Assessing time-varying characteristics of epidemiological parameters and improvement in impact assessment of public health and social measures for COVID-19 in Hong Kong and mainland China

Abstract

<p><strong>Background & aims of study</strong></p><p>In general, assessing the changes in the activity and transmissibility of SARS-CoV-2 virus, the impact of the public health and social measures (PHSMs) against COVID-19 was evaluated. Where, along with daily infection activity, the serial interval distribution (time between successive symptom onsets in a transmission chain) is required to evaluate the transmissibility (e.g., effective or instantaneous reproduction number) at temporal scale. In practice, serial interval distributions found to vary over time and true epi-curve is not observed at real time. Therefore, our aim to assess the true epi-curve and serial interval distributions at temporal scale to evaluate impact of COVID-19 PHSMs.</p><p><strong>Methods & results</strong></p><p>We first reconstructed the transmission pairs from the line-list information for ancestral variant of COVID-19 in Hong Kong and mainland China. We evaluated the time-varying effective serial interval disributions, using 10-14 day of sliding window for fitting a series of distribution including normal distribution via. MCMC. We also used a likelihood data augmentation framework to predict the true epi-curves and estimated the instantaneous reproduction number (R_t) incorporating the effective serial interval distributions. We used a series of multivariable regression models to evaluate the impact of different PHSMs and finally compared with traditional setup of Rt. We found the serial interval distribution was varying significantly across the waves and locations and ranged in (1.8 days (sd 4.3 days) to 7.8 day (sd 5.2 days)) with decreasing and increasing trend over the pandemic span. The instantaneous reproduction number ranged from 2.4 (95% CI: 3.1, 2.0) to 0.5 (0.9, 0.3). We found that case isolation, other PHSMs and the case profile were the significant factors for explaining the variance in serial intervals and hence in transmissibility. We estimated up to 12%-35% excess variance could be explained by these PHSMs when evaluated under effective serial intervals and true epi-curves compared to that of traditional constant serial intervals.</p><p>Conclusions: Our findings highlight the need of evaluating temporal changes in the serial interval distribution and true epi-curves to predict real time transmissibility, to improve assessing transmission dynamics and impact of control measures.</p>