Researchers’ predictive algorithm to forecast coronavirus outbreaks taps clinicians’ searches in UpToDate from Wolters Kluwer
Wolters Kluwer, Health today announced that a new predictive algorithm is using UpToDate® search activity from clinicians at the point of care—together with data streams from social media, internet search trends and mobility data from smartphones—to serve as a coronavirus early-warning system that could forecast an outbreak two to three weeks in advance. The algorithm is presented in a pre-published paper on arXiv.com.
Researchers, including Dr. Mauricio Santillana, Director of the Machine Intelligence Lab, Boston Children’s Hospital, and an Assistant Professor of Pediatrics and Epidemiology at Harvard Medical School, created a novel model to combine information from clinicians’ searches in UpToDate, as well as COVID-related Twitter posts that included geotag location information, Google searches on COVID topics, anonymous mobility data from smartphones, predictions from a model developed by researchers at Northeastern University, and temperature readings from the Kinsa Smart Thermometer app.
“UpToDate search data is an important part of this algorithm because it gives us insight into what information clinicians need right now,” said Dr. Santillana. “A COVID-19 early warning system that analyzes multiple data streams, including clinician searches at the point of care could help providers and public authorities to anticipate new epidemic waves and implement effective interventions.”
In order to test the ability of the algorithm to forecast COVID-19 outbreaks, researchers compared how each data stream correlated with case counts and deaths in each state in March and April. For example, an abrupt increase in tweets about COVID-19 appeared over a week before reported cases spiked dramatically in New York in the middle of March. Related Google searches and Kinsa temperature readings increased several days before the spike as well. The hybrid algorithm estimated outbreaks by an average of 21 days.
“Covid-19 created a global need for reliable information with unparalleled urgency. Our team responded by synthesizing evidence-based guidance that addressed the many complex and dynamic clinical questions facing healthcare professionals and patients,” said Peter Bonis, Chief Medical Officer for Clinical Effectiveness at Wolters Kluwer, Health. “Our COVID-19 topics have been viewed more than 10 million times by front-line providers around the world, and we are pleased that an added benefit from such information-seeking is the ability to track the pandemic.”
Read the full study here.