Visualizing flu forecast data with Viziflu
This article originally appeared on BioQuest and was co-authored by Andrea Brennen, Dylan George, and George S.
Viziflu is a visualization tool that displays multiple predictions about the timing of “peak week,” the week with the highest number of flu cases nationwide. By displaying the outputs of multiple influenza models and allowing users to compare the uncertainty across those models, Viziflu can help make influenza forecasts more actionable for decision-makers.
Infectious disease forecasting is becoming an important analytic tool for managing outbreaks, such as seasonal influenza. The Centers for Disease Control and Prevention (CDC) hosts an annual influenza forecasting challenge called FluSight, during which participating teams develop models to predict the timing of onset and peak week, the number of cases at peak, and the duration of the outbreak.
As described in the FluSight documentation, “accurate and timely forecasts for the peak week can be useful for planning and promoting activities to increase influenza vaccination prior to the bulk of influenza illness. For healthcare, pharmacy, and public health authorities, a forecast for the peak week can guide efficient staff and resource allocation.”
To the untrained eye, however, the output of forecasting models can be challenging to interpret, leaving key messages unclear. Well-designed visualizations can make forecasting data more actionable and support operational decision making.
IQT Labs designed & developed Viziflu to support interpretation of the output of influenza forecasting models. Viziflu runs in a browser and shows the forecasted temporal progression of seasonal influenza in the United States. The display highlights predictions about the timing of “peak week,” the week during which the highest number of flu cases is expected to occur.
Viziflu displays several important aspects of seasonal influenza forecasts in one graphical depiction:
- Timeline: the duration of the flu season is represented as a horizontal timeline from October through May.
- Multiple models: Multiple forecasting models are listed alphabetically on the left side of the display. The set of models currently shown were submitted by teams participating in the 2018–2019 CDC FluSight Challenge.
- Predicted probabilities of each model: A color gradient indicates each models’ predicted probability that flu will peak in each week of the flu season.
- Previous forecasting results: As the flu season progresses, the models are updated week-by-week. Along the bottom of the Viziflu display, a series of numbered tabs allows the viewer to toggle through these revised versions, to see how they evolve over the course of the flu season. A vertical black line indicates when in the season the forecasts were submitted to CDC.
Viziflu was developed following a B.Next workshop about how technology can advance infectious disease forecasting for outbreak management. Participants in that workshop identified the need for improved communication about public health data, including new visualization techniques to help convey risks and uncertainties to decision-makers and the public.
Viziflu is written in JavaScript and built using a number of open source libraries and resources, including D3.js, Colorbrewer 2.0, and visavail.js. The CDC’s Influenza Division provided domain expertise and guidance on the visualization design.
If you would like to explore Viziflu peak week visualizations, you can click here, to access a fully working demo of the 2017–2018 season or here, to see forecasts for the 2018–2019 season. We will soon update these public demos to include 2019–2020 seasonal forecasts.
Although Viziflu was designed for flu forecasts, we imagine the tool could be adapted for other applications where viewers would want to compare forecasted probabilities of occurrence for the timing of any event, as predicted by multiple models. If you are interested in using or adapting Viziflu, please check out this code repo and readme, which are available to the public for reuse/modification under the Apache 2.0 License.
The forecasting models presented in Viziflu are not official CDC forecasts and are not endorsed by either CDC or IQT Labs.
