The sting in the tail

How did a Covid-19 case slip through New Zealand’s quarantine system?

Those with a taste for language may have given a thought to the irony of ‘quarantine’ meaning ‘fourteen days’ in our current pandemic. The word is derived from the Italian – and before that the Latin – word for forty, in reference to a Venetian policy of making ships wait forty days before they could dock, to ensure that there were no plague cases aboard. This was a very cautious quarantine period, given that the incubation period for plague is usually under seven days. However, it’s hard to blame them for their caution, given the devastation that the plague had caused in Europe and across Asia and their poor understanding of how it was spread. Quarantine, from the very start, was always a balance between science and pragmatism.

In fact, the period of forty days had its origins much earlier, in ancient Greece. A part of their definition of plague was that it manifested itself within forty days. Forty was also a number which was considered to have mystical properties. At least in part, it was probably for these properties that the number was chosen.

The fourteen day quarantine, though, does have an ancient precedent. In the book of Leviticus, isolation for fourteen days, with an examination by a priest at seven and fourteen days, was used to distinguish ‘leprous diseases’ from other skin disorders. (Incidentally these ‘leprous diseases’ were not leprosy. It’s unclear whether they were even infectious, since isolation for various periods of time was required for all sorts of uncleanliness, many of them not obviously associated with infectious disease.)

A fourteen day quarantine period for Covid-19 was in use before the disease had been known for a month. In late January, when one of my colleagues had returned to New Zealand from a trip to China, she was asked to stay home and self-isolate for two weeks. At this point, it wasn’t an official requirement for all travellers, but it was required by our employer. New Zealand’s Ministry of Health did not have a recommended quarantine period at this time, however they did note that anyone becoming unwell within a month of arriving from overseas should seek medical advice. This time period pre-dates Covid-19, and is part of the general advice to everyone arriving in New Zealand.

At this time, the World Health Organisation (WHO) was using at least two different time periods in its official advice. It was suggesting exposed health workers keep a record of any potential symptoms for at least 21 days. A few days previously it had suggested family members of an infected person should monitor their health for fourteen days.

So, by late January, a fourteen day quarantine period had already started appearing in the official advice, even though, at that stage, it wasn’t the only suggested time period. But it was far too soon for this advice to be based on direct knowledge of Covid-19.

The virus which is causing us so much trouble right now is officially known as SARS-CoV-2, and in that name there is a clue to why the fourteen day quarantine period was in use so quickly – because there is a SARS-CoV-1, more usually just known as SARS-CoV, or just SARS. SARS emerged in China in late 2002, and spread to almost 30 countries, infecting more than 8000 people and killing nearly 800. Covid-19 has another cousin as well, the virus known as MERS-CoV. This virus is best known for causing the outbreak in South Korea in 2015 which infected 186 people and killed 32. However MERS has its origin as a disease of camels and all known human cases can be traced to the Arabian Peninsula.

Both SARS and MERS have incubation periods which are typically under a week, but can be up to fourteen days. This incubation period is not standard for all viruses – influenza, for example, has an incubation period of 1-4 days, while rabies has an incubation period which can range from one week to one year. So, the reason that a two week quarantine period appeared so quickly for Covid-19 is almost certainly based on the incubation periods of its closest relatives.

But, by now, we have vastly more experience of Covid-19 than we do of either SARS or MERS, and that begs the question  whether we are still justified in using a quarantine period based on diseases which have, in total, infected fewer than 11,000 people.

The answer seems to be both yes and no.

Questions about whether Covid-19 had the same incubation periods as SARS and MERS began to arise in February. But it’s difficult to do a careful study in the middle of a crisis. Early reports used small numbers of people; a study with 10 people found an average incubation time of 5.2 days, a study with 88 found 6.4 days and a study with 50 found 4.9 days. While there were initial indications that the incubation time for Covid-19 might be longer than SARS or MERS, one of these studies re-examined the original data on incubation times for MERS, and found something different. In that data were a few cases with incubation times greater than 14 days – up to 21 days. When the longer incubation times for MERS were included in the comparisons, the average incubation time for Covid-19 didn’t significantly differ from MERS or SARS.

But numbers in biology are tricky things. Subtle differences in the order of specific molecules making up an individual’s DNA can result in extraordinary variation. Even viruses, which aren’t quite living things even though they are built from the same DNA (or its cousin RNA), have this variation in the order of molecules. The variation means that whenever we measure something about a living creature (or a virus), we find a slightly different number. One human is 160 cm tall, another is 161, another 171. One person has blonde hair, another black hair, another has hair just a shade lighter which we might call black or might call brown, depending on the light. One person gets a sore throat five days after coming into contact with Covid-19, another loses their sense of smell after six days, then later develops a cough and fever.

This variation makes it difficult to accurately measure anything in nature, and it also means how we express the measured numbers is important. For example, the study on Covid-19 incubation times which looked at 50 people found an average incubation time of 4.9 days. More precisely, this average was the mean. The mean is the number you get when we add up all the incubation times and then divide by the number of cases, in this case, 50. But there are other numbers which are important when we look at incubation times. After all, it’s not the average case we want to catch with our quarantine period. We want to catch all of the cases if we want quarantine to work as well as possible. And to do this we need numbers which describe variation.

Averages are no help in understanding variation. If we take the numbers 4 and 6, or 1 and 9, then calculate the average, we get the same result. But if those numbers are incubation times and we are using them to determine a quarantine period, we would use different times depending on which set of numbers we looked at. If we take the first set, we might use 6 days, and for the second set, we might use 9. That’s assuming, of course, that we didn’t add any sort of safety margin, which we probably should if the size of our sample is just 2.

The simplest way to describe variation is to give the range – with the example above the ranges would be 4-6 and 1-9. Using the range is also the best way to determine an effective quarantine period, because we need to know the maximum time that Covid-19 can remain latent before causing illness. But we will never know the incubation time of every case of Covid-19. And even if we did, that wouldn’t mean that we could predict future incubation times with complete certainty. We also need to balance up other factors before we make a decision. We need to listen to the science, but we do need also to consider the disruption to travel, the cost of keeping people in quarantine versus the cost of an outbreak, and confusion or non-compliance if we differ from international norms. Just as Venice did when it imposed a 40 day quarantine for the plague, we need to make a decision that considers both science and pragmatism.

My first hint that New Zealand’s quarantine period might not be completely effective came when Belgium announced that they were moving from a 14 day self-isolation period to a 7 day self-isolation period for people exposed to Covid-19. In an interview with the BBC, Belgian virologist Professor Steven van Gucht explained that they had measured the effectiveness of a 7 day quarantine with testing at day 7, and found that it was ‘nearly’ as good as a 14 day quarantine. The precise figure he gave for the 7 day quarantine with testing was 88% effective. That number didn’t sound too good to me – it means that twelve cases out of every hundred would get missed. But if the goal is to keep the number of cases low enough so that the health system can cope, rather than to eliminate the virus, perhaps that is enough. And, as van Gucht explained, if it meant that more people complied with the less restrictive rule, it may be just as good.

At that point the interviewer asked a question – “how does that compare with the 14 day quarantine? – and his answer hit me with a physical jolt.

“Well, with 14 days you get 96% of cases, even then there are a few cases with longer incubation periods.”

When I heard those words, I was walking in the park with my dog, enjoying the fresh morning air and listening to the BBC World Service playing through my phone. There was nothing I could do to replay the interview or check the veracity of his statement at that moment. But, later, I searched for a copy of the interview online and checked that I hadn’t misheard. Then I began to look much more closely at Covid-19 incubation periods. Was it really possible that New Zealand’s quarantine system could be missing four Covid-19 cases out of every hundred infected people who arrived here?

The answer, it turned out, was yes and no.

One of the papers I found, with a sample size just over 400, reported a mean of 5.2 days and a 95th percentile of 12.5 days. The mean is consistent with the early papers I mentioned above, but it’s the 95th percentile that’s of interest here. What it means is that 95% of incubation times were less than or equal to 12.5 days. But it also means that 5% of incubation times were more than 12.5 days, which begs the very important question – how much more?

The answer was not reassuring. The authors didn’t give the raw data, but they did give a graph showing the incubation times. The high point of the graph, representing the most commonly reported incubation time, was around the 3-4 day mark. After that, the graph dropped away rapidly, and by 14 days it was nearly flat, but not quite. The line of the graph didn’t reach zero until it hit 21 days.

Figure 1. Incubation period of Covid-19, graph from Qun Li et al (2020) in N Engl J Med; 382:1199-1207

Another paper, this time looking at nearly 2000 people and published in mid-March, was even more alarming. This study, unlike many of the earlier studies, didn’t just include people who were in hospital with severe symptoms. It also included people who suffered a much milder illness, including 99 children. It found that the mean incubation period was longer than that reported in other sources – 7 days – and that for children it was even longer – 9 days. But their truly startling finding was that nearly 12% of cases had incubation times longer than 14 days. That is, if 100 people infected with Covid-19 were in quarantine, 12 could walk out at the end of their two weeks and still become ill later – as much as 33 days later, in fact.

This paper, however, seems to be something of an anomaly. It is the only paper I found to indicate such a long incubation period, and it was a ‘pre-print’ which means a draft which hadn’t yet been peer reviewed. Right now, many researchers are releasing unreviewed manuscripts on various websites – in an emergency, the faster information can be shared among scientists, the better. But it does mean some scepticism is needed. So far, I haven’t been able to find a later, reviewed, copy of the paper. It is quoted by other publications, but it doesn’t appear in a later systematic review on incubation times – a paper which has collated results from a number of different studies, evaluated their quality and then provided further analysis of the conclusions.

So how much credence should we give the study which found that nearly 12% of cases had an incubation time greater than 14 days, when the World Health Organisation is still recommending 14 days, and no other publication has found a similar result? That’s a difficult question. The answer that I give may differ from yours – whatever the evidence says, we always apply our own concerns and values to how we interpret that evidence. For example, airlines and people in other industries dependent on travel and tourism may say that we shouldn’t take the results of an unreviewed paper too seriously. People who manage nursing homes or who themselves fall into a high-risk category may urge more caution.

Another study, published in July and based on cases from China outside the state of Hubei, did provide more confidence in the 14 day quarantine. Out of nearly 3000 cases for which incubation times could be determined, more than 95% had incubation times of less than 13 days. However, once again, there were cases with longer incubation times – the longest was 24 days.

Again, though, there are reasons to wonder about the results, because there was something rather odd in the data for this paper. Over time, from the cases reported in January through to the ones reported in February, there was a gradual increase in the reported incubation times. This was particularly marked in people who had a history of travel, and were thought to have acquired the virus in a different city from where they were diagnosed. There’s no obvious biological reason that this could result from a change in the virus, but there is one reasonably likely explanation. If the disease was more prevalent than being officially reported in some cities, and if the numbers of undiagnosed cases were increasing, people may have become infected later than was assumed based on contact tracing. The authors did acknowledge that incubation periods could be overestimated if people had become infected from an unknown source later than their last reported contact with the disease. They didn’t suggest it as an explanation for increasing incubation times, but, to my mind, that is a plausible explanation.

The papers I’ve quoted are only a small subset of the papers considering incubation times, but there is, in general, a pretty clear pattern. The 96% effectiveness of quarantine that Professor van Gucht quoted in his BBC interview doesn’t look wrong.

So, where does this leave our 14 day quarantine? Will it pick up all the cases or not? Is it good enough?

The honest answer here is that we don’t know. Most of the time, it will work. Sometimes, given the huge amount of variability that this very new disease has shown, it probably won’t. Any graph showing incubation times for Covid-19 has what statisticians call a ‘long tail’. That is, even though a very high proportion of cases will have shown up within a week, a small number of cases may not show symptoms until quite some time later. Given that multiple studies have found incubation times of more than 14 days and given the huge number of cases globally, it’s almost certain that there will be cases that slip through quarantine.

But we do have another layer of protection – we are testing people, whether symptomatic or not, at both day 3 and day 12. And this should give a lot more confidence. But there’s a sting in the tail of testing data as well, and if we really want to figure out if our quarantine system is good enough, we have to understand that too.

But that’s a question for part two of this article, which I will publish next week.

The sting in the tail: part two