Inference of seasonal and pandemic influenza transmission dynamics

Abstract

The inference of key infectious disease epidemiological parameters is critical for characterizing disease spread and devising prevention and containment measures. The recent emergence of surveillance records mined from big data such as health-related online queries and social media, as well as model inference methods, permits the development of new methodologies for more comprehensive estimation of these parameters. We use such data in conjunction with Bayesian inference methods to study the transmission dynamics of influenza. We simultaneously estimate key epidemiological parameters, including population susceptibility, the basic reproductive number, attack rate, and infectious period, for 115 cities during the 2003-2004 through 2012-2013 seasons, including the 2009 pandemic. These estimates discriminate key differences in the epidemiological characteristics of these outbreaks across 10 y, as well as spatial variations of influenza transmission dynamics among subpopulations in the United States. In addition, the inference methods appear to compensate for observational biases and underreporting inherent in the surveillance data.

Keywords: asymptomatic infections; data assimilation; influenza; spatial patterns; transmission dynamics.

Fig. 1.

Comparison of key epidemiological variables/parameters over 2003–2004 through 2012–2013 flu seasons: (A) susceptibility at the onset, (B) attack rate over the entire season/pandemic wave, (C) the basic reproductive number R₀, (D) the maximum effective reproductive number R_e, (E) infectious period D estimated at the maximal epidemic forcing, and (F) γ at the maximal epidemic forcing. Each subplot shows the pattern of differences over the 10 y computed by Tukey's honest significant difference test. The three segments on each horizontal line denote the mean and the 95% confidence intervals of difference between that season and season 2003/2004 (the first season in this study), determined by an ANOVA test. A horizontal line segment intersecting the vertical dash line (x = 0) indicates the parameter estimate for that season is not significantly different from season 2003/2004 and vice versa; similarly, overlapping between two horizontal line segments indicates estimates for those two seasons are not significantly different and vice versa. The mean estimate (absolute value) is listed next to each season in red. pdm denotes pandemic.

Fig. 2.

Correlation of R₀ over the 2003–2013 period between cities. A total of 62 cities have complete ILI+ records during 2003–2013; correlations of R₀ over the nine interpandemic seasons and two pandemic waves between every pair of cities were computed. The heat map shows the correlation of chronological R₀ between each city pair (names of cities are shown on the x and y axes). Black lines divide the cities by HHS regions.