Early Detection of COVID-19 Epidemic Using Search Engine Query Data

Background: Early warnings of emerging infectious disease are crucial to prevent epidemics. However, in the early stage of the COVID-19 pandemic, traditional infectious disease surveillance failed to deliver a warning alert. The aim of this work is to develop search-engine-based surveillance methods for the early warning and prediction of COVID-19 outbreaks. Methods: By using more than 444 million Baidu search queries from China as training set, we collected 32 keywords from the Baidu Search Index that may related to COVID-19 outbreak from 18 December 2019 to 11 February 2020. The Beijing Xinfadi outbreak from 30 May 2020 to 30 July 2020 was used as independent test set. A multiple linear regression was applied to model the relationship between the daily query frequencies of keywords and the daily new cases. Findings: Our results show that 11 keywords in search queries were highly correlated to the daily numbers of confirmed cases (r =0.96, P <0.01). An abnormal initial peak (1.46 times the normal volume) in queries appeared on 31 December 2019, which could have served as an early warning signal for an outbreak. Of particular concern, on this day, the volume of the query “Wuhan Seafood Market” increased by over 240 times (from 10 to 2410), the volume of the query “Wuhan outbreak” increased by over 622 times (from 7 to 4359), and 17.5% of China’s query volume originated from Hubei Province, 51.15% of which was from Wuhan city. The quantitative model using four keywords (“Epidemic”, “Masks”, “Coronavirus” and “Clustered pneumonia”) successfully predicted the daily numbers of cases for the next two days, and detected an early signal during the Beijing Xinfadi outbreak (R2 =0.80). Interpretation: Our study demonstrates the ability of search engine query data to detect COVID-19 outbreaks, and suggests that abnormalities in query volume can serve as early warning signals.