Methods for Estimating Disease Burden of Seasonal Influenza

Influenza is a common respiratory infection with substantial disease and economic burdens. Due to the threat of
another global pandemic, significant resources have been devoted to increase influenza surveillance, laboratory
capacity and pandemic preparedness worldwide since 2009. Disease burden estimates are critical for evaluating
vaccine benefits, for communicating prevention and control messages, and for developing evidence-based
policies for resource allocations. There are several major analytical challenges in estimating influenza disease
burden. First influenza symptoms are non-specific and testing is conducted at the discretion of healthcare
providers. Severe complications (e.g., pneumonia and cardiovascular events) may occur weeks after infection
when influenza viruses are no longer detectable or the patient’s symptoms may not suggest influenza. Second,
policy-relevant evaluation of influenza burdens at the national or global scales are often limited by the availability
of high-quality surveillance data. A common approach is to create multipliers for extrapolating available burden
estimates to other locations or larger populations, while introducing considerable uncertainties. There is a
pressing need to develop methods and tools to support burden estimation that will increase accuracy, improve
precision, enhance multi-partner collaboration, and quantify uncertainty appropriately. In this 2-year exploratory
project, we will examine the use of state-of-the-art approaches from epidemiology and evidence synthesis to
influenza burden estimation. In Aim 1, we will develop single-site time-series models for attributing counts of
adverse respiratory health outcomes to influenza. Our models will address several commonly encountered
analytic challenges, including residual temporal autocorrelation, overdispersion, and unmeasured temporal
confounders. By leveraging a unique multi-state emergency department (ED) visits database and three national
influenza surveillance systems, these methods will be applied to estimate season-specific influenza-associated
ED visits for 102 U.S. during the period 2005 to 2018. We will estimate burdens for specific age groups, sex and
influenza types. In Aim 2, we will develop data integration models for combining information across multiple sites
and perform predictions to sites without burden estimates. This involves the use of privacy-preserving, distributed
algorithms for multi-site analyses that can incorporate individual participant data, improve accuracy, account for
reporting bias, and potentially encourage participation. Methods will be applied to (1) estimate annual season-
specific influenza-associated ED visits in the U.S. nationally, and (2) estimate global burden of influenza-
associated hospitalization as part of an ongoing collaboration with the U.S. Centers for Disease Control and
Prevention. Anticipated outcomes from this project include (1) feasibility and performance evaluations of the
proposed time-series and data integration models; and (2) substantive findings on influenza-associated morbidity
as measured by ED visits and hospitalization for respiratory disease. Moreover, models developed in this project
are also widely applicable to other respiratory pathogens.

Grant Number: 5R21AI167418-02
NIH Institute/Center: NIH
Principal Investigator: Howard Chang