Use of Google Trends to investigate anosmia: power and pitfalls of infodemiology.

Abstract

To the Editor: We read the recent study by Walker et al.1 with great interest. The authors describe using Google Trends (GT) relative search volume (RSV) data to correlate increases in online searches for smell loss to coronavirus disease 2019 (COVID-19) cases and deaths in several countries. The study illustrates the potential power of infodemiology or digital epidemiology to use big data from online search engines and web browsers for public health use in the identification and response to infectious disease outbreaks. The authors found significant correlations between search volumes for anosmia and new COVID-19 cases and deaths in Italy, Spain, the United Kingdom, Iran, the United States, Germany, France, and the Netherlands. The presumption is that patients with COVID-19 and anosmia will search for this symptom or browse related articles and web pages. The authors note that infodemiology approaches are limited by media coverage or news events that may distort online search trends, because it cannot weed out those individuals who search for the term merely out of curiosity. For the case of anosmia and COVID-19, the Ear, Nose, Throat surgery–United Kingdom (ENT-UK) press release on March 20, 2020 was such an event.2 Except for Iran, Italy, and Spain, most countries studied in this article suffered the major portion of their COVID-19 outbreaks after this event. The correlation data presented in Table 1 does include the time surrounding the ENT-UK press release and subsequent media coverage may skew the presented results. The authors did state that they investigated the confounding effect of the ENT-UK statement by analyzing Spain, Italy, and the United Kingdom only prior to March 19, 2020, and found preserved strong correlations. In our previously published infodemiology study,3 we similarly found strong correlations in most countries RSV for anosmia and new COVID-19 cases and deaths. However, in performing autoregressive integrated moving average (ARIMA) modeling and calculating lag correlations, we found that anosmia had positive lag, meaning that the searches lagged behind actual cases or deaths. The lag directly correlated to the influx of media coverage over this new symptom. Interestingly, the 1 country that did not have a strong correlation was China. Only by looking regionally at Wuhan did we find significant correlation between smell/taste and new COVID-19 cases. This finding speaks to the importance of analyzing local and regional data. Large countries such as the United States or China may give misleading results. We also found that other less publicized symptoms such as headaches, myalgia, and chest pain were, perhaps, better predictors. The relationship between anosmia and COVID-19 is now broadly recognized by medical professionals and lay persons alike because of previous work from the authors and others. The utility of infodemiology in determining if online searches related to anosmia can be used to track new cases of COVID-19 may be distorted by media coverage and related searches. It might be fruitful to look again at these data well after the widespread dissemination of this information. Infodemiology may be best utilized for new or less publicized symptoms in emerging diseases or in well recognized symptoms for more established diseases.