Revisiting early-stage COVID-19 strategy options

Introduction

When I first submitted this paper in May 2020, I wrote with some urgency because I believed the issue of asymptomatic transmission of COVID-19 was not receiving the attention it deserved, and updated the paper before any reviews were in. All the versions of the paper, read in conjunction of the reviews, form a living review, showing progress in my understanding of the issue and progress in research into asymptomatic transmission.

As the COVID-19 pandemic has spread, its outcomes have differed by locality. In some, it has been contained quickly. In others, the rapid growth has slowed but not stopped. In many, the rapid growth has driven health systems to the point of collapse. New York state is the epicenter of the United States epidemic. What adds urgency to the search for alternative containment strategies is the fact that mortality rate (deaths as a fraction of population) in New York state on 22 April 2020 surpassed 1000 per million. If scaled to the entire country, this would be over 300,000 deaths¹. At time of writing the second version of this paper, US deaths were at almost 180,000 and that count exceeded 500,000 by the third version, illustrating that there is still work to be done at containment.

Since the pandemic is still playing out it is useful to reflect on the positive and negative outcomes and to try to map a way ahead for localities where it has not gone past the stage where it can easily be contained. Given that no known remedy exists for the disease, and that it is spreading too fast to rely on a vaccine to avoid major health or economic problems, non-pharmaceutical interventions (NPI) are the most critical thing to get right. Now that the pandemic has progressed beyond its initial stages, understanding asymptomatic transmission remains important as the question of universal mask wearing remains controversial¹ with inadequate testing of informal masks nonetheless at least justifying their use in terms of the precautionary principle². One study shows that viral shedding starts 2 days or more before symptoms show and that infectiousness peaks 1–2 days prior to symptoms and presymptomatic transmission is 37% to 48% and that this figure can be as high as 62% without adequate case finding. There are no definitive studies of asymptomatic transmission, despite evidence that it is real³.

While understanding asymptomatic transmission alone does not fill the gap in investigating the efficacy of informal masks, it strengthens the case for applying the precautionary principle pending such a study. Studying informal masks is inherently difficult as there is so much variability, which is why I propose an initial focus on asymptomatic infection. Asymptomatic infection can also assist with understanding the highly variable progression of the disease. At a later stage of the pandemic, understanding transmission from milder cases could also be of value in understanding whether vaccines prevent further transmission.

A number of NPIs have been tried from social distancing to complete lockdowns. The consequence of acting too slowly is the risk of crashing a health system, which has hit some of the best in the world (Italy for example has historically been near the top of the World Health Organization’s ranking⁴; and the epicentre of the COVID-19 outbreak is in the north⁵, which has Italy’s strongest health resources⁶. Though the pandemic at time of writing is still developing, it is worth reviewing options for countries able to avoid runaway exponential growth.

By way of example, I looked at options for South Africa, which embarked on a 21-day lock down⁷ that started at about the time when 1,000 cases were reported (midnight, 26 March 2020; subsequently extended by another two weeks to 30 April), in the first version of this paper. At this much later stage of progress of the pandemic, countries where cases are declining have the same opportunity to use spare testing capacity for a study I propose in this paper.

I examine case studies in other localities encompassing the variability in outcomes and assess likely contributing factors to this variance. Given the shortness of time to make decisions, I do not attempt to develop a rigorous model but rely on extracting meaning from these case studies.

In my opinion, ignoring asymptomatic spread is a major error; a relatively simple experiment in a country like South Africa at the early stage of spread (or now at a later stage as outlined above) could validate this opinion. If proved correct, many lives could be saved. If proved incorrect, the cost is low relative to the benefit. I therefore urge that the experiment be carried out as a matter of urgency.

It is natural at an early stage of a rapidly-expanding pandemic to focus on the most serious cases as these are the ones where interventions make the biggest difference. Now that there is more time to assess evidence, there is also a case to do detailed studies of less serious cases to understand better what predicts progression to the worst effects. Identifying asymptomatic cases in particular could aid with this as they represent the extreme of the mild form of the disease.

In the remainder of the paper, I describe significant unknowns, work through cases studies and examine other factors, leading to the conclusion that a study of asymptomatic spread was the most urgent gap in early knowledge that should have been addressed to assess options for containment containment – and could still be of use at later stages of the pandemic. I propose a strategy for identifying and investigating asymptomatic cases.

Significant unknowns

Because testing started under pressure, standards are not consistent⁸. That means statistics like case fatality rate are problematic to compare across localities.

Another big unknown is the true number of infections since many that were not serious enough to require hospitalization may have resolved without being counted where testing was inadequate; if the asymptomatic fraction is as high as claimed in some instances, that also skews the case fatality rate high.

All of these factors also result in difficulty in establishing an accurate value for the basic reproduction rate R₀, the initial mean number of infections per infected case. The value of R₀ matters for computing the herd immunity level, widely reported in the mass media as 60% of the population, the number used to justify the initial British response of allowing it to run through the population⁹. As infections spread, the effective reproduction number at time t, R_t, will decline below 1, the herd immunity threshold¹⁰. NPIs artificially force R_t to drop, faking the effect of a less contagious disease. However, if NPIs are relaxed before herd immunity is reached, another round of rapid increase can ensue as R_t again rises above 1. It is for this reason that estimation of R_t is useful¹¹.

Herd immunity occurs when the fraction of the population immunised (either by vaccination or by acquiring immunity post-infection) exceeds the threshold P_herd in Equation 1,

Figure 1 illustrates how P_herd varies with R₀ (assuming E = 1; in the absence of a vaccine, this means that any recovered cases cannot be reinfected). For seasonal influenza, if R₀ = 1.3, the herd immunity threshold is 23%. For R₀ = 2.5, the herd immunity threshold is 60%. However, if the true COVID-19 R₀ value is significantly higher, so is the the herd immunity threshold. For example, if R₀ = 4, P_herd is 75%.

Figure 1. Herd immunity varying with basic reproduction rate R₀.

If R₀ for COVID-19 is 2.5, herd immunity occurs after 60% of the population is infected (green arrow). The mean R₀ value for seasonal influenza results in herd immunity at about 23% of the population (red arrow).

Even for influenza, R₀ can vary widely depending on the strain. For the H1N1 strain, R₀ was estimated at 1.4–1.6; for the 1918 flu, the estimated R₀ range is 1.4–2.8 and even seasonal flu has a wide R₀ range of 0.9–2.1¹³.

One study reports COVID-19 R₀ values varying from 1.4 to 3.8¹⁴. Another narrows the range to 2.24 to 3.58¹⁵.

A model with R₀ = 2.68 yields a doubling time of 6.4 days¹⁶. (which would hold good until the fraction susceptible dropped enough to reduce R_t, unless NPIs artificially reduced R_t – it is this phase of expontial growth that puts health systems under pressure). The doubling time was far shorter than this during peak growth in places where it was not under control. In the United States, for example, doubling time was less than 3 days 1–24 March 2020 (see Figure 2, based on the rule-of-70¹⁷). That rapidity of growth suggests that R₀ is on the high rather than low side of published estimates, but the low level of testing in the USA at early stages of the pandemic make it difficult to derive robust measures.

Figure 2. Doubling time in the US: 1–31 March 2020.

For much of the month, doubling time is 2–3 days, implying that R_t is not varying significantly.

With so much uncertainty, relying on herd immunity is folly, as was discovered in the UK⁹.

Broadly speaking, countries with a tradition of personal liberty and rejecting authority have found it hard to adapt to NPIs like social distancing¹⁸. Informal settlements and other high-density dwellings of the poor also make social distancing hard¹⁹. For this reason, it is important to explore all alternatives including those that were missed in early stages so that other countries that are being hit later can learn the right lessons.

Methods

I use Python 3.7.4 within the Jupyter Notebook environment version 6.0.3 to do anyalysis. All the code used is archived on GitHub²⁰ and is labelled as version 1.1.1.

Data is from the Worldometers web site and papers cited; data used in analysis is embedded in the GitHub archive²⁰. A snapshot of the Worldometers site containing the data used is archived at Webarchive (updated in Webarchive for Version 2 of the paper).

Herd immunity is calculated based on the standard formula assuming exponential growth using Equation 2.

I calculate doubling time for March 2020 using the rule-of-70, which assumes exponential growth and is accurate under that assumption:

Case studies

I consider three cases: instances where the disease has run its course, examples where containment has slowed the growth to the point of being manageable and finally examples where growth has severely stressed health systems.

These examples are not meant to be exhaustive but provide archetypes of different types of outcome.

Other factors

An important factor to consider is comorbidities. South Africa has high rates of tuberculosis and HIV infection, both of which are significant risk factors for any pulmonary disease³². In one study in Italy, out of 355 patients who died, only 3 (0.8%) had no prior condition. 25.1% had one condition, 25.6% had two and 48.5% had three or more²⁷.

Balanced against this is early statistical evidence that differences in national coverage of the Bacillus Calmette-Guerin (BCG) tuberculosis vaccine explain differences in case fatality rates³³. This study is not peer reviewed and therefore should not be relied on too strongly. A pure statistical study without direct causal evidence points to the need to establish causality rather than signifying causality.

Weighed against relying on BCG coverage before further evidence emerged is that Iceland, which does not have mandatory BCG vaccination³⁴, was very successful in slowing the spread by an aggressive testing programme, including quarantining everyone who had contacted a known case³⁵.

More recent evidence since early versions of this paper justifies caution: there is at best mild evidence of a protective effort and no evidence that BCG slows the spread or reduces serious cases³⁶.

A broader lesson arises out of the early claims about BCG: a statistical study without causality could turn out to be coincidence. Another example is a study of invermectin in Peru that appears to show a close link between ivermectin use and reduced mortality³⁷. Yet one of the first reasonably rigorous peer-reviewed studies of ivermectin shows it has no significant therapeutic effect³⁸.

Finally there is indirect evidence of asymptomatic spread in the apparent efficacy of masks in slowing spread. While early WHO advice was against the asymptomatic mask wearing, this was part of an advisory that aimed to prevent a run on medical-grade masks³⁹. More recently there is support for cloth masks being worn by the public² and experimental evidence that surgical masks block aerosol transmission of coronaviruses and influenza viruses⁴⁰. While some still doubt the evidence, the fact that mandatory mask wearing was a factor in reducing the spread of the 1918 influenza pandemic⁴¹ supports the case for encouraging mask-wearing as a protection for the COVID-19 pandemic provided that this does not deplete supplies of medical-grade masks and there is a programme of public education on use of masks and their role in the context of other measures like distancing and hygiene.

Given the high risk associated with comorbidities, it is premature to place too much reliance on any mitigating factors other than slowing transmission. NPIs are the main game until pharmaceutical options – including vaccines – become viable on a large scale.

Proposed research

Since earlier versions of this paper, a systematic meta-analysis and review of viral load dynamics, shedding and infectiousness has shed some light on asymptomatic transmission, indicating that asymtomatic infection is contagious but not for as long as symptomatic infection⁴². One study estimates the relative infectiousness of asymptomatically infected cases to be 0.27 (if on a small number of cases).

Since one of the biggest unknowns was the prevalence and contagiousness of asymptomatic infection, I proposed a project in earlier versions of this paper to identify such cases early and identify informative features. Features of interest include:

To identify these cases in time to measure viral shedding from the start of infection, comprehensive contact tracing of a representative sample of the infected population is necessary. Any who test positive out of this cohort can be followed up by a further round of contact tracing; if this finds more positive cases that have been in contact with only asymptomatic cases, that would provide a measure of asymptomatic contagion. If however the programme prevents secondary infections, viral shedding will still be a useful indicator of contagiousness. In one study, rapid identification and isolation of asymptomatic cases prevented secondary infections¹¹ but this should not be taken as indicative that asymptomatic cases are not contagious.

This is a project that could be carried out with a modest burden on testing capacity – as long as the pandemic is not expanding at full pace.

Since the earlier versions of this paper, there have been some advances in understanding asymptomatic infection, including studies that show asymptomatic cases develop antibodies, if at a lower level than symptomatic cases^44,45. However, given that we are no longer at the early stage of the pandemic, another related concern needs to be added to the mix: how contagious a vaccinated person could be if they are infected – despite not developing symptoms. I conjecture that this could follow the same pattern as asymptomatic infection without a vaccine: a lower rate or shorter duration of viral shedding, or both. An early study of the Pfizer–BioNTech vaccine (BNT162b2) supports this conjecture but more systematic studies over a range of vaccines will be useful to characterize the extent to which transmission is suppressed, as opposed to the primary endpoints of vaccine trials, reducing serious illness and death⁴⁶. There is considerable variability in primary endpoints in vaccine trials though the public health interest focuses on severe disease so vaccine trials mostly focus on symptomatic infection with various definitions of “severe”⁴⁷.

The last question I raise on T cell variability has been addressed – even asymptomatic cases who are anti-body seronegative have a robust SARS-CoV-2-specific T Cell response⁴⁸. This bodes well for vaccine efficacy.

Conclusions

The most significant finding out of this review is that we do not know enough about asymptomatic transmission.

This is a problem easily remedied in early-stage spread – or at a later stage when cases are reducing – and there is spare test capacity.

The proposed project would give a clearer picture of the potential for asymptomatic spread and add to the evidence for universal mask wearing.

If the experiment to measure asymptomatic spread shows that it is a significant factor, that signifies a change in testing strategy. Should any asymptomatic positives be discovered, more aggressive action should follow: they and all their contacts should be quarantined and released when they no longer test positive, taking into account time for incubation.

South Africa, at the date of the first version of the paper, had less than 3500 cases; at time of writing the second version, the daily number of cases had declined to 25% of the peak. Going back to test all contacts was doable at the outset; with test demand 25% lower than at peak, testing all contacts of a representative sample is practical. While it is possible that despite a significant asymptomatic fraction, asymptomatic infection is not contagious, the cost of finding this out is very low compared with the cost of not containing the spread. Iceland does not do widespread BCG vaccination³⁴ and yet was successful in containing the spread. The Iceland experience indicates that there is no need to rely on questionable evidence such as that the BCG vaccine provides protection: act early and catch all cases including the asymptomatic and the spread can be curtailed³⁵. However, if we know exactly how contagious aymptomatic cases are, that will further inform NPI strategy.

The cost of taking the step I advocate is far lower than the cost of allowing asymptomatic spread to get out of control. If it leads to an effective containment strategy, it will also remove the need for an economically damaging extended lockdown. Even at a later stage of the pandemic, informing policy choices like mask wearing is useful, as is better understanding of the drivers in variabililty of the disease. And economically damaging NPIs can be better avoided if we understand the disease better – how it spreads, what makes people more vulnerable.

In the time since the first draft of this paper, studies have pointed to a lower fraction of asymptomatic transmission than earlier reports but the fraction is still high enough to be a significant factor in transmission. However, I am still of the view that studying asymptomatic transmission is important as a clue to what differentiates milder cases, and could also shed light on the extent to which vaccines stop transission, as opposed to restricting infections to casuing a milder illness.

As with the early stage of the pandemic, milder to asymptomatic cases are attracting less interest in vaccine trials because they do not impose a direct public health burden. However, I argue that they impose an indirect public health burden as long as they may be contagious.

Data availability