An article from The Economist on October 25 proposed to use cell phone positioning data to understand people movements and help curb the spread of Ebola. That got us thinking that SGI with its expertise and large memory systems can make this happen.
Within days, we assembled a team of people from SGI, GIS Federal, and the University of Minnesota to tackle the problem. The team developed a plan to model the spread of the disease using a database containing anonymized cell phone numbers/unique id, timestamp, latitude & longitude.
The solution takes advantage of:
- The SGI® UV™ 300 system with terabytes of cache coherent main memory and flash memory, with hundreds of CPUs and 16 GPUs, able to hold the movements of potentially hundreds of millions of people in main memory for fast processing and real time what-if scenarios
- GIS Federal’s GPU accelerated in-memory database, GPUdb, which performs near real-time processing of big data streams with native geospatial visualization support and is optimized for the SGI UV system
- The University of Minnesota Epidemiology Research team and algorithms for modeling the spread of the disease
At SC14 in mid-November, we demonstrated a prototype of the solution as a tool for health organizations to be able to react quickly to a new disease, and today in particular Ebola.
The tool contains data on people movement, allows flagging one or more people as infected, backtracking where those people have been, and show other possible infections through contact.
Examples of how this tool may be used include:
- Notification of potentially infected people so they take precautions.
- Predict the spread of the disease based on movement of infected people to assist with preparedness.
Current efforts to detect Ebola cases in the U.S. are largely based on individuals, who were potentially exposed to or contracted the disease, voluntarily reporting their suspicion to health carriers, agencies, or practitioners. For that reason, surveillance efforts are mostly, passive.
By using information provided by cell phone networks, it is possible to identify potential risk contacts based on the history of movements. Knowledge of these contacts would allow for more effective active surveillance mechanisms at a relatively low cost.