29 April 2013

April Literature Review: Data Visualizations


ISDS’s Research Committee hosted another great Literature Review this month!


The ISDS Literature Review calls take place bi-monthly and are an opportunity to discuss journal articles related to biosurveillance. Recently, the Research Committee decided that, in order to generate more focused discussion, the Literature Review calls would be centered around topics of interest to the community. These topics are decided on by the Research Committee about two months prior to a Literature Review. The group’s Zotero library is then searched for keywords relating to the topic. Finally, these articles are sent to the Research Committee. This month the Literature Review was focused on data visualization.

Call Highlights:
Three articles were summarized and discussed on the April Literature Review Call:
While, the call participants found this to be an interesting presentation of data, there was interest in assessing reactions among public health practitioners (i.e., ‘Would practioners find this visualization useful?’).
  • Hafen RP, Anderson DE, Cleveland WS, et al. Syndromic surveillance: STL for modeling, visualizing, and monitoring disease counts. BMC Medical Informatics and Decision Making. 2009;9(1):21. Here the authors present a statistical outbreak detection method along with visualizations of data decomposition introducing a method they call STL, for “seasonal-trend decomposition procedure based on loess smoothing.” The authors compared the STL-derived detector to EARS methods C1, C2, and C3 and to a regression detector based on general linear modeling (GLM) for small, medium, and large-scale Emergency Department (ED) time series for detection of artificial lognormal signals at 3 levels. The STL method outperformed the EARS methods for in each situation. It outperformed the GLM method significantly for the smaller signals and slightly for the larger ones.
Again, there was general interest in this method from a theoretical viewpoint, but the question of practical utility to public health practitioners was raised. Essentially, would practitioner be interested in this standardized technique? If so, it may be worth developing into a standardized tool.
  • Ji X, Chun S, Geller J. Epidemic Outbreak and Spread Detection System Based on Twitter Data. In: He J, Liu X, Krupinski E, Xu G, eds. Health Information Science.Vol 7231. Lecture Notes in Computer Science. Springer Berlin / Heidelberg; 2012:152–163. This study discusses the design and utility of the New Jersey Institute of Technology’s Epidemics Outbreak and Spread Detection System (EOSDS). EOSDS utilizes publicly available information from Twitter to generate three different visualizations of the space and time dimensions of a spreading epidemic: static map, distribution map, and filter map. As an example of the utility of this system they considered a time period during which there was a severe outbreak of listeria in North America and compared the EOSDS visualizations (collecting a test dataset by specifying the keyword “listeria,” and monitoring the period from “09-26-2011” to “09-28-2011”) and CDC official reports – a gold standard. They reported that all three visualizations showed good correlation with CDC reports and concluded that EOSDS may be used as an effective early warning system.
While there is still debate about the utility of social media for disease surveillance, the study proposed geocoding and visualization methods that may provide information on health events prior to official reports. To see this system in action, visit the EOSDS website.

To view the full article summaries, please visit the ISDS April 2013 Summaries Wiki.
You may also review the Literature Review archives here.


Written by ISDS Program Manager, Tera Reynolds.

No comments:

Post a Comment