Effective real-time transmission estimations incorporating population viral load distributions amid SARS-CoV-2 variants and pre-existing immunity

Background: Population-level viral load distribution, measured by cycle threshold (Ct), has been demonstrated to enable real-time estimation of Rt for SARS-CoV-2 ancestral strain. The generalizability of the framework under different circulating variants and pre-existing immunity remains unclear.

Methods: We obtained the first Ct record of local COVID-19 cases from July 2020 to January 2023 in Hong Kong. The log-linear regression model, fitting on daily Ct mean and skewness to the Rt estimated by case count, was trained using data from wave 3 (i.e., ancestral strain with minimal population immunity), and we predicted Rt for wave 5, 6 and 7 (i.e., Omicron subvariants with > 70% vaccine coverage). Cross-validation was performed by training on the other 4 waves. Stratification analysis by disease severity was conducted to retrospectively evaluate the impact of the changing severity profiles.

Results: Trained with the ancestral dominated wave 3, our model can accurately estimate whether Rt was above 1, with the area under the receiver operating characteristic curve of 0.98 (95% confidence interval: 0.96, 1.00), 0.62 (95% CI: 0.53, 0.70) and 0.80 (95% CI: 0.73, 0.88) for three Omicron dominated waves 5 to 7, respectively. Models trained on the other four waves also had discriminative performance. Stratification analysis suggested the potential impact of case severity on model estimation, which coincided with the fluctuation of sampling delay.

Conclusions: Our findings suggested that incorporating population viral shedding can provide timely and accurate transmission estimation with evolving variants and population immunity. Model application needs to account for sampling delay.