Whither, COVID-19 in India?

In my last post on Jan 27, I had expressed cautious optimism that the SARS-CoV2 virus would be contained soon. My assessment was based on the very focal and recent origin of the virus as determined by genomic studies. How things have changed since. There are now more than 400,000 documented cases and more than 18,000 deaths. The virus has turned out to be far nimbler and nastier than one would have thought.

The nimbleness of SARS-CoV2 is best reflected by its so called serial interval, i.e., how long it takes from when an infected person shows symptoms to when another person infected by this original person show symptoms. The longer this interval, the more opportunity there is to intervene and prevent spread. For instance, Ebola's serial interval was 10-16 days on average, and SARS' was ~8 days. In contrast, estimates from cases in China prior to Feb 8 indicate that COVID-19's serial interval could be as short as ~4 days. That leaves little time to intervene. Even more surprising is the fact that ~12% of the time, the serial interval is negative; in other words, if X transmits the infection to Y, then Y might show symptoms even while X remains yet asymptomatic. In conclusion, at least 12% of infection transmissions occur from completely asymptomatic individuals. The fact that the average time from infection to symptoms is more than 4 days (5.1 days) also suggests that asymptomatic transmission does indeed occur.

Given this, when and how does one intervene? The only way to determine if an asymptomatic person is infected is by testing them for the virus. Testing involves swabbing the nose or the back of the throat, extracting RNA from the swab, and checking to see if this RNA indeed belongs to the SARSCoV-2 virus whose entire RNA sequence we have known since mid-Jan. The swabs have to be collected by medical professionals wearing protective equipment. They have to be sent to a central lab and results often take a day or two (or longer, given systems are still just falling in place). One may be deep into or even past the serial interval by then. Or, equally, one may be testing too early, and would need to repeat the test a few times to be sure. There is an additional catch as well. Even in confirmed COVID-19 patients, samples from the upper respiratory tract seem to test positive no more than 65%. of the time. Samples from the lower tract yield greater sensitivity, but are harder to access. Perhaps the lack of detection from the upper tract indicates lack of infective capability in these patients because infections are typically transmitted from the upper tract, but we don't know that for sure today. Given all these complications, is testing of asymptomatic individuals a viable strategy at all?

The only answer we have to this question comes from South Korea, which leads the way in testing asymptomatic individuals with over 300,000 tests run so far via 633 drive-through collection centres returning results in under 24hrs (they also added a quick and cheap screening test that looks for antibodies to the virus prior to running the RNA test above). Those who test positive are isolated, and their close contacts rigorously traced and quarantined. Additionally, an app that warns of the presence of a positive individual within 100 metres allows people to avoid contact. Even with all the caveats of the previous paragraph, this strategy has arrested growth from an addition of 800 cases on Feb 28 to an addition of just 150 cases on Mar 19, without any lockdowns or large scale distancing whatsoever. Could this strategy work in India?

India's population in 25 times that of South Korea. We don't know how many asymptomatic individuals we need to test at that scale before we can gain control. I am also not sure why the entire population didn't queue up for testing in South Korea. Perhaps only those with at least mild symptoms did, or maybe the quick and cheap screening test eliminated several unaffected individuals rapidly. In any case, if we were to use the same algorithm here and assume, perhaps naively, that we need 25 times as much testing as South Korea, we would need 375,000 tests a day. This is challenging but not impossible, and would cost billions of dollars—a large number, but perhaps far smaller than the alternatives that follow, and far less painful. The efficacy of this approach needs to be better understood though. Until that happens, the government has taken a more conservative approach.

This more conservative approach is to test only those with a recent history of international travel and with symptoms severe enough to seek medical help . This is followed by isolation, contact tracing and quarantines. Can this stop the spread of the virus? This simulation gives us an indication. It starts with a certain number of infected individuals, say 184 (which is the number of documented positive cases known to have arrived from abroad thus far). Each is simulated to develop symptoms after a certain period of time reflective of the incubation period (~5.1 days on average). Then, after a further delay that accounts for the time to seek professional advice and implement isolation (~3.5 days on average), each is considered no longer capable of infection. However, prior to isolation, each individual is considered to have infected ~2.5 others on average among the several contacts they could have interacted with. 15% of these transmissions are considered to have occurred prior to the development of symptoms, as we saw earlier. Some of these contacts are then considered to be traceable and isolated immediately upon the development of symptoms, if any, while the rest are considered untraceable. The fraction of traceable contacts is key. Simulations show that more than 80% of the contacts have to be traceable to have even a 50% chance of controlling the outbreak. That seems like a long shot.

What makes the shot even longer is this claim from a clever attempt at modelling the spread of the virus in China that most cases went undocumented. This model begins by estimating the number of cases in the epicentre Wuhan on Jan 10 using information obtained by sequencing the virus genome. My earlier post described how a comparison of the various virus genomes sequenced from multiple patients suggests a focal origin for the virus in Nov. The first humans were infected around then but no one realised this until much later. Given the doubling rate of the virus in the absence of any distancing measures is ~6.4 days, the estimated number of cases in Wuhan by Jan 10 would have been ~1000. Between Jan 10 and Jan 23, the Spring festival period, roughly half a billion travel events were recorded in China (data here). These events would have spread the virus to other cities. Simulating these events starting with the seed cases in Wuhan on Jan 10 suggests that the only way the virus could have spread to so many different cities is if the number of undocumented cases from Jan 10 to Jan 23 was ~6x the documented number, with much of the spread attributable to these undocumented cases. A large proportion of these undocumented cases probably had milder symptoms. And that confounds the government's approach of starting with just the severe cases alone.

Of the ~80,000 passengers who arrived in India every day in Feb and part of Mar, how many undocumented cases might there be? Going by the previous paragraph, perhaps ~6x the 184 confirmed thus far, or ~1100. Further transmissions from these over the last month would have reached several thousands. A lot of energy has gone into the mammoth task of contact tracing for the documented 184 cases, but that might be akin to fighting in one corner of the battlefield while the battle rages all over.

Cognisant of the above possibility, the Indian government has made an attempt to determine if there are indeed undocumented cases beyond those it has identified solely by testing severe cases with a link to foreign travel or to another confirmed positive case. They did so by testing 826 samples randomly drawn of people suffering from severe acute respiratory infection/influenza like illnesses, collected between Feb 15 and Mar 15. None of these samples was found positive. What can we conclude from this?

The confounder here are severe pneumonia cases on account of bacterial infections. The number of such cases in India is ~4m annually, or about 330,000 cases each month. In this ocean, a simple probability calculation shows that the 826 randomly chosen samples will miss picking any severe COVID-19 cases with a 50% chance if the number of COVID-19 cases in the community is ~7500 or lower (this calculation assumes that 3.7% of COVID-19 cases are severe, as per demographics in India). 7500 is still ~18x the number of documented cases to date. Therefore, it is wrong to conclude from this random sample that there has been no community spread. Larger sample sizes are needed, which means more testing capacity.

Realising this, the government took a significant step on 21st Mar to open up testing to the private sector and simultaneously expanded testing criteria to include ALL hospitalised patients with severe acute respiratory illness (fever AND cough and/or shortness of breath). This is a significant relaxation of the previous criteria, which required an association with a positive case or with recent international travel. Broader testing of severe hospitalised cases will be a key tool for identifying hotspots and tracing and isolating contacts in the next few months, or even much longer.

Given all the above, the inevitable conclusion is that we are confronted with a nimble and nasty virus that has probably already infected many more than the documented figure of ~400 (several thousands would be my guess). Contact tracing and isolation of severe symptomatic individuals with a connection to international travel or to another positive individual is highly insufficient, and perhaps a losing battle from the outset in the absence of other measures. We are not yet prepared to test in large numbers as South Korea did, and we don't fully understand how effective that might be anyway. Meanwhile, the several thousands of undocumented cases might march through the crowds in India to a hundred thousand in April and a million or two in May.

Sensing that the battle might slip away, the government righty announced an unprecedented countrywide lockdown today for three weeks. Perhaps, rigorously quarantining the ~80,000 passengers who landed everyday in India from Feb onwards would have been a less expensive proposition. This would be ~1.12m individuals at any given time, most wealthy enough to pay for themselves. But that window has passed. Now 1.3 billion people will be locked down for three weeks, and a large fraction might struggle for their livelihood. How long should this lockdown continue for it to be effective? And what happens when it is lifted?

A complete lockdown that isolates societies into small enough households for a few weeks can in theory eliminate the virus. In practice, compliance can be challenge, as also the fact that households in India need not be insulated from neighbouring households. A recent much publicised simulation looks at the effect of partial lockdowns in Great Britain and in the US on virus spread. By a partial lockdown, they mean social distancing of the entire population so contacts outside the household reduce by 75% (perhaps a reasonable reflection of an Indian lockdown), closure of schools and colleges, isolation of symptomatic cases, and quarantining of family members. It suggests that the number of cases peaks three weeks from lockdown and then gradually reduces. That is the good news. The bad news is that the number of cases starts rising again when the lockdown is lifted. Given the drastic impact that lockdowns have on livelihoods, the simulation suggests an alternative where social distancing of the entire population and closure of schools and colleges is applied intermittently, so people have a chance to catch up with their lives before retreating again. These intermittent lockdowns are triggered when the count of severe cases goes beyond a certain threshold. And this cycle needs to go on for quite a while. Will this be the fate of our lockdown as well?

There is little doubt that the three-week lockdown will stall the spread of the virus and perhaps even reduce the number of cases substantially. But will it eliminate the virus altogether? The above simulation suggests that it may not, unless compliance levels are super high. China's experience perhaps suggests otherwise, with no new suspected or confirmed cases in Wuhan or Hubei for the last four days.

In any case, we have bought three weeks to get our act in place. By the time the lockdown is relaxed, testing will have to be ramped up dramatically so we have a chance of knowing whether or not the lockdown succeeded. What is bothersome though is the thought that if Wuhan could explode from just a few cases that remained un-isolated for two months, the same could happen again in India. This is not an unlikely proposition at all unless the virus weakens with time or with the summer, but both appear to be wishful thoughts at the moment. We probably have to prepare for the outbreak to resume, or for it to subside temporarily and return.

And this leads to several questions. We cannot do full lockdowns repeatedly. Would partial lockdowns work in an Indian setting, where people live and commute in high density environments with their own peculiarities. What would be the right algorithm? How would we know what is right if we cannot estimate the undocumented cases? Would broader scale testing along the lines of the South Korean model be more efficient than the cost of lockdowns? At the very least, would it provide a better estimate of the undocumented case load? We don't have the answers yet, but we should expect answers to come out in the next few weeks.

In any case, there is a good chance that this might be a bit of a long haul. Is there light at the end of this tunnel? Previous pandemics might shed some light on this question. Almost exactly a century ago, the Spanish Flu caused by the H1N1 virus had a devastating effect, infecting ~1/3 of the then population and killing 17-50m worldwide in three waves from Jun 1918 to May 1919. The waves subsided thereafter.

Those who got the infection but survived acquired long lasting immunity (even without a.vaccine; none existed then). With a substantial fraction of the surviving population now immunized and no longer capable of transmitting infections, even those not immunized received indirect protection. In other words, herd immunity set in.

The challenge with herd immunity is that a majority (say 60%) of the population must get immunized for it to be effective. A vaccine is still 18 months away at the earliest and may or may not be fully effective. This means that the only source of immunity might be for, say, 60% of the population to get the infection and for most to recover from it. That will come at a unacceptably steep cost: an estimated 29m hospitalisations, 5.5m ICU admissions, and 3.1m deaths, as per the demographic impact proportions here. We need to find another way.

Our task then, with the greatest urgency, is to use the set of tools that we have—modeling and simulation to predict the spread of the virus, rapid and transparent data gathering to feed these models, testing to establish whether or not our predictions are working and to identify whom to isolate, distancing measures to keep repeated outbreaks under check, and various technology enablers that enable distancing, increase compliance, and help monitor use of key resources like ventilators—to slow the growth of the virus without freezing livelihoods and stalling the economy altogether.

If we achieve this, we will buy enough time for vaccines and better drugs to be developed, and get back to business as usual. We would have to be (and can be) nimbler than the virus though.