, a significant fraction of infections are identified through testing and recorded as cases. In others, almost all infections pass under the radar.
Seroprevalence survey data from Chhattisgarh and Bihar indicated that about one in every 15 infections had been detected in the districts surveyed in Chhattisgarh, whereas only about one in every 140 infections had been detected in those surveyed in Bihar.
A similar disparity seemed to hold for deaths too. Madhubani, a rural district in Bihar, experienced a huge epidemic – but it was almost entirely invisible in both case and death data.
Surveillance of deaths may be as variable as of infections – a fact that the Indian Council of Medical Research appeared to acknowledge in a paper on the first national seroprevalence survey.
Even within a single locality, there can be variable detection, reflecting in part variable access to healthcare and tests. A striking example comes from Mumbai, where COVID-19 cases disproportionately represented non-slum areas.
Careful analysis suggests that over a period of about five weeks, from mid-April to late May, weekly cases in Mumbai’s slums increased by over 13-times. But weekly cases from the city as a whole went up only about fivefold. Poor detection in the slums meant that the very rapid slum-surge was obscured in city-level data.
So official case- and death-counts reflect both epidemic size and quality of surveillance. To use an analogy, the size of what we see depends both on the object we are looking at, and on the power of our telescope.
Also read: The Problem With Spinning Simple Stories Based on India’s COVID-19 Numbers
The national slowdown
Variable spread and detection mean we should be careful about what we read into national case and death data. The national decline in cases and deaths almost certainly reflects a real drop in infections – but where?
States such as Maharashtra, Andhra Pradesh and Karnataka have seen large drops in the numbers of cases and deaths. Others, including Madhya Pradesh, Rajasthan and Gujarat, have witnessed recent surges.
By late November daily cases nationally had dropped nearly 60% from their mid-September peak. But if we remove six states with high case-loads and big declines – Maharashtra, Andhra Pradesh, Karnataka, Tamil Nadu, Uttar Pradesh and Odisha – the rest of the country has seen only a modest decline, of about 28% from mid-September peak values.
So the sharp national decline reflected rapidly decreasing cases in some badly hit states. Unravelling the story of each state would be a major task, but even glancing at the data raises interesting questions.
Karnataka’s data is heavily dominated by one district: Bengaluru Urban. While almost all districts reported sharp declines in cases, the statewide decline largely reflected the massive decline in Bengaluru’s cases.
By contrast, Andhra Pradesh, with a case load similar to that of Karnataka, has more even data. Most districts contributed roughly equally to its cases. Do these differences reflect different patterns of development and hence of disease spread? Or different patterns of surveillance?
Prospect of new surges
Most states have pictures closer to that of Karnataka than of Andhra Pradesh – that is, a few districts tend to dominate the state-level data. But the detail can change over time. In Odisha, Ganjam and Cuttack districts saw similar-sized peaks, but the first occurred fully two months before the second.
The variable speeds and staggered timing of outbreaks from different localities often explain why epidemics appear to get stuck after a peak, or even surge again. As some outbreaks die down, others pick up.
This is true nationally as well. Kerala’s late-starting epidemic, along with Delhi’s late surge, Maharashtra’s plateau, Haryana’s and Rajasthan’s second waves, and West Bengal’s very slow decline, all help explain why case numbers nationally have not fallen further.
Will India see a major resurgence as in several European countries? We have seen many city-level resurgences: in Ahmedabad, Delhi, Indore, Jodhpur, Kolkata, Ludhiana, Mumbai and Pune, for example. Some experts have mentioned colder weather and pollution as possible drivers of more severe disease and increased spread.
Until the country rolls out a national vaccination programme, we should indeed expect new upswings – some of them major. But whether they will add up to a national resurgence is difficult to predict. This depends partly on where the upswings take place. Remember that surges from regions with poor surveillance can be hidden in national data.
Waiting for a vaccine
Although the disease might naturally spread at different speeds in a city slum, a gated colony or a network of villages, there is probably no corner of the country that is ‘naturally’ safe from outbreaks. What will slow or prevent new outbreaks is a combination of mitigation and some level of population immunity. And these will play different roles in different areas.
People denying the severity of the COVID-19 pandemic often choose to ignore mitigation – especially where it isn’t very obvious or coercive – and focus only on immunity.
When an epidemic winds down, they generally suggest herd immunity has been reached. For example, there was a chorus of incorrect, politically motivated claims about Sweden reaching herd immunity after its first wave.
Read more: The Problem With Using Herd Immunity To Control COVID-19
In India, COVID-19 has changed ways of living – probably everywhere within the country. This is a key factor in why many local epidemics have wound down even as lockdowns have ended. It implies that most places remain vulnerable to new surges if there is a full return to ‘normal’. Tracking the extent of this vulnerability, for example with local serosurveying and random testing, is crucial at the moment.
Alongside mapping vulnerability to new outbreaks, a careful study of what drives transmission locally could help fashion more focused strategies to prevent disease spread.
People in an area need to know: how much active infection is there locally? Are large gatherings driving spread? Is it occurring mainly in workplaces? While travelling? During social activity? What are high risk activities? Contact tracing data, if shared, can provide valuable insights into these issues.
Less propaganda, more data and above all more transparency – these are required to limit the pandemic’s toll while we wait for a vaccine.
(COVID-19 case and death data at the state and district levels are from covid19india.org.)
Murad Banaji is a mathematician with an interest in disease modelling.