Novel Approaches: “Nowcasting” Suicide Trends
CDC’s scientists tapped into new methods and data sources to create real-time dashboards
Suicide is a major cause of mortality in the United States and is a priority topic for CDC’s National Center for Injury Control and Prevention (NCIPC). However, understanding suicide trends in real time is challenging because mortality data can take about a year to be finalized.
To address this challenge, the NCIPC Data Science Team needed to do two things: they needed to find additional types of data that could give insights on suicide deaths, and then they needed to develop new methods to analyze and share the results quickly within CDC.
New data, new methods
The first step of the project involved bringing together data from multiple new data sources that could help widen the lens on suicide. These novel data sources include real-time emergency department chief complaint data*, search data, social media data, and online forum data.
The NCIPC Data Science Team then established an innovative way to understand current trends using these data. They applied new methods to the data, such as machine learning and prediction modeling, to create a real-time prediction and visualization platform. They also automated the whole process to reduce manual labor.
The result is an internal-facing, interactive dashboard CDC’s scientists can use to “nowcast” suicide death trends nationally on a week-to-week basis.
A first-ever process holds promise
This work is the first time that the team has used a machine learning approach to predict injury trends. The good news is that they have successfully been able to reduce the time it takes to understand suicide death trends by over a year. In a recent suicide cluster investigation in Iowa, the dashboard was used by CDC staff to provide real-time national context to the investigation.
The benefits of this work are many. For example, tapping into these new datasets will open the doors for more innovative projects in the future. The process that was established is now being adapted for other injury outcomes, such as firearm violence and overdoses. Automating the process also required the exploration and understanding of many new tools, including CDC’s new Enterprise Data Analytics and Visualization platform, and represents the promise of CDC Data Modernization efforts.
*ED data from NSSP and Consumer Product Safety Commission’s NEISS-AIP, collected by medical abstractors for injury ED data from a nationally representative sample of 60+ hospitals
- JAMA publication on the methodology: Development of a Machine Learning Model Using Multiple, Heterogeneous Data Sources to Estimate Weekly US Suicide Fatalities