Inferring generation time accounting temporal incubation periods and effective serial intervals for COVID-19

Abstract

Background: The generation time (GT) distribution, the time between successive infections in transmission chains, is often proxied by the serial interval distribution. However, when there is significant pre-symptomatic transmission, some observed serial intervals (SI) can be negative. In theory, unlike estimates of incubation period (IP) of infector and infectee can modulate the estimates of GT. Further, sampling biases could affect the estimates of incubation period and GT. Therefore, we aimed to correct these estimates for sampling bias and estimate temporal GT accounting the effects of NPIs. Methods: We first constructed the transmission pairs and exposure widows from the line-list information in mainland China during January-March 2020. We estimated the temporal incubation period and serial interval distributions for by using likelihood-based fitting approach and corrected for sampling bias. We developed an inferential framework to estimates using these corrected estimates incorporating a simple decomposition as: forward_GT = forward_SI + backward_IP of infector – forward_IP of the infectee. We also extended the joint framework to estimate the serial interval, incubation period and GT simultaneously. Results: The mean backward incubation period for infector fell in the range (4.2, 10.0) days with increasing trend over time, while forward incubation period of infectee found to be constant and ranged between 6.3 to 7.1 days and the forward SI was decreasing in (2.7, 8.2) days across the epidemic. After correcting the backward IP for infector, the estimates of forward GT found to be slightly decreased over time ranged from 4.5 to 6.9 days due to the effect of NPIs. Backward GT construction showed similar results. We estimated the reproduction number around 2.2 by using both forward SI and forward GT during pre-peak, assuming growth rate=0.14. Conclusions: Our result suggest to use both incubation period along with serial interval to estimate generation time distribution and hence reproduction number.