Serial interval and incubation period of COVID-19: a systematic review and meta-analysis

We identified relevant studies through searching electronic databases and gray literatures. Additionally, we were searched from the reference lists of all the included studies to identify any other studies that may have been missed by our search strategy.

We followed the preferred reporting items for systematic review and meta-analysis (PRISMA) checklist for this study [12]. A comprehensive search strategy was performed from international electronic databases (Google Scholar, PubMed, Science Direct, Web of Science, CINAHL, and Cochrane Library) by two experienced review (MAA and DBK) authors between 1st of June and the 31st of July 2020. The following searching terms are used from the above databases: “serial interval” OR “generation time” AND “incubation period” OR “infectious period” AND “COVID-19” OR “SARS-CoV-2” OR “novel coronavirus”.

Articles that were not fully accessed after at least two email contacts of the principal investigator were excluded. In addition, we excluded case reports, letters, and review articles.

The eligibility assessment was undertaken by two (WG and TYB) authors, independently. The disagreement between two reviewers were fixed by consensus.

This study has two outcome variables. The first is the average estimate of serial interval. The serial interval is defined as the time from illness onset in the primary case to illness onset in the secondary case. It also measured from pairs of cases with a clear infector–infectee relationship. The second outcome variable is the average estimate of the incubation period. Incubation period is defined as the time from infection occurred to the onset of signs and symptoms. It was measured with cases of a well-defined period of exposure and symptom onset. Screening of studies and all essential data from the included studies were extracted independently by two (MA and LY) of the authors. This form includes the last name of the first author, country, data collection period, sample size, average estimate, standard deviation, and 95% confidence intervals. The same data extraction form was used for both outcomes. Discrepancies between the two reviewers was resolved by consensus involving all authors.

Two experienced reviewers (MA and DBK) were assessed the risk of bias of the included articles. The Newcastle-Ottawa Scale (NOS) adapted for cross-sectional studies was used to evaluate the quality of studies [13]. This tool includes three categories with a maximum score of 9 points. The first is the “selection” category, which accounts for a maximum of 4 points, the second is the “Comparability” category, which accounts for a maximum of 2 points, and the third is “outcome” which accounts a maximum of 3 points. Based on the composite score from this three categories, the studies were classified as good quality if the score ≥ 6 points, fair quality 2 to 5 points inclusively and poor quality ≤1 point.

A meta-analysis of continuous outcomes was employed for this study. We analyzed the data sets for each outcome variable (serial interval and incubation period). After extracting all essential data using Microsoft Excel, data were exported to R 4.0.2 statistical software for meta-analysis. In order to pool the results of included studies in a consistent format, we estimated the sample mean and standard deviation for studies that report median and interquartile range [14]. To determine the extent of variation between the studies, we did a heterogeneity test using the Higgins method, that was quantified by I² value [15]. Weighted average using the inverse variance method was used to estimate the pooled average. A random-effect meta-analysis with an estimation of DerSimonian and Laird method was performed. The publication bias was also assessed using a funnel plot and Egger’s tests [16]. The pooled average estimates with 95%CI confidence interval was presented using forest plots.

Figure 1 indicates the overall flow of study selection, literature search and number of the included studies. During electronic literature search 14,247 articles were identified and 14, 140 duplicated articles were removed. After meticulous review of the whole articles, 28 studies that fulfill the suitability standards were included. From the included studies, a single study might report both outcomes (serial interval and incubation period). Accordingly, a total of 23 and 14 studies were combined to estimate the mean serial interval, and incubation period of COVID-19, respectively.

All the included studies are cross-sectional, and half of them were preprints. Majority of studies included in this study are conducted in China. We included a total of 23 articles to pool the mean of serial interval of COVID-19. The minimum and maximum pairs of COVID-19 patients among the included studies were 6 [17] and 1407 [18], respectively. Among the included studies, the mean serial interval of COVID-19 was ranged from 4. 2 days [19] to 7.5 days [17] (Table 1).

Similarly, to pool the mean incubation period of COVID-19, a total of 14 articles were included. Among those, the minimum sample size was 10 [17] and the maximum was 183 [20]. The mean incubation period of COVID-19 ranged from 4.8 days [20] to 9 days [19] (Table 2).

In this study, a total of 3924 pairs of COVID-19 patients were included to pool the mean serial interval. Accordingly, the weighted overall mean serial interval of COVID-19 was 5.2 (95%CI: 4.9–5.5) days (Fig. 2). Likewise, a total of 1, 453 COVID-19 patients were included to pool the overall incubation period of COVID-19. Consequently, the weighted pooled mean incubation period of COVID-19 was 6.5 (95%CI: 5.9–7.1) days (Fig. 3).

Of the included studies to pool the mean serial interval of COVID-19, our summary quality assessment showed that nearly three-fourth (73.9%) of the studies had a good quality (Table S1). Similarly, among the included studies to pool the mean incubation period of COVID-19, about 71.4% of studies had a good quality (Table S2). We assessed the issue of publication bias by visual inspection of funnel plot and by using Egger’s regression test. Though the funnel plot looks asymmetrical the Egger’s test showed that no relationship between the effect size and its precision (Fig. 4).

The current study has a number of limitations. Firstly, the overall estimate of serial interval and incubation period were computed with in a considerable heterogeneity. The source of heterogeneity might be difference in study population, data collection period, and method of analysis. Secondly, the majority of the included studies had relatively small study participants which may decrease the power of the study. Thirdly, the review was limited to only articles published in the English language. Lastly, since the included articles are limited to few countries, it may not represent the global figure.