The results of this analysis indicate that searches for sarcoidosis are higher in the spring months, and significantly less so in the winter, particularly in the Northern Hemisphere. This suggests that the condition may be more commonly diagnosed in the spring. Overall, this seems to support findings of higher incidence in winter-spring, as observed in previous studies. Both the Minnesota and the Taiwanese studies showed lowest incidence in autumn, which was statistically significant [6, 7]. The Taiwanese study also found that almost two-thirds of individuals presented with sarcoidosis in winter or early spring [7].
Our study assumes that patients only search for sarcoidosis online after contact with the healthcare system and when they are most unfamiliar with the term, that is, when they are first diagnosed. In addition, there is often a delay between when patients develop symptoms and diagnosis; a delay would systematically skew results. If sarcoidosis truly is most common in the winter-spring and least common in autumn, a few weeks’ delay would shift search patterns forward in time, making winter appear to be less popular, reflecting lower incidence in autumn, followed by the spring peak, corresponding to incidence in winter-early spring. Once they have been diagnosed, we expect that patients would preferably consult other sources, such as their physician, or resources provided to them by their physician, when their disease relapses. There are cases of seasonal variation in hypercalcemia in sarcoidosis patients, so it is plausible that this could also impact Google search trends in a similar fashion [14].
Another important assumption was that the countries included follow a traditional four-season structure. We would expect opposing results when comparing southern hemisphere countries to those in the northern hemisphere; that is, that June–August would have the least search term popularity and September–November, the most. The results of the comparison of means in the southern countries showed mixed trends, none of which met statistical significance. December was less popular overall, but a smaller peak in interest was observed in September and October. The December phenomenon suggests that the variation may be due, in part, to a non-weather-related factor, but the latter peak is more consistent with the pattern observed in the Northern Hemispheric countries and true seasonal variation.
The main limitation of our study was the nature of the data obtained from Google Trends. In the interest of maintaining privacy, Google Trends only provides relative search volume, and it is not possible to obtain absolute values. This makes it difficult to ascertain that the data is sufficient to make valid conclusions. Other countries with less access to internet or using search engines other than Google would be excluded, despite a relatively high prevalence of sarcoidosis. Lastly, the search terms used in this study were English words, which probably limited non-Anglophone countries from fulfilling the inclusion criteria.
As mentioned earlier, the results obtained from Google Trends varies day-by-day; this is because Google Trends only analyzes a sample of the total searches [9]. With billions of Google searches daily, it would take too long to analyze all of these searches [9]. This causes fluctuations that could become significant, especially in this case, where the term of interest is less popular. When search volume is too low, a value of zero is designated; zeroes were seen in our datasets, including those from Ireland, despite being noted as being the most popular country for both search terms in the worldwide data [9].
In Moccia et al.’s study on trends in Multiple Sclerosis, the country selection criteria were more stringent, with a minimum relative search volume of 70 and population of 20 million to avoid “noise” from areas of less search interest [12]. This was not feasible for our study since the prevalence of sarcoidosis is lower than that of Multiple Sclerosis [15, 16]; utilization of this criteria would have resulted in exclusion of more than half of the countries analyzed. However, higher prevalence may also lead to more searches by individuals other than patients, reducing the ability to correlate search volume with disease patterns, the major confounder in the Multiple Sclerosis study [12].
The strongest data to support the overall results was obtained from the United Kingdom, which showed very significant data in every data set. This probably swayed the overall statistics for the Northern Hemisphere countries towards significance. The incidence of sarcoidosis is about to 11.5/100,000 in Sweden [17], but only estimated at 7 per 100,000 in Great Britain [18]. Yet, Sweden did not meet the relative volume requirements for this analysis, so the data does not always represent burden of disease. One article looking at the use of Google Trends in epidemiology concluded that data “seems to be more influenced by the media clamor than by true epidemiological burden” [19]. The study looked at several medical conditions, including Ebola: it showed no concordance with the geographic or temporal patterns of this disease [19]. Most notably, there were two significant peaks in search interest for “Ebola” in Northern Italy in 2014 despite not a single case there [19]. Though this is an extreme example, it does not negate the need to be cautious with Internet-based data.