South Korea is experiencing the largest outbreak of
Middle East respiratory syndrome coronavirus infections
outside the Arabian Peninsula, with 166 laboratory-confirmed
cases, including 24 deaths up to 19
June 2015. We estimated that the mean incubation
period was 6.7 days and the mean serial interval 12.6
days. We found it unlikely that infectiousness precedes
symptom onset. Based on currently available
data, we predict an overall case fatality risk of 21%
(95% credible interval: 14–31).
South Korea is experiencing the largest outbreak of
Middle East respiratory syndrome coronavirus (MERSCoV)
infections outside the Arabian Peninsula. Up to 19
June 2015, there have been 166 laboratory-confirmed
cases, including 24 deaths, 30 recovered individuals
discharged from hospital, and 112 still remaining in
hospital [1]. The aim of our study was to conduct a preliminary
epidemiological assessment of the MERS-CoV
outbreak in South Korea in order to further describe
and update key epidemiological determinants of MERSCoV
outbreaks.
Primary case
The ongoing outbreak in South Korea began when the
primary case developed respiratory illness on 11 May
after returning on 4 May from Bahrain (18 April–2 May)
via Qatar (2–3 May). Further epidemiological investigation
showed that the primary case had also travelled
to the United Arab Emirates (29–30 April) and Saudi
Arabia (1–2 May) during their stay in Bahrain [2]. Feeling
unwell after returning to South Korea, the primary
case visited a local clinic (Hospital A) in Pyeongtaek,
Gyeonggi province on 12, 14 and 15 May and was hospitalised
in Hospital B from 15 to 17 May*. However,
this patient did not initially report their recent travel
in the Middle East. Upon discharge from Hospital B,
the patient visited another clinic (Hospital C) and was
admitted to a general hospital (Hospital D) in Seoul
on 17 May, where the patient was later diagnosed with
MERS-CoV on 20 May. Since then, the patient has been
isolated and treated in another hospital designated by
the Korean government to treat MERS patients.
Sources of data
We retrieved publicly available data from multiple
sources, including the Korea Centers for Disease
Control and Prevention (Korea CDC), the Korean
Ministry of Health and Welfare (MoH), the WHO and
local Korean news reports to compile a line list of all
confirmed cases reported by 19 June 2015. In case of
any data discrepancy between the different sources,
we used the most up-to-date information from official
reports published by the Korea CDC and MoH on
a daily basis during the outbreak. The official reports
were only available in Korean language and included
a brief description of each confirmed case, including
demographic characteristics (e.g. age and sex), date of
exposure and onset of symptoms, as well as possible
linkage with confirmed cases and the associated hospital
cluster (e.g. Hospital A to P).
Statistical analysis
We fitted parametric distributions to the time intervals
(i) from infection to onset (i.e. the incubation period)
and (ii) from illness onset to case confirmation. We also
fitted a nonparametric distribution on the incubation
period. The exact dates of infection were not known for
most cases, but exposure windows were available, and
we accounted for the consequent interval censoring in
the likelihood function [9] and the possibility of infectiousness
before illness onset (details on the methodology
are available from the corresponding author on
request). We used survival models to fit alternative
parametric distributions including log-normal, Weibull
and gamma distributions, and compared the goodness
of fit of these parametric distributions using the
Bayesian information criterion. We observed that the
delay from illness onset to confirmation shortened as
the epidemic progressed, so we fitted two separate
survival curves for onset before and after 28 May. We
used the same approach to estimate the serial interval distribution, based on data on illness onset times for
linked cases. We calculated the 95% credible interval
(CrI) by bootstrapping.
To estimate the case fatality risk (CFR) allowing for the
uncertain clinical outcomes of those who remained in
hospital on the date of analysis (19 June 2015), we used
the methods proposed by Garske et al. which adjusts
the fatality risk based on the time-to-death distribution
[10]. We assumed that the time from onset to death
followed a log-normal distribution, and used Markov
chain Monte Carlo methods to estimate the parameters
in a Bayesian framework, setting an informative
prior for the time from onset to death with a mean of
14 days [11], and non-informative priors for the other
parameters. All statistical analyses were conducted in
R version 3.0.2 (R Foundation for Statistical Computing,
Vienna, Austria).
Outbreak description
The number of laboratory-confirmed cases increased
rapidly until 7 June, when 23 cases were confirmed on
a single day but appears to have subsided since then
(Figure 1A). Figure 1B shows the epidemic curve by
date of illness onset for 110 cases with available data.
It should be recognised that while the outbreak has not yet ended, our preliminary assessment shows that the
epidemic to date may have peaked on 1 June when 15
cases reported illness onset. Median age of the 166
cases was 56 years, 101 of 166 (61%) were male, and
30 of 166 (18%) were healthcare personnel (Table 1).
Transmission chains
Figure 2 shows a summary sketch of the transmission
chain (additional material** showing the detailed
chains is available at: http://sph.hku.hk/bcowling/
eurosurveillance2015appendix.zip). 119 cases were
identified by Korea CDC as having had contact with a
confirmed case in the period before their illness onset,
and three of these cases had contact with more than
one confirmed case. A total of 27 secondary cases in
a single hospital have been traced back to the primary
case (excluding six cases with an unclear linkage), and
two of these, Cases 14 and 16, led the second wave
of the outbreak by infecting at least 73 and 24 tertiary
cases, respectively, following the initial outbreak generated
by the primary case in Hospital B (Figure 2). In
particular, Case 14 infected at least 70 cases between
27 and 29 May while being treated in the emergency
room in Hospital D, one of the five largest hospitals
located in Seoul with 3,980 healthcare professionals
and more than 8,000 outpatient visits per day [12].
According to the press conference given at Hospital D
on 7 June, at least 893 patients and visitors were potentially
exposed to the virus during this period [13], which
explains a significant increase in the number of cases
confirmed and notified between 6 and 11 June. Since
12 June, when the first fourth-generation case was confirmed,
10 more potential fourth-generation cases have
been reported. Because of the marked heterogeneity in
transmissibility, with the vast majority of cases associated
with just these three superspreading events in the
nosocomial setting, it would be misleading to summarily
characterise the transmissibility of the virus in this
ongoing outbreak with a single average value of the
reproductive number [14]. The mean serial interval was
12 to 13 days in each of four epidemiological clusters
associated with Cases 1, 14, 15 and 16.
Epidemiological parameters
We found that a gamma distribution had the best fit to
the incubation period distribution and was very similar
to the nonparametric estimate (Figure 3A). The fitted
gamma distribution had a median of 6.3 days (95% CrI:
5.7–6.8), a mean of 6.7 days (95% CrI: 6.1–7.3) and a
95th percentile of 12.1 days (95% CrI: 10.9–13.3). Using
data on 99 cases with single identified infectors, we
found that a gamma distribution with a mean of 12.6
days (95% CI: 12.1–13.1) and standard deviation of 2.8
days (95% CI: 2.4–3.1) provided best fit to the serial
interval distribution (Figure 3B). The mean duration
of illness onset to laboratory confirmation was 8.1
days for cases with illness onset before May 28, and
substantially shorter (mean: 4.4 days) for cases with
illness onset after that date (Figure 3C). We used a lognormal
regression model for the time from illness onset
to laboratory confirmation to estimate that healthcare
worker status was not significantly associated with
time to confirmation (beta=− 0.05; 95% CI: − 0.34 to
0.25), with the point estimate signifying a 5% reduction
in time to confirmation in healthcare workers.
Presymptomatic infectiousness
It appeared that a small number of cases might have
been infected before their infectors became symptomatic.
Furthermore, Cases 37 and 39 were epidemiologically
linked to multiple confirmed cases. To account
for the possibility of presymptomatic infectiousness
and the uncertainty of who infected Cases 37 and 39
when estimating the incubation period, we (i) simultaneously
inferred the incubation period of the infector
of Case 37, (ii) assumed that Case 39 was equally likely
to be infected by all cases to whom he had been epidemiologically
linked, namely Cases 9, 11, 12 and 14
(because the infector of Case 39 was not statistically
identifiable), and (iii) introduced a parameter Y to represent
the time interval between onset of symptoms and
onset of infectiousness For example, if cases become
infectious two days before onset of symptoms, then
Y=2 days. For a given value of Y, the dates of exposure
of a case must not precede the date of symptom onset
of the case’s infector by more than Y days. The data
were adjusted accordingly during the estimation of the
incubation period. Furthermore, we excluded Case 40
when performing the estimation because their exposure
and onset date were the same, which was implausible.
We used Markov chain Monte Carlo methods to
estimate the parameters of this model in a Bayesian
framework.