The serial interval of COVID-19 from publicly reported confirmed cases

Abstract

We estimate the distribution of serial intervals for 468 confirmed cases of COVID-19 reported in 93 Chinese cities by February 8, 2020. The mean and standard deviation are 3.96 (95% CI 3.53–4.39) and 4.75 (95% CI 4.46–5.07) days, respectively, with 12.6% of reports indicating pre-symptomatic transmission.

One sentence summary

We estimate the distribution of serial intervals for 468 confirmed cases of COVID-19 reported in 93 Chinese cities by February 8, 2020.

A new coronavirus (COVID-19) emerged in Wuhan, China in late 2019 and was declared a public health emergency of international concern by the World Health Organization (WHO) on January 30, 2020 (1). As of February 19, 2020, the WHO has reported over 75,204 COVID-19 infections and over 2,009 COVID-19 deaths (2), while key aspects of the transmission dynamics of COVID-19 remain unclear (3). The serial interval of COVID-19 is defined as the time duration between a primary case (infector) developing symptoms and secondary case (infectee) developing symptoms (4,5). Obtaining robust estimates for the distribution of COVID-19 serial intervals is a critical input for determining the reproduction number which can indicate the extent of interventions required to control an epidemic (6). However, this quantity cannot be inferred from daily case count data alone (7).

To obtain reliable estimates of the serial interval, we obtained data on 468 COVID-19 transmission events reported in mainland China outside of Hubei Province between January 21, 2020, and February 8, 2020. Each report consists of a probable date of symptom onset for both the infector and infectee as well as the probable locations of infection for both cases.

The data include only confirmed cases that were compiled from online reports from 18 provincial centers for disease control and prevention (Table S3).

Notably, 59 of the 468 reports indicate that the infectee developed symptoms earlier than the infector. Thus, pre-symptomatic transmission may be occurring, i.e., infected persons may be infectious before their symptoms appear. In light of these negative-valued serial intervals, we find that COVID-19 serial intervals better resemble a normal distribution than more commonly assumed gamma or Weibull distributions (8,9) that are limited to strictly positive values (see Supplement). We estimate a mean serial interval for COVID-19 of 3.96 [95% CI 3.53–4.39] with a standard deviation of 4.75 [95% CI 4.46–5.07], which is considerably lower than reported mean serial intervals of 8.4 days for SARS (9) and 12.6 days (10) - 14.6 days (11) for MERS. The mean serial interval is slightly but not significantly longer when the index case is imported (4.06 days [95% CI 3.55–4.57]) versus locally infected (3.66 days [95% CI 2.84–4.47]); it is slightly shorter when the secondary transmission occurs within a household (4.03 days [95% CI 3.12–4.94]) versus outside of the household (4.56 days [95% CI: 3.85–5.27]). Combining these findings with published estimates for the early exponential growth rate COVID-19 in Wuhan (12,13), we estimate a basic reproduction number (R₀) of 1.32 [95% CI 1.16–1.48] (6), which is lower than published estimates that assume a mean serial interval exceeding seven days (13–15).

These estimates reflect reported symptom onset dates for 752 cases from 93 Chinese cities, who range in age from 1 to 90 years (mean 45.2 years and SD 17.21 years). Recent analysis of COVID-19 case data from mainland China, Taiwan, Hong Kong, Vietnam, South Korea, Germany and Singapore have reported average serial intervals of 7.5 days [95% CI 5.3–19] (13), 4.4 days [95% CI 2.9–6.7] (16) and 4.0 days [95% CrI 3.1–4.9] (17) based on considerably smaller samples of 6, 21 and 28 infector-infected pairs, respectively. Whereas none of these studies report negative serial intervals in which the infectee developed symptoms prior to the infector, 12.6% of the serial intervals in our sample are negative.

We note four potential sources of bias in our estimates, three of which are likely to cause underestimation of COVID-19 serial intervals. First, the data are restricted to online reports of confirmed cases and therefore may be biased towards more severe cases in areas with a high-functioning healthcare and public health infrastructure. The rapid isolation such cases may have prevented longer serial intervals, potentially shifting our estimate downwards compared to serial intervals that might be observed in an uncontrolled epidemic. Second, the distribution of serial intervals varies throughout an epidemic, with the time between successive cases contracting around the epidemic peak (18). To provide intuition, a susceptible person is likely to become infected more quickly if they are surrounded by two infected people rather than just one. Since our estimates are based primarily on transmission events reported during the early stages of outbreaks, we do not explicitly account for such compression and interpret the estimates as basic serial intervals at the outset of an epidemic. However, if some of the reported infections occurred amidst growing clusters of cases, then our estimates may reflect effective (compressed) serial intervals that would be expected during a period of epidemic growth. Third, the identity of each infector and the timing of symptom onset were presumably based on individual recollection of past events. If recall accuracy is impeded by time or trauma, cases may be more likely to attribute infection to recent encounters (short serial intervals) over past encounters (longer serial intervals). In contrast, the reported serial intervals may be biased upwards by travel-related delays in transmission from primary cases that were infected in Wuhan or another city before returning home. If their infectious period started while still traveling, then we may be unlikely to observe early transmission events with shorter serial intervals. Indeed, the mean serial interval is slightly higher for the 218 of 301 unique infectors reported to be imported cases.

Given the heterogeneity in type and reliability of these sources, we caution that our findings should be interpreted as working hypotheses regarding the infectiousness of COVID-19 requiring further validation as more data become available. The potential implications for COVID-19 control are mixed. While our lower estimates for R₀ suggest easier containment, the large number of reported asymptomatic transmission events is concerning.

Figure. Estimated serial interval distribution for COVID-19 based on 468 reported transmission events in China between January 21, 2020 and February 8, 2020.

Bars indicate the number of infection events with specified serial interval and blue lines indicate fitted normal distributions for (a) all infection events (N = 468) reported across 93 cities of mainland China by February 8, 2020 and (b) the subset infection events (N = 122) in which both the infector and infectee were infected in the reporting city (i.e., the index case was not an importation from another city). Negative serial intervals (left of the vertical dotted lines) suggest the possibility of COVID-2019 transmission from asymptomatic or mildly symptomatic cases.

Author Bio

Dr. Du is a postdoctoral researcher in the Department of Integrative Biology at the University of Texas at Austin. He develops mathematical models to elucidate the transmission dynamics, surveillance, and control of infectious diseases.