The details of data sources can be found in Multimedia Appendix 1. For China, all provinces, autonomous regions, and municipalities (including mainland China, Hong Kong, Macau, and Taiwan) that had reported cases of COVID-19 were identified according to the daily updates by the National Health Commission of China [24]. Subsequently, the official websites and WeChat accounts (if any) of local governments and health authorities (eg, Municipal Health Commission, Center for Disease Control and Prevention, Department of Health) were checked manually through April 8, 2020, to identify and download the reports on laboratory-confirmed cases of COVID-19. Google and Baidu were searched to identify public media reports written based on official press releases. Chinese words for the following terms were used to perform the search: (“family” OR “household”) AND (“cluster” OR “dinner” OR “party”) AND “infection.”

Other countries reporting COVID-19 cases were identified according to the COVID-19 situation reports published by the World Health Organization [25] and the data were searched through April 8, 2020. For Japan and Singapore, information regarding confirmed cases was retrieved from their respective Ministry of Health. We also searched for individual cases from relevant departments and public media of the United States, the United Kingdom, Canada, and Australia, but failed to find any with details allowing for parameters estimation for this study. Typically, the dates of exposure and symptom onset were lacking (eg, see reference [26]). Owing to the language barrier, we did not do a comprehensive search for other countries.

PubMed was searched to identify relevant publications by using the following terms: “coronavirus,” “2019-nCov,” “SARS-CoV-2,” and “COVID-19.” The China National Knowledge Infrastructure (CNKI) was searched to identify publications in Chinese journals using “novel coronavirus” (“Xin Xing Guan Zhuang Bing Du” in Chinese pinyin), which is the official Chinese name for SARS-CoV-2. Both databases were searched from December 2019 through April 8, 2020. The reference lists of eligible publications were also checked to see if there were other eligible studies not found by previous searches. We also kept an eye on the COVID-19 studies disseminated by public media and the official WeChat accounts of various academic entities in China and those recommended by experts in this field to the authors of this study.

To be eligible, a report had to contain individual data that allowed for estimation of at least one of the following parameters of laboratory-confirmed cases of COVID-19: incubation period, serial interval, and symptoms-to-transmission time, which was defined as the day of an infectee’s contact with the infector relative to the latter’s symptom onset date, with a negative value meaning that the transmission occurred before the infector developed symptoms (ie, presymptomatic transmission).

To obtain an accurate estimate of incubation period, only the cases with an exposure period spanning 3 days or less were included in the analysis. For those exposed for three continuous days and those exposed on two dates with one day apart (ie, exposed on the first and third days), the second day was uniformly used as the exposure date in estimations. For those exposed for two continuous days, the first day was uniformly used as the exposure date in estimations. This approach ensured the upper limit of error in the estimated incubation period be smaller than 1 day for the cases with a 2- or 3-day exposure, regardless of when exactly (ie, first, second, or third day) the transmission occurred. The actual overall error was bound to be much smaller than 1 day, as most included cases were exposed for only 1 day, which would dilute the error caused by 2- or 3-day exposures.

Serial interval was estimated based on the symptom onset dates of infector-infectee pairs, which were typically from cluster infections. For two cases to qualify as an infector-infectee pair and be included in this study, the following two criteria must both be fulfilled. First, there must be evidence that the presumed infector had been exposed outside the cluster (eg, close contact with a confirmed case, travel history to Hubei, exposure to a person who returned from Hubei) before he/she attended the group activities (eg, family gathering, business conference) that led to cluster infections. Second, in the 14 days prior to symptom onset, the presumed infectee was exposed to the presumed infector only, without other exposure histories.

The estimation of symptoms-to-transmission time also involved determination of exposure date and judgement about transmission chain, hence the above principles applied in estimating incubation period and serial interval were followed as well.

In total, 6 researchers were involved in data collection. The reports retrieved through the above searches were scrutinized one by one according to the inclusion criteria specified above. The following data were extracted from eligible reports by using a standard extraction form which was pilot tested with the reports from Liaoning province of China: the geographical location concerned, age, sex, type of exposure, first date and period (if applicable) of exposure, date of symptom onset, initial symptoms, and whether the case was from a cluster. For a clustering case, the generation he/she belonged to, the exposure date, and symptom onset date were also recorded. The retrieved reports were split into six parts, with each researcher responsible for one part. For each part, the eligibility of and data extracted from reports were first determined by one researcher and then cross-checked by another three researchers. All uncertainty and disagreements were discussed among the researchers. If no consensus could be reached after discussion, the concerned cases would be excluded to ensure the correctness and accuracy of data. For example, if it could not be determined who was infected through attending a group activity (ie, infectee) and who had been infected before he/she attended the activity (ie, infector), then all clustering cases related to the activity had to be excluded.

The basic characteristics of included cases were summarized descriptively. Lognormal, Weibull, and gamma distributions were fitted to the data sets of incubation period and serial interval and the one with the smallest Akaike information criterion (AIC) score was used for the final analysis. For serial interval, which had negative values, shifted distributions were fitted, with the best-fitting values determined by maximum likelihood. The key parameters were estimated by using the maximum likelihood approach. The data on symptoms-to-transmission time was roughly symmetrical according to visual inspection and thus fitted by normal distribution. For each parameter, the range, median, selected percentiles, mean, and standard deviation were estimated. The 95% confidence intervals of mean and standard deviation were estimated by using the bootstrap technique. R₀ was estimated by the well-studied Euler-Lotka equation [19]: R₀ = 1/M[−r|h(∙)], where r is the exponential growth rate, h(∙) is the estimated distribution of the generation time, and the function M[∙] is the Laplace transform (ie, moment generating function known for statisticians) for h. As generation time is hard to observe directly, it was approximated by serial interval in this study. Compared with generation time, serial interval is expected to have the same mean but a larger variance due to possibly differing incubation periods of infectors and infectees. The exponential growth rate was obtained directly from a previous study reporting incidence data of the early stage of the epidemic [1], while the other parameters came from this study.

Sensitivity analyses were conducted to examine the robustness of the two parameters involving exposure date. Specifically, the third day for those whose exposure period spanned three days and the second day for those exposed for two continuous days were used as their exposure dates in sensitivity analyses. The difference between the estimates from sensitivity analysis and those from primary analysis represents the largest possible error in the latter. All statistical analyses were conducted with SAS software (Version 9.4, SAS Institute).

No ethical approval was needed as the data used in this study was publicly available, either released by health authorities or retrieved from electronic databases.

Figure 1 shows the process of data collection. We systematically retrieved and screened 20,658 case reports and 9942 publications. The 43 publications included in our analysis can be found in Multimedia Appendix 2. In total, 1591 cases were included for the final analysis, including 1211 (76.12%) from China, 301 (18.92%) from Japan, 60 (3.77%) from Singapore, and 19 from five other countries. The cases from China covered 156 cities in 31 provinces of 34 in total.

The age of included cases ranged from 5 days to 95 years (Table 1). There were 133 cases (10.74%) with a travel history to Hubei province, where Wuhan is the capital city. The source of exposure of 1302 cases (81.99%) was known confirmed cases. The exposure period was exactly 1 day in 508 cases (53.08%), spanned 2-3 days in 227 cases (23.72%), and spanned more than 3 days in the other cases (which were not included in the estimation of parameters involving exposure date). The top 5 initial symptoms were fever (730/1009, 72.35%), cough (257/1009, 25.47%), fatigue (69/1009, 6.84%), sore throat (57/1009, 5.65%), and malaise (54/1009, 5.35%). Clustering cases accounted for 84.92% (1351/1591) of all, and most of them were second generation.

Incubation period was estimated from 687 cases, of whom 482 (69.55%) were exposed for 1 day; the remainder were exposed for 2 or 3 days. The incubation periods of individual cases ranged from 0 to 23 days (Table 2), with a median of 6 days; in total, 5.82% (n=40) of incubation periods were longer than 14 days. The incubation period is best described by a gamma distribution (Multimedia Appendix 3, Figure 2A), with a mean of 7.04 days (95% CI 6.74-7.33) and standard deviation of 4.27 days (95% CI 3.92-4.43).

The serial interval was estimated for 1015 infector-infectee pairs and ranged from −5 to 29 days (Table 2), with a median of 6 days. For 32 pairs (3.15%), the serial interval was less than 0; for these 32 pairs, the mean was −2.50 days, meaning that the infectee showed symptoms 2.50 days earlier than the infector. The serial interval is best described by a shifted gamma distribution (Multimedia Appendix 3, Figure 2B), with a mean of 6.49 days (95% CI 6.15-6.80) and standard deviation of 4.90 days (95% CI 4.68-5.25). Based on an exponential growth rate of 0.10 per day [1] and the mean and standard deviation of the serial interval, R₀ was estimated at 1.85 (95% CI 1.37-2.60). The country-specific R₀ for China, Japan, and Singapore was 1.83 (95% CI 1.37-2.54), 1.83 (95% CI 1.37-2.53), and 1.98 (95% CI 1.41-2.97), respectively.

The symptoms-to-transmission time was estimated from 296 infector-infectee pairs and ranged from −9 to 14 days (Table 2), with a median of 0 day. Presymptomatic transmission occurred in 129 infector-infectee pairs (43.58%, 95% CI 37.93%-49.23%); for these 129 pairs, the mean symptoms-to-transmission time was −2.88 days, meaning that the transmission occurred 2.88 days before the infectors showed symptoms. Fitted by a normal distribution (Figure 2C), the mean of this parameter was −0.07 days (95% CI −0.43 to 0.31) and the standard deviation was 3.31 days (95% CI 2.96 to 3.66).

The symptoms-to-transmission time can also be inferred by mean serial interval minus mean incubation time (−0.55 days), which is consistent with the direct estimate (−0.07 days) as both suggest the mean symptoms-to-transmission time to be around the day before primary cases’ symptom onset. Based on the above estimates, the timeline of infection for an “average” infector-infectee pair in a transmission chain is demonstrated in Multimedia Appendix 4.

In sensitivity analyses, the estimates of incubation period and symptoms-to-transmission time remained stable, with a difference of 0.24-0.3 days from those in primary analysis (Table 2), which represents the largest possible error caused by inclusion of cases with an exposure period spanning 2 or 3 days. The results of the stratified analysis are summarized in Multimedia Appendix 5. The difference between strata was statistically significant in four of the analyses (see footnote of Multimedia Appendix 5). Briefly, the incubation period was longer for those exposed for only 1 day (as compared with those exposed for 2-3 days), while serial interval and symptoms-to-transmission time were longer for those at lower generations of clustered infection. The serial interval was also longer for those infected through nonhousehold contact.