Journal Pre-proof

Serial interval of novel coronavirus (COVID-19) infections

Hiroshi Nishiura, Natalie M. Linton, Andrei R. Akhmetzhanov

PII: S1201-9712(20)30119-3
DOI: https://doi.org/10.1016/j.ijid.2020.02.060
Reference: IJID 4006

To appear in: International Journal of Infectious Diseases

Received Date: 14 February 2020

Revised Date: 25 February 2020
Accepted Date: 27 February 2020

Please cite this article as: Nishiura H, Linton NM, Akhmetzhanov AR, Serial interval of novel
coronavirus (COVID-19) infections, International Journal of Infectious Diseases (2020),
doi: https://doi.org/10.1016/j.ijid.2020.02.060

This is a PDF file of an article that has undergone enhancements after acceptance, such as
the addition of a cover page and metadata, and formatting for readability, but it is not yet the
definitive version of record. This version will undergo additional copyediting, typesetting and
review before it is published in its final form, but we are providing this version to give early
visibility of the article. Please note that, during the production process, errors may be
discovered which could affect the content, and all legal disclaimers that apply to the journal
pertain.

© 2019 Published by Elsevier.

 Title: Serial interval of novel coronavirus (COVID-19) infections

Running title: Serial interval of COVID-19

Authors:

Hiroshi Nishiura1,2, Natalie M. Linton1, Andrei R. Akhmetzhanov1

Affiliations:

f
oo
1
Graduate School of Medicine, Hokkaido University, Kia 15 Jo Nishi 7 Chome,

Kitaku, Sapporo, 060-8638, Japan; 2CREST, Japan Science and Technology

pr
Agency, Honcho 4-1-8, Kawaguchi, Saitama, 332-0012 Japan;
e-
Correspondence to: H. Nishiura, Graduate School of Medicine, Hokkaido
Pr

University, Kia 15 Jo Nishi 7 Chome, Kitaku, Sapporo, 060-8638, Japan Tel:

+81 11 706 5066 Fax: +81 11 706 7819, E-mail: [email protected]

al

Article type:
n

Original article
ur
Jo

Highlights

 The serial interval of novel coronavirus (COVID-19) infections was

estimated from a total of 28 infector-infectee pairs.

-1-
  The median serial interval is shorter than the median incubation period,

suggesting a substantial proportion of pre-symptomatic transmission.

 A short serial interval makes it difficult to trace contacts due to the rapid

turnover of case generations.

Abstract

Objective: To estimate the serial interval of novel coronavirus (COVID-19)

f
from information on 28 infector-infectee pairs.

oo
Methods: We collected dates of illness onset for primary cases (infectors) and

secondary cases (infectees) from published research articles and case

pr
investigation reports. We subjectively ranked the credibility of the data and
e-
performed analyses on both the full dataset (n=28) and a subset of pairs with

highest certainty in reporting (n=18). In addition, we adjust for right truncation

Pr

of the data as the epidemic is still in its growth phase.

Results: Accounting for right truncation and analyzing all pairs, we estimated
al

the median serial interval at 4.0 days (95% credible interval [CrI]: 3.1, 4.9).

Limiting our data to only the most certain pairs, the median serial interval was
n

estimated at 4.6 days (95% CrI: 3.5, 5.9).

ur

Conclusions: The serial interval of COVID-19 is close to or shorter than its

Jo

median incubation period. This suggests that a substantial proportion of

secondary transmission may occur prior to illness onset. The COVID-19 serial

interval is also shorter than the serial interval of severe acute respiratory

syndrome (SARS), indicating that calculations made using the SARS serial

interval may introduce bias.

-2-
 Keywords: coronavirus; outbreak; illness onset; generation time; statistical

model; epidemiology; viruses

Introduction

The epidemic of novel coronavirus (COVID-19) infections that began in

China in late 2019 has rapidly grown and cases have been reported worldwide.

An empirical estimate of the serial interval—the time from illness onset in a

f
oo
primary case (infector) to illness onset in a secondary case (infectee)—is needed

to understand the turnover of case generations and transmissibility of the disease

pr
[1]. Estimates of the serial interval can only be obtained by linking dates of onset

for infector-infectee pairs, and these links are not easily established. A recently
e-
published epidemiological study used contact tracing data from cases reported in

Hubei Province early in the epidemic to estimate the mean serial interval at 7.5
Pr

days [2], which is consistent with the 8.4-day mean serial interval reported for

severe acute respiratory syndrome (SARS) from Singaporean household contact

al

data [3]. However, there were only six infector-infectee pairs in this dataset, and
n

sampling bias may have been introduced to the variance and mean. To further
ur

assess the serial interval of COVID-19 infections we compiled a dataset of 28

publicly shared infector-infectee pairs and calculated the serial interval from
Jo

these data.

Materials and Methods

We scanned publicly available information published in research articles

and quoted from official reports of outbreak investigations to obtain our dataset.

-3-
 The date of illness onset was defined as the date on which a symptom relevant to

COVID-19 infection appeared and was determined by the reporting

governmental body. We subjectively ranked the credibility of the ascertained

pairs into “certain” and “probable,” where the former was used for pairs and

dates of illness onset were clearly defined in an academic article and the latter

was applied to pairs and dates of illness onset that were clearly defined but

quoted from outbreak investigation reports. Pairs of cases that cannot be

scientifically linked were classified as “uncertain” and removed from our

f
oo
analysis, and the verification of illness onset date was ensured following an

erratum report of presymptomatic transmissions in Germany [4]. Estimates were

obtained for certain and probable pairs combined (n=28) as well as for the certain

pairs alone (n=18).

pr
e-
The interval censored data were handled in units of days. We employed

a Bayesian approach with doubly interval censored likelihood to obtain estimates

Pr

of the serial interval [5]:

𝐸𝑅.𝑖 𝑆𝑅.𝑖
𝐿(𝛩𝑔 ; 𝑫) = ∏ ∫ ∫ 𝑔(𝑒)𝑓(𝑠 − 𝑒)d𝑠d𝑒 ,
al

(1)
𝑖 𝐸𝐿.𝑖 𝑆𝐿.𝑖

where i represents the identity of each pair, (ER,i, EL,i) is the interval for symptom
n

onset of the infector and (SR,i, SL,i) is the interval for symptom onset of the
ur

infectee. Intervals are allowed for illness onset event so that the serial interval
Jo

can be inferred even when we do not have exact dates of illness onset for infetor-

infectee pairs. Here, g(.) is the probability density function (p.d.f.) of exposure

following a uniform distribution, because there is no indication of time-

dependence in the frequency of illness onset among primary cases. f(.) is the

p.d.f. of the serial interval, assumed to be governed by three different

-4-
 distributions—lognormal, gamma, and Weibull. We sampled the posterior

distributions using CmdStan version 2.22.1 [6]

(http://github.com/aakhmetz/nCoVSerialInteval2020).

As the epidemic will continue to grow beyond our data collection cutoff

point of 12 February 2020, it is possible that the naïve likelihood (1)

underestimates the serial interval as sampling during the early stage of the

epidemic preferentially excludes infector-infectee pairs with longer serial

intervals. We adjusted for this selection bias—called right truncation—in our

f
oo
model. The alternative p.d.f. that accounts for right truncation during the

exponential growth phase of the epidemic is written as:

𝑓′(𝑠 − 𝑒, 𝑒) =
𝑓(𝑠 − 𝑒)
𝑇−𝑒 𝑟exp(−𝑟𝑢)
pr
∫0 1 − exp⁡(−𝑟𝑢) 𝐹(𝑇 − 𝑒 − 𝑢)du
(2)
e-
where r is the exponential growth rate estimated at 0.14 [7] and T is the latest
Pr

time of observation (12 February 2020). We account for the exponential growth

rate of cases, because recently infected individuals are more likely to be sampled

during the exponential growth phase of an epidemic. The widely applicable

al

information criterion (WAIC) was used to compare between distributions and the
n

model with the minimal WAIC value was selected as the best-fit model for each
ur

set of estimates with and without right truncation.

Results
Jo

We were able to obtain data on 28 infector-infectee pairs (see

Supplementary Table). Of these, 12 pairs were family clusters. Accounting for

right truncation and analyzing all pairs, the model using the lognormal

distribution was selected as the best-fit model (WAIC=224.0), while no

-5-
 significant differences from other models were identified. The median serial

interval was estimated at 4.0 days (95% credible interval [CrI]: 3.1, 4.9) while

the mean and standard deviation (SD) of the serial interval were estimated at 4.7

days (95% CrI: 3.7, 6.0) and 2.9 days (95% CrI: 1.9, 4.9), respectively. Without

truncation, the model using the lognormal distribution was also the best-fit model

(WAIC=128.0) with the median serial interval was estimated at 3.9 days (95%

CrI: 3.1, 4.8).

Limiting our dataset to only certain observations, the median serial

f
oo
interval of the best-fit Weibull distribution model was estimated at 4.6 days (95%

CI: 3.5, 5.9) with a mean and SD of 4.8 days (95% CrI: 3.8, 6.1) and 2.3 days

pr
(95% CrI: 1.6, 3.5), respectively. Without truncation, the best-fit model used the

lognormal distribution and estimated the median serial interval at 4.1 days (95%
e-
CrI: 3.2, 5.0). Figure 1 shows the best-fit distributions overlaid with a published

distribution of the SARS serial interval [3].

Pr

Discussion
al

Our estimate of the median serial interval as 4.0 days indicates that

COVID-19 infection leads to rapid cycles of transmission from one generation of

n

cases to the next. The shorter serial interval compared to SARS implies that
ur

contact tracing methods must compete against the rapid replacement of case
Jo

generations, and the number of contacts may soon exceed what available

healthcare and public health workers are able to handle. The difference between

these distributions suggests that using serial intervals estimates from SARS data

will result in overestimation of the COVID-19 basic reproduction number.

-6-
 More importantly, the estimated median serial interval is shorter than the

preliminary estimates of the mean incubation period (approximately 5 days)

[2,8]. As illustrated in Figure 2, when the serial interval is shorter than the

incubation period, pre-symptomatic transmission is likely to have taken place

and may even occur more frequently than symptomatic transmission. A

substantial proportion of secondary transmission occurring before illness onset

indicates that many transmissions cannot be prevented solely through isolation of

symptomatic cases, as by the time contacts are traced they may have already

f
oo
become infectious themselves and generated secondary cases [9]. It is possible

that serial intervals were shortened due to case isolation, especially outside of

China, but even the subset of data in China alone can indicate that the mean is

shorter than 7.5 days.

pr
e-
Correct ascertainment of dates of illness onset is critical to the calculation of

the serial interval. Considering the overall mild nature of the infection [10] it is
Pr

possible that different reporting jurisdictions have different criteria for

determining what qualifies as illness onset for COVID-2019 cases, which is a

al

potential bias we are unable to account for. However, the present study addresses
n

the issue of data quality of the reported pairs in two ways. First, our data include
ur

the updated information from a recent report of pre-symptomatic transmission in

Germany [4] where it was later found that the primary case was already
Jo

symptomatic while in contact with persons who later became infected

(Supplementary Material in [4]). Second, classification of the credibility of the

data and comparing analyses including and excluding less certain (but

nonetheless highly probable) pairs allowed us to determine that our results using

-7-
 all pairs (and therefore a greater sample size) did not differ significantly from the

results using only the most credible data.

In conclusion, we have estimated the median serial interval of COVID-19 at

4.0 days, which is close to or shorter than the disease’s median incubation period

indicating that rapid cycles of transmission and substantial pre-symptomatic

transmissions are occurring. Thus, containment via case isolation alone is likely

to be very challenging.

f
oo
Funding source

H.N. received funding support from Japan Agency for Medical Research and

pr
Development [grant number: JP18fk0108050] the Japan Society for the
e-
Promotion of Science (JSPS) Grants-in-Aid for Scientific Research (KAKENHI

in Japanese abbreviation) grant nos. 17H04701, 17H05808, 18H04895 and

Pr

19H01074, and the Japan Science and Technology Agency (JST) Core Research

for Evolutional Science and Technology (CREST) program [grant number:

al

JPMJCR1413]. NML received a graduate study scholarship from the Ministry of

Education, Culture, Sports, Science and Technology, Japan. The funders had no
n

role in study design, data collection and analysis, decision to publish, or

ur

preparation of the manuscript.

Jo

Ethical approval

This study was based on publicly available data and did not require ethical

approval.

-8-
 Conflict of interest

The authors declare no conflicts of interest.

Declaration of interests

☒ The authors declare that they have no known competing financial interests or
personal relationships that could have appeared to influence the work reported in
this paper.

f
oo
pr
e-
Pr
n al
ur
Jo

-9-
 References

1. Fine PE. The interval between successive cases of an infectious disease. Am

J Epidemiol 2003;158:1039-1047.

2. Li Q, Guan X, Wu P, Wang X, Zhou L, Tong Y, Ren R, Leung KSM, Lau

EHY, Wong JY, Xing X, Xiang N, Wu Y, Li C, Chen Q, Li D, Liu T, Zhao J,

Li M, Tu W, Chen C, Jin L, Yang R, Wang Q, Zhou S, Wang R, Liu H, Luo

f
Y, Liu Y, Shao G, Li H, Tao Z, Yang Y, Deng Z, Liu B, Ma Z, Zhang Y, Shi

oo
G, Lam TTY, Wu JTK, Gao GF, Cowling BJ, Yang B, Leung GM, Feng Z.

Early Transmission Dynamics in Wuhan, China, of Novel Coronavirus-

pr
Infected Pneumonia. N Engl J Med. 2020; in press. doi:

10.1056/NEJMoa2001316
e-
3. Lipsitch M, Cohen T, Cooper B, Robins JM, Ma S, James L, Gopalakrishna

G, Chew SK, Tan CC, Samore MH, Fisman D, Murray M. Transmission

Pr

dynamics and control of severe acute respiratory syndrome. Science.

2003;300(5627):1966-70.
al

4. Rothe C, Schunk M, Sothmann P, Bretzel G, Froeschl G, Wallrauch C,

Zimmer T, Thiel V, Janke C, Guggemos W, Seilmaier M, Drosten C,

n

Vollmar P, Zwirglmaier K, Zange S, Wölfel R. Transmission of 2019-nCoV

ur

infection from an asymptomatic contact in Germany. N Eng J Med. 2020; in

press. doi:10.1056/NEJMc2001468.
Jo

5. Reich NG, Lessler J, Cummings DA, Brookmeyer R. Estimating incubation

period distributions with coarse data. Stat Med 2009;28:2769-2784.

6. Stan Development Team. Stan Modeling Language Users Guide and

Reference Manual. 2014, URLhttp://mc-stan.org/manual.html.

- 10 -
 7. Jung S, Akhmetzhanov AR, Hayashi K, Linton NM, Yang Y, Yuan B,

Kobayashi T, Kinoshita R, Nishiura H. Real time estimation of the risk of

death from novel coronavirus (2019-nCoV) infection: Inference using

exported cases. medRxiv, http://dx.doi.org/ 10.1101/2020.01.29.20019547v1

8. Linton NM, Kobayashi T, Yang Y, Hayashi K, Andrei, AR, Jung S, Yuan B,

Kinoshita R, Nishiura H. (2020). Epidemiological characteristics of novel

coronavirus infection: A statistical analysis of publicly available case data.

medRxiv. http://dx.doi.org/10.1101/2020.01.26.20018754

f
9. Fraser C, Riley S, Anderson RM, Ferguson NM. Factors that make an

oo
infectious disease outbreak controllable. Proceedings of the National

Academy of Sciences of the United States of America. 2004;101(16):6146–

6151.

pr
10. Nishiura H, Kobayashi T, Yang Y, Hayashi K, Miyama T, Kinoshita R,
e-
Linton NM, Jung S, Yuan B, Suzuki A, Akhmetzhanov AR. The rate of

underascertainment of novel coronavirus (2019-nCoV) infection: Estimation

Pr

using Japanese passengers data on evacuation flights. J Clin Med. 2020; in

press. doi:10.3390/jcm9020419.
n al
ur
Jo

- 11 -
 Figure Legends

f
oo
Figure 1. Serial interval of novel coronavirus (COVID-19) infections.

The solid line shows the estimated serial interval distribution of COVID-19

pr
infections using the best-fit lognormal distribution with right truncation. A
e-
distribution based on a published estimate of the serial interval for severe acute

respiratory syndrome [3] is overlaid as a dashed line for comparison.

Pr
n al
ur
Jo

Figure 2. The relationship between the incubation period and serial interval.

- 12 -
 If the transmission takes place during the symptomatic period of the primary

case, the serial interval is longer than the incubation period. However, this

relationship can be reversed when pre-symptomatic transmission takes place.

Furthermore, it is possible that the secondary case may even experience illness

onset prior to onset in their infector.

f
oo
pr
e-
Pr
n al
ur
Jo

- 13 -