Periodic Trends in Internet Searches for Ocular Symptoms in the US

Abstract

Purpose:

To identify periodic trends in eye-related internet-based search queries and to determine the seasonal peaks and troughs.

Methods:

This cross-sectional study examined publicly available Google trends data from the United States (01/01/2015 to 12/31/2019). A list of common ocular symptoms was compiled from the American Academy of Ophthalmology Eye Health website and Wills Eye Manual. Eye-related symptoms were stratified into categories involving vision change, eye pain, or eye redness. The search volume over time for each term was modeled using periodic regression functions and the goodness-of-fit was reported. Fisher’s exact tests were used to compare the characteristics of periodic vs. non-periodic query terms.

Results:

Seasonal trends were demonstrated by 45% (48/106) of the eye-related symptoms included in this investigation. Search terms with best fit to the periodic model included stye (r² = 0.89), pink eye (r² = 0.82), dry eye (r² = 0.76), blurry vision (r² = 0.72), and swollen eye (r² = 0.71). Periodic search terms were more likely to involve eye redness (21% vs. 11%, p = 0.014) and less likely to involve vision change (11% vs. 36%; p < 0.001). Periodic queries involving eye redness most often peaked in the spring and those involving eye pain peaked in the summer.

Conclusion:

Eye-related symptom queries directly reflect seasonal trends for allergic eye disease and ocular trauma. Search query analyses can serve as accurate epidemiological tools with research and real-world clinical applications.

Introduction:

There is high variability in the presentation patterns of individuals with common eye complaints. Patients exhibiting minor symptoms are intuitively less likely to present to an eye specialist for evaluation. Those symptomatic enough to seek care often initially visit a primary care or urgent care physician. The fragmentation of the initial presentation setting makes it challenging to understand the temporal incidence of highly prevalent ocular conditions.

Internet-based searches are increasingly used by patients to self-assess eye-related symptoms before evaluation by a physician. The analysis of Google search trends has been studied extensively in health care research^1–3 with applications including tracking disease outbreaks,^4,5 surveying lifestyle diseases,⁶ and identifying patterns in the pursuit of medical care.⁷ The Google Trends database has been used in ophthalmology to demonstrate the seasonality of eye-related search terms particularly associated with allergic and viral eye disease.^8–10 However, seasonal trends in other categories of ocular symptoms have not been well characterized.

We hypothesize that internet-based search queries can be leveraged to better understand the epidemiologic patterns of common ocular symptoms that may otherwise not present to an eye specialist. In this study, we model the periodic trends of a large array of ocular chief complaints to increase our understanding of temporal incidence patterns.

Materials and Methods:

This was a cross-sectional study of search queries performed using Google Trends data from the United States over a 5-year period (01/01/2015 to 12/31/2019). This study adhered to the guidelines of the Declaration of Helsinki and involved the use of publicly available, aggregated data that was not considered non-human subjects research and exempt from Institutional Review Board approval.

Search term database:

A database of common ocular symptom terms potentially used by individuals to self-evaluate eye problems was compiled. Eye-related symptoms were parsed from the American Academy of Ophthalmology (AAO) Eye Health website, an educational tool designed for patients, containing a list of common ocular chief complaints (https://www.aao.org/eye-health, accessed June 2020).¹¹ Common symptoms and diagnoses were also extracted from the Wills Eye Manual (6th Edition, Chapter 1, Differential Diagnosis of Ocular Symptoms), a reference manual for eye care providers (eSupplement 1).¹² We excluded any queries containing $>$ 3 words as these are generally low-frequency queries which are not reported in Google Trends (Figure 1). Each term included in the analysis was stratified based on symptom category involving change in vision, eye redness, or eye pain.

Figure 1:

Flow chart summarizing the methodology for selecting eye-related symptoms used in analysis. AAO = American Academy of Ophthalmology Eye Health, Wills = Wills Eye Manual.

Google Trends query:

The weekly search interest for each eye-related term in the database was extracted, and the primary outcome was the relative search interest, defined as the percentage of the maximum number of searches for a given term in any one week interval during the study period. For privacy reasons, Google Trends does not provide the specific frequency of infrequently searched terms and instead outputs a relative interest equal to zero. Therefore, to explore only the most frequently searched terms, we excluded any terms with low search frequency defined as $\le$ 5 % of weekly queries with a relative interest equal to zero.

Regression model:

Generalized linear models were used to fit the relative interest of each search query term over time. We modelled periodicity using a combination of sine and cosine functions based on a Fourier series and included a linear slope term to account for any potential increase in annual search volume during the study interval. The periodic function was defined as:

$$f(t)\hspace{0.17em}=\hspace{0.17em}{A}_0\hspace{0.17em}+\hspace{0.17em}{A}_{1}t\hspace{0.17em}+\hspace{0.17em}{A}_2\hspace{0.17em}\cos (\omega t)\hspace{0.17em}+\hspace{0.17em}{A}_3\hspace{0.17em}\sin (\omega t)$$ 1

Where t is time, $\text{ω}$ is period, A₀ is the y-intercept and A₁ is the slope. Given our interest in annual seasonal trends, we set the period equal to $\text{2}\pi / 365\text{.25}$. A₂ and A₃ are constants used to calculate the amplitude of the periodic function using the formula:

$$amplitude\hspace{0.17em}=\hspace{0.17em}\sqrt{{A_2}^2\hspace{0.17em}+\hspace{0.17em}{A_3}^2}$$ 2

Where amplitude is defined as the difference between the mean query frequency and nearest peak/trough. We assessed goodness-of-fit using coefficients of determination (r²). We classified the query pattern of each search term as either periodic or non-periodic. Query terms with a fitted regression model with r² > 0.10 and amplitude > 2.5% were considered periodic terms, whereas all others were considered non-periodic terms. We compared the types of symptom categories of periodic and non-periodic terms using Fisher’s exact tests. We described temporal trends with seasons defined by the northern hemisphere astronomical equinox and solstice dates. All statistical tests were two-sided and p < 0.05 was considered statistically significant. Statistical analysis was performed using R, version 4.1.0 (R Core Team, 2021)¹³ with the gtrendsR package, version 1.4.6.¹⁴

Results:

The ocular symptoms (N = 106) in this study included terms derived from the AAO Eye Health portal (N = 57), Wills Eye Manual (N = 77), and excluded queries containing > 3 terms (N = 9) and low-frequency queries (N = 19) (Table 1). The regression analysis identified 45% (48/106) periodic search terms and 55% (58/106) non-periodic terms.

Table 1:

Differences in ocular chief complaint category between periodic and non-periodic search terms.

	Eye-related search terms: N (%)
Category of term	Non-periodic	Periodic	p-value
Vision change	38 (35.8)	12 (11.3)	<0.001
Redness/irritation	12 (11.3)	22 (20.8)	0.012
Pain	8 (7.5)	14 (13.2)	0.06

Individual terms with the best fit to the periodic model included stye (r² = 0.89), pink eye (r² = 0.82), dry eye (r² = 0.76), blurry vision (r² = 0.72), and swollen eye (r² = 0.71). The terms with greatest change in frequency between peak and trough included eye allergy (15%), pink eye (13%), eye sand (12%), eye discharge (9%), and eye pus (9%) (Figure 2 and eSupplement 2) Non-periodic search terms were more likely to involve vision change category (36% vs. 11%; p < 0.001) and less likely to involve eye redness or irritation categories (11% vs 21%, p = 0.014) (Table 1).

Figure 2:

Eye-related symptom queries with suspected periodicity (N = 48) above threshold r² > 0.10 and amplitude > 2.5%. Arranged top-left to bottom-right in descending order of r². Blue bars indicate the winter seasons.

Periodic queries in the eye redness category peaked in the summer (67%) and those involving eye pain peaked in the summer (57%). Fourteen (29%) of periodic eye-related search queries peaked in the winter, 23 (48%) in the spring, 11 (23%) in the summer and 0 (0%) in the fall. The most query peaks were 11 (23%) in May, 8 (17%) in March, and 8 (17%) in June. Conversely, 8 (17%) of periodic queries experienced a trough in the winter, 1 (2%) in the spring, 17 (35%) in the summer, and 22 (46%) in the fall (Table 2). The most query troughs were 11 (23%) in November, 10 (21%) in August, and 9 (19%) in October (Figure 3 and eSupplement 2).

Table 2:

Seasonal differences in the peak timing of each category of ocular symptom.

	Season: N (%)
	Winter	Spring	Summer	Fall
Vision change	4 (8.3)	6 (12.5)	2 (4.2)	0 (0.0)
Redness/irritation	7 (14.6)	14 (29.2)	1 (2.1)	0 (0.0)
Pain	3 (6.3)	3 (6.3)	8 (16.7)	0 (0.0)

Figure 3:

Scatter plot of amplitudes of suspected periodic queries as a function of peak (amplitude > 0) and trough (amplitude < 0) times. Search terms are used in place of points. Colours represent seasons: Blue = Winter, Green = Spring, Red = Summer, Purple fall.

Discussion:

In this study, we used Google Trends data to identify periodic trends in many common ocular symptoms. Our regression-based approach found that nearly half of the eye-related search terms included in the study demonstrated some evidence of periodicity. Peaks in search queries related to eye redness were most common in the spring, and peaks in eye pain queries occurred most frequently in the summer season. Eye-related symptom queries reflect seasonal trends for allergic eye disease and ocular trauma.

Our results support previous work using Google Trends that has demonstrated seasonal trends in allergic and infectious eye disease,^9,10 and identifies several new periodic eye-related search terms. Query terms related to eye redness were most likely category of symptom to demonstrate periodic trends in search volume. The eye redness category included symptoms related to allergic and infectious aetiologies including pink eye, eye discharge, itchy eye, and eye redness. Notable periodic trends identified in the eye pain category include symptoms related to trauma and infection such as corneal abrasion, eye scratch, swollen eye, eye swelling, and corneal ulcer, or eyelid infection/inflammation such as stye, chalazion, and eyelid bump. Finally, periodic trends identified in the vision change category include symptoms related to glare and refractive error.

We found that the spring season, and specifically the month of May, to have the most peaks of the search terms. This is consistent with the traditional time of presentation of allergic eye disease, with many of the symptoms peaking at this time (eye tearing, eye discharge, eye redness) often present in allergic eye disease. Conversely, we found the fall season and specifically the month of November to have the most troughs of the search terms we queried. None of the terms with periodic trends based on our model had a peak in the fall season, which likely reflects that many of the symptoms typically associated with the fall, such as those related to allergic eye disease, are significantly more prevalent in the spring.

Interestingly, the term with the best fit to our regression model was stye with r² = 0.89 with a peak during the summer. The seasonality of the term stye has been previously reported, though with a lower goodness-of-fit (r² = 0.19) and association with temperature-related environmental factors.¹⁰ The study by Leffler and colleagues was conducted using Google Trends data from 2004–2008, and we suspect that the increase in use of Google searches over the past decade may have contributed to a greater number of queries for the word stye-thereby decreasing the noise and revealing the periodic nature of the query (eSupplement 3). Internet-based surveillance tools such as Google Trends, become more powerful at revealing clinically meaningful patterns the more frequently they are used. Clinically, these findings may allow primary care practitioners and eye-care providers to recommend increasing simple interventions such as warm compresses and lid hygiene as we approach the summer months for patients with a predilection for styes.

There are several limitations of our study design and analytic approach. First, although query frequency has been shown to be associated with clinical diagnosis in cases of conjunctivitis,⁹ it may not be correlated with the incidence of other types of ocular disease. Furthermore, even in the event of correlation, one must consider the clinical relevance of relative increases in query frequency, which can be initiated for any number of reasons aside from personal experience of that symptom. Second, although our modelling approach offers considerable interpretability for seasonal trends, other functions incorporating additional periodic or non-linear terms such as splines may provide better overall fit to the patterns observed in this study. Third, we define a threshold for suspected periodicity using both goodness-of-fit and amplitude of the function with a conservatively low threshold to capture any potential periodic terms with lower frequency that may otherwise be obstructed by the noise in the dataset. We chose the amplitude because it reflects at minimum 5% change in search volume between the peak and trough, which we felt would be the minimum change of interest for a term. Naturally, a more stringent threshold would yield fewer suspected periodic queries. Although this will not affect any of the high amplitude, high fit terms, the terms near the thresholds should be interpreted with consideration of these cut-offs. Third, it is important to consider the demographics of population that is being studied using the Google Trends database. We must be cognizant of potential sampling bias in the demographic and socioeconomic status of the individuals this database represents (internet access, computer/mobile device availability, computer literacy), as well as the cultural and linguistic barriers to using this technology to self-assess eye symptoms. The latter is particularly important given that approximately 10 % of the US population is considered limited English proficient according to the US Census Bureau.¹⁵ As such, this approach may struggle to capture the eye symptoms of specific groups including children, the elderly, individuals with disabilities, and non-English speaking individuals. We restricted our searches geographically to the United States and English language given anticipated challenges of compiling synonymous terms across languages or expressions across different regions of the world.

There are many exciting applications of search query analysis in ophthalmology. Studies have already used this tool to explore non-infectious/allergic conditions such as the severity dry eye symptoms and correlation with weather fluctuations⁸ and even used it as a tool to gauge public interest in ophthalmic laser procedures.¹⁶ Our results support the use of search query analysis to provide insight into the periodic trends of many common eye symptoms, which may not otherwise present to an eye specialist. Search query analyses can serve as valuable epidemiological tools to better prepare physicians and health systems for the management of seasonal eye conditions.

Supplementary Material

Supplementary Material

eSupplement 1: Eye-related symptoms and diagnoses included in this study obtained from the American Academy of Ophthalmology Eye Health Website and Wills Eye Manual.

eSupplement 2: Summary of search terms with suspected periodicity (N = 48) above threshold r² > 0.10 and amplitude > 2.5%. Parameters include chief complaint category, goodness of fit (r²), amplitude, rate of increase and peak/trough times.

eSupplement 3: Stye search queries from 2004 to 2020 illustrating how the periodic nature of the query becomes more apparent with increased search volume over time.