Feasibility of controlling COVID-19 outbreaks by isolation of cases and contacts

Abstract

Background
Isolation of cases and contact tracing is used to control outbreaks of infectious diseases, and has been used for coronavirus disease 2019 (COVID-19). Whether this strategy will achieve control depends on characteristics of both the pathogen and the response. Here we use a mathematical model to assess if isolation and contact tracing are able to control onwards transmission from imported cases of COVID-19.

Methods
We developed a stochastic transmission model, parameterised to the COVID-19 outbreak. We used the model to quantify the potential effectiveness of contact tracing and isolation of cases at controlling a severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2)-like pathogen. We considered scenarios that varied in the number of initial cases, the basic reproduction number (R0), the delay from symptom onset to isolation, the probability that contacts were traced, the proportion of transmission that occurred before symptom onset, and the proportion of subclinical infections. We assumed isolation prevented all further transmission in the model. Outbreaks were deemed controlled if transmission ended within 12 weeks or before 5000 cases in total. We measured the success of controlling outbreaks using isolation and contact tracing, and quantified the weekly maximum number of cases traced to measure feasibility of public health effort.

Findings
Simulated outbreaks starting with five initial cases, an R0 of 1·5, and 0% transmission before symptom onset could be controlled even with low contact tracing probability; however, the probability of controlling an outbreak decreased with the number of initial cases, when R0 was 2·5 or 3·5 and with more transmission before symptom onset. Across different initial numbers of cases, the majority of scenarios with an R0 of 1·5 were controllable with less than 50% of contacts successfully traced. To control the majority of outbreaks, for R0 of 2·5 more than 70% of contacts had to be traced, and for an R0 of 3·5 more than 90% of contacts had to be traced. The delay between symptom onset and isolation had the largest role in determining whether an outbreak was controllable when R0 was 1·5. For R0 values of 2·5 or 3·5, if there were 40 initial cases, contact tracing and isolation were only potentially feasible when less than 1% of transmission occurred before symptom onset.

Interpretation
In most scenarios, highly effective contact tracing and case isolation is enough to control a new outbreak of COVID-19 within 3 months. The probability of control decreases with long delays from symptom onset to isolation, fewer cases ascertained by contact tracing, and increasing transmission before symptoms. This model can be modified to reflect updated transmission characteristics and more specific definitions of outbreak control to assess the potential success of local response efforts.

Funding
Wellcome Trust, Global Challenges Research Fund, and Health Data Research UK.

Full text

www.thelancet.com/lancetgh Published online February 28, 2020 https://doi.org/10.1016/S2214-109X(20)30074-7
1
Articles
Lancet Glob Health 2020
Published Online
February 28, 2020
https://doi.org/10.1016/
S2214-109X(20)30074-7
*Contributed equally
†Contributed equally
Centre for the Mathematical
Modelling of Infectious
Diseases, Department of
Infectious Disease
Epidemiology, London School
of Hygiene & Tropical Medicine,
London, UK (J Hellewell PhD,
S Abbott PhD, A Gimma MSc,
N I Bosse BSc, C I Jarvis PhD,
T W Russell PhD,
J D Munday MSc,
A J Kucharski PhD,
Prof W J Edmunds PhD,
S Funk PhD, R M Eggo PhD)
Correspondence to:
Dr Rosalind M Eggo, Department
of Infectious Disease
Epidemiology, London School of
Hygiene & Tropical Medicine,
London, UK
r.eggo@lshtm.ac.uk
Feasibility of controlling COVID-19 outbreaks by isolation of
cases and contacts
Joel Hellewell, Sam Abbott*, Amy Gimma*, Nikos I Bosse, Christopher I Jarvis, Timothy W Russell, James D Munday, Adam J Kucharski,
W John Edmunds, Centre for the Mathematical Modelling of Infectious Diseases COVID-19 Working Group, Sebastian Funk†,
Rosalind M Eggo†
Summary
Background Isolation of cases and contact tracing is used to control outbreaks of infectious diseases, and has been
used for coronavirus disease 2019 (COVID-19). Whether this strategy will achieve control depends on characteristics
of both the pathogen and the response. Here we use a mathematical model to assess if isolation and contact tracing
are able to control onwards transmission from imported cases of COVID-19.
Methods We developed a stochastic transmission model, parameterised to the COVID-19 outbreak. We used the
model to quantify the potential eectiveness of contact tracing and isolation of cases at controlling a severe acute
respiratory syndrome coronavirus 2 (SARS-CoV-2)-like pathogen. We considered scenarios that varied in the number
of initial cases, the basic reproduction number (R0), the delay from symptom onset to isolation, the probability that
contacts were traced, the proportion of transmission that occurred before symptom onset, and the proportion of
subclinical infections. We assumed isolation prevented all further transmission in the model. Outbreaks were deemed
controlled if transmission ended within 12 weeks or before 5000 cases in total. We measured the success of controlling
outbreaks using isolation and contact tracing, and quantified the weekly maximum number of cases traced to measure
feasibility of public health eort.
Findings Simulated outbreaks starting with five initial cases, an R0 of 1·5, and 0% transmission before symptom onset
could be controlled even with low contact tracing probability; however, the probability of controlling an outbreak
decreased with the number of initial cases, when R0 was 2·5 or 3·5 and with more transmission before symptom
onset. Across dierent initial numbers of cases, the majority of scenarios with an R0 of 1·5 were controllable with less
than 50% of contacts successfully traced. To control the majority of outbreaks, for R0 of 2·5 more than 70% of contacts
had to be traced, and for an R0 of 3·5 more than 90% of contacts had to be traced. The delay between symptom onset
and isolation had the largest role in determining whether an outbreak was controllable when R0 was 1·5. For R0 values
of 2·5 or 3·5, if there were 40 initial cases, contact tracing and isolation were only potentially feasible when less than
1% of transmission occurred before symptom onset.
Interpretation In most scenarios, highly eective contact tracing and case isolation is enough to control a new
outbreak of COVID-19 within 3 months. The probability of control decreases with long delays from symptom onset to
isolation, fewer cases ascertained by contact tracing, and increasing transmission before symptoms. This model can
be modified to reflect updated transmission characteristics and more specific definitions of outbreak control to assess
the potential success of local response eorts.
Funding Wellcome Trust, Global Challenges Research Fund, and Health Data Research UK.
Copyright © 2020 The Author(s). Published by Elsevier Ltd. This is an Open Access article under the CC BY-NC-ND
4.0 license.
Introduction
As of Feb 5, 2020, more than 24 550 cases of coronavirus
disease 2019 (COVID-19) had been confirmed, including
more than 190 cases outside of China, and more than
490 reported deaths globally.1 Control measures have
been implemented within China to try to contain the
outbreak.2 As people with the infection arrive in
countries or areas without ongoing transmission, eorts
are being made to halt transmission, and prevent
potential outbreaks.3,4 Isolation of confirmed and
suspected cases, and identification of contacts are a
crucial part of these control eorts; however, whether
these eorts will achieve control of transmission of
COVID-19 is unclear.
Isolation of cases and contact tracing becomes less
eective if infectiousness begins before the onset of
symptoms.5,6 For example, the severe acute respiratory
syndrome (SARS) outbreak that began in southern China
in 2003, was eventually able to be controlled through
tracing contacts of suspected cases and isolating
confirmed cases because the majority of transmission
occurred after symptom onset.7 These interventions also
play a major role in response to outbreaks where onset of
symptoms and infectiousness are concurrent—eg, Ebola

Articles
2
www.thelancet.com/lancetgh Published online February 28, 2020 https://doi.org/10.1016/S2214-109X(20)30074-7
virus disease,8,9 Middle East respiratory syndrome
(MERS),10,11 and many other infections.12,13
The eectiveness of isolation and contact-tracing
methods hinges on two key epidemiological parameters:
the number of secondary infections generated by each
new infection and the proportion of transmission that
occurs before symptom onset.5 In addition, successful
contact tracing and reducing the delay between
symptom onset and isolation are crucial, because,
during this time, cases remain in the community where
they can infect others until isolation.6,14 Transmission
before symptom onset could only be prevented by
tracing contacts of confirmed cases and testing (and
quarantining) those contacts. Cases that do not seek
care, potentially because of subclinical infection, are a
further challenge to control.
If COVID-19 can be controlled by isolation and contact
tracing, then public health eorts should be focused on
this strategy; however, if this is not enough to control
outbreaks, then additional resources might be needed
for additional interventions. Several key characteristics
of the transmissibility and natural history of COVID-19
are currently unknown—eg, whether transmission can
occur before symptom onset. Therefore, we explored a
range of epidemiological scenarios that represent
potential transmission properties based on current
information about COVID-19 transmission. We assessed
the ability of isolation and contact tracing to control
disease outbreaks in areas without widespread
transmission using a mathematical model.6,15–17 By
varying the ecacy of contact-tracing eorts, the size of
the outbreak when detected, and the promptness of
isolation after symptom onset, we show how viable it is
for countries at risk of imported cases to use contact
tracing and isolation as a containment strategy.
Methods
Model structure
We implemented a branching process model, in which
the number of potential secondary cases produced by each
individual is drawn from a negative binomial distribution
with a mean equal to the reproduction number, and
heterogeneity in the number of new infections produced
by each individual.6,15,17–19 Each potential new infection was
assigned a time of infection drawn from the serial interval
distribution. Secondary cases were only created if the
person with the infection had not been isolated by the
time of infection. As an example (figure 1), a person
infected with the virus could potentially produce three
secondary infections (because three is drawn from the
negative binomial distribution), but only two trans-
missions might occur before the case is isolated. Thus, in
the model, a reduced delay from onset to isolation would
reduce the average number of secondary cases.
We initialised the branching process with five, 20, or
40 cases to represent a newly detected outbreak of varying
size. Initial symptomatic cases were then isolated after
symptom onset with a delay drawn from the onset-to-
isolation distribution (table). Isolation was assumed to be
100% eective at preventing further transmission;
therefore, in the model, failure to control the outbreak
resulted from the incomplete contact tracing and the
delays in isolating cases rather than the inability of
isolation to prevent further transmission. Either 100% or
90% of cases became symptomatic, and all symptomatic
cases were eventually reported.
Research in context
Evidence before this study
Contact tracing and isolation of cases is a common intervention
for controlling infectious disease outbreaks. It can be effective
but might require intensive public health effort and
cooperation to effectively reach and monitor all contacts.
Previous work has shown that when the pathogen has
infectiousness before symptom onset, control of outbreaks
using contact tracing and isolation is more challenging. Further
introduction of coronavirus disease 2019 (COVID-19) to new
territories is likely in the coming days and weeks, and
interventions to prevent an outbreak following these
introductions are a key mitigating strategy. Current planning is
focused on tracing of contacts of introduced cases, and rapid
isolation. These methods have been used previously for other
novel outbreaks, but it is not clear if they will be effective for
COVID-19.
Added value of this study
We use a mathematical model to assess the feasibility of
contact tracing and case isolation to control outbreaks of
COVID-19. We used disease transmission characteristics specific
to the pathogen and give the best available evidence if contact
tracing and isolation can achieve control of outbreaks.
We simulated new outbreaks starting from 5, 20, or 40
introduced cases. Contact tracing and isolation might not
contain outbreaks of COVID-19 unless very high levels of
contact tracing are achieved. Even in this case, if there is
subclinical transmission, or a high fraction of transmission
before onset of symptoms, this strategy might not achieve
control within 3 months.
Implications of all the available evidence
The effectiveness of isolation of cases and contacts to control
outbreaks of COVID-19 depends on the precise characteristics of
transmission, which remain unclear at the present time. Using
the current best understanding, around 80% of symptomatic
contacts must be traced and isolated to control over 80% of
outbreaks in the model. Future research on the transmission
characteristics could improve precision on control estimates.

Articles
www.thelancet.com/lancetgh Published online February 28, 2020 https://doi.org/10.1016/S2214-109X(20)30074-7
3
Each newly infected case was identified through
contact tracing with probability ρ. Secondary cases that
had been traced were isolated immediately on becoming
symptomatic. Cases that were missed by contact tracing
(probability 1–ρ) were isolated when they became
symptomatic, with a delay drawn from the onset-to-
isolation distribution.
In addition, each case had an independent probability
of being subclinical, and was therefore not detected
either by self-report or contact tracing. New secondary
cases caused by a subclinical case were missed by contact
tracing and could only be isolated on the basis of
symptoms. The model included isolation of symptomatic
individuals only—ie, no quarantine, so isolation could
not prevent transmission before symptom onset. In the
model, subclinical cases were never isolated, whereas
symptomatic cases might transmit before symptoms
appear, but were eventually isolated. Quarantining
contacts of cases (ie, individuals who are not yet
symptomatic, and might not be infected) requires a
considerable investment in public health resources, and
has not been widely implemented for all contacts of
cases.3 However, some countries have adopted a
quarantine or self-quarantine policy for airline travellers
who have returned from countries with confirmed
COVID-19 transmission.23
Transmission scenarios
We ran 1000 simulations of each combination of the
proportion of transmission before symptom onset (R0),
onset-to-isolation delay, the number of initial cases, and
the probability that a contact was traced (table).
We explored two scenarios of delay (short and long)
between symptom onset and isolation (figure 2). The
short delay was estimated during the late stages of the
2003 SARS outbreak in Singapore,18 and the long delay
was an empirical distribution calculated from the early
phase of the COVID-19 outbreak in Wuhan.23 We varied
the percentage of contacts traced from 0% to 100%,
at 20% intervals, to quantify the eectiveness of contact
tracing.
The incubation period for each case was drawn from a
Weibull distribution. A corresponding serial interval for
each case was then drawn from a skewed normal
Figure 1: Example of the simulated process that starts with person A being infected
After an incubation period, person A shows symptoms and is isolated at a time drawn from the delay distribution table. A draw from the negative binomial
distribution with mean reproduction number (R0) and distribution parameter determines how many people person A potentially infects. For each of those, a serial
interval is drawn. Two of these exposures occur before the time person A is isolated. With probability ρ, each contact is traced; with probability 1–ρ they are missed by
contact tracing. Person B is successfully traced, which means that they will be isolated without delay when they develop symptoms. They could, however, still infect
others before they are isolated. Person C is missed by contact tracing. This means that they are only detected if and when symptomatic, and are isolated after a delay
from symptom onset. Because person C was not traced, they infected two more people (E and F), in addition to person D, than if they had been isolated at symptom
onset. A version with subclinical transmission is given in the appendix (p 12).
IsolatedInfects
person B
Infects
person C
Infects
person E
Infects
person F
Infects
person D
SymptomsInfected
Symptoms
+
isolated
Incubation Delay from onset to isolation
No infection
(isolated)
Serial intervals
Incubation
Serial intervals
No infection No infection
Incubation Delay from onset to isolation
Traced (ρ)
Not traced (1–ρ)
Serial intervals
Symptoms Isolated
AB
C
x
x
x
D
E
F
Person A
Person B
Person C
Infected
Infected
Value Reference
Sampled
Delay from onset to isolation (short) 3·43 days (2·02–5·23) Donnelly et al20
Delay from onset to isolation (long) 8·09 days (5·52–10·93) Li et al21
Incubation period 5·8 days (2·6) Backer et al22
Serial interval Incubation period (2) Assumed
Fixed
Initial cases 5, 20, and 40 Public Health England11 and
Klinkenberg and colleagues14
Percentage of contacts traced 0%, 20%, 40%, 60%, 80%, 100% Tested
Reproduction number (R0; low, central,
high estimate)
1·5, 2·5, 3·5 Kucharski et al17 and Imai et al18
Overdispersion in R0 (SARS-like,
influenza-like)
0·16 Lloyd-Smith et al19
R0 after isolation 0 Assumed
Cases isolated once identified 100% Assumed
Isolation effectiveness 100% Assumed
Subclinical infection percentage 0%, 10% Tested
Data are median (IQR) or mean (SD), n, or %. Sampled values are probabilistically sampled during the simulation, and
fixed values remain constant during the simulation. The mean of the short and long delays are 3·83 and 9·1,
respectively. SARS=severe acute respiratory syndrome.
Table: Parameter values for the model
See Online for appendix

Articles
4
www.thelancet.com/lancetgh Published online February 28, 2020 https://doi.org/10.1016/S2214-109X(20)30074-7
distribution with the mean parameter of the distribution
set to the incubation period for that case, an SD of 2, and
a skew parameter chosen such that a set proportion of
serial intervals were shorter than the incubation period
(meaning that a set proportion of transmission happened
before symptom onset; figure 2). This sampling approach
ensured that the serial interval and incubation period for
each case was correlated, and prevented biologically
implausible scenarios where a case could develop
symptoms soon after exposure, but not become infectious
until very late after exposure and vice versa.
There are many estimates of the reproduction
number for the early phase of the COVID-19 outbreak
Figure 2: Probability distributions used in simulations
(A) The short and long delay distributions between the onset of symptoms and
isolation (mean marked by line). Parameter values and references are given in
the table. (B) The incubation distribution estimate fitted to data from the
Wuhan outbreak by Backer and colleagues.22 (C) An example of the method used
to sample the serial interval for a case that has an incubation period of 5 days.
Each case has an incubation period drawn from the distribution in (B), their
serial interval is then drawn from a skewed normal distribution with the mean
set to the incubation period of the case. In (C), the incubation period was 5 days.
The skew parameter of the skewed normal distribution controls the proportion
of transmission that occurs before symptom onset; the three scenarios explored
are less than 1%, 15%, and 30% of transmission before onset.
Figure 3: Effect of isolation and contact tracing on controlling outbreaks and
on the effective reproduction number
(A) The percentage of outbreaks that are controlled for scenarios with varying
reproduction number (R0), at each value of contacts traced. The baseline scenario
is R0 of 2·5, 20 initial cases, a short delay to isolation, 15% of transmission before
symptom onset, and 0% subclinical infection. A simulated outbreak is defined as
controlled if there are no cases between weeks 12 and 16 after the initial cases.
Other scenarios are presented in the appendix (p 2). (B) Effective reproduction
number in the presence of case isolation and contact tracing. Median, and 50%
and 95% intervals are shown.
Short delay
Long delay
<1%
15%
30%
Proportion of transmission
before symptoms
0
5
10
15
20
100
Probability density (%)
A
0
5
10
15
20
100
Probability density (%)
B
0 5 10 15
0
10
20
30
40
100
Probability density (%)
Time since infection (days)
C
0
20
40
60
80
100
Simulated outbreaks controlled (%)
A
0 20 40 60 80 100
0
0·5
1·0
3·0
4·5
4·0
3·5
2·5
2·0
1·5
5·0
Effective reproduction number (n)
Contacts traced (%)
B
1·5
2·5
3·5
Reproduction number (n)

Articles
www.thelancet.com/lancetgh Published online February 28, 2020 https://doi.org/10.1016/S2214-109X(20)30074-7
5
in Wuhan, China,15,17,18,21,24–28 and therefore we used the
values 1·5, 2·5, and 3·5, which span most of the
range of current estimates (table). We used the
secondary case distribution from the 2003 SARS
outbreak,19 and tested the effect of lower heterogeneity
in the number of secondary cases29 as a sensitivity
analysis (appendix pp 2–5). We calculated the effective
reproduction number (Reff) of the simulation as the
average number of secondary cases produced by each
infected person in the presence of isolation and
contact tracing. We present results in relation to the
baseline scenario of R0 of 2·5,21 20 initial cases, a short
delay to isolation,20 15% of transmission before
symptom onset,30 and 0% subclinical infection.31
Values of the natural history represent the current
best understanding of COVID-19 transmission, and
we used 20 index cases and a short delay to isolation
to represent a relatively large influx into a setting of
high awareness of possible infection.23
Definition of outbreak control
Outbreak control was defined as no new infections
between 12 and 16 weeks after the initial cases. Outbreaks
that reached 5000 cumulative cases were assumed to be
too large to control within 12–16 weeks, and were
categorised as uncontrolled outbreaks. Based on this
definition, we reported the probability that an outbreak
of a severe acute respiratory syndrome coronavirus 2-like
pathogen would be controlled within 12 weeks for each
scenario, assuming that the basic reproduction number
remained constant and no other interventions were
implemented.
The probability that an outbreak is controlled gives a
one-dimensional understanding of the diculty of
achieving control, because the model placed no
constraints on the number of cases and contacts that
could be traced and isolated. In reality, the feasibility of
contact tracing and isolation is likely to be determined
both by the probability of achieving control, and the
resources needed to trace and isolate infected cases.32 We
therefore reported the weekly maximum number of
cases undergoing contact tracing and isolation for each
scenario that resulted in outbreak control. New cases
require their contacts to be traced, and if these numbers
are high, it can overwhelm the contact-tracing system
and aect the quality of the contact-tracing eort.33 It is
likely that the upper limit on contacts to trace varies from
country to country.
All code is available as an R package.
Role of the funding source
The funders of the study had no role in study design,
data collection, data analysis, data interpretation, writing
of the Article, or the decision to submit for publication.
All authors had full access to all the data in the study and
were responsible for the decision to submit the Article
for publication.
Results
To achieve control of 90% of outbreaks, 80% of contacts
needed to be traced and isolated for scenarios with a
reproduction number of 2·5 (figure 3). The probability of
control was higher at all levels of contact tracing when
the reproduction number was 1·5, and fell rapidly for a
reproduction number of 3·5. At a reproduction number
of 1·5, the eect of isolation was coupled with the chance
of stochastic extinction resulting from overdispersion,19
which is why some outbreaks were controlled even at 0%
contacts traced.
Isolation and contact tracing decreased transmission,
as shown by a decrease in the eective reproduction
number (figure 3). When the basic reproduction number
was 1·5, the median estimate rapidly fell below 1, which
indicated that control was probable. For the higher
transmission scenarios, a higher level of contact tracing
was needed to bring the median eective reproduction
number below 1. The eect of isolation without contact
tracing can be seen at 0%, where the eective
reproduction number was lower than the simulated basic
reproduction number because of rapid isolation (and
ceasing transmission) of cases.
The number of initial cases had a large eect on the
probability of achieving control. With five initial cases,
there was a greater than 50% chance of achieving control
in 3 months, even at modest contact-tracing levels
Figure 4: Achieving control of simulated outbreaks under different transmission scenarios
The percentage of outbreaks controlled for the baseline scenario, and varied number of initial cases (A), time from
onset to isolation (B), percentage of transmission before symptoms (C), and proportion of subclinical
(asymptomatic) cases (D). The baseline scenario is a reproduction number (R0) of 2·5, 20 initial cases, a short delay
to isolation, 15% of transmission before symptom onset, and 0% subclinical infection. Results for R0=1·5 and 3·5
are given in the appendix. A simulated outbreak is defined as controlled if there are no cases between
weeks 12 and 16 after the initial cases.
0 20 40 60 80 100
0
20
40
60
80
100
Simulated
outbreaks controlled (%)
Contacts traced (%)
0 20 40 60 80 100
Contacts traced (%)
CTransmission before symptoms DSubclinical infections
0
20
40
60
80
100
Simulated outbreaks controlled (%)
AInitial cases BOnset to isolation delay
5
20
40
Short
Long
<1%
15%
30%
0%
10%
For code see https://github.com/
cmmid/ringbp

Articles
6
www.thelancet.com/lancetgh Published online February 28, 2020 https://doi.org/10.1016/S2214-109X(20)30074-7
(figure 4). More than 40% of these outbreaks were
controlled with no contact tracing because of the
combined eects of isolation of symptomatic cases and
stochastic extinction. The probability of control dropped
as the number of initial cases increased—eg, for 40 initial
cases, 80% contact tracing did not lead to 80% of
simulations controlled within 3 months.
The delay from symptom onset to isolation had a
major role in achieving control of outbreaks (figure 4).
At 80% of contacts traced, the probability of achieving
control fell from 89% to 31%, with a long delay from
onset to isolation. If no transmission occurred before
symptom onset, then the probability of achieving
control was higher for all values of contacts traced
(figure 4). The dierence between 15% and 30% of
transmission before symptoms had a marked eect on
probability to control. We found this eect in all
scenarios tested (appendix p 5). In scenarios in which
only 10% of cases were asymptomatic, the probability
that simulations were controlled by isolation and contact
tracing for all values of contact tracing decreased
(figure 4). For 80% of contacts traced, only 37% of
outbreaks were controlled, compared with 89% without
subclinical infection. These figures show the eect of
changing one model assumption at a time; all
combinations are given in the appendix, in comparison
to the baseline scenario (appendix pp 2–5).
In many scenarios, between 25 and 100 symptomatic
cases occurred in a week at the peak of the simulated
outbreak (figure 5). All of these cases, and their contacts,
would need to be isolated. Large numbers of new cases
can overwhelm isolation facilities, and the more contacts
that need to be traced, the greater the logistical task of
following them up. In the 2014 Ebola epidemic in Liberia,
each case reported between six and 20 contacts,8 and the
number of contacts seen in MERS outbreaks is often
higher than that.10 20 contacts for each of 100 cases means
2000 contacts traced to achieve control. Uncontrolled
outbreaks typically had higher numbers of cases
(appendix p 13). The maximum numbers of weekly cases
Figure 5: The maximum weekly cases requiring contact tracing and isolation in scenarios with 20 index cases that achieved control within 3 months
Scenarios vary by reproduction number and the mean delay from onset to isolation. 15% of transmission occurred before symptom onset, and 0% subclinical
infection. The percentage of simulations that achieved control is shown in the boxplot. This illustrates the potential size of the eventually controlled simulated
outbreaks, which would need to be managed through contact tracing and isolation. *The interval extends out of the plotting region.
R0=1·5 R0=2·5 R0=3·5
0
50
100
150
Maximum weekly cases in controlled outbreaks (n)
83% 98% 100% 100% 18%
42%
87% 100%
10%
40%
99%
100806040
0
50
100
150
Maximum
weekly cases in controlled
outbreaks (n)
Contacts traced (%)
100806040
Contacts traced (%)
100806040
Contacts traced (%)
39% 66% 95% 100%
*
32%
100%
97%
Long isolation delay
Short isolation delay

Articles
www.thelancet.com/lancetgh Published online February 28, 2020 https://doi.org/10.1016/S2214-109X(20)30074-7
7
in figure 5 might appear counterintuitive, because a
lower maximum number of weekly cases is not associated
with higher outbreak control. This occurs because with
better contact tracing it becomes possible to control
outbreaks with higher numbers of weekly cases.
Discussion
We determined conditions in which case isolation,
contact tracing, and preventing transmission by contacts
who are infected would be sucient to control a new
COVID-19 outbreak in the absence of other control
measures. We found that in some plausible scenarios,
case isolation alone would be unlikely to control
transmission within 3 months. Case isolation was more
eective when there was little transmission before
symptom onset and when the delay from symptom onset
to isolation was short. Preventing transmission by
tracing and isolating a larger proportion of contacts,
thereby decreasing the eective reproduction number,
improved the number of scenarios in which control was
likely to be achieved. However, these outbreaks required
a large number of cases to be contact traced and isolated
each week, which is of concern when assessing the
feasibility of this strategy. Subclinical infection markedly
decreased the probability of controlling outbreaks within
3 months.
In scenarios in which the reproduction number
was 2·5, 15% of transmission occurred before symptom
onset, and there was a short delay to isolation, at least
80% of infected contacts needed to be traced and isolated
to give a probability of control of 90% or more. This
scenario echoes other suggestions that highly eective
contact tracing will be necessary to control outbreaks in
other countries.16 In scenarios in which the delay from
onset to isolation was long, similar to the delays in the
early phase of the outbreak in Wuhan, the same contact
tracing success of 80% achieved a probability of con-
taining an outbreak of less than 40%. Higher pre-
symptomatic transmission decreases the probability that
the outbreaks were controlled, under all reproduction
numbers and isolation delay distributions tested.
Our model did not include other control measures
that might decrease the reproduction number and
therefore also increase the probability of achieving
control of an outbreak. At the same time, it assumed
that isolation of cases and contacts is completely
eective, and that all symptomatic cases are eventually
reported. Relaxing these assumptions would decrease
the probability that control is achieved. We also make
the assumption that contact is required for transmission
between two individuals, but transmission via fomites
might be possible. This type of transmission would
make eective contact tracing challenging, and good
respiratory and hand hygiene would be crucial to reduce
this route of transmission, coupled with environmental
decontamination in health-care settings. For this reason,
we used contact-tracing percentage intervals of 20% to
avoid indicating more precision in the corresponding
probability of control than the model could support.
We simplified our model to determine the eect of
contact tracing and isolation on the control of outbreaks
under dierent scenarios of transmission; however, as
more data becomes available, the model can be updated
or tailored to particular public health contexts. The
robustness of control measures is likely to be aected both
by dierences in transmission between countries, but also
by the concurrent number of cases that require contact
tracing in each scenario. Practically, there is likely to be an
upper bound on the number of cases that can be traced,
which might vary by country, and case isolation is likely to
be imperfect.34 We reported the maximum number of
weekly cases during controlled outbreaks, but the capacity
of response eorts might vary. In addition to the number
of contacts, other factors could decrease the percentage of
contacts that can be traced, such as cooperation of the
community with the public health response.
We explored a range of scenarios informed by the latest
evidence on transmission of COVID-19. Similar analyses
using branching models have already been used to
analyse the Wuhan outbreak to find plausible ranges for
the initial exposure event size and the basic reproduction
number.15,18 Our analysis expands on this work by
including infectiousness before the onset of symptoms,
case isolation, explicit modelling of case incubation
periods, and time to infectiousness. A key area of
uncertainty is whether, and for how long, individuals are
infectious before symptom onset, and whether
subclinical infection occurs; both are likely to make the
outbreak harder to control.22 Whether, and how much,
transmission occurs before symptoms is dicult to
quantify. Under-reporting of prodromal symptoms, such
as fatigue and mild fever, is possible; thus, transmission
might not truly be occurring before symptoms, but
before noticeable symptoms. There is evidence of
transmission before onset,30 and so we used 15%.
Increased awareness of prodromal symptoms, and
therefore short delays until isolation—as seen in the
SARS outbreak in Beijing in 200335—would increase
control of outbreaks in our model. If contact tracing
included testing of non-symptomatic contacts, those
contacts could be quarantined without symptoms, which
would decrease transmission in the model. Costs
associated with additional testing might not be possible
in all contexts.
The model could be modified to include some
transmission after isolation (such as in hospitals), which
would decrease the probability of achieving control. In
addition, we defined an outbreak as controlled if it
reached extinction by 3 months, regardless of outbreak
size or number of weekly cases. This definition might be
narrowed where the goal is to keep the overall caseload of
the outbreak low. This might be of concern to local
authorities for reducing the health-care surges, and
might limit geographical spread.

Articles
8
www.thelancet.com/lancetgh Published online February 28, 2020 https://doi.org/10.1016/S2214-109X(20)30074-7
Our study indicates that in most plausible outbreak
scenarios, case isolation and contact tracing alone is
insucient to control outbreaks, and that in some
scenarios even near perfect contact tracing will still be
insucient, and further interventions would be required
to achieve control. Rapid and eective contact tracing can
reduce the initial number of cases, which would make the
outbreak easier to control overall. Eective contact tracing
and isolation could contribute to reducing the overall size
of an outbreak or bringing it under control over a longer
time period.
Contributors
RME conceived the study. JH, AG, SA, WJE, SF, and RME designed the
model. CIJ, TWR, and NIB worked on statistical aspects of the study.
JH, AG, SA, and NIB programmed the model, and, with RME, made the
figures. AJK and JDM consulted on the code. All authors interpreted the
results, contributed to writing the Article, and approved the final version
for submission.
Declaration of interests
We declare no competing interests.
Data sharing
No data were used in this study. The R code for the work is available at
https://github.com/cmmid/ringbp.
Acknowledgments
JH, SA, JDM, and SF were funded by the Wellcome Trust (grant
number 210758/Z/18/Z), AG and CIJ were funded by the Global
Challenges Research Fund (grant number ES/P010873/1), TWR and
AJK were funded by the Wellcome Trust (grant number
206250/Z/17/Z), and RME was funded by HDR UK (grant number
MR/S003975/1). This research was partly funded by the National
Institute for Health Research (NIHR) (16/137/109) using UK aid from
the UK Government to support global health research. The views
expressed in this publication are those of the authors and not
necessarily those of the NIHR or the UK Department of Health and
Social Care. This research was partly funded by the Bill & Melinda
Gates Foundation (INV-003174). This research was also partly funded
by the Global Challenges Research Fund project RECAP managed
through Research Councils UK and Economic and Social Research
Council (ES/P010873/1). We would like to acknowledge (in a
randomised order) the other members of the London School of
Hygiene & Tropical Medicine COVID-19 modelling group, who
contributed to this work: Stefan Flasche, Mark Jit, Nicholas Davies,
Sam Cliord, Billy J Quilty, Yang Liu, Charlie Diamond, Petra Klepac,
and Hamish Gibbs. Their funding sources are as follows:
Stefan Flasche and Sam Cliord (Sir Henry Dale Fellowship [grant
number 208812/Z/17/Z]); Mark Jit, Yang Liu, and Petra Klepac (BMGF
[grant number INV-003174]); Nicholas Davies (NIHR [grant number
HPRU-2012–10096]); Billy J Quilty (grant number NIHR [16/137/109]);
Charlie Diamond and Yang Liu (NIHR [grant number 16/137/109]); and
Hamish Gibbs (Department of Health and Social Care [grant number
ITCRZ 03010]).
References
1 WHO. Novel coronavirus (2019-nCoV) situation report 16. World
Health Organization, 2020. Ahttps://www.who.int/docs/default-
source/coronaviruse/situation-reports/20200205-sitrep-16-ncov.
pdf?sfvrsn=23af287f_2 (accessed Feb 5, 2020).
2 WHO. Novel coronavirus (2019-nCoV) situation report 2. World
Health Organization, 2020. https://www.who.int/docs/default-
source/coronaviruse/situation-reports/20200122-sitrep-2-2019-ncov.
pdf?sfvrsn=4d5bcbca_2 (accessed Jan 22, 2020).
3 European Centre for Disease Prevention and Control. Public
health management of persons having had contact with novel
coronavirus cases in the European Union. European Centre for
Disease Prevention and Control, 2020. https://www.ecdc.europa.
eu/en/publications-data/public-health-management-persons-
having-had-contact-novel-coronavirus-cases (accessed
Feb 2, 2020).
4 Quilty B, Cliord S, CCMID nCoV working group, Flasche S,
Eggo RM. Eectiveness of airport screening at detecting travellers
infected with 2019-nCoV. 2020. https://cmmid.github.io/ncov/
airport_screening_report/airport_screening_preprint_2020_01_30.
pdf (accessed Feb 5, 2020).
5 Fraser C, Riley S, Anderson RM, Ferguson NM. Factors that make
an infectious disease outbreak controllable. Proc Natl Acad Sci USA
2004; 101: 6146–51.
6 Peak CM, Childs LM, Grad YH, Buckee CO. Comparing
nonpharmaceutical interventions for containing emerging
epidemics. Proc Natl Acad Sci USA 2017; 114: 4023–28.
7 Glasser JW, Hupert N, McCauley MM, Hatchett R. Modeling and
public health emergency responses: lessons from SARS. Epidemics
2011; 3: 32–37.
8 Swanson KC, Altare C, Wesseh CS, et al. Contact tracing
performance during the Ebola epidemic in Liberia, 2014–2015.
PLoS Negl Trop Dis 2018; 12: e0006762.
9 WHO. Implementation and management of contact tracing for
Ebola virus disease. World Health Organization. 2015.
http://www.who.int/csr/resources/publications/ebola/contact-
tracing/en/ (accessed Feb 4, 2020).
10 Kang M, Song T, Zhong H, et al. Contact tracing for imported case
of Middle East respiratory syndrome, China, 2015.
Emerging Infect Dis 2016; 22: 9.
11 Public Health England. MERS-CoV close contact algorithm.
Public health investigation and management of close contacts of
Middle East Respiratory Coronavirus (MERS-CoV) cases
(v17 29 January 2019). 2019.https://assets.publishing.service.gov.uk/
government/uploads/system/uploads/attachment_data/file/776218/
MERS-CoV_Close_contacts_algorithm.pdf (accessed Feb 6, 2020).
12 European Centre for Disease Prevention and Control. Risk
assessment guidelines for diseases transmitted on aircraft. Part 2:
Operational guidelines for assisting in the evaluation of risk for
transmission by disease. 2011. https://www.ecdc.europa.eu/sites/
default/files/media/en/publications/Publications/1012_GUI_
RAGIDA_2.pdf (accessed Feb 6, 2020).
13 Hoang TTT, Nguyen VN, Dinh NS, et al. Active contact tracing
beyond the household in multidrug resistant tuberculosis in
Vietnam: a cohort study. BMC Public Health 2019; 19: 241.
14 Klinkenberg D, Fraser C, Heesterbeek H. The eectiveness of
contact tracing in emerging epidemics. PLoS One 2006; 1: e12.
15 Abbott S, Hellewell J, Munday J, CMMID nCoV working group,
Funk S. The transmissibility of novel coronavirus in the early stages
of the 2019–20 outbreak in Wuhan: exploring initial point-source
exposure sizes and durations using scenario analysis.
Wellcome Open Res 2020; 5: 1 7.
16 Imai N, Cori A, Dorigatti I, et al. Report 3: transmissibility of
2019-nCov. 2020. https://www.imperial.ac.uk/media/imperial-
college/medicine/sph/ide/gida-fellowships/Imperial-2019-nCoV-
transmissibility.pdf (accessed Jan 25, 2020).
17 Kucharski AJ, Eggo RM, Watson CH, Camacho A, Funk S,
Edmunds WJ. Eectiveness of ring vaccination as control strategy
for Ebola virus disease. Emerg Infect Dis 2016; 22: 105–08.
18 Imai N, Dorigatti I, Cori A, Riley S, Ferguson NM. Estimating the
potential total number of novel coronavirus cases in Wuhan City,
China. 2020. https://www.imperial.ac.uk/media/imperial-college/
medicine/sph/ide/gida-fellowships/2019-nCoV-outbreak-
report-22-01-2020.pdf (accessed Jan 22, 2020).
19 Lloyd-Smith JO, Schreiber SJ, Kopp PE, Getz WM. Superspreading
and the eect of individual variation on disease emergence. Nature
2005; 438: 355–59.
20 Donnelly CA, Ghani AC, Leung GM, et al. Epidemiological
determinants of spread of causal agent of severe acute respiratory
syndrome in Hong Kong. Lancet 2003; 361: 1761–66.
21 Li Q, Guan X, Wu P, et al. Early transmission dynamics in Wuhan,
China, of novel coronavirus-infected pneumonia. N Engl J Med
2020; published online Jan 29. DOI:10.1056/NEJMoa2001316.
22 Backer JA, Klinkenberg D, Wallinga J. The incubation period of
2019-nCoV infections among travellers from Wuhan, China. medRxiv
2020; published online Jan 28. DOI:10.1101/2020.01.27.20018986
(preprint).
23 BBC News. Coronavirus patient is University of York student.
Feb 1, 2020. https://www.bbc.com/news/health-51337400 (accessed
Feb 19, 2020).

Articles
www.thelancet.com/lancetgh Published online February 28, 2020 https://doi.org/10.1016/S2214-109X(20)30074-7
9
24 Riou J, Althaus CL. Pattern of early human-to-human transmission
of Wuhan 2019 novel coronavirus (2019-nCoV), December 2019 to
January 2020. Eurosurveillance 2020; 25: 2000058.
25 Bedford T, Neher R, Hadfield J, Hodcroft E, Ilcisin M, Müller N.
Genomic analysis of nCoV spread. Situation report 2020-01-30.
2020. https://nextstrain.org/narratives/ncov/sit-rep/2020-01-30
(accessed Jan 30, 2020).
26 Majumder M, Mandl KD. Early transmissibility assessment of
a novel coronavirus in Wuhan, China. Rochester, NY: Social Science
Research Network. 2020 https://papers.ssrn.com/abstract=3524675
(accessed Jan 30, 2020).
27 Read JM, Bridgen JR, Cummings DA, Ho A, Jewell CP. Novel
coronavirus 2019-nCoV: early estimation of epidemiological
parameters and epidemic predictions. medRxiv 2020; published
online Jan 28. DOI:10.1101/2020.01.23.20018549 (preprint).
28 Zhao S, Ran J, Musa SS, et al. Preliminary estimation of the basic
reproduction number of novel coronavirus (2019-nCoV) in China,
from 2019 to 2020: a data-driven analysis in the early phase of the
outbreak. Int J Infect Dis 2020; 92: 214–17.
29 Dugas AF, Jalalpour M, Gel Y, et al. Influenza forecasting with
Google Flu Trends. PLoS One 2013; 8: e56176.
30 Yu P, Zhu J, Zhang Z, Han Y, Huang L. A familial cluster of
infection associated with the 2019 novel coronavirus indicating
potential person-to-person transmission during the incubation
period. J Infect Dis 2020; published online Feb 18. DOI:10.1093/
infdis/jiaa077.
31 Heymann DL, Shindo N. COVID-19: what is next for public health?
Lancet 2020; 395: 542–45.
32 Dhillon RS, Srikrishna D. When is contact tracing not enough to
stop an outbreak? Lancet Infect Dis 2018; 18: 1302–04.
33 Olu OO, Lamunu M, Nanyunja M, et al. Contact tracing during an
outbreak of Ebola virus disease in the Western area districts of
Sierra Leone: lessons for future Ebola outbreak response.
Front Public Health 2016; 4: 130.
34 Huang C, Wang Y, Li X, et al. Clinical features of patients infected
with 2019 novel coronavirus in Wuhan, China. Lancet 2020;
395: 497–506.
35 Pang X, Zhu Z, Xu F, et al. Evaluation of control measures
implemented in the severe acute respiratory syndrome outbreak in
Beijing, 2003. JAMA 2003; 290: 3215–21.