The data in electronic health records can be harnessed to better predict, in near real-time, regional estimates of flu outbreaks in the U.S., according to an article in Scientific Reports.
Influenza is a leading cause of death in the U.S.; detecting and monitoring it can enable health officials and providers to better combat it. While the Centers for Disease Control and Prevention (CDC) collects this data, there's a time lag of up to two weeks. The data also tends to be more nationally skewed, as opposed to locally or regionally, where it could be more useful.
The researchers, from Harvard and athenahealth Research, developed an algorithm called AutoRegressive Electronic health record Support vector machine (ARES). It was used in combination with historical epidemiological information and athenahealth's EHR data from its 72,000 provider users, which included claims data from more than 64 million lives and EHRs of over 23 million lives. It pulled information about the flu, such as patient flu visits and vaccinations, studying records from June 28, 2009 through Oct. 15, 2015.
ARES's estimates of flu outbreaks were more accurate and timely than CDC estimates. It correctly estimated the timing and magnitude of the national peak week of three flu seasons. It also outperformed Google's Flu Trends estimator, which was shut down in summer of 2015.
"EHR data in combination with historical patterns of flu activity and a robust dynamical machine learning algorithm provide a novel and promising way of monitoring infectious diseases at the national and local level," the researchers said.
To learn more:
- here's the study