Plane risk on
long-haul flights
Health Warning: the following is both long and messy, and in parts, reflects my lack of clarity.
Freedman studied [1]
all peer-reviewed or public health publications in a 6-month period of
possible, likely or unproven in-flight transmission of covid-19. Freedman
argues that the absence of large numbers of confirmed cases of in-flight
transmission of covid-19 is encouraging – but not definitive evidence that
fliers are safe. He also notes that a few cases have been noted of infections
having occurred at more than 3 rows away from the index case, so the standard
‘two row rule’ may need to be re-examined.
Doucleff [2] cites
Freedman’s evidence [1] that
Emirates airlines has a strict masking policy and that even though they
transported covid-positive patients nobody else got infected because everyone
rigorously wore masks (Emirates staff ensured this policy was implemented).
Barnett & Wilder-Smith [3a] found that the odds of getting the
virus in a standard 2-hour flight are about 1/4300, and the odds of getting the
virus are about half that, 1/7700, if airlines leave the middle seat empty.
More recently, based
on data from late September 2020 [3b],
Barnett & Fleming revised the odds to about 1 in 3,900 for full flights and
1 in 6,400 when middle seats are kept empty.
.
“In making his estimates, Barnett approximated the probability
that a given airline passenger has COVID-19, the probability that universal
masking could prevent a contagious passenger from spreading the disease, and
how risk of infection changes based on the locations of the infected and
non-infected passengers. [4]
Barnett [3] assumed
that everyone on a plane is wearing masks (all U.S. airlines have
mandated mask policies), and that the primary risk to passengers comes from
others in the same row and, to a lesser degree, the rows behind or in front of
the passenger. Seatbacks provide some measure of protection from passengers
in other rows, Barnett said. “Other passengers do not pose as much of a risk
because of the air purification systems on airplanes, he said.” [5]
Barnett’s estimate of
1/(4,300) was for the U.S. (1 in 6,500 people confirmed positive daily);
his estimate for the U.K. was about 10 times lower i.e. about 1/(40,000) [6], because of the lower prevalence
rate of infection in the U.K. at the time (1 in 60,000 people confirmed
positive daily).
An infected couple flew
from China to Canada on 22 January, none of the other 350 passengers on the 15-hour
flight were infected, probably because masks were worn.
The air in planes is replaced
every 3 to 5 minutes, and the air that is recirculated goes through HEPA
filters that should remove almost all droplets containing viruses [6]. “The ventilation systems on planes
are very effective in reducing the overall concentration of any airborne
pathogen exhaled by passengers,” says Dr.Julian Tang (University of Leicester).
The main risk may be face-to-face conversations where air can be exchanged
before being pulled away – along with any conversations before or after the
flight.
Barnett
argues that: "…three things have to go wrong for you to get infected (on a
flight). There has to be a Covid-19 patient on board and they have to be
contagious," he says. "If there is such a person on your flight, assuming
they are wearing a mask, it has to fail to prevent the transmission. They also
have to be close enough that there's a danger you could suffer from the
transmission" [7]. Barnett says
he took all of these probabilities into account before determining an overall
transmission risk.
Barnett
[7] states that there isn't much of
a difference in terms of risk between passengers sitting in an aisle seat on a
full flight and those in the window seat. However, the chances of becoming
infected are ever so slightly higher for those in aisle seats, because they
simply have more people around them.
Barnett
[7] states that because of his age
(72) he will not travel by plane soon – but he advises that for any high-risk
person, one should wear not merely a mask but also a face shield – to prevent
aerosols from entering in due to mask in-leakage.
Additive:
Dr Henry Wu [8] associate
professor at Atlanta's Emory School of Medicine, said the findings were
inconclusive on their own because the minimum infective dose remains unknown,
and risks increase in step with exposure time.
Boeing tests concluded that sitting beside an infected economy
passenger is comparable to seven-foot distancing in an office, posing an
acceptably low risk with masks [8]. Airbus
showed similar findings, while Embraer tested droplet dispersal from a
cough. Some 0.13% by mass ended up in an adjacent passenger's facial area,
falling to 0.02% with masks [8].
"It's simply additive," said Wu, who would prefer
middle seats to be left empty. "A 10-hour flight will be 10 times riskier
than a one-hour flight" [8].
If the risk of infection in a 2 hr flight is 1/4300 (as
calculated by Barnett [3]), for a 10
hr flight it is 5/4300 i.e. about 0.116% according to Wu’s logic.
Exponential:
a) If the risk is constant per unit time,
then the risk for 1 hr is:
(1/4300)1/2 = 0.0152; compared to: 1/4300 = 2.32x10-4
The
chance of not getting infected in 1 hr is 1
- 0.0152 = 0.9847.
So
the chance of not getting infected in 10 hrs is: (0.9847)10 =
0.8571.
So
the chance of getting infected in 10 hrs is 1 – 0.8571 = 0.1429.
That
is, about 14%.
This method is wrong because the risk
for 1 hr has to be less than the risk for 2 hrs!
b)
For 1 hr:
[1 –
(1/4300)]1/2 = 0.99988
For
10 hrs: 1 - (0.99988)10 = 0.00116
i.e. 0.116%
(Whereas:
For 5 hrs: 1 - (0.99988)5 = 0.0006
i.e. 0.06%
So
this method is correct).
This
gives the same result as Wu’s additive method. Probably because of the binomial
approximation:
1 -
(1 - t)n @ nt.
Usually
the condition for the approximation to be valid is: t <<
1.
In this
case, when n can be very large: nt << 1 (n is large when t is given
as a rate per minute instead of a rate per hour).
Recent
CDC guidelines suggest that a total of 15 minutes exposure in a day to a
virus-infected person is enough to cause infection; it need not be a continuous
15 minute exposure to an infected person [9].
The
critical viral load for infection is still not known. But since the virus can
survive up to 9 hours on human skin [10],
and aerosols from an infected person are being continuously emitted, the
‘additive’ model seems less likely to be correct than the constant rate
(exponential) model.
c)
Hertzberg
[11] has used the risk of
infection as 1.8% per minute on the basis of an incident in 1977 in which
38 out of 54 passengers stuck in a plane on the tarmac for 4.5 hrs without air
circulation were infected with an influenza-like illness (ILI). However, bio-mathematician
Howard Weiss (Penn State Univ) – and a co-author - has criticized this
study by saying that nobody knows these probabilities for SARS or covid-19, and
that assuming an omnidirectional flow of virions is a crude assumption.
Nevertheless, it is interesting that using this number for a 10 hr flight gives
the risk of infection as: 1 – (1 – 0.018)600 = 0.99998, or about 100%.
Instead if we assume 1.8% per hour, then we get: 1 – (1 – 0.018)10
= 0.17, or 17%. Hertzberg does mention that the number 1.8% per minute is
‘conservatively high’ and based on an incident when the aircraft ventilation
system was off (the plane was on the ground).
Figure
from [12a]: The
airflow in the plane is also designed to keep any droplets spewed by a passenger
from floating around the cabin. The air flows from the top of a passenger’s
head and is collected down by their feet, which keeps whatever we exhale from
spreading too far horizontally.
d)
A
study (partly funded by airlines) [12a,
12b,14] showed that it would take 54 flight hours for an individual sitting
next to an infected individual to get the assumed infectious viral dose. The
assumed dose has not been confirmed by virologist because it is not known for
covid-19. “Mannequins with and without face masks sat in various seats on
the aircraft while fluorescent tracer particles were released at intervals of
two seconds to simulate breathing for a minute during ground and in-flight
tests. Real-time fluorescent particle sensors were placed throughout the
aircraft at the breathing zone of passengers to measure concentration over
time,” in a Boeing 767. The other issue is that aerosols in covid-19 may not be
the same as the tracer particles used. Also: the researchers assumed only
one person aboard the plane was infected, that everyone was wearing masks at
all times, and that the infected passenger (mannequn!) never turned their
head and sat facing forward for the entire flight.
Anyway, just for fun assume 1.8% per hour, with 54 hrs:
1 – (1 – 0.018)54 = 0.625, i.e. 62.5% chance of
getting infected.
e)
But Barnett [3b],
estimated the transmission risk on different flights per minute in the range of
0.0005 to 0.05.
(0.9995)120 =
0.942: for 2 hours; low transmission
risk t: 6% (Tel Aviv-Frankfurt)
(0.95)120 = 0.0021: for 2 hours; high transmission
risk t: 99.8% (Vietnam flight)
Assuming the Tel Aviv for 10 hours:
1 – (0.9995)600 = 0.259. That is: 26%.
Assume masking reduces transmission risk by 50% from 0.05 to
0.025.
Then: 1 - (0.975)600 @ 1
How does one explain the range of t?
An IHME study [12] by
Mokdad was a meta-analysis of the effect of mask wearing. Mokdad concluded that
if 95% of people work masks this could avert 30% of the deaths. This is based
on a particular case of prevalence in the U.S. where the degree of transmission
goes down, decreasing the R0. IHME made a similar prediction for
India that 95% mask use could prevent 200,000 deaths [13]. This analysis is disputed by others (such as Natalie Dean [12]), who doubt whether the model
captures real world effects. However, it is clear that it does not account for
the variation in transmission coefficient (0.0005 to 0.05!), which can only be
the result of variations in prevalence in the group of actual flyers and in
their behavior (social distancing, hand-washing etc.).
Using Wu’s additive method for 10 hrs: (600)(0.0005) = 0.3
Most likely risk for 10 hrs: t =
1/(3,900) for a full-flight of 2 hrs with everyone wearing masks [3b]. So for 5 hrs, the risk of
infection is: 5/(3,900) or 0.13%.
Note: this risk (as pointed out
by Barnett) depends upon the assumed prevalence. The risk may go down if the
prevalence is lower than the assumed value, and go up if it is higher.
The TRANSCOM report [14]
- not peer reviewed - uses the figure of 4,000 virions/hr, but
acknowledges the fact that this is only indicative. Another study by Jianxin et
al [15] is quoted by them as
mentioning a figure of 1,03,000 to 22,500,000 viral particles/hr produced by
infected individuals in exhaled breath by 14 out of 52 patients – but the
remainder had no detectable virus in their breath. Further, these are particles
of viral copy numbers derived from RT-PCR – but not all of these viral copies
are infectious, and the ratio of infectious copies to total viral copies may be
as low as 1/300 [14]. This becomes a
range of 350-75,000 virus/hr (infectious).
The study by the airlines (Airbus, Boeing & Embraer) [16] said that the HEPA filters and
airflow patterns in an airplane meant that 1 foot of separation on an airplane
corresponds to 6 feet of separation on the ground in open air.
Sophie Bushwick [17]
points out that the TRANSCOM study [14]
assumes that passengers are always wearing surgical masks, whereas people often
take them off for meals and talking, and does not account for movement within
the plane. She also quotes two peer-reviewed studies by Sebastian Hoehl said: “An
airplane cabin is probably one of the most secure conditions you can be in“ [18,19].
In the NEJM letter [18] Hoehl et al found 2 out of 114 patients tested positive by
RT-PCR (1.8%) and cell culture indicated potential infectivity, in a German air
force evacuation of mostly German passengers from Wuhan. In the JAMA study [19], Hoehl et al looked at passengers
in a 4 hr 40 min flight from Tel Aviv to Frankfurt and found that 7 out of 24
members of a tourist group tested positive by RT-PCR and concluded that there
were likely 7 index cases and two potential in-flight covid-19 transmissions,
that may also have occurred before the flight, but both infected passengers sat
within two rows of an index case. No one wore masks, so this low risk could
have been further reduced had all passengers worn masks.
A 2011 study [20] by
A.Ruth Foxwell et al found a 3.6% increased risk of contracting H1N1 from a
symptomatic passenger within two rows in a plane, and a 7.7% risk if the index
case was even closer: 2 seats in front, 2 seats behind or 2 seats either side.
Is this risk per hour? Hertzberg [11] assumed 1.8% per minute!
The size of the covid-19 virus is thought to be between 0.06 and
1.4 microns [21], whereas the HEPA
filters in the plane remove 99.97% of particles above the size of 0.3 microns.
What about particles smaller than 0.3 microns? Not clear, but presumably a
lower percentage.
Khanh et al [22]
studied a few in-flight transmission clusters during a long-haul (10 hours) commercial
flight from London to Hanoi in March 2020. 16 persons were later found to be
infected, of whom 12 had been sitting in business class with just one
symptomatic person. Seating proximity to an infected person was strongly
correlated with risk of infection risk (risk ratio 7.3: 11 persons (out of 12
infected, 92%) were sitting within 2 rows of the index case, and just 1 (out of
12, 13%) more than 2 rows away. The most likely mode of infection were aerosol
or droplet transmission from the index case. Face masks were not recommended,
or widely used, in March 2020 – and the authors say there is no data on whether
they were used in this flight.
One study (quoted in [23])
suggests that “infrared thermal image scanners for mass screening of travellers
at airport have a specificity of 71% and sensitivity of 86% to detect fever,
but there are variations depending on where the camera is positioned, which
part of the body is being scanned, and other environmental and individual
factors that can affect the precision of these thermal scanners.” Neither the
sensitivity nor the specificity seems to be particularly high! This probably
explains why no one relies exclusively on IR thermal scanners, apart from the
additional problem that many patients infected with covid-19 get a fever for a
limited time period – or not at all.
Currently airlines require a negative RT-PCR test conducted
within 72 hours of travel. This certainly would help reduce the numbers of
escapees, but would not bring it down to zero because even RT-PCR tests do not
have 100% sensitivity, and one might get infected at any point after the test
was taken. Still, this is a useful step taken by airlines and public health authorities.
Lastly, one must mention a recent Korean study by Kwon et al [24] of infection that occurred in a
restaurant. The index case was 6.5 metres away and the person who got infected
was exposed for just 5 minutes, as determined from CCTV footage. The infection
happened because the airflow due to the AC system happened to direct the
viruses from the index case to the person who got infected. It is not clear if
the index case was coughing, sneezing or speaking in a loud voice. Kwon et al
quote the CFD study by Dbouk and Drikakis [25]
to the effect that droplets could travel 6 metres with a 4 km/hr wind, whereas
they measured the airflow as between 3.6-4.3 km/hr.
If we take the lower limit of 1,03,000 viral particles/hr [15] mentioned above, and the 5 mins
exposure time [24], the critical viral load for infection may be as low as 8,600 viral particles. Silcott [14] states that the literature has
values as low as 300 and as high as several thousand as the infectious dose
(compared with 280 for SARS-COV-1).
Bottomline:
Barnett [6] has
calculated the risk of infection as 1/4,300 for a full flight (middle seat
occupied) that lasts 2 hours in a high prevalence area (1 positive case in
6,500 per day). He calculated 1/4,300 based on an average of the per capita rate
of infection per week in Texas (high, 1/184) and New York (low, 1/1,000) [3a] as 1/310.
Wu pointed out that the risk for a 10 hour flight is 5X greater
(since the risk is low, using the binomial approximation, as above).
Barnett added that in low prevalence areas (1 positive case in
60,000 per day) the risk is even lower (~1/40,000).
Barnett also has assumed the probability of infection as a function of the
distance d between the index case and the infectee is given by:
T = 0.13 exp(- 0.69d)
based on a meta-analysis by Chu et al [26].
Barnett very rightly emphasizes that these are only estimates,
because of the large number of assumptions made in the calculations.
Currently in the US (25th December 2020), the number
of cases every day is roughly 200,000 based on worldometers data [27]. On a per capita basis, that is
about 1/1650 per day. This is about 4X of the value quoted above (1 in 6,500
per day). Thus the risk is about 5/4,300. Take a 10 hour flight, and the risk
is 20/4,300 or about 1/215 or about 0.45%.
The TRANSCOM study [14]
estimated 54 hours of flight time to reach as infectious dose. Just as a check
we can now calculate with current US covid-19 numbers of 200,000 per day:
1 – {1 – [4/4300]}54 = 1 - 0.95 = 0.05
I.e. a 5% chance of getting covid. Note that the TRANSCOM study
predicts a higher risk (maybe 95%?) because an aerosol particle generator
(proxy for an infected person) is definitely present, unlike possibly
present as in Barnett’s calculations [3].
My conclusion is that today a 10 hr flight from the U.S. would
give you about a 0.45% chance of catching covid in-flight.
References:
1. D.Freedman & A.Wilder-Smith J.Travel
Med. doi: 10.1093/jtm/taaa178 (18th Sep.2020)
3. a). Arnold Barnett & doi:
https://doi.org/10.1101/2020.07.02.20143826.this version posted August 2, 2020
b)
Arnold Barnett & Keith Fleming doi: https://doi.org/10.1101/2020.07.02.20143826 22nd Oct.2020
6. Michael LePage https://www.newscientist.com/article/2252152-how-likely-are-you-to-be-infected-by-the-coronavirus-on-a-flight/
7. Tamara Hardingham-Gill https://edition.cnn.com/travel/article/odds-catching-covid-19-flight-wellness-scn/index.html
8. Laurence Frost https://in.mobile.reuters.com/article/amp/idUSKBN27411C
10.
https://www.livescience.com/amp/coronavirus-survives-9-hours-on-skin.html
11. V.S.Hertzberg
et al PNAS 115 (2018) 3263-67
12. a) Julia
Belluz & Brian Resnick: https://www.vox.com/21525068/covid-19-airplane-risk-coronavirus-pandemic-airports
b) https://www.ustranscom.mil/cmd/panewsreader.cfm?ID=C0EC1D60-CB57-C6ED-90DEDA305CE7459D
14. David Silcott et al ,”TRANSCOM/AMC Commercial Aircraft Cabin
Aerosol Dispersion Tests” (final TRANSCOM report).
15. Jianxin Ma et al Clinical Infectious Diseases, 28Aug.2020, ciaa1283, https://doi.org/10.1093/cid/ciaa1283
16. Adam Rogers in Wired (18th
Nov.2020) https://www.wired.com/story/can-you-get-covid-19-on-an-airplane-yeah-probably/
17. Sophie Bushwick et al Sci.Am. (19thNov.2020) https://www.scientificamerican.com/article/evaluating-covid-risk-on-planes-trains-and-automobiles2/
18.Sebastian Hoehl et al NEJM letter 382 (26th
Mar.2020)
19. Sebastian Hoehl et al (18th Aug.2020) JAMA
Network Open. 2020;3(8):e2018044. doi:10.1001/jamanetworkopen.2020.18044
20.A.Ruth Foxwell et al EID 17 (2011) 1188-94
21. Sandee LaMotte CNN, 16th Dec.2020
https://edition.cnn.com/travel/article/air-travel-risk-covid-19-wellness/index.html
22.N.C.Khanh et al EID 26 (Nov.2020) 2617
23. Aisha Khatib et al J.Travel Medicine (2020) doi:
10.1093/jtm/taaa212
24. K.-S.Kwon et
al J Korean Med Sci. 2020 Nov
30;35(46):e415
https://doi.org/10.3346/jkms.2020.35.e415
25. T.Dbouk & D.Drikakis Phys. Fluids 32, 053310 (2020);
https://doi.org/10.1063/5.0011960
26. D.Chu et al The
Lancet, 395(10242), pp. 1973-1987, June 27, 2020
https://doi.org/10.1016/S0140-6736(20)31142-9
27. https://www.worldometers.info/coronavirus/country/us/
No comments:
Post a Comment