Internet Searches for Acute Anxiety During the Early Stages of the COVID-19 Pandemic

Paper by John W. Ayers et al: “There is widespread concern that the coronavirus disease 2019 (COVID-19) pandemic may harm population mental health, chiefly owing to anxiety about the disease and its societal fallout. But traditional population mental health surveillance (eg, telephone surveys, medical records) is time consuming, expensive, and may miss persons who do not participate or seek care. To evaluate the association of COVID-19 with anxiety on a population basis, we examined internet searches indicative of acute anxiety during the early stages of the COVID-19 pandemic.Methods

The analysis relied on nonidentifiable, aggregate, public data and was exempted by the University of California San Diego Human Research Protections Program. Acute anxiety, including colloquially called anxiety attacks or panic attacks, was monitored because of its higher prevalence relative to other mental health problems. It can lead to other mental health problems (including depression), it is triggered by outside stressors, and it is socially contagious. Using Google Trends ( we monitored the daily fraction of all internet searches (thereby adjusting the results for any change in total queries) that included the terms anxiety or panic in combination with attack (including panic attacksigns of anxiety attackanxiety attack symptoms) that originated from the US from January 1, 2004, through May 4, 2020. Raw search counts were inferred using Comscore estimates (

We compared search volumes after President Trump declared a national COVID-19 emergency on March 13, 2020, with expected search volumes if COVID-19 had not occurred, thereby taking into account the historical trend and periodicity in the data. Expected volumes were computed using an autoregressive integrated moving average model,4 based on historical trends from January 1, 2004 to March 12, 2020, to predict counterfactual trends for March 13, 2020 to May 9, 2020. The expected volumes with prediction intervals (PIs) and ratio of observed and expected volumes with bootstrap CIs were computed using R statistical software (version 3.5.3, R Foundation). The results were similar if we varied our interruption date plus or minus 1 week….(More)”.