key: cord-265603-3we40x62
authors: Casey, M.; Griffin, J.; McAloon, C. G.; Byrne, A. W.; Madden, J. M.; McEvoy, D.; Collins, A. B.; Hunt, K.; Barber, A.; Butler, F.; Lane, E. A.; O Brien, K.; Wall, P.; Walsh, K. A.; More, S. J.
title: Estimating pre-symptomatic transmission of COVID-19: a secondary analysis using published data
date: 2020-05-11
journal: nan
DOI: 10.1101/2020.05.08.20094870
sha: 
doc_id: 265603
cord_uid: 3we40x62

Background: Understanding the extent of virus transmission that can occur before symptom onset is vital for targeting control measures against the global pandemic of COVID-19. Objective: Estimation of (1) the proportion of pre-symptomatic transmission of COVID-19 that can occur and (2) timing of transmission relative to symptom onset. Design: Secondary analysis of published data Data sources: Meta-analysis of COVID-19 incubation period and a rapid systematic review of serial interval and generation time, which are published separately. Methods: Simulations were generated of incubation period and of serial interval or generation time. From these, transmission times relative to symptom onset were calculated and the proportion of pre-symptomatic transmission was estimated. Results: A total of 23 estimates of serial interval and five estimates of generation time from 17 publications were included. These came from nine different data source categories (presented here in descending order of the proportion of pre-symptomatic transmission): Hong Kong, Tianjin, pooled data from Hong Kong and Shenzhen, Singapore, Mainland China excluding Hubei, mixed sources, Shenzhen, northern Italy and Wuhan. Transmission time relative to symptom onset ranged from a mean of 2.05 days before symptom onset for Hong Kong to 1.72 days after symptom onset for Wuhan. Proportion of pre-symptomatic transmission ranged from 33.7% in Wuhan to 72.7% in Hong Kong. Based on individual estimates, transmission time relative to symptom onset ranged from mean of 2.95 days before symptom onset to 1.72 days after symptom onset and proportion of pre-symptomatic transmission ranged from 33.7% to 79.9%. Simple unweighted pooling of estimates based on serial intervals resulted in a mean time of transmission of 0.67 days before symptoms, and an estimated 56.1% of transmission occurring in the pre-symptomatic period. Conclusions: Contact rates between symptomatic infectious and susceptible people are likely to influence the proportion of pre-symptomatic transmission. There is substantial potential for pre-symptomatic transmission of COVID-19 in a range of different contexts. Our work suggests that transmission of SARS-CoV-2 is most likely in the day before symptom onset whereas estimates suggesting most pre-symptomatic transmission highlighted a mean transmission times almost 3 days before symptom onset. These findings highlight the urgent need for extremely rapid and effective case detection, contact tracing and quarantine measures if strict social distancing measures are to be eased.

Strengths and weaknesses of this study • We contribute evidence about the extent and variation of pre-symptomatic transmission of COVID-19 across a range of contexts. This provides key information for development and targeting of control policies.

• This is a secondary analysis using simulations based on published data, some of which is in pre-print form and not yet peer-reviewed. There is overlap in the contact tracing data that informed some of our source publications. We partially addressed this by summarising data at source location level as well as at study level.

• A strength of our approach is that it builds a picture of pre-symptomatic transmission from a range of estimates in the literature, facilitates discussion for the drivers of variation between them, and highlights the consistent message that consideration of pre-symptomatic transmission is critical for COVID-19 control policy.

What is already known about this topic • COVID-19 has spread rapidly, prompting calls for understanding about key transmission characteristics.

• Case-reports highlight the potential for people to transmit COVID-19 before they develop symptoms.

• Virological studies report high viral loads at the time of symptom onset and suggest the possibility for substantial viral shedding before symptom onset.

• Quantitative analyses of contact tracing data show that pre-symptomatic transmission of COVID-19 occurs in certain contexts.

• We estimate the proportion and timing of pre-symptomatic transmission from a range of different contexts.

• Despite the variation between estimates based on different studies, we highlight the consistent finding that a substantial amount of pre-symptomatic transmission can occur.

• We clarify the conceptual difference between asymptomatic and pre-symptomatic transmission which has not always been clear in the literature, and has relevance for control measures, disease transmission modelling and further research efforts.

. CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

There is currently a global pandemic of COVID-19, a recently emerged and rapidly spreading infectious disease that is caused by the novel coronavirus SARS-CoV-2. There are large direct impacts of COVID-19 amongst known cases. As of 6 th of May 2020, the World Health Organization has reported 3,588,773 confirmed cases and 247,503 deaths due to COVID-19

[1]. In China, 14% and 5% of cases were classified as severe and critical, respectively [2], and a report from Italy showed that 18% of cases required intensive care [3] . There are also major indirect impacts of COVID-19 and its control measures on other aspects of health care As there is currently no COVID-19 vaccine ready for widespread use, primary control measures entail reducing transmission from infectious individuals. These include case isolation, contact tracing and quarantine, physical distancing and hygiene measures [9] .

Infectious people are predominantly identified by reported symptoms of COVID-19.

In absence of active surveillance, infectious people without symptoms may not be quarantined, and therefore may have more contacts with susceptible people resulting in increased COVID-19 transmission. Therefore, quantifying the transmission potential of COVID-19 before or in the absence of symptoms will inform disease control measures and predictions of epidemic progression.

Characteristics of pre-symptomatic and asymptomatic transmission are potentially different, and separate approaches may be required to understand them. We aimed to capitalise upon the considerable information about pre-symptomatic transmission that can be inferred from contact tracing studies. Therefore, we focus here on transmission from people before they develop symptoms rather than that from people who never develop symptoms. This addresses the urgent need for more data on extent of pre-symptomatic transmission which has been highlighted by those developing models to inform policies [10].

The pre-symptomatic transmission potential of COVID-19 has been highlighted by case reports [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] . The potential for pre-symptomatic transmission was also suggested by detection of viral genome in upper respiratory samples prior to symptoms [21-23].

. CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 11, 2020 Here, using secondary analysis of data collated in meta-analysis [33] and a rapid systematic review [34] that are published separately, we apply a standardised methodology to estimate the proportion and timing of pre-symptomatic transmission of COVID-19 in a range of different contexts.

If generation time, the duration in days between time of infection of a secondary case (infectee) and that of its primary case (infector), is longer than incubation period, the time between infection and symptom onset in the infector, transmission will have occurred after symptom onset (Scenario A in Figure 1 ). If generation time is shorter than incubation period, pre-symptomatic transmission will have occurred (Scenarios B and C in Figure 1 ). If an infector and infectee incubation periods are taken to be independent and identically distributed, serial interval, the time between infector and infectee symptom onset, can be taken as an approximation of generation time [35, 36] , although serial interval will have more variation [26] . Table 1 contains definitions relevant to our analysis.

Data about incubation period, serial interval and generation time were sourced through on our separately published metanalysis of incubation period [33] and rapid systematic review of serial interval and generation time [34] . As described fully elsewhere [33, 34] , literature searches covered publication dates between the 1 st of December 2019 and 8 th of April 2020 for incubation period, extending to the 15 th of April 2020 for generation time / serial interval.

A dedicated team searched for publications on the electronic databases PubMed [37], Google . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 11, 2020. . https://doi.org/10.1101/2020.05.08.20094870 doi: medRxiv preprint 6 Scholar [38] , MedRxiv [39] and BioRxiv [40] with the following keywords: "Novel coronavirus" OR "SARS CoV 2" OR "2019-nCoV" OR "COVID-19" AND "serial interval" OR "latent period" OR "incubation period" OR "generation time" OR "infectiousness" OR "pre-symptomatic" OR "asymptomatic"). The dynamic curated PubMed database "LitCovid" [41, 42] was also monitored. In addition, publicly available reports from the World Health Organization [1, 9, 43] , European Centre for Disease Prevention and Control [44] and Centres for Disease Control and Prevention Morbidity and Mortality Weekly Reports [45] were monitored, as well as curated summaries on relevant topics from the American Association for the Advancement of Science [46] and the Nature Journal [47].

Both our meta-analysis [33] and rapid systematic review [34] completed checklists to show fulfilment of Preferred Reporting Items for Systematic reviews and Meta-Analysesextension for Scoping Reviews (PRISMA-ScR) [48] .

Based on the estimates reported by our meta-analysis [33] and rapid systematic review [34] , we simulated data for incubation period, serial interval and generation time. We subtracted incubation period from serial interval or generation time to infer transmission time relative to onset of symptoms and to estimate the proportion of pre-symptomatic transmission.

All analyses were conducted in the R statistical environment [49] . For each publication included in our analyses, parameters describing the distributions of generation time or serial interval, and the location and dates of collection of the contact tracing data from which they were generated were collated.

If not defined directly, gamma and Weibull parameters were estimated from the reported mean and standard deviation using the "epitrix" [50] and "mixdist" [51] R packages, respectively.

If lognormal distribution parameters (Meanlog and SDlog) were not directly reported, they were estimated from the reported mean and standard deviation (SD) as follows:

. CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 11, 2020. . https://doi.org/10.1101/2020.05.08.20094870 doi: medRxiv preprint

The incubation period that we used from the meta-analysis [33] had a lognormal distribution (Mean = 5.8, Median = 5.1 , SD = 3, Meanlog = 1.63, SDlog = 0.5).

As serial interval has more variation than generation time [26], we considered them separately for plotting and summary purposes. For the two studies [26,27] that estimated generation time, we also generated serial interval simulations to allow a more direct comparison with the other estimates based on serial interval. One of these, [26], related serial interval to generation time with the following approach: Serial interval of an infectee can be expressed as generation time of the infectee plus the difference between the incubation periods of the infectee and the infector ( Figure 1 ). That is, the incubation period used for the generation time estimation was simulated twice to generate two samples ("inc 1" and "inc2").

The extra variation in serial interval compared to generation time was then simulated by:

We repeated this estimation to simulate the serial interval for study [26] and cross-checked the simulation against the summary statistics were reported in that publication. We then estimated a serial interval from the generation time of study [27], using the same methodology and the same incubation period as was used to infer generation time in study [27] .

One further study [52] did not directly supply enough information to simulate a distribution, but supplied their code and data from which they estimated the serial interval. For this study, we fitted various distributions to their data by maximum likelihood estimation with the "fitdistrplus" package [53] and chose one based on AIC values and visual cross-checking by plotting.

We simulated data (n = 100,000 samples) for incubation period and generation time or serial interval from each study. The distributions were plotted, summarised (Supplementary Table   1 ) and cross-checked against the summary statistics and plots reported in the papers. The . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 11, 2020. . https://doi.org/10.1101/2020.05.08.20094870 doi: medRxiv preprint 8 incubation period sample was subtracted from the generation time or serial interval sample to give a resultant distribution indicating transmission time relative to onset of symptoms. This was plotted and summarised and the proportion of samples of transmission times relative to symptom onset that were negative (indicating transmission before symptoms was reported) was calculated.

As there are known drivers of variation of generation time and serial interval [34] , and therefore of transmission time relative to symptom onset estimated based upon them, estimates were plotted and summarised individually to show this variation. Estimates were also grouped, plotted and summarised at the level of source location for the contact tracing data that they were inferred from. A simple unweighted pooled overall estimate was presented for interpretation in the context of the variation between the individual results.

Of the 19 studies reporting serial interval or generation time in our rapid systematic review [34] , 17 were included in this study. We excluded the study [54] as they defined the start of the exposure window for the infectee as the time of symptom onset in the infector, excluding the possibility of transmission before symptom onset. We excluded one further study, [20] , pending clarification from the authors, as we could not replicate the distribution they described for serial interval.

A total of 28 estimates from 17 studies were included. Transmission time relative to symptom onset estimates were based on five estimates of generation times and 23 estimates of serial interval. Several studies generated more than one estimate. This was due to separate estimates for different locations [25, 26] , different models used to infer generation time [26], sub-setting of data depending on confidence in transmission pair identification and exposure windows [52, 55] , and estimation of both generation times and serial intervals from the same papers [26, 27] . Of the two models used in [26] , one only allowed positive serial intervals to be inferred for missing data whereas a second model allowed negative serial intervals for missing data. Table 2 shows the counts of estimates and studies that came from specific locations or mixed sources under nine different data source categories (Mixed sources, Tianjin, Singapore, . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 11, 2020. .

Mainland China excluding Hubei, Hong Kong, northern Italy, pooled data from Hong Kong and Shenzhen, and Wuhan). Table 3 show the variation in transmission time relative to symptom onset amongst the 28 estimates ranging from a mean (median) of 2.95 (2.91) days before symptom onset to 1.72 (1.68) days after symptom onset. Proportion of pre-symptomatic transmission associated with the 28 different estimates ranged from 33.7% to 79.9%. Table 3 is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 11, 2020. . https://doi.org/10.1101/2020.05.08.20094870 doi: medRxiv preprint symptom onset and peaked at 0.7 days before symptom onset [28] . Our study, analysing all estimates from a variety of locations has produced a similar estimate (mean from pooled estimate of 0.67 days before symptom onset).

We estimated that the proportion of pre-symptomatic transmission ranged from 33.7% to 79.9% depending on the study analysed. Despite this understandable variation in estimates from different contexts, our work consistently suggests that much pre-symptomatic transmission is occurring. This means that a person presenting with COVID-19 symptoms has potentially been infectious to others for . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 11, 2020. . https://doi.org/10.1101/2020.05.08.20094870 doi: medRxiv preprint several days. Therefore, in absence of severe social distancing measures, extremely effective and rapid contact tracing and quarantine will be required to control the spread of COVID-19.

This potential of the COVID-19 spread to be too fast to be controlled by conventional contact tracing been highlighted with Ferretti et al.'s model [27] . This model suggests that presymptomatic transmission alone can account for a basic reproductive number of 0.9 (47% of the overall reproductive basic number), almost enough to sustain an epidemic on its own. However, this estimate may be influenced by the low level of asymptomatic infectiousness (10% relative to a symptomatic case) assumed by that model. This uncertainty highlights the need for transmission from asymptomatically infected people to be more fully understood, and to be considered as having potentially distinct characteristics compared to the presymptomatic transmission that we report on in this paper.

We used a straightforward approach of simulating large numbers of samples from both the incubation period distribution from our meta-analysis [33] and the distributions of 28 serial interval or generation time estimates from our rapid systematic review [34] , and subtracting Tianjin. An alternative method, using maximum likelihood estimation, was used by He et al. . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 11, 2020. . https://doi.org/10.1101/2020.05.08.20094870 doi: medRxiv preprint Table 2 shows the potential overlap in contact tracing data that the studies we analysed are based on. Therefore, a limitation of our study is that all our data sources cannot be considered completely independent. We partially address this by grouping estimates by the source location of the contact-tracing data upon which they were based. A strength of our approach is that it builds a picture of pre-symptomatic transmission from a range of estimates in the literature, facilitates discussion for the drivers of variation between them, and highlights the consistent message that consideration of pre-symptomatic transmission is critical for COVID-19 control policy. The important insights into COVID-19 transmission gleaned from the studies that contributed to our analyses, that used publicly available transmission pair data, highlights the immense value of allowing public access to anonymised transmission pair data.

Although contact rates between symptomatic infectious and susceptible people are likely to influence the proportion of pre-symptomatic transmission, our study highlights substantial potential for pre-symptomatic transmission of COVID-19 in a range of different contexts.

Our work suggests that transmission of SARS-CoV-2 is most likely in the day before symptom onset, whereas estimates suggesting most pre-symptomatic transmission highlighted a mean transmission times almost 3 days before symptom onset. These findings highlight the urgent need for extremely rapid and effective case detection, contact tracing and quarantine measures if strict social distancing measures are to be eased.

Dublin, the Irish Department of Agriculture, Food and the Marine (DAFM), or the Irish Health information and Quality Authority (HIQA). No additional funding was obtained for this research.

All authors have completed the ICMJE 508 uniform disclosure form at www.icmje.org/coi_disclosure.pdf and declare: no support from any organisation for the submitted work; no financial relationships with any organisations that might have an interest . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 11, 2020 . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 11, 2020 . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 11, 2020 systematic review and meta-analysis of observational research. medRxiv . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 11, 2020. . https://doi.org/10.1101/2020.05.08.20094870 doi: medRxiv preprint medium%3Demail&utm_source=membership&utm_campaign=tbc-2020-acq-fosglobal-lead-list-2&utm_content=email1&dmc=P2BSO1EMAQL2&et_rid=35079646&et_cid=328571 8 (accessed 26 Apr 2020).

Nature. Nature Coronavirus updates. 2020.

. CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 11, 2020. . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 11, 2020. . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 11, 2020. . https://doi.org/10.1101/2020.05.08.20094870 doi: medRxiv preprint Table 1 : Definitions referred to in this review.

An infected person who never develops symptoms of the disease

An infected person before they develop symptoms of the disease.

The time interval in days during which an infectious agent may be transferred directly or indirectly from an infected person to another person [58] .

The time interval in days between invasion by an infectious agent and appearance of the first signs or symptoms of the disease in question [58] .

The duration in days between symptom onset of a secondary case and that of its primary case.

The duration in days between time of infection of a secondary case (infectee) and that of its primary case (infector).

An infected person (infector) and a person who they transmit the pathogen to (infectee).

The period from the point of infection to the beginning of the state of infectiousness [59] . This period corresponds to the "exposed" (E) compartment of a susceptible-exposed-infectiousrecovered/removed (SEIR) model.

The time of transmission of an infectious agent from an infector to an infectee in days relative to the onset of symptoms in the infector.

The proportion of all transmission events that occur before the onset of symptoms in the infector. . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 11, 2020 is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint interval/generation time is shorter than incubation period, transmission occurs prior to symptom onset. Scenario C: A negative serial interval is possible if symptoms manifest in the infectee before the infector. If incubation period is assumed to be independent and identically distributed, mean serial interval will approximate mean generation time. Table 3 ). Mixed = data came from multiple countries.

. CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 11, 2020. . https://doi.org/10.1101/2020.05.08.20094870 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 11, 2020. . https://doi.org/10.1101/2020.05.08.20094870 doi: medRxiv preprint

Active Monitoring of Contacts for the Mitigation of COVID-19: a modelling study

A family cluster of SARS-CoV-2 infection involving 11 patients in Nanjing, China

Presymptomatic Transmission of SARS-CoV-2

Potential Presymptomatic Transmission of SARS-CoV-2

COVID-19 Transmission Within a Family Cluster by Presymptomatic Carriers in China

A Locally Transmitted Case of SARS-CoV-2

Infection in Taiwan

A familial cluster of infection associated with the 2019 novel coronavirus indicating potential person-to-person transmission during the incubation period

Transmission of 2019-nCoV Infection from an Asymptomatic Contact in Germany

A familial cluster of COVID-19 indicating virus can be transmitted by asymptomatic carriers. Bull World Heal Organ Published Online First

Suppression of COVID-19 outbreak in the municipality of Vo

A rapid review of available evidence on the serial interval and generation time of COVID-19

Serial Intervals of Respiratory Infectious Diseases: A Systematic Review and Analysis

A note on generation times in epidemic models

org -the preprint server for Health Sciences

bioRxiv. bioRxiv.org -the preprint server for Biology. 2020

Keep up with the latest coronavirus research

Bulletin of the World Health Organization. WHO Published Online First: 2020

Centre for Disease Prevention and Control Rapid Risk Assessments

PRISMA-ScR): Checklist and Explanation

R: A language and environment for statistical computing. R Foundation for Statistical Computing

Package 'epitrix" Small Helpers and Tricks for Epidemics Analysis

Package 'mixdist' : Finite Mixture Distribution Models

The time-varying serial interval of the coronavirus disease (COVID-19) and its gender-specific difference: A data-driven analysis using public surveillance data in Hong Kong and Shenzhen

fitdistrplus: An R Package for Fitting Distributions

High transmissibility of COVID-19 near symptom onset

Serial interval of novel coronavirus (COVID-19) infections

Failed detection of the full-length genome of SARS-CoV-2 by ultra-deep sequencing from the recovered and discharged patients retested viral PCR positive Authors. medRxiv Published Online First

Inferred duration of infectious period of SARS-CoV-2: rapid scoping review and analysis of available evidence for asymptomatic and symptomatic COVID-19 cases

Systematic review on the incubation and infectiousness/ shedding period of communicable diseases in children

Estimating clinical severity of COVID-19 from the transmission dynamics in Wuhan, China

Strongly heterogeneous transmission of COVID-19 in mainland China: local and regional variation

Evolving epidemiology and transmission dynamics of coronavirus disease 2019 outside Hubei province, China: a descriptive and modelling study

Estimating the serial interval of the novel coronavirus disease (COVID-19): A statistical analysis using the public data in Hong Kong from

Epidemiology and transmission of COVID-19 in 391 cases and 1286 of their close contacts in Shenzhen, China: a retrospective cohort study