A robust parameter estimation method for estimating disease burden of respiratory viruses

Abstract

Background: Poisson model has been widely applied to estimate the disease burden of influenza, but there has been little success in providing reliable estimates for other respiratory viruses. Methods: We compared the estimates of excess hospitalization rates derived from the Poisson models with different combinations of inference methods and virus proxies respectively, with the aim to determine the optimal modeling approach. These models were validated by comparing the estimates of excess hospitalization attributable to respiratory viruses with the observed rates of laboratory confirmed paediatric hospitalization for acute respiratory infections obtained from a population based study. Results: The Bayesian inference method generally outperformed the classical likelihood estimation, particularly for RSV and parainfluenza, in terms of providing estimates closer to the observed hospitalization rates. Compared to the other proxy variables, age-specific positive counts provided better estimates for influenza, RSV and parainfluenza, regardless of inference methods. The Bayesian inference combined with age-specific positive counts also provided valid and reliable estimates for excess hospitalization associated with multiple respiratory viruses in both the 2009 H1N1 pandemic and interpandemic period. Conclusions: Poisson models using the Bayesian inference method and virus proxies of age-specific positive counts should be considered in disease burden studies on multiple respiratory viruses.