You are here

Forecasting Influenza Occurrence to Improve ED Operations

The dashboard shows flu map for the state of Louisiana, flu forecast for the city of Baton Rouge (top right) along with environmental conditions including temperature, humidity and precipitation at that time (bottom right)

Given how the virus strain changes every year and related environmental factors seasonal influenza forecasting is a challenging problem. Researchers at the Center for Visual and Decision Informatics (CVDI) developed a novel big data based real-time seasonal influenza forecasting technique projects titled "Visual Analytic Approaches for Mining Large-Scale Dynamic Graphs."

This novel influenza forecasting model that uses a two-stage vectorized time series model that captures the influence of local environmental weather conditions (based on frequent associations between the flu severity and weather conditions). It can be used to forecast patients visiting emergency departments for influenza type illness. To forecast future flu occurrences, the impacts of environmental conditions and spatiotemporal flu spread characteristics are integrated into the vectorized time series model.

The United States Centers for Disease Control (CDC) monitors weekly flu projections and provides data that is anywhere from a week to two weeks old. There are also several real-time flu surveillance systems for flu monitoring based on search keys from Google, and social media trends from Twitter. These models only provide real-time monitoring capabilities rather than forecasting.

There is considerable evidence in literature about the influence of environmental factors (temperature, humidity, precipitation) on influenza virus survivability, and patterns of spread in space and time. However, the influence of environmental factors had not previously been captured adequately for real-time forecasting of influenza. The resulting model from CVDI researchers outperforms much better accuracy performance compared to existing time series based influenza forecasting.

The seasonal influenza forecasting model that was developed from this research is currently being adopted by the Schumacher Group. The influenza prediction model will be used to forecast emergency department visits for influenza type illness. It is expected to be deployed across the 130 emergency department facilities in the United States that the Schumacher Group manages. The predictive analytics capabilities of this model will help Schumacher Group and their partnering hospitals to better manage emergency department resources, to better staff emergency department facilities, and to more effectively allocate resources.

Economic Impact:

The ability to accurately predict influenza volume contributes to the organization’s ability to prepare for staffing and other operational impacts. Proactive resource management impacts hospitals’ abilities to make most efficient use of its resources, including providers. Given the high percentage of expenses in emergency departments (EDs) that are associated with direct provider cost, modest reductions to provider cost can be expected to result in significant and positive impact on ED operating costs.



For more information, contact Raju Gottumukkala at the University of Louisiana at Lafayette, raju@louisiana.edu, http://www.nsfcvdi.org/peopleprofiles/raju-gottumukkala/, 337.482.0632.

PDF icon CVDI-2016.pdf