PreprintPDF Available

Outlines of a probabilistic evaluation of possible SARS-CoV-2 origins

Authors:
Preprints and early-stage research may not have been peer reviewed yet.

Abstract and Figures

A probabilistic treatment can be very useful when trying to discover the most probable causes that are consistent with the available information at the time. In particular in such a treatment all assumptions and all probability estimates are explicit and are open for investigation. Here we explore the relative probabilities of a lab-related accident against a non-lab-related zoonotic event being at the root of the current COVID-19 pandemic. In doing so we use estimates of the relevant probabilities published in the specialized literature, especially estimates of the risk of a lab-acquired infection (LAI) and of the subsequent community outbreak risk. We show that, based on present knowledge, the relative probability of a lab-related accident against a non-lab related zoonotic event is not negligible across a wide range of defensible input probabilities. For instance, under a reference set of input probabilities, the relative probabilities are at least 55% for a lab-related event against 45% at most for a non-lab-related zoonotic event. Even under a particularly conservative set of assumptions the relative probability of the lab-related accident is still 6% (to 94% for the non-lab related zoonotic event). Through a review of the Chinese specialized literature, we further show that our underlying estimate for the probability of lab-acquired infection is consistent with risk assessments from Chinese authorities and specialists. We then review a list of common probabilistic misunderstandings that are often associated with discussions about COVID-19 origins and conclude by discussing how such a probabilistic treatment can also offer a way to properly guide an investigation into the causes of the pandemic while being able to embrace different estimates of the underlying probabilities.
Content may be subject to copyright.
page 1 of 44
Outlines of a probabilistic evaluation of possible
SARS-CoV-2 origins
Gilles Demaneuf1, Rodolphe de Maistre2
1. Data Science - BNZ (Ingénieur ECP, Applied Maths)
2. Ex Auditor IHEDN (MBA INSEAD, Chinese studies at University of Paris & Sichuan University)
Correspondence: Gilles Demaneuf, contact@demaneuf.com
ABSTRACT
A probabilistic treatment can be very useful when trying to discover the most probable causes that
are consistent with the available information at the time. In particular in such a treatment all
assumptions and all probability estimates are explicit and are open for investigation. Here we
explore the relative probabilities of a lab-related accident against a non-lab-related zoonotic event
being at the root of the current COVID-19 pandemic. In doing so we use estimates of the relevant
probabilities published in the specialized literature, especially estimates of the risk of a lab-acquired
infection (LAI) and of the subsequent community outbreak risk.
We show that, based on present knowledge, the relative probability of a lab-related accident against
a non-lab related zoonotic event is not negligible across a wide range of defensible input
probabilities. For instance, under a reference set of input probabilities, the relative probabilities are
at least 55% for a lab-related event against 45% at most for a non-lab-related zoonotic event. Even
under a particularly conservative set of assumptions the relative probability of the lab-related
accident is still 6% (to 94% for the non-lab related zoonotic event).
Through a review of the Chinese specialized literature, we further show that our underlying estimate
for the probability of lab-acquired infection is consistent with risk assessments from Chinese
authorities and specialists. We then review a list of common probabilistic misunderstandings that
are often associated with discussions about COVID-19 origins and conclude by discussing how
such a probabilistic treatment can also offer away to properly guide an investigation into the causes
of the pandemic while being able to embrace different estimates of the underlying probabilities.
KEYWORDS
accident analysis, risk assessment, risk disclosure, laboratory escape, laboratory acquired infection,
bayesian statistics, biosafety and biosecurity issues, COVID-19, SARS-CoV-2
INTRODUCTION
Despite considerable efforts, the exact origin of the current COVID-19 pandemic has to this date not
been asserted. An initial theory of a zoonotic event at awildlife market [1,2]has been found
wanting and is now considered unlikely [3,94]. Various conspiracy theories have emerged in the
meantime in the public debate [4], some heavily politicized [5], at times exactly mirroring earlier
conspiracy theories involving SARS [6]. At the same time the scientific community is doing its best
to explore the probable exact origins of the pandemic [7,8,9,93]while focussing first on finding
possible treatments and designing effective containment measures.
page 2 of 44
In that difficult context we believe that a probabilistic treatment of the possible origins of Covid-19
can help. Such a treatment is suited to a situation where different origins are probable. It does not
require taking position for one specific probable origin but instead assigns a probability to each
probable origin based on the available information at the time. In particular it allows for a
constructive debate between parties who may estimate the input probabilities (the ‘priors’)
differently. Such a probabilistic treatment offers thus a method to potentially bridge differences
between expert opinions [10]and to keep updating these input probabilities as either more
information or a consensus emerges.
We attempt here a probabilistic treatment of the main two probable origins of Covid-19: pure
random zoonotic event and lab-related accident. While duly acknowledging that any such analysis
must rely to a large extent on uncertain data and uncertain factors, we shall try to base our
treatment on conservative values for the key input probabilities - conservative values which we
believe provide a good basis to initiate a reasoned discussion of the resulting relative probabilities of
the probable origins. We shall additionally consider alternative sets of values for these input
probabilities so as to observe the variability of the results under such a range of plausible
assumptions.
In the course of this analysis we do not get into any controversy about Gain-of-Function (GOF) and
whether SARS-CoV-2 (the virus that causes Covid-19) is a virus that first came from nature or was
man-made. That controversy is irrelevant to the scope of this paper. We shall instead simply
suppose that SARS-CoV-2 is nature-made, from which point we can then consider whether the
outbreak itself is nature-made or man-made.
Nor do we wish to get into any controversy about the so-called ‘Wuhan P4 lab’ (strictly meaning the
National Biosafety Laboratory located in the Zhengdian Park of Wuhan Institute of Virology, which
also hosts BSL-2 and BSL-3 labs [11]). So as to avoid any such controversy, this paper simply
ignores the BSL-4 lab component of the Wuhan National Biosafety Laboratory in its risk estimates.
Nor do we wish to get into any controversy about possible intentional release vs. possible accidental
release. We fully trust that Ockham’s razor has common enough applications to not have to
suppose any malicious intent.
Last, in the hope that such a treatment may inform a larger audience, we shall intentionally keep the
mathematical approach as simple as possible.
ESTIMATION OF THE ODDS
1. Hypotheses under consideration
When faced with a pandemic such as COVID-19 an essential question with huge implications for
public policy is
‘How probable is it for the initial COVID-19 outbreak in Wuhan to be linked to coronavirus
lab activities in Wuhan against the alternative explanation of a purely natural zoonotic
origin?'
We will call the two hypotheses:
Hacc: The COVID-19 community outbreak that was first observed in Wuhan was caused by an
accident linked to a Wuhan lab (be it collection, transport or lab accident, including leak)
Hrand
:The COVID-19 community outbreak that was first observed in Wuhan was caused by a
random zoonotic event unrelated to a lab, somewhere in China
page 3 of 44
2. Probabilities for each hypothesis
a. COVID-19 Random Zoonotic hypothesis (Hrand)
We need to estimate the probability of:
Hrand
:The COVID-19 community outbreak that was first observed in Wuhan was caused by a
random zoonotic event unrelated to a lab, somewhere in China
The underlying event behind Hrand is:
Erand: arandom SARS-like zoonotic event somewhere in China leading to a first community
outbreak in Wuhan
The probability of that event is difficult to evaluate directly. As explained in Annex A, a more
practical route is to consider a more general event:
Grand: arandom SARS-like zoonotic event somewhere in China leading to a first community
outbreak somewhere in China
A SARS-like community outbreak is a real risk in China, especially in the context of changes in
human population patterns and land use patterns [12,13] close to natural reservoirs of animal
carriers such as bats and other possible intermediate hosts. While the risk of the bat-host-human
infection path is well understood, the practical risk of direct bat-human infection has so far eluded a
precise answer [see Box 2]. Nevertheless the resulting risk of epidemic and then pandemic is clearly
compounded by increasing movements of people around the country, particularly between
countryside and cities [14]. Some major work and progress in understanding this risk has been
done, often involving leading Chinese research institutions such as the WIV but also international
organizations [15, 16, 17].
The last human coronavirus (HCoV) community outbreak that originated in China before COVID-19
was SARS in 2003. Between that SARS epidemic and the COVID-19 epidemic 16 years and ahalf
have elapsed. Since we do not know yet if COVID-19 is an event unrelated to alab or not, it means
that - whatever the theoretical risk debates -we have observed at most 2 SARS-like community
outbreaks in 16.5 years in China caused by a random zoonotic event.
It is very difficult to precisely estimate a probability from 2 data points (especially if the second one
is tentative), but we shall start with an indicative probability of non-lab related community outbreak
due to a SARS-CoV-like virus in China as being of 1 every 10 years and will later consider
alternative values. The motivations for such an initial estimate and its intrinsic uncertainties are
discussed in Annex A.
Probability of non-lab related SARS-like community outbreak in China ~ 1 every 10 years
or to use some more standard notations:
(Grand) 0.1 per yearP
Additionally we know that the COVID-19 outbreak was first observed in Wuhan, with all the viral
strains to date linking back to the Wuhan genomes published in the early days of the outbreak [18].
So the probability we need to estimate
is the probability of a random SARS-CoV-like zoonotic event
leading to a first community outbreak in Wuhan (against any other place in China).
In order to estimate this probability let’s consider a few scenarios that should map all possibilities:
Eloc: A natural zoonotic event (possibly involving a host animal) of a SARS-CoV-like virus
which happens in a given place in China can only lead to a first community outbreak
in close proximity to that place.
page 4 of 44
Eprov
: A natural zoonotic event (possibly involving a host animal) of a SARS-CoV-like virus
is most likely to happen and cause a first community outbreak in a province with
known human SARS-CoV-like viruses.
Eany
: A natural zoonotic event (possibly involving a host animal) of a SARS-CoV-like virus
which happens in a given place in China can lead to a first community outbreak
anywhere in China with no preference for any particular place.
Let’s review each scenario and see what they imply for the probability of a random SARS-CoV-like
zoonotic event leading to a first community outbreak in Wuhan (against any other place in China)
Scenario Eloc:
Supposing first that Eloc
holds (,let’s evaluate the probability of a first community (Eloc) 1))P=
outbreak in Wuhan due to a natural zoonotic event in close proximity to the city.
1 P(zoonotic outbreak in W uhanGrandEloc)p=
1 P(zoonotic outbreak in W uhan | GrandEloc) × P(GrandEloc)p=
and as in this scenario we are supposing (Eloc) 1P=
1 P(zoonotic outbreak in W uhan | Grand) × P(Grand)p= E loc
is effectively our rescaling factor for ,the larger(zoonotic outbreak in W uhan | Grand)PEloc (Grand)P
distribution of human SARS-like random zoonotic events somewhere in China leading to an
outbreak in the country.
We note that there are no known animal carriers reservoirs in the city of Wuhan (either bats or
intermediate hosts) [19]and that a zoonotic event is thus more likely to happen in the countryside,
close to bat cave reservoirs or in a farming environment involving possible intermediate hosts.
Accordingly, under Eloc aWuhan citizen is less at risk of being part of an initial SARS-like outbreak
due to a zoonotic event than the ‘average’ Chinese citizen, since the overall Chinese population
encompasses not only cities but also countryside.
The relative population of Wuhan compared to the whole of China is 0.79%, as 11 mln over 1,400
mln, which we shall round up as 1%. Hence based on the above:
(zoonotic outbreak in W uhan | Grand) 1%PEloc < [conservative]
and as (Grand) 1 in 10 yearP =
1 1% of 1 in 10 yearp <
Scenario Eprov:
Supposing first that Eprov
holds (,let’s evaluate the probability of a first community (Eprov) 1))P=
outbreak in Wuhan due to a natural zoonotic event in Hubei province.
2 P(zoonotic outbreak in W uhan | GrandEprov) × P(GrandEprov)p=
and as we are supposing (Eprov) 1P=
2 P(zoonotic outbreak in W uhan | Grand) × P(Grand)p= E prov
is effectively our rescaling factor for .Let’s first(outbreak in W uhan | Grand)PEprov (Grand)P
consider a zoonotic event via an intermediate host and let’s try to determine how likely a Hubei
citizen is to be infected by an intermediate host compared to an ‘average’ Chinese citizen.
Unfortunately at this stage little is known about possible animal hosts, but domestic animals,
page 5 of 44
chicken, pigs, ducks and pangolins seem to have been discounted while ferrets and bamboo rats
(for instance) are still being considered [20,21,91]. Based on the limited knowledge available, at
best what we can state is that the provinces with the strongest interfaces between bat populations
and animal farming generally seem to be in the Southern province corridor (Yunnan, Guizhou,
Guangxi, Guangdong, Fujian, up to Zhejiang), with Hubei sitting on the edge of that corridor [22, 92].
Box 1: Bat species richness in China:
Map extracted from Feijó et al [
92
], with overlays of the Mojiang and Wuhan locations.
If we consider a zoonotic event with a direct bat-human interface, the bat populations with known
SARS-CoV-like virus seem to be in Yunnan, Guanxi, Zhejiang, but also with some incidences in
Hubei itself and neighbouring Shanxi. However, the (limited) known bat populations with
SARS-CoV-like viruses that make use of the human ACE2 receptor (essential for direct bat-human
infection) are in Yunnan. All of this must nevertheless be taken carefully; for instance there could be
a historical sampling bias for Guangdong and Yunnan with other provinces having been less
systematically surveyed. Our present knowledge is still very patchy on these essential questions
and at best what we can say at this stage is that Hubei seems less likely than Yunnan for such a
direct bat-human zoonotic event, but also more likely than the average Chinese province [22]. This
is also confirmed by the absence of any SARS-CoV2 related virus in any samples collected in
Wuhan or Hubei to date. [94]
If we then try to translate the qualitative assessments (with and without intermediate host) into the
resulting risks in terms of population, we further note that most of China’s population is in the
provinces along the East coast, with the South East coast being generally more at risk than Hubei,
and the North East (including Beijing) at par or actually under Hubei. Hence a citizen of Hubei
seems at most a bit more at risk under the Eprov
scenario than the average Chinese citizen.
However, we are not just considering the probability of an outbreak in Hubei but more specifically
the probability of an outbreak in Wuhan. The continued absence of any detected initial case out of
Wuhan nine months after the initial outbreak and the discarding of the early wet-market animal-host
page 6 of 44
theory both play against the scenario of aHubei provincial zoonotic event leading to a first outbreak
in Wuhan (with or without intermediate host). In the end we shall consider that the two factors (at
most slightly higher risk for an average Hubei citizen compared to a population-adjusted average
Chinese citizen, continued absence of any detected early case out of Wuhan) work to cancel each
other, leading us to assume that the resulting probability is still in line with the previous population
argument :
(zoonotic outbreak in W uhan | Grand) 1%PEprov
giving .2 1% of 1 in 10 yearp
Scenario Eany:
Supposing instead that Eany
holds (,let’s evaluate the probability of a first (Eany) 1))P=
community outbreak in Wuhan due to a natural zoonotic somewhere else in China.
3 P(zoonotic outbreak in W uhan | GrandEany) × P(GrandEany)p=
and as we are supposing (Eany) 1P=
3 P(zoonotic outbreak in W uhan | Grand) × P(Grand)p= E any
is our rescaling factor for .Per Eany
,Wuhan shall(outbreak in W uhan | GrandEany)P (Grand)P
be treated exactly like any other place in China with 1% of the population. Hence:
(zoonotic outbreak in W uhan | Grand) 1%PEany [exact]
and 3 1% of 1 in 10 yearp
For the sake of clarity, similarly to what we noted with Eprov
, such an Eany scenario is rather
unlikely due to the total absence of any detected early case out of Wuhan 9 months after the original
outbreak. However, as shown below, this won’t matter.
Retained probability:
The above analysis shows that, based on the information presently available, the probability of a
zoonotic outbreak in Wuhan seems reasonably well approximated by a simple population argument
under a range of scenarios that should map all possibilities. So without having to consider how to
weight these scenarios, we can simply retain their common upper value:
(zoonotic outbreak in W uhan | Grand) P1%
Given the uncertainties attached to the Eprov scenario, we will nevertheless later consider an
alternative value of 2% for . For now we have:(zoonotic outbreak in W uhan | Grand)P
(zoonotic outbreak in W uhan | Grand) × P(Grand) 1% of 1 in 10yP
(zoonotic outbreak in W uhan) 0.1% per yearP
Or using the notation for the hypothesis:
(Hrand) 0.1% per yearP
which can be also stated as a ‘once in 1,000 years’ event.
Box 2: A review of assessments of the direct spillover risk from natural bat reservoirs:
In the wake of SARS and given the role that an intermediate host animal is generally considered to
have played, perceptions of the risk of direct transmission of a coronavirus from bats to humans were
page 7 of 44
initially rather low, the dominant assumption being that an intermediate host was required [23]. These
perceptions changed with the discovery of a SARS-CoV-like virus that uses the ACE2 receptor in a
Yunnan bat colony in Oct 2013 [24].
However, primacy should really go to the publication in May 2013 of the less well known but essential
MS thesis on the Mojiang ‘miners’ severe pneumonia cases [25], following prolonged clearing out of
bat guano in an abandoned hillside mine. That MS thesis, drawing on the diagnostic of the top SARS
expert in China (Dt Zhong Nan Shan), notes in its conclusion:
‘With the Kunming Institute of Zoology, we confirmed that the six patients were exposed to
Chinese Rufous horseshoe bat, which caused the disease. However, a paper published in
Science magazine in 2005 by Scientist Shi Zheng Li and Zhang Shu Yi from Wuhan Institute
of Virology under the Chinese Academy of Science [see
23], concluded that the
SARS-like-CoV carried by bats is not contagious to humans. This contradiction indicates the
importance of these six cases: the severe pneumonia caused by the unknown virus and the
bats in the cave merit further investigation and research.’
Nevertheless, even today for many specialists the actual risk in normal circumstances still remains low.
For instance in Feb 2010, Lin Fa Wang, a top specialist on bat coronavirus and a frequent collaborator
of the WIV rated the risk a low:
Still, very few bat viruses are ready to transmit directly to humans, said Wang, who has been
studying bat origins of human viruses for decades and works with a group of researchers
sometimes dubbed ‘The Bat Pack.’ “I always say that if they could do that, then the human
population would have been wiped out a long time ago because bats have been in existence
for 80-to-100 million years -- much older than humans” [
26
]
Not that long ago (Dec 2017), the WIV scientists who regularly do bat samplings in the wild voiced a
similar opinion:
‘These SARS-like viruses usually stay quietly among wild animals in nature. They have never
attacked humans. The problem always first comes from humans. So the method is very
simple. If you don't touch or disturb wild animals such as bats and civet cats, the virus will
naturally not spread to humans.’[
27
]
In contrast to these low risk estimates, a pandemic scenario by USAID-PREDICT [28] published
around 2014 may be seen as a high point in the risk evaluation of a possible direct bat-to-human
coronavirus infection. This was done in the context of bat guano collection, a possibility highlighted in a
study by PREDICT a bit earlier in 2013 [29]. That pandemic scenario insisted on the risk of direct
infection while collecting bat guano from caves and further gave an indicative estimate of the
probability of a subsequent pandemic of 96%. These alarmist estimates seem high; first as far as the
risk of infection is involved some of the bat guano collectors in the case studied had been doing so for
40 years without any issue [30], and - secondly - as far as the risk of subsequent pandemic is involved,
that scenario came out around a year after the Mojiang ‘miners’ accident with its suspected CoV
infections while clearing up bat guano (albeit after long exposure times of 4 to 14 days) which did not
actually lead to any community transmission from any of the 6 cases [31]. Additionally, some of the
people involved in that USAID-PREDICT study stated in subsequent papers (including one published
in Nov. 2019) that a coronavirus spillover in communities living close to bat colonies is nevertheless a
‘rare event’, with mostly ‘subclinical or [...] only mild symptoms’ [19, 88].
All things considered it is quite possible that the actual risk of direct bat-to-human transmission is still
rather low as long as bat colonies are not under environmental stress, including human encroaching
and land change use [12,13]. In contrast, what has most definitely changed over the recent past, has
added substantially to the risk and is unlikely to change, is the increased possibility of a local outbreak
turning in an epidemic and then into a pandemic, due to the important developments in national and
international travelling patterns [32]. That in itself should certainly not invite complacency.
page 8 of 44
b. COVID-19 Lab-related accident hypothesis (Hacc)
We need to estimate the probability of:
Hacc: The COVID-19 community outbreak that was first observed in Wuhan was caused by an
accident linked to a Wuhan lab (be it collection, transport or lab accident, including leak)
In order to estimate Hacc, let’s start by considering the underlying event:
Eacc: An accident involving a SARS-like coronavirus linked to aWuhan lab (be it collection,
transport or lab accident, including leak)
Let’s then decompose this accident probability between collection, transport and lab accidents:
Accident during collection of a virus:
Thousands of SARS-like coronaviruses have been found in bats populations in particular in South
China caves. In less than 10 years the research teams at the WIV have collected 15,000 bat
samples mostly from China (a lesser part being from Africa and other countries), which have
delivered so far around 1,500 types of virus strains and more than 60,000 individual virus strains
(individual occurrence of a virus strain type in a sample) [33,34].While these numbers are already
impressive, potentially there could be even more viruses waiting to be identified in these samples as
it is not clear if all the samples have been fully tested as the tools to do so are still being tested and
refined [35].
The risk of infection of a worker through these collections is not negligible, especially considering
some of the collection conditions that have been reported [36]while the real risk of direct infection of
a SARS-like coronavirus from bats to humans has been recognized since 2013 [24,16,37] even if
the actual likelihood of such an event under normal circumstances is still heavily debated [see Box
2].
Yet we do not have a precise estimate for such a risk. At best we can show that even with a very
small risk per virus strain contained in a sample, we shall still end up with a non-negligible risk of
Collection-Acquired Infection over the unprecedented sheer quantity of virus strains being handled -
effectively at no other time in history have so many bat viruses been handled, amongst which some
are likely to have a potential for a human jump. For instance if we suppose an a-priori low 0.0001%
(one in a million) risk of infection per virus strain thus detected, counting around 50,000 identified
virus strains collected in China itself (out of the 60,000), this still sums up to a5% risk of a
Collection-Acquired Infection over the full collection over the years.
Additionally we do not know how many strains and samples were collected in the few months prior
to the start of the outbreak (which is really what matters here). There is unfortunately no open
record on this.
While from the above we can reasonably conjecture that the cumulative probability of a
Collection-Acquired Infection being the cause of the outbreak is not exactly null and may not even
be negligible, we must concede that it is very difficult to estimate that number even approximately
and we shall not attempt it here - leaving it instead as a possible refinement of our probability
estimates. Instead we shall simply conservatively ignore the risk of an accident during collection
leading to some worker getting infected:
(CollectionAcquired Inf ection) 0% per yearP [conservative]
Accident during transport a virus:
page 9 of 44
Next there is the probability of an infection during transport of virus samples to Wuhan. We know
little about these transport conditions, and equally little about the number of strains and samples
that were collected in the few months preceding the outbreak.
So we shall simply assume that the virus samples are safely transported according to best practices
and that the risk of infection during transport is quasi null. We will again invite a proper assessment
as a possible refinement of our probability estimates, knowing that this anyway puts us on the
conservative side:
(T ransportAcquired I nf ection) 0% per yearP [conservative]
Accident directly involving a lab:
Last we need to consider two possible accident scenarios directly involving a lab:
Lab-Acquired Infection (LAI): a lab worker gets infected in the lab and passes on that
infection to the community.
Lab Leakage: the virus escapes the lab without first infecting a lab worker, for instance due
to an issue with the treatment of solid, liquid or gaseous wastes [90].
The two scenarios can be both described as ‘Lab Escapes’. However it is much easier to find
records of Lab-Acquired Infections than of Lab Leakages. LAIs are actually not that uncommon and
are typically recorded by international organizations [38], while Lab Leakages are not necessarily
even detected [39], especially since not all Lab Leakages would necessarily result in an infection in
the community. Additionally such accidents may simply not be reported to authorities even if
detected by the laboratory itself [40].
For the reason just given we shall simply ignore the contribution of Lab Leakages to the probability
of aLab Escape. The probability of aLab Escape via a Lab-Acquired Infection can then be
estimated through official records and then checked against a few reference points.
In doing so we shall only consider those labs in Wuhan that we know were actively working on
SARS-like coronaviruses. Work on SARS and SARS-like coronaviruses started in China just after
the 2002 epidemic [92], with many samples being collected, tested and sequenced, and key papers
being published - especially after the discovery of large natural reservoirs of coronaviruses in South
China bat colonies in 2005 [23], and again following the discovery of the potential of some bat
coronavirus to infect humans without any intermediate hosts in 2013 [24,25]. From 2003 to 2017 all
that work was without any doubt done at BSL-3 or lower (some at BSL-2 [41,94]) since the BSL-4
suite at the WIV (Zhendian site) would only open in 2017.
In any case the revised guidelines specifically aimed at SARS-CoV-2 [42]that were published in
January 2020 stipulate that BSL-3 is the suitable level for work on the live virus ([A]BSL-3 for animal
experiments) while BSL-2 is the suitable level for work on uncultured SARS-CoV-2 infectious
materials, which is fully consistent with the standard biosafety levels for this type of pathogen
[Annex D]. Hence to this day most of the work involving live coronaviruses culture in Wuhan is still
being done at [A]BSL-3, often by the same teams in the same labs [Annex E].
With this in mind we shall conservatively ignore the BSL-4 suite at the Wuhan Institute of Virology
(VIW) and the various Wuhan BSL-2 labs involved (which should only handle uncultured
coronaviruses), focusing purely on BSL-3 labs where the cultured strains were normally handled.
It is important to note that there is no easy consensus on an estimate of LAIs for BSL-3 labs [Annex
B]. First such estimates depend on many variables (level of activity of the lab, level of expertise of
lab personnel, physical characteristics of the lab, characteristics of the virus being handled, type of
work on these viruses, etc), which we are already difficult to obtain for known LAIs in US BSL-3
labs, and even more difficult if not impossible to obtain for the Chinese BSL-3 labs of interest.
page 10 of 44
Secondly, even knowing these variables many differences in assumptions and methodology behind
these estimates may remain.
In the end, the inherent limitations of trying to rigorously evaluate the risk of LAIs may be best
illustrated by the case of the planned National Bio- and Agro-Defense Facility in Manhattan, Kansas,
for which the Department of Homeland Security released a first assessment of LAI escape risk at
2.4% per year and afinal risk assessment of around 0.002% per year, immediately criticized by the
National Research Council.
With all these limitations in mind, we shall follow the analysis of Klotz [43-46]who derives an
estimate of the risk of aLab Escape via an Lab-Acquired Infection in a BSL-3 lab-complex of 0.2%
per year, based on a US CDC report [47]. Further estimates will be considered in section 4 below.
As is shown in Annex E, 3 Wuhan lab-complexes with either BSL-3 or ABSL-3 labs are most
definitely actively involved in the study of SARS-like coronaviruses. So considering 3 BSL-3
lab-complexes that were very active working on coronaviruses over the last few years, we get:
(W uhan Lab Escape | 3 BSL–3) 1 (1 .2%) per yearP = − 0 3
which in first order (thereafter systematically used) becomes:
(W uhan Lab Escape | 3 BSL–3) 3 × 0.2% per yearP
(W uhan Lab Escape | 3 BSL–3) 0.6% per yearP
Summing over the 3 types of possible accidents (collection, transport and lab), we get:
(W uhan Lab Related Accident | 3 BSL–3) .6% per yearP ≈ 0
Probability of a community outbreak following a Lab-Related SARS-like Infection:
As further explained in Annex C and illustrated in Annex D,an isolated infection - or even a few
concurrent cases of infections due to aLab Escape - will not necessarily lead to a community
outbreak. Here we shall refer to Klotz [48] (building on Lipsitch et al [49]and Merler et al [39]) who
uses an intermediate estimate of 25% for the probability of an outbreak given a Lab-Acquired
Infection. Merler et al shows that the 25% outbreak probability is consistent with an infectious
disease with an R0of around 1.75, under a specific scenario of urban lab escape followed by closure
of the laboratory closure and quarantine of the households of laboratory workers. We further note
that such an R0is on the lower side of available estimates of the R0for COVID-19 which are
generally between 2and 2.5 [50]. If we were instead to use a common estimate of 2.2 for COVID-19
R0, the probability of an outbreak estimated by Merler et al
would become around 50%.
With this in mind, in order to remain conservative we shall retain a value of 20% as reference value,
slightly less than the 25% used by Klotz.
(labrelated outbreak | inf ection due to a W uhan labrelated accident) 20%P
Strictly speaking this is the probability of a community outbreak due to aWuhan Lab Escape, but
that outbreak itself could happen in Wuhan or elsewhere. Nevertheless such a Wuhan Lab Escape
would most likely cause the outbreak to happen locally in Wuhan, and much less likely cause a
distant first outbreak away from the escaped lab. So at most a slight curtailment of the 20% may be
needed to allow for that unlikely alternative of a distant first outbreak. This won’t change the
page 11 of 44
probabilities significantly and since we are already sitting on the conservative side overall, we shall
simply take:
(labrelated outbreak in W uhan | inf ection due to a W uhan labrelated accident) 20%P
Hacc - SARS-like community outbreak due to a Wuhan lab-related accident:
Putting the above probabilities together we get:
(labrelated outbreak in W uhan | Eacc with 3 BSL–3) 0.20 0.6% per yearP ×
(labrelated outbreak in W uhan | Eacc with 3 BSL–3) 0.12% per yearP
Or using the notation for the hypothesis:
(Hacc | 3 BS L–3) 0.12% per yearP
This can be also stated as a ‘once in 833 years’ event.
3. Resulting Odds
Now let’s calculate the odds or relative probabilities.
We found that:
(Hrand) 0.10% per yearP
(Hacc | 3 BS L–3) 0.12% per yearP
and since the random zoonotic hypothesis is in no way linked to any lab:
= (Hrand)P(Hrand | 3 BSL–3)P
Hence we have:
(Hacc | 3 BS L–3) .2 (H rand | 3 BSL–3)P1 × P
Since we are only considering the two hypotheses Hacc and Hrand:
(H acc | 3 BSL–3) P(Hrand | 3 BSL–3) 1P+ =
hence:
(Hacc | 3 BS L–3) 54.5%P
(Hrand | 3 BSL–3) 45.5%P
Said otherwise under conservative assumptions, the probability that the COVID-19 community
outbreak first observed in Wuhan is linked to some Wuhan lab activity is at least 54.5% (given by
0.12/(0.12+0.10)) and the probability of the alternate purely natural origin is at most 45.5%.
In odds terms one would formulate that as saying that, under conservative assumptions, the odds of
a lab-induced origin to a purely natural origin - given a first observed community outbreak in Wuhan
- are at least 6 to 5 on.
dds(Hacc vs. H rand | 3 BS L–3) 1.2 6 to 5 onO =
How conservative are these odds of a lab-induced origin?
A number of conservative assumptions were made during the derivation of the odds:
page 12 of 44
- We did not take into account the collection and transport risks
-We only considered 3 Wuhan [A]BSL-3 lab-complexes which we can ascertain were actively
working on coronaviruses. We ignored 2 other BSL-3 lab-complexes that were also known to be
working to some degree on SARS-like coronaviruses .
-We totally ignored the many BSL-2s (including the one at the Wuhan CDC) and one BSL-4 lab
in Wuhan that were either storing or actively working on SARS-like coronaviruses [51, 41].
-The reference Lab Escape probability of 0.2% only considers a Lab-Acquired Infection (LAI),
meaning a lab worker being infected and spreading the virus to the community. It does not
include the very possible risk of Lab Escape without LAI (for instance via a waste treatment
problem) which is more difficult to tabulate [52].
-The Lab-Acquired Infection probability of 0.2% per year per BSL-3 lab-complex is conservative.
As shown in Annex B,the Department of Homeland Security used instead a reference
probability of 2.4% per year in its first assessment of the risk for the planned National Bio and
Agro-Defense Facility (NBAF, to be opened in 2022).
-That 0.2% is calibrated on US data, when it could be argued the safety of Chinese BSL-3 labs is
on average (and not in all cases) still lagging behind the US ones (see point 3of the Discussion
below).
-The 20% probability of outbreak given a Lab Escape was arrived at for an infectious disease
with R0of around 1.75 in a scenario of active countermeasures (including closure of the urban
lab and quarantine of the lab workers’ households) [39]. COVID-19 had an initial R0closer to 2.2
[50]and aLab Escape may not necessarily be met with countermeasures, especially if
undetected for a while.
-The 6 recorded SARS Lab-Acquired Infections (including 4 in China, some with community
transmission), in only 2 years following the 2002 SARS outbreak [53], show that the risk of an
LAI when working with highly dangerous coronavirus is likely higher than this 0.2% per year
baseline.
-We had to assume that COVID-19 is a non-lab related accident to estimate the mean time
interval between purely zoonotic SARS-like community outbreaks in China. From this we
actually find a non-negligible probability that COVID-19 is a lab-related accident. So our
estimate of that mean interval is favouring a non-lab origin, which means that the odds should
be even more in favour of the lab-related accident.
4. Variations on Estimate Input Probabilities
The main issue when trying to come up with reasonable estimates for the key input probabilities is
that there is not much data to work with. For instance there has been a limited number of
human-coronaviruses community outbreaks in China over the last 20 years, making it difficult to
estimate the arrival process (a Poisson process) - in this case the data is available but not dense.
Or we do not have available data where it should theoretically be possible to have some (such as
the precise numbers of live coronavirus worked on in Wuhan BSL-3 labs, type of activity involving
them, the total durations and types of exposures, lab conditions, biosafety training of employees,
even general statistics about LAIs in Chinese labs, etc).
So while we started with some specific choices for the key input probabilities (our Reference
scenario), it must be clear that these priors are just educated estimates, based on available
information at the time and our understanding. Any such estimate is partly arbitrary and complex
mathematical models - while they may be able to deliver structural insights [54] - cannot solve this
fundamental lack-of-data issue. Still a redeeming grace of such probabilistic outline is that we do not
page 13 of 44
need exact estimates to engage in a constructive discussion about probable origins, since using
conservative estimates may be enough to assess whether the possibility of a lab-Induced
community outbreak is negligible or not.
With this in mind we shall explore alternative sets of input probabilities beyond our Reference
scenario. We will consider a Base scenario which uses ‘raw’ values for these probabilities, and aDe
Minimis where values in favour of the purely zoonotic event hypothesis (Hrand) are systematically
used. These scenarios are presented in Box 3 and the resulting relative probabilities in Table 1
below.
We note that even under the very conservative De Minimis scenario, the relative probability of alab
induced accident being the origin of the COVID-19 community outbreak is not negligible, at 6%.
Box 3: Base, Reference and De minimis scenarios:
P(Grand): probability of occurrence of a purely zoonotic human outbreak in China of a SARS-like
coronavirus, per year
Base: we retain the MLE
(Maximum Likelihood Estimate)
for the mean of the Poisson
process associated with the
occurrence of the event of
interest. Specifically SARS and
COVID-19 occurred at an
interval of 16.5 years, which
gives one additional event
every 16.5 years, so around
6.06% per year (with the initial
SARS event starting the clock).
Reference: we use 1 additional
event in 10 years as an
estimate of the mean of the
Poisson process associated
with the occurrence of an event
of interest. We can show that
there is around a 45%
probability that we should
observe one or no additional
event in 16.5 years (on top of
the initial SARS event) if that
estimate is indeed the true
mean.
De minimis: we use 2.2
additional events in 10 years as
an estimate of the mean of the
Poisson process associated
with the occurrence of an event
of interest. We can show that
there is only around a 10%
probability that we should
observe one or no additional
event in 16.5 years (on top of
the initial SARS event) if that
estimate is indeed the true
mean.
P(Wuhan | China): rescaling factor applied to P(Grand) to get the probability of occurrence of a
purely zoonotic community outbreak in Wuhan of a SARS-like coronavirus, per year
Base: we use the population
proportion as described in 2.a.
Wuhan has around 11mln
inhabitants and China as a
whole around 1,400mln.
11/1,400 gives us 0.79%.
Reference: we round up the
base value to 1% which will
favour the non-lab induced
zoonotic event hypothesis.
De minimis: in order to
account for uncertainties
around bat populations carrying
SARS-CoV-like viruses,
especially those with the ability
to directly infect humans, we
use twice the Reference value.
P(Active-Lab Acquired Infection | 1 BSL–3): probability of a Lab-Acquired Infection with a human
SARS-like coronavirus for one BSL-3 lab complex actively working on these (cell cultures or animal
experiments).
Base: Based on the structural
issues with some Chinese labs
reported in point 3 of the
Discussion, we increased the
0.2% per BSSL-3 complex (that
was calibrated on US labs) to
0.25% - which likely still does
not properly reflect the relative
risk level.
Reference: We use the
estimate of 0.2% per BSL-3
complex per year discussed in
Annex B.
De minimis: We use a low
estimate of 0.1% per BSL-3
complex per year which would
mean that either our base
estimate is much too high
compared to the actual safety
of US BSL-3 labs, or that the
average Wuhan lab of interest
is much safer than the average
US lab.
page 14 of 44
P(community outbreak | escape due to a Wuhan lab–related accident)
Base: we use the 25%
discussed in Annex C.
Reference: we use an
intermediate 20%.
De minimis: we use a lower
10% to reflect the difficulty of
generalizing from a specific
simulation with its specific
assumptions.
Table 1: Relative probabilities for the two hypotheses
page 15 of 44
DISCUSSION
1. Understanding the odds: A simple analogy
A simple analogy may provide a good summary of our key findings about the odds.
Let’s suppose that the risk of an epidemic starting due to a SARS-like coronavirus escaping a lab is
a lottery. Let’s also suppose that the risk of an epidemic starting due to a SARS-like purely random
zoonotic event is another lottery.
We showed that while the risk of a lab-related epidemic is less than the risk of a pure zoonotic event
epidemic when considering the whole of China
(as is often correctly mentioned in the SARS-CoV-2
origins debate), for the city of Wuhan itself
the balance of risks is actually very different.
Based on the fact that Wuhan has some the most active labs in China working on these SARS-like
coronaviruses but has only 1% of the population of China, following the lottery analogy we observed
that Wuhan has bought a large chunk of the lab-related epidemic lottery tickets but less than 1% of
the purely zoonotic epidemic lottery tickets.
It then followed that - based on rather conservative assumptions
-Wuhan should be expected to win
the purely-zoonotic epidemic lottery every 1,000 years and to win the lab-related epidemic lottery
every 833 years (on average). Said otherwise, a lab-related epidemic is more likely to first break out
in Wuhan than a purely zoonotic-based epidemic.
From there, knowing that Wuhan is today in possession of awinning lottery ticket, we considered
the question: ‘Which lottery did Wuhan likely win?’. Our answer to that question is based purely on
what we know of the two lotteries, but it is clear: using conservative assumptions, there is at least a
54% chance of Wuhan having won the lab-related epidemic lottery and at most a 46% chance of
Wuhan having won the purely zoonotic event lottery.
2. A rebuttal of common misunderstandings
Misunderstanding #1:
‘Since we know that a SARS-like epidemic in China is much more likely to be triggered by a
natural encounter with some animal rather than by any lab accident, saying that the recent
epidemic may have been caused by alab accident is simply unscientific and not worth
discussing.’
This misunderstanding is often repeated in the current debate [55,56]. It is true that for China as a
whole the risk of a SARS-like community outbreak triggered by a purely random zoonotic event is
likely higher than a lab-induced one, using the 0.2% per year baseline. With the probabilities used
for the Reference odds and supposing another 6 BSL-3 lab-complexes actively working on
SARS-like coronaviruses beyond Wuhan (for a total of 9 lab-complexes doing such work in China),
we indeed get:
(community outbreak in China | Grand) 10% per yearP =
and
(community outbreak in China | Eacc with 9 BSL–3) 9 0.20 0.2% per yearP = × ×
(community outbreak in China | Eacc with 9 BSL–3) 0.36% per yearP =
so that:
page 16 of 44
dds(Hrand vs. Hacc | 9 BS L–3) 28 to 1 onO
which effectively is ahuge weighting of the odds towards the random zoonotic event when
considering the whole of China, in full opposition to the Wuhan odds that we previously calculated:
dds(Hacc vs. H rand | 3 BS L–3) 6 to 5 onO >
or dds(Hrand vs. Hacc | 3 BS L–3) 5 to 6 onO <
Said otherwise, the odds totally pivot in favour of the lab-induced accident once we know that the
outbreak started in Wuhan. Using again the lottery analogy, Wuhan bought a large portion of the
lab-induced ‘SARS-like community outbreak’ lottery tickets (as per above 3 out of every 9 tickets, so
a third of the tickets), but it purchased less than 1% of the random zoonotic community outbreak
tickets.
As a result Wuhan is more likely to be holding a‘winning ticket’ from the lab lottery than from the
natural encounter lottery. From this we can see that the relative probabilities (natural vs. lab
induced) for a community outbreak in China as awhole do not extend to a community outbreak that
actually started in Wuhan -quite the contrary. Hence it is unfortunately misleading to generalize the
China odds to the Wuhan odds.
Additionally there is good circumstantial evidence to believe that the 0.2% baseline is very
conservative when applied to ahighly transmissible coronavirus and to variable lab safety
conditions. Indeed it is impossible to explain otherwise how 6 SARS LAIs could have happened in
only 2 years (2003-04), with 4 incidents in Chinese labs, if these LAIs were universally governed by
such a low baseline probability.
Misunderstanding #2:
‘The virus could have emerged naturally somewhere else in China and before causing the
Wuhan community outbreak - hence the odds are wrong because they do not consider the
possible emergence out of Wuhan'.
The possibility of natural emergence out of Wuhan with a first detected community outbreak in
Wuhan is already in the odds as we explicitly reviewed that possibility and included it in our estimate
of . See the discussion on Eloc
, Eprov
and Eany
.(community outbreak in W uhan | Grand)P
Misunderstanding #3:
‘Considering that the risk of a purely zoonotic event is linked to the population size of a city
or region makes no sense because all you need is just one infected person to start an
infection'.
While it is correct that only one initial carrier (the ‘patient zero’) is needed to start an outbreak of a
contagious disease, this does in no way invalidate the population size argument. The chance of that
initial carrier being present in a certain population is still linked to that population size as well as to
the proximity of that population to natural reservoirs of the responsible pathogen. These two aspects
were covered in our estimate of .(community outbreak in W uhan | Grand)P
Misunderstanding #4:
‘If you suppose that a community outbreak happens in China, then by definition it must
happen somewhere. So there is no point saying that there was a1% chance that it
page 17 of 44
happened in a particular place after the fact. It had to happen somewhere and it just
happened in Wuhan by chance.'
An easy way to see why this is incorrect is to notice that for the Wuhan community outbreak to be a
purely neutral after-the-fact random observation, then the rest of China must look like Wuhan.
Hence given that we are considering 3 BSL-3 lab-complexes in Wuhan actively working on
coronaviruses and given that Wuhan has 1% of China’s population, this means that the argument
would be correct if China had at least 300 BSL-3 lab-complexes actively working on coronaviruses.
With 300 BSL-3 working on these SARS-like coronaviruses:
(labrelated outbreak in C hina | Eacc with 300 BSL–3) 300 .2% 0% per yearP × 0 × 2
(labrelated outbreak in C hina | Eacc with 300 BSL–3) 0.12 per yearP
This 0.12 per year is to be compared with the 0.10 per year for the pure zoonotic event epidemic
expectation over China. Hence under the conditions necessary for the logic behind this
misunderstanding to be correct, we are effectively back to the same ‘6 to 5on’ odds, this time over
the whole of China.
Interestingly, if told that 300 lab-complexes were actively working on SARS-like coronaviruses in
China, most people at this stage would not intuitively consider the odds of a lab-related origin for a
SARS-like community outbreak somewhere in China to be negligible (compared to a purely random
zoonotic event) without even needing a more detailed inspection of individual probabilities. But
crucially this is exactly the same odds as when considering the probability of an observed first
community outbreak in Wuhan with 3 active BSL-3 lab-complexes against a purely random zoonotic
community outbreak there.
Misunderstanding #5:
‘There is still nothing proving that the COVID-19 community outbreak was caused by a
lab-related accident, whatever the probabilities. So it makes no sense to talk about a
possible lab accident.’
This misunderstanding seems to be surprisingly common in the debate about COVID-19. It is easy
to see why it is wrong: there is simply nothing proving that the outbreak is actually a purely random
zoonotic event either.
The too-often accompanying assertion that, when considering China as whole, a natural origin
SARS-like community outbreak is anyway much more likely than a lab-induced SARS-like
community outbreak - so that the probabilities are actually as good as a proof - offers no support at
all here since it is based on another misunderstanding (see Misunderstanding #1).
When faced with this kind of situation where there is no definite proof for any of the possible causes,
or even no dominating probability for any of the probable causes, all we can do is to try to evaluate
the relative probabilities as we did here, to use these probabilities to inform the debate and a
reasoned investigation, and then to keep updating these probabilities as more insights are collected
[57].
Misunderstanding #6:
‘You suppose a 1 in 10y probability for a random SARS-like community outbreak in China
but we know that coronaviruses outbreaks are more common (MERS, SARS pig, etc) across
the world. We also know that populations living close to bat colonies in China carry
page 18 of 44
antibodies for SARS-like coronaviruses, so this is only the tip of the iceberg and outbreaks
involving SARS-like coronaviruses are much more common than that’.
This misunderstanding is based on three possible confusions. There is first a confusion on the
probability of interest, which is the probability of a non-lab related (1) community outbreak of a (2)
human (3) SARS-like coronavirus (4) in Wuhan. As discussed in Annex A,the only one of these 4
attributes that we can reasonably relax to get more events and still be able to carefully rescale to the
probability of interest is (4) ‘in Wuhan’.
If we consider the attribute ‘community outbreak’ for instance, there is no point considering a
probability based on local non-outbreaks because there is no meaningful way to translate that
denser probability distribution into the probability of a proper SARS-like outbreak. These local
non-outbreaks are fundamentally different: they are effectively only detected through some
antibodies in a small fraction of rural populations living close to bat colonies [94] (2.7% in the study
reported in Ning Wang et al [19], 0.7% in the study reported by Hongying Li et al
[88]) -antibodies
which are not only conspicuously absent in the Wuhan population but also suggest that ‘infections
were subclinical or caused only mild symptoms’ [19].
The second confusion is a logical one. Local non-outbreaks in these populations living close to bat
colonies, as inferred by antibodies, are by definition local. So one would have to contrive an
exclusively directed scenario where someone in such a community got infected with SARS-CoV2
(or a an early strain of it), for some reason remained asymptomatic, did not create a local outbreak
but somehow led to an outbreak in Wuhan and nowhere else along the way, and particularly not
back home if home that was. This would have to involve some very directed travelling from such a
local community to Wuhan. Interestingly the people who do such directed travelling between these
communities and Wuhan are quite likely often involved with bat coronavirus studies.
The third confusion is a cognitive confusion between risk awareness and the actual level of risk.
Specifically, the knowledge that bat-colonies are natural reservoirs of SARS-like coronaviruses and
that some people living close to bat colonies often have antibodies for SARS-like coronaviruses has
certainly dramatically increased our awareness of the possible mechanisms of a SARS-CoV-like
spillover, but it has not in itself proportionally increased the risk level itself [see Box 2]. In the same
way (taking a much more extreme example) that our tracking of asteroids over the last 30 years has
not increased the risk of the earth being hit by one. What may have much more impact on the actual
risk level are the trends governing the intensity of the possible contacts between populations and
bats (possibly through an intermediate animal host), land use and the development of transport
links.
Most importantly, whatever the theoretical debates, in the end we have only at most two non-lab
related occurrences of human SARS-like community outbreaks in China over 16.5 years. Based on
the above discussion we consider that these 2 events provide the best available signature of the
actual outbreak distribution applicable to Wuhan via a rescaling argument. We further investigate
the estimation issues caused by such a small sample in Annex A.
Misunderstanding #7:
‘You start by estimating the risk of aWuhan lab-induced community outbreak of a human
SARS-like coronavirus to 1in 833 years, and your conclusion is that most likely the outbreak
in Wuhan is due to alab escape. But such a chance is so remote, as a 1 in 833 years event,
that it just makes no sense. The whole argument is suspect.’
Let’s first notice that the probability of a community outbreak starting in Wuhan due to a pure
random zoonotic origin is actually even smaller at less than 1 every 1,000 years. So the question
page 19 of 44
being really asked is how one can intuitively reconcile the Wuhan community outbreak to such small
probabilities (‘1 in 1000 years’ and ‘1 in 833 years). Given that the outbreak must be explained by
either one of the two hypotheses - dismissing any conspiracy theory [5,6] - most likely one or
maybe both probabilities are too low.
Let’s then note that even supposing that a SARS-like community outbreak could happen every
single year in China (instead of every 10 years), we still get a probability of less than 1in 100 years
for an outbreak in Wuhan. Said otherwise using this clearly excessive probability of 1per year for a
pure random SARS-like zoonotic community outbreak in China, we are still left with a very small and
intuitively unsatisfactory probability of that community outbreak happening in Wuhan against any
other place in China.
The only alternative to try to intuitively reconcile the Wuhan community outbreak to the small
probabilities we used is that the ‘1 in 833 years’ lab-induced community outbreak probability is
underestimated. We already noticed that that probability is conservative and we listed some of the
reasons. But actually the best insight into why this probability is likely seriously underestimated may
be provided by some Chinese assessments of the actual risk as detailed in the following section.
3. Evaluation of the Lab Escape risk by the Chinese authorities
A review of Chinese scientific papers and government-aligned publications shows that the relevant
Chinese supervising authorities and the Chinese government itself have consistently evaluated the
Lab Escape risk as all too real. Their declarations and writings are therefore consistent with the
scale of the Lab Escape risk highlighted in this paper. Here is a quick review of such evaluations:
Yang Zhanqiu’s evaluation of the risk in Chinese BSL labs (16th Feb 2020):
Yang Zhanqiu, adeputy director of the pathogen biology department at Wuhan University, was
recently quoted by the Global Times, a Chinese newspaper considered as strongly aligned with the
government [52]. The article shows a clear understanding of the risks:
‘The Ministry of Science and Technology issued new rules [--] that experts said could fix
chronic inadequate management issues [--]. The release of the guideline deals with chronic
loopholes at laboratories [--]
“Laboratories in China have paid insufficient attention to biological disposal”, Yang said.
Lab trash can contain man-made viruses, bacteria or microbes with apotentially deadly
impact on human beings, animals or plants.
“Some researchers discharge laboratory materials into the sewer after experiments without
a specific biological disposal mechanism”, Yang explained.
Medical staff and experts have long been asking for better regulation and supervision of
biological research institutes in China, but with mixed results.‘
‘Notice on Strengthening the Biosafety Management of Pathogenic Microorganism
Laboratories’ (9th Feb 2020)
On the 9th Feb 2020, 6 government offices (of the Ministry of Agriculture and Rural Affairs, Ministry
of Education, Ministry of Science and Technology, the National Health Commission, the Customs
Administration, the National Forestry and Grassland Administration) and the Chinese Academy of
Sciences together issued a notice detailing new rules to strengthen the security of Chinese bio-labs
[58].
page 20 of 44
That notice starts by mentioning that in recent years the safety of bio-labs has ‘significantly
improved’ but that some ‘problems and risks’ still remain. It then calls for
cooperation in the review of new, developed or expanded BSL-3 and BSL-4 labs
increased sharing of information relative to BSL-3 and BSL-4 labs across departments
improved regulation of the attributions of authorisations for experimental activities
suspension of any lab activity if the lab cannot meet the approval or review criteria
It further calls for a strict supervision and enforcement of the laws and regulations, especially with
regards to labs activities which have not received the relevant authorisations and stipulates that any
scientific results from such irregular activities shall not be recognized (which may tentatively be
interpreted as meaning that the need to publish is a common factor behind such activities).
It also reminds that only specialized institutions and laboratories designated by the Ministry of
Agriculture and Rural Affairs are allowed to keep stocks of bacteria and viruses, either isolated
strains or samples. It further asks for the supervising administrations to either destroy offendings
stocks and samples according to the relevant regulations or to send them to a specialized
institution.
If then calls for the bio-laboratories to properly implement relevant safety guidelines, covering
transport, reception and use of the pathogens, putting particular emphasis on the transport and
shipment of these.
The collection of pathogens is also addressed, stressing that this must be done according to
relevant regulations and that the exact sources, collection samples and methods should be properly
documented.
It then asks laboratories to improve the process for disposal of wastes from experimental activities
(in particular as to proper sterilization), to reinforce their organization and management, to
implement information and record management and develop better training and biosafety
awareness.
Last, it explicitly asks all [A]BSL-3 and BSL-4 labs to proactively engage with relevant public offices
and to fully accept their supervision and guidance - which may seem to suggest that afew BSL-3s
may not always have been exactly cooperative in this regard.
Yuan Zhiming’s evaluation of the risk in Chinese BSL-2 and BSL-3 labs (Oct 2019):
Agood introduction to the very real risk of alab related accident is provided by Yuan Zhiming - the
director of the WIV (the Wuhan P4 lab) and atop CCP representative there. In October 2019, the
Journal of Biosecurity and Biosafety published an article by Y. Zhiming [59]that highlighted major
structural issues with Chinese labs, including lack of funding, lack of training, lack of standard
operating procedures:
‘[...] due to different investment sources, affiliations, and management systems, the
implementation of these laboratories faces difficulties converging objectives and cooperation
workflows. This scenario puts laboratory biosafety at risk since the implementation efficiency
and timely operations are relatively compromised.
[...] several high-level BSLs have insufficient operational funds for routine yet vital
processes. Due to the limited resources, some BSL-3 laboratories run on extremely minimal
operational costs or in some cases none at all.
Currently, most laboratories lack specialized biosafety managers and engineers. In such
facilities, some of the skilled staff is composed by part-time researchers. This makes it
page 21 of 44
difficult to identify and mitigate potential safety hazards in facility and equipment operation
early enough.’
Yuan Zhiming & al evaluation of the risk in Chinese BSL labs (2016):
This 2019 assessment above essentially repeats the one offered in a2016 paper [60] co-authored
by Yuan Zhiming (with an additional insight on issues at the BSL-4), thus showing perduring chronic
issues:
[translation from the original document]
‘China has certain problems in the construction and management of high-level biosafety
laboratory systems.
At present, only one BSL-4 laboratory has been built in the country, and the management
and maintenance of its key equipment and the personnel’s mastery of the standardized
operating procedures (SOP) of Level 4 laboratories are not mature enough.
Among the BSL-3 laboratories that have been built, the distribution of laboratories across
the country is uneven, and many laboratories have low utilization rates due to insufficient
construction, operation and maintenance funds.
On the whole, the problems of China's high-level biosafety laboratory system are mainly
manifested in:
(1) In terms of overall layout, the industry and economic development and the needs of
special fields are not fully considered. [..]
(2) In terms of funding and operating mechanism, long-term stable maintenance funding,
incomplete sharing and cooperation mechanism, lack of stable operating funding, and the
disconnection between construction and operation, resulting in some laboratories not
completing construction or being difficult to operate normally after completion.
(3) In terms of management and support system development, the laws, regulations and
standard system of high-level biosafety laboratories need to be further improved, and the
construction of supporting research conditions such as information resources and
experimental data is somewhat lagging behind. The confluence of technology, management
and strategy research needs to be strengthened.’
Mainstream article in the China Daily mentioning the risk of working with dangerous
pathogens in labs (2015):
Discussing the challenges faced by China in its biosafety laboratories was not just limited to a circle
of experts. The China Daily, an English language newspaper owned by the CCP and often used as
aguide to Chinese government policy, published an article in February 2015 titled Be ready to fight
potential risks from P4 lab
’ [61].
The article welcomes the opening of the first Chinese P4 lab but ends this with a clear reminder
about the existing issues with management, maintenance and supervision of high biosafety level
labs, with a rather dramatic illustration that would be unthinkable in the current charged context.
‘But the government will also have to tighten supervision and monitoring of research on
dangerous and exotic pathogens, and strengthen the