In this study, we retrieved GT data on the keywords coronavirus (as a virus and search term), cough (as a topic), anosmia (as a disease), and ageusia (as a topic).

With the exception of coronavirus, no other nontopic or nondisease search terms were used. In fact, we tested the search terms loss of smell, hyposmia, olfaction, dysgeusia, and loss of taste [8] using translations of the terms into native languages of the studied countries (using double quotation marks when searching for keywords containing more than one word); however, the data retrieved with these queries were not consistent or of sufficient quality.

We obtained country-level GT data for all analyses except for the worldwide analysis of the last five years. We retrieved GT data for the following time periods:

Categories and subcategories were not selected when searching for keywords.

After plotting worldwide GT data on the selected keywords for the last five years, we retrieved GT data for the year 2020 and assessed the trends of those queries in the 17 countries where searches for anosmia were sufficient to perform an analysis.

To further assess the impact of media coverage on COVID-19–related GT data, and to assess whether the GT data correlated with COVID-19 cases, we focused on 8 countries in different stages of the COVID-19 pandemic: France, Germany, Italy, Portugal, Spain, the United Kingdom, Brazil, and the United States. For each country, we plotted the weekly GT data for selected keywords together with weekly data on new COVID-19 cases (numbers retrieved from official sources).

Subsequently, we performed an analysis restricted to the time period prior to March 16, 2020, the date that the media started reporting that anosmia and ageusia can be symptoms of COVID-19. In fact, from that date onward, GT data could largely reflect interest in media coverage rather than searches for symptoms that patients were experiencing. Therefore, for each country, between the date of the first confirmed COVID-19 case and March 15, 2020, we assessed the correlation (by means of the Pearson correlation coefficient, r) between the daily average of GT for anosmia and ageusia (herein reported as anosmia/ageusia) and daily data on new COVID-19 cases. Similar analyses were performed for new COVID-19 deaths (in the time frame from the date of the first COVID-19 death to March 15, 2020).

To facilitate plot reading, we plotted normalized weekly data on COVID-19 cases and deaths. That is, we plotted the total number of new COVID-19 cases and deaths as percentages of the respective maximum weekly values observed during the defined time period.

The results of this study suggest that COVID-19–related GT queries do not necessarily follow the evolution of the epidemic and, in particular for anosmia and ageusia, are more closely related to media coverage.

Using a stepwise approach based on 1- and 5-year perspectives, we showed that search peaks not only for coronavirus but also for anosmia/ageusia appeared for the first time in 2020; also, there may be a relationship between the two peaks. This is different from the cough search term, for which searches were detected for all years but which also showed a peak coincident with the coronavirus peak.

We then assessed countries with an identifiable anosmia peak in 2020 in the northern and southern hemispheres. Surprisingly, in all countries, peaks for coronavirus, cough, and anosmia/ageusia all occurred simultaneously, irrespective of the pandemic stage. A simple interpretation is that this is unlikely to be associated with COVID-19 incidence. However, the time of onset differed for coronavirus or cough versus anosmia or ageusia; the latter coincided with the timing with which media news covered information on these symptoms (Table 4).

We subsequently studied the peaks for coronavirus, cough, and anosmia/ageusia. The peak for anosmia/ageusia is delayed compared to that for cough, which is a major symptom of COVID-19. The peaks were usually short (1 week), confirming that most of the queries were driven by media coverage. Prior studies have also pointed out that GT data are highly influenced by media [23,24]; due to media coverage, aberrant ragweed pollen peaks were observed during the grass pollen season [25]. In fact, one important limitation of demand-based infodemiological studies is the difficulty of distinguishing the effects of a true biological epidemic from what generates interest or apprehension in internet users [2,5]. In that sense, complementing search data with click data has been suggested as a partial solution to overcome this limitation [2].

The correlation between anosmia/ageusia and deaths or new cases of COVID-19 varied substantially among countries. Depending on the country, there was a high correlation or no correlation at all. Prior to March 16, in countries with higher COVID-19 infection or death rates, there were moderate to good correlations between queries on anosmia/ageusia and new COVID-19 cases or deaths. This suggests that in the absence of substantial changes in media coverage and in the presence of a sufficiently high COVID-19 incidence, GT data mostly reflect searches for symptoms patients are experiencing. Thus, the strong correlations found by Walker et al [15] may reflect the facts that they analyzed GT data for anosmia/ageusia only up to March 25, 2020 (ie, up to the week before searches for anosmia/ageusia started to decrease); that their analyses on the associations between COVID-19 cases/deaths and premediatic coverage of anosmia GT data were restricted to three countries (the United Kingdom, Spain, and Italy); and that this premediatic coverage was considered by the authors to have occurred up to March 20, 2020 (ie, searches between March 16 and 19 were misclassified because they had already occurred under the potential influence of media coverage).

Our study has some potentially relevant limitations. We used data at national levels, which may have not captured within-country heterogeneity on COVID-19 incidence or GT data; different results may have been obtained if the data were assessed at a more granulated level. Another relevant limitation concerns the fact that by March 16, 2020, the incidence of COVID-19 was still low in most Western countries; with the exception of Italy and Spain, the remaining Western countries had fewer than 100 confirmed COVID-19 cases per million inhabitants. The possibility of assessing a larger number of countries with higher numbers of COVID-19 cases would have allowed us to more confidently assess anosmia and ageusia search patterns (and their association with COVID-19 epidemiology) before and after media coverage on those symptoms.

Another important GT limitation concerns the representativeness of internet users [26]. Internet use is lowest among older persons, who constitute the age group with the highest COVID-19 morbidity. Finally, GT provides relative rather than absolute numbers, which may limit across-country comparisons. However, as expected, similar correlation coefficients were obtained when comparing GT data with relative or absolute numbers of COVID-19 cases/deaths.

At least in the initial stages of the SARS-CoV-2 pandemic, COVID-19–related web searches may more closely reflect media coverage (and subsequent users’ interest or apprehension) than epidemiological trends. The use of Google Trends has increased dramatically in the last decade; whereas in the past, the focus had been on surveillance and monitoring, the focus of research has now shifted to forecasting changes [27]. It appears to be important to link GT with other sources of data to overcome the limitations of using search information alone.