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

Conceptual model of population infection dynamics

Infectious period was contextualised in relation to a working conceptual model of COVID-19 disease dynamics (online supplementary figure S1, online supplementary material 1). From this conceptual model, three parameters were identified as important in context of this study:

• T2 defined as: duration of the total infectious period for asymptomatic cases, postlatent to recovery (‘recover’ in this context relates to clearing of infection).

• T3 defined as: duration of presymptomatic infectious period for those infected individuals who subsequently develop symptoms (ie, postlatent to onset of symptoms).

• T5 defined as: duration from onset of symptoms to recovery (recovery was inferred as either the first of two clear RT-PCR tests or hospital discharge after admission from COVID-19-related symptoms) or death.

‘Asymptomatic’ case definition was interpreted pragmatically following Davies et al ^{14 15} and may include very mild symptoms that may occur but are unnoticed.

T2, T3 and T5 represent readily measurable parameters but may be upper limits of infectious period, as patients may be non-infectious for a period before recovery or death. We also review evidence where infectiousness is inferred from viral shedding and contract tracing (transmission).

Literature search

A survey of the literature between 1 December 2019 and 1 April 2020 for all countries was implemented using the following search strategy. Publications on the electronic databases PubMed, Google Scholar, MedRxiv and BioRxiv were searched with the following keywords: ‘Novel coronavirus’ OR ‘SARS‐CoV‐2’ OR ‘2019-nCoV’ OR ‘COVID-19’ AND ‘infectious’. Additionally, national and international government reports were monitored. No restrictions on language or publication status were imposed so long as an English abstract was available. Articles were evaluated for data relating to the aim of this review; all relevant publications were considered for possible inclusion. Bibliographies within these publications were also searched for additional resources.

Manual searches of the literature was undertaken using daily updated COVID-19 collections from the National Centre for Biotechnology Information and MedRxiv servers (https://connect.medrxiv.org/relate/content/181), respectively, searching specifically for papers relating to ‘infectious period’ or ‘infectious duration’ from both empirical and modelling studies.

Finally, we used the complementary work undertaken by the Health Information and Quality Authority (HIQA) of Ireland, specifically the evidence summaries relating to asymptomatic transmission and viral load.^{16 17} The protocol for the evidence synthesis is published on the HIQA website.¹⁸ Briefly, the evidence synthesis process included searching databases from 30 December 2019 to 27 March 2020 (PubMed, Embase, ScienceDirect, NHS evidence, Cochrane, medRxiv and bioRxiv, and HRB open), screening, data extraction, critical appraisal and summarising the evidence.

Our aim was to have as great a breadth for an evidential base as possible to clarify what evidence was available to inform on the infectious period of COVID-19 and to identify key characteristics of the data sources and their interpretation. Therefore, our approach is a scoping review (following¹⁹). However, due to the emergent nature of COVID-19, this work is considered a rapid review.²⁰ This paper follows the Preferred Reporting Items for Systematic Reviews and Meta-Analyses—Extension for Scoping Reviews (PRISMA-ScR) checklist. In accordance with the PRISMA-ScR checklist, the electronic search strategy can be found in the supplementary material (online supplementary material 2).

Inclusion criteria were for papers that provided data to inform duration of infectious period based on: time from symptoms to recovery; time from symptoms to death; time from symptoms to diagnostic test clearance (≥two clear tests, defined as at least two consecutive negative reverse transcriptase PCR (RT-PCR) tests conducted 24 hours apart); presymptomatic infectious period; and time from first diagnostic test to diagnostic test clearance (≥two clear tests) for presymptomatic/asymptomatic cases. Inclusion criteria for viral dynamics were papers that reported viral load via cycle threshold (Ct) values from RT-PCR testing over repeated sampling of infected patients and studies that additionally reported viral isolation.

For quality control, studies were: (1) selected and screened initially by three members of the team from search terms outlined above (ÁBC, KH and FB), with parameters identified and recorded. (2) This was reviewed and supplemented by manual search by a different two team members (AWB and DM), again with parameters identified and recorded. (3) Finally, the review was then internally reviewed by an additional two members of the team (CMc and MC) and cross-referenced with other parameter synthesis documents being worked on by the group (all authors).

Parameter comparison

Viral dynamics

A narrative comparison of reported viral dynamics from studies that undertook serial viral load estimates from patients over their period of observation was undertaken. Trends in the literature, strength and weaknesses were identified, and a conceptual model was illustrated.

Parameter comparison

Overall, 65 parameter estimates were harvested from 48 papers (tables 1–3).

Viral load dynamics

Viral load was reported from 21 papers using real-time RT-PCR testing, generally postsymptomatic monitoring.3 29 40–59 Qualitatively, the viral dynamics described early increase in viral load, peaking around onset or within 2–4 days of symptom onset (figure 5 for a theoretical model), before decreasing gradually over the next 1–3 weeks postsymptom onset. Maximum duration of detection ranged from approximately 20–49 days, with the longest duration associated with faecal samples (see Discussion). The duration where RNA was recoverable by RT-PCR may have been truncated due to insufficient follow-up in some cases. Studies that have investigated blood samples have provided some evidence for an association with severity of infection,^{16 60} though it is not clear whether this is a consistent feature of SARS-CoV-2 infection.⁴⁰

It should be noted the lack of data on presymptomatic or asymptomatic cases with regards viral load. An exception was Kam et al,⁶¹ who describe a presymptomatic case in an infant. In another study, Zou et al ⁵³ undertook serial RT-PCR testing from nasal and throat swab samples from 14 imported cases and 4 secondary cases, in Guangdong, China. The dynamics of the infection in terms of Ct values and RNA copy number were described; Ct values of 30.76, 27.67, 24.56 and 21.48 corresponding to 1.5×10⁴, 1.5×10⁵, 1.5×10⁶ and 1.5×10⁷ copies per millilitre. Hence, lower Ct values infer higher viral loads. The authors report on a patient without symptoms, but with positive nasal swabs (Ct values: 22–28) and throat swabs (Ct values: 30–32) testing positive on days 7, 10 and 11 after contact. Importantly, the authors suggest ‘the viral load that was detected in the asymptomatic patient was similar to that in the symptomatic patients’. Furthermore, Kimbell et al ⁶² report that Ct values between asymptomatic (21.9–31.0), presymptomatic (15.3–37.9) and symptomatic cases (18.6–29.2) within a nursing home environment did not differ significantly. To et al ⁵⁹ present data on temporal profile of viral load from saliva samples and found that median initial and peak viral loads in severe cases were non-significantly higher (p>0.5) by approximately 1 log10 higher than those in mild cases. Liu et al ⁵⁸ present data showing viral load being 60 times greater for severe cases relative to mild cases.

This lack of presymptomatic data may result in left truncation of the risk distribution associated with viral load and shedding. Therefore, the typical timing of peak viral shedding (whether prior to, at or after onset), and its impact on transmission, is still uncertain. He et al ²⁹ reported highest viral load at symptom onset from patients sampled in a hospital in China. Furthermore, the author’s estimate using a separate infector–infectee dataset (n=77) that 44% (95% CI 25% to 69%) of infectee cases were infected during the presymptomatic stage of the infector. Separately, a modelling paper by Ferretti et al ⁶³ also appears to support this, estimating that 47% (0.9/2) of total transmission contributing R₀, an overall measure of transmission during an infection, was presymptomatic (also see ref 33).

Wölfel et al ⁵⁰ provides important data on a cohort of nine ‘mild’ cases that were serially tested using sputum, swabs (throat and nasopharyngeal), urine and faecal samples over time. Importantly, the virus was isolated, and inferences on viral replication could be made. Viral isolation and insights into viral replication improve inference around viral dynamics and transmission risk. The study suggested high viral loads shortly after symptom onset, which declined thereafter over time. Positive cultures were found from day 3–8 postsymptom onset (online supplementary figure S3), and the minimum 5% isolation success was achieved up to 9.8 (95% CI 8.5 to 21.8) days postonset from throat and lung samples but not faeces, blood or urine.

Overall duration findings

There are few data for the precise definition of the asymptomatic infectious period (T2) parameter. Some reported asymptomatic cases can actually be presymptomatic, when cases are subject to follow-up (eg, ref 66; see Discussion). However, Hu et al ⁷ do provide the data for asymptomatic cases (that remain asymptomatic) across their presumed infectious period. Therefore, in the first instance, a parameter mimicking their data is probably the best available data over the period of the present study. Note there is a large variation in this data parameter, and a gamma distribution of a shape alpha 3, beta 2, mean 6, may be appropriate for the initial model runs. Despite these being the primary informative data, caution is required, given the uncertainty around the relationship between RT-PCR results and infectiousness. Overall, an informed central tendency of ~6 days, with very low probability draws for durations >20 days for the T2 parameter may be considered given the current state of knowledge.

The presymptomatic period is sometimes referred to as ‘preclinical infectious’ period (parameter T3). This has been estimated from several papers, and the central tendency of these estimates vary from <1 to 4 days, cautiously approximating to 2 days, on average. Current models have used central tendency estimates of 0.5–2.4 days.14 15 26 39 The relative consistency around the duration of this period allows for some confidence of its distribution. Current understanding of viral dynamics of infection suggest that viral load and shedding increases during postlatent phase, peaking around onset (for symptomatic cases), before declining.^{29 50 53} This aspect of the natural history of infection may be important when attempting to model transmission dynamics.

Length of infectious period in symptomatic cases that do not isolate (T5 parameter) has also been rarely directly measured in the literature, as serial monitoring of patients in terms of symptoms or viral load (RT-PCR) generally occurs after diagnosis and/or after admission to hospital (from a modelling perspective, this means cases are censored as they are assumed to no longer contribute to transmission). If natural progression of infection after diagnosis or hospital admission mimics the course of infection for those who do not isolate, the review of the literature describing time to two clear tests is informative. Symptom onset to serial testing clearance (assessed the time to first of two RT-PCR clear tests) averaged 13.4 days from our meta-analysis. In the maximal case, where patients succumb or fully recover from infection, time from symptoms to death or discharge may be informative. Studies that collated such information suggest mean durations of 18.07 days but with time to discharge being 4.96 days shorter on average than time to death. These values may represent an overestimation of the infectious period; one study suggested that there was on average 2.5 days between end of infectiousness and ‘removal’ (recovery or death).³⁷

Cheng et al ³³ provided evidence of transmissibility, based on attack rate from primary to secondary cases, at around symptom onset. The authors estimate cumulative infectiousness from onset, which suggests that 67% of total infectiousness potential occurs by the first day postonset. Most of the total infectiousness occurs within 5 days (86.9%) postonset, with the remaining infectiousness potential (13.1%) being distributed up to day 30 (this truncation is an assumption by the authors). It is possible that presymptomatic transmission occurred during this study, but the authors do not estimate what proportion of transmissions occurred during a presymptomatic infectious period or its potential duration.

A model by He et al ²⁹ is informative for overall symptomatic duration (T3+T5), using 77 infector–infectee pairs where COVID-19 transmission occurred in China. The study reported that infectiousness was apparent on average 2.5 days prior to symptoms, reached a peak in risk at 0.6 days before symptoms and decline up until 7 days after onset (9.5 days total infectious period). The proportion of transmission before symptom onset (area under the curve) was estimated as 44% (95% CI 25% to 69%) based on inferences on incubation period. The authors suggest their data supported the view that transmission risk decline substantially after 7 days postsymptoms onset.

Model estimates used for infectious period parameter appears to be shorter than virological studies tracking RNA viral load over time. For example, Li et al ²⁷ fitted a flat prior distribution for mean duration (D) fixed to vary between: 2≤D ≤5 days, and Lavezzo et al ⁶⁴ fixed infectious period to 2 days in their epidemic model, whereas viral repeat testing studies provide evidence to suggest high viral loads can be detected to up 20 days (eg, pharyngeal swabs) and potentially longer from faecal samples (3–4 weeks or longer postsymptom onset)). Oral–faecal transmission risk is currently unknown, but some doubt has been raised about studies that have reported positive RT-PCR test results (see ref 68; but there may be some evidence of the risk among children⁶⁹). Wölfel et al ⁵⁰ has produced an important study that provides some data on viral replication and the site and duration over which this may be taking place. Their data suggest that viral replication, with high viral loads, occur in the upper respiratory tract, over the first week of symptoms peaking in day 4. Virus could not be isolated from faecal samples, despite high RNA concentration. Furthermore, virus was not isolated from blood or urine in that study.⁵⁰

It should be noted that some of the virological and tracing studies reviewed had small sample sizes (see Study limitations) and potentially biased towards more severe cases or clusters of infection. It is unknown as to whether these cases are representative of infectious duration generally across populations. However, if symptom severity is linked to infectious duration, one could speculate that this bias could help to explain the some of the difference between model and empirical duration estimates.

Study limitations

Overall, the studies included were of good quality, though due to the rapid need for information from the global research community, many papers are preprints that have yet to be reviewed (at time of writing). Many papers were limited in terms of sample sizes, with several papers being case studies of one patient or single cluster outbreaks. There was a diversity of methods employed to infer dynamics of infectiousness across studies, and therefore the evidential base was variable. Some issues around nomenclature were noted, including definitions of asymptomatic, infectious period, latent and incubation period. It is possible the same data may have been used across different studies, especially where publicly available data were used.

There was significant heterogeneity across study findings, and this was related to diversity of clinical findings and methods employed. The meta-analysis employed for one parameter (T5) using virological studies, where cross-study comparisons could be made, suggested that the heterogeneity was high. Fu et al ⁷⁰ cautions against combining studies to give an overall estimate without exploring subgroup or meta-regression analysis, which we have done here. The meta-regression was based on a small number of studies (n=12–13). Cochrane’s handbook suggests 10 studies for each level of a meta-regression; however, in practice much lower numbers have been used to test hypotheses,²² as is the case here. Fu et al ⁷⁰ recommend a minimum of four studies per category, and therefore we dichotomised our predictor variables to ensure we met this minimum. Aggregating our categories resulted in crude findings.

Another limitation is that a systematic review was not undertaken to inform this research, hence there is a possibility that some relevant studies were overlooked. However, two independent research groups conducted comprehensive search strategies as part of a broader epidemiological parameters project for COVID-1912 13 71–73 to inform this research, hence limiting the potential for missing key studies.