A Novel Epidemiological Approach to Geographically Mapping Population Dry Eye Disease in the United States through Google Trends
Daniel B. Azzam, Nitish Nag, Julia Tran, Lauren Chen, Kaajal Visnagra, Kailey Marshall, Matthew Wade
AA Novel Epidemiological Approach toGeographically Mapping Population Dry EyeDisease in the United States through Google Trends
Daniel B. Azzam B.S.
School of MedicineUniversity of California, Irvine
Irvine, [email protected]
Nitish Nag, Ph.D.
Donald Bren School of Information and Computer SciencesUniversity of California, Irvine
Irvine, [email protected]
Julia Tran, B.S.
School of MedicineUniversity of California, Irvine
Irvine, USA
Lauren Chen, B.S.
School of MedicineUniversity of California, Irvine
Irvine, USA
Kaajal Visnagra B.S.
School of MedicineUniversity of California, Irvine
Irvine, USA
Kailey Marshall, O.D.
School of MedicineUniversity of California, Irvine
Irvine, USA
Matthew Wade, M.D.
School of MedicineUniversity of California, Irvine
Irvine, [email protected]
Abstract —Dry eye disease (DED) affects approximately half ofthe United States population. DED is characterized by drynesson the corena surface due to a variety of causes. This studyfills the spatiotemporal gaps in DED epidemiology by usingGoogle Trends as a novel epidemiological tool for geographicallymapping DED in relation to environmental risk factors. Weutilized Google Trends to extract DED-related queries estimatinguser intent from 2004-2019 in the United States. We incorpo-rated national climate data to generate heat maps comparinggeographic, temporal, and environmental relationships of DED.Multi-variable regression models were constructed to generatequadratic forecasts predicting DED and control searches. Ourresults illustrated the upward trend, seasonal pattern, environ-mental influence, and spatial relationship of DED search volumeacross US geography. Localized patches of DED interest werevisualized along the coastline. There was no significant differencein DED queries across US census regions. Regression model1 predicted DED searches over time (R =0.97) with significantpredictors being control queries (p=0.0024), time (p=0.001), andseasonality (Winter p=0.0028; Spring p¡0.001; Summer p=0.018).Regression model 2 predicted DED queries per state (R =0.49)with significant predictors being temperature (p=0.0003) andcoastal zone (p=0.025). Importantly, temperature, coastal status,and seasonality were stronger risk factors of DED searches thanhumidity, sunshine, pollution, or region as clinical literaturemay suggest. Our work paves the way for future explorationof geographic information systems for locating DED and otherdiseases via online search query metrics. Index Terms —population health, Google Trends, healthylifestyle, personalized health, precision health.
I. I
NTRODUCTION
Google Inc. first initiated Google Trends in 2006 as a sourceof data that compares the popularity of search terms on GoogleSearch in various regions and languages from around the world[2], [20]. Google Trends is a tool that is free, readily availableworldwide, and updated daily at https://Google.com/trends/. In2009, the Centers for Disease Control and Prevention (CDC) partnered with Google and developed the first ever use ofdata from internet search engines to successfully predict fluoutbreaks in advance [15]. To this day, Google Trends hassuccessfully predicted trends in the utbreaks of many viruses,including West Nile virus, norovirus, varicella, influenza, andHIV [15], [19], [17], [40], [56]. More recently, Google Trendshas been used to study an increasing variety of healthcaredomains, such as diabetes [35], [51]. Dry eye disease ishistorically one of the most commonly seen conditions in theophthalmology clinic, with a significant detrimental impact onpatients quality of life [14], [36]. Epidemiological data forDED has traditionally been collected via surveys requiringexcessive time and resources, while providing only limiteddata specific to the time and population that is studied, suchas the Womens Health Study, the Physicians Health Studies,and the 2013 National Health and Wellness Survey [45], [44],[12]. The 2017 Tear Film and Ocular Surface Society Dry EyeWorkshop II (TFOS DEWS II) reported that in the last 10years there have been no population studies examining DEDprevalence in any region of the world south of the equator[8], [3]. In this study, we provide the most current estimatesof dry eye disease prevalence across the United States usingGoogle Trends data. Our analysis fills the gaps in current dryeye disease epidemiology, which was largely derived fromsurveys, by reaching beyond the limitations of time and space.We provide an analysis of how internet search intent can beextrapolated to map DED geographically and how this canbe compared in relation to environmental risk factors suchas temperature and humidity. This work serves as a proof ofconcept of using internet epidemiological tools along with GISdata from the environment as a mapping technique for diseasesurveillance models. This GIS approach can be universallyapplied to population diseases across the globe. Clinically,Google search analysis of dry eye search intent may one day a r X i v : . [ q - b i o . P E ] J un e able to power ophthalmology clinic logistics according toreal time changes in dry eye risk factors such as climate orseason. II. MATERIALS AND METHODSThe experiments were done using public domain data andanalysis tools. These are described below. A. Google Trends
The following methodologies were designed based on pub-lished methods [48]. The Google Trends search interest outputnumbers are relative to the peak point of popularity for themost popular search term within the query in the given regionand time [2]. The numbers range from 0 to 100, with a valueof 100 meaning peak popularity and a value of 50 meaningthat the term is half as popular at that time point. Thesevalues were then normalized relative to control search termsfor the given region and time and are represented as arbitraryunits (a.u.) from 0 a.u. to 100 a.u. On August 5, 2019, wecompleted a series of queries in Google Trends with queryfilters confined to within the region of the United States overthe time period of January 1, 2004 to August 1, 2019 underall categories within the web search Google property. Wedownloaded data for a list of DED symptoms and treatmentsthat are frequently encountered in clinical practice: dry eyes,irritated eyes, scratchy eyes, watery eyes, burning eyes, grittyeyes, eye drops, artificial tears, Restasis, Systane, Oasis tears,Thera tears, Sooth, Blink tears, Visine, Clear eyes, Xiidra, andpunctal plugs. The scale from 0 to 100 a.u. was maintained thesame by including the most popular search term, eye drops,in each query. Control search terms included: news, weather,sports, and Google. We summed the search interest of all theDED terms together at each time point and all the controlterms together at each time point, then normalized DED searchinterest relative to control search interest at each time pointon a scale of 0 to 100 a.u.
B. Environmental Factors
C. Statistical Analysis
The values for normalized DED search intent relative tocontrol search intent, temperature, and humidity were cat-egorized by United States (US) census regions based onthe US Census Bureau. Results between US census regionsof West, Midwest, South, and Northeast were analyzed fordifferences using unpaired Brown-Forsythe One-Way ANOVA(GraphPad Prism 5.0; GraphPad Software, CA, USA). Sta-tistical significance was determined at p¡0.05. The Brown-Forsythe ANOVA was applied here because the standard deviations were not equal between regions. We generatedgeographic heat maps in order to compare the geographic,temporal, and environmental relationships of dry eye diseasequeries. Additionally, we conducted linear regression analysisto compare the relationship between DED search queries withtemperature, and separately, with humidity. For predictiveanalytics, multi-variable regression models were constructedto generate quadratic forecasts to predict dry eye search intentand control search intent.III. RESULTSIn the following section we describe various relationshipsbetween user search query trends and factors that are under-stood to cause dry eye disease.
A. Temporospatial Trends in DED Search Intent in Relationto Temperature and Humidity
The Google Trends data from January 2004 to August2019 illustrated a significant upward trend such that dry eyesearch intent grew 157%, while control search intent grew106%, with a significant pattern of seasonality [Figure 1].Looking forward, dry eye search intent is forecasted to growan additional 57% to 145.47 a.u. (95% CI, 138.30,152.63 a.u.)by July 2025 and 441% to 500.76 a.u. (95% CI, 493.60, 507.92a.u.) by July 2040.The data for DED search intent relative to control searchintent, temperature, and humidity were graphed by US censusregion [Figure 2]. There was no significant difference in drysearch queries across the regions of West, Midwest, South,and Northeast (Brown-Forsythe One-Way ANOVA, p=0.35).However, temperature and humidity were significantly differ-ent across census regions (Brown-Forsythe One-Way ANOVA,p¡0.0001 and ¡0.01, respectively), with the Northeast being thecoldest (47F) and the most humid (57%) region. The hottestregion was the South (60F), while the driest region was theWest (46%).The spatial relationship of dry eye search queries wasmapped across the United States geography, illustrating thehighest relative dry eye search intent in California and theleast in Wyoming [Figure 3].Localized hot spots for interest in dry eye exist, mostlyalong the coastline. The average temperature was mappedacross the United States, with the highest temperature inFlorida (70.7F) and the lowest in Alaska (26.6F) [Figure 4].The average afternoon humidity was mapped across theUnited states, with the highest humidity in Alaska (64%) andthe lowest in Arizona (25%) [Figure 5].Temperature was compared to dry eye search queries inFigure 6, which showed a positive linear correlation withmoderate strength (r = 0.56). Humidity was compared to dryeye search queries in Figure 6, which illustrated no linearrelationship (r = 0.11). The distribution of the data for dryeye search intent, temperature, and humidity across the USregions was plotted in Figure 7. There were 3 outlier statesfor dry eye search intent, including California (100 a.u.) andWyoming (11 a.u.) in the West and Florida (94 a.u.) in the ig. 1.
Time series plot of Search Intent for dry eye and control terms in the United States from 2004 to 2019 with forecasts to 2025:
The data aregenerated from Google Trends data for search queries related to dry eye disease and control queries. Quadratic forecasts were generated using multi-variableregression models, as shown in Table III, to predict dry eye search intent and control search intent. A: Time series plot illustrating dry eye and control searchintent from 2004 to 2019 with forecasts to 2025. There is a significant upward trend and seasonal pattern in dry eye search intent over time. B: Time seriesplot illustrating dry eye search intent overlaid with the dry eye quadratic forecast. C: Time series plot illustrating control search intent overlaid with the controlquadratic forecast in arbitrary units (a.u.) where Google sets the highest search intent during 2004-2019 to 100 a.u.
South Figure 7. There were no outlier states for temperatureor humidity.
B. Multi-variable Regression Models and Predictive Analytics
Table I demonstrates the multi-variable regression modelusing state-specific data such as geography and environmentalrisk factors to predict DED search intent relative to controlper state, and Table II shows the corresponding correlationcoefficients for all variables that were considered. Variablesthat were significant risk factors of DED search intent weretemperature (r = 0.56, p¡0.001) and coastal zone (r = 0.43,p=0.025). Using the model in Table I, there is a 1.32 a.u. (95%CI, 0.64, 1.99 a.u.) increase in DED search intent with each1 degree Fahrenheit increase in temperature, and DED search intent in coastal states is 13.28 a.u. (95% CI, 1.74, 24.82 a.u.)higher than non-coastal states. No relationship exists betweenDED search intent and humidity (r = 0.11, p=0.48) or UScensus regions (West r = -0.21, p=0.45; South r = 0.28, p=0.15;Northeast r = 0.02, p=0.64).Table III demonstrates the multi-variable regression modelusing time, season, and environmental risk factors to pre-dict DED search intent over time, and Table IV shows thecorresponding correlation coefficients for all variables thatwere considered. Variables that were significant predictorsof DED search intent were control search intent (r = 0.85,p=0.0024), time (r = 0.96, p¡0.001), time2 (r = 0.97, p¡0.001),and seasonality (Winter r = -0.04, p=0.0028; Spring r = 0.10,p¡0.001; Summer r = 0.05, p=0.018). Using the model in Table ig. 2.
Bar graph for comparisons of dry eye Google search queries, temperature, and humidity by United States census region:
The data for dry eyesearch queries, temperature, and humidity were averaged across the four US census regions. These sets of data were separately compared in Brown-ForsytheANOVA statistical analysis. p-values less than 0.05 were considered significant, with ** denoting p¡0.01, and **** denoting p¡0.0001.Fig. 3.
Geographic heat map of dry eye search intent on Google Trends in the United States:
This map was generated with the data from GoogleTrends relative to the control search terms. Localized patches of interest can easily be visualized in various hot spots across the country, with the majority ofthese hot spots being along the coast.
Variables Coefficients Standard error t-Statistic P-value
Intercept -7.12 27.31 -0.26 0.796Temperature 1.32 0.34 3.93 0.001Humidity -0.25 0.35 -0.71 0.479Rank in Low Pollution 0.25 0.17 1.46 0.152Coastal 13.28 5.72 2.32 0.025West region -5.05 6.57 -0.77 0.447South region -10.24 6.91 -1.48 0.146Northeast region 3.41 7.21 0.47 0.639
Rˆ2 = 0.49
TABLE IM
ULTI - VARIABLE REGRESSION MODEL USING ENVIRONMENTAL RISKFACTORS AND EYE SEARCH INTENT RELATIVE TO CONTROL . T
HIS MODELPREDICTS AVERAGE DRY EYE SEARCH INTENT RELATIVE TO CONTROLFOR EACH STATE BASED ON THAT STATES AVERAGE TEMPERATURE , HUMIDITY , AIR POLLUTION , COASTAL ZONE , AND US CENSUS REGION .T HE REGIONS ARE RELATIVE TO THE M IDWEST . III, there is a 0.16 a.u. (95% CI, 0.097, 0.22 a.u.) increase eachmonth in DED search intent; DED search intent in Winter,Spring, and Summer are 3.68 a.u. (95% CI, 1.28, 6.07 a.u.),6.47 a.u. (95% CI, 4.96, 7.98 a.u.), and 2.83 a.u. (95% CI,0.49, 5.17 a.u.) higher than in Fall, respectively.IV. DISCUSSIONTo our knowledge, this is the first study to use internetsearch intent to geographically map dry eye disease in com-parison to the unique regional characteristics of the UnitedStates by using statistical analysis of national Google Trendsand environmental data. Traditional prevalence data for dryeye disease has been collected via surveys, such as the 2013National Health and Wellness Survey, that require an excessive ig. 4.
Geographic heat map of average annual temperatures (F)in the United States:
This map was generated using data from the United States NationalClimatic Data Center.
Variables Correlation coefficient, r
Temperature 0.56Humidity 0.11% Sun 0.10Sunshine hours 0.06Wind speed -0.46Rank in low pollution 0.31Coastal 0.43Atlantic coast 0.21Pacific coast -0.03Gulf coast 0.28Great Lakes 0.30West region -0.21Midwest region -0.09South region 0.27Northeast 0.02
TABLE IIC
ORRELATION COEFFICIENTS RELATING DRY EYE SEARCH INTENTRELATIVE TO CONTROL WITH VARIOUS VARIABLES FOR EACH STATE . Variables Coefficients Standard error t-Statistic P-value
Intercept 31.11 2.85 10.93 0.001Control Searches -0.14 0.04 -3.08 0.002Temperature 0.06 0.05 1.18 0.24AQI value -0.02 0.02 -0.84 0.403Timeˆ2 0.00 0 10.75 0.001Time 0.16 0.03 5.04 0.001Winter 3.68 1.21 3.03 0.003Spring 6.47 0.76 8.47 0.001
Summer
Rˆ2 = 0.97
TABLE IIIM
ULTI - VARIABLE REGRESSION MODEL USING ENVIRONMENTAL RISKFACTORS , TIME ( MONTH AND YEAR ), AND SEASON TO PREDICT DRY EYESEARCH INTENT . T
HIS MODEL PREDICTS DRY EYE SEARCH INTENT FOREACH MONTH BASED ON CONTROL SEARCH INTENT , TEMPERATURE , AQI
VALUE , TIME , AND SEASON OF THE YEAR . S
EASONS ARE RELATIVE TO F ALL . Variables Correlation coefficient, r
Control Searches 0.85Temperature 0.06AQI value -0.11Time2 0.97Time 0.96Winter -0.04Spring 0.10Summer 0.05January 0.00February 0.00March 0.03April 0.06May 0.06June 0.03July 0.02August 0.02September -0.05November -0.06December -0.06
TABLE IVC
ORRELATION COEFFICIENTS RELATING DRY EYE SEARCH INTENTRELATIVE TO CONTROL WITH VARIOUS VARIABLES FOR EACH STATE . amount of time and resources to conduct [45], [12], [49],[47], [46], [24], [23], [7], [6], [4]. Additionally, these typesof surveys are limited in space and time to the participantsexamined in a certain geographic region within a certain yearand are laden with subjective biases and observer effects.According to the 2007 Dry Eye Workshop (DEWS) report,we face a recurring issue that there is still a need to conductepidemiological studies for dry eye in different geographicalpopulations [49]. The most recent TFOS DEWS II report ig. 5. Geographic heat map of average afternoon humidity (%) in the United States:
This map was generated using data from the United StatesNational Climatic Data Center. published in 2017, stated that there have been zero populationstudies examining DED prevalence in the southern hemispherein the last 10 years, with most of the DED prevalence studiesfocusing the attention on Asia and Europe [8], [3]. Thesolution to this scarcity of epidemiological data may be digitalepidemiology: the use of real time monitoring of diseasethrough internet data, which overcomes the issues of resources,time, and geographic region. In the world of today, this digitaldata method is becoming a reality. Internet data can be usedto instantly track data from all over the world over any periodof time to include all people who have access to internet.Furthermore, the internet data is accessible to researchers atno additional cost. An example of the successful use of internetdata for epidemiology can be seen in the field of infectiousdiseases. In 2009, the CDC partnered with researchers fromGoogle in order to develop the first ever use of data frominternet search engines to predict infectious disease trends[15], [19]. Using terms for flu-like symptoms, the CDC andGoogle were capable of successfully predicting flu outbreaksweeks in advance of the CDCs US Influenza Sentinel ProviderSurveillance Network. After this project made internationalheadlines, Google made its program, Google Trends, publiclyavailable. To this day, Google Trends has successfully pre-dicted trends in the outbreaks of many viruses, including WestNile virus, norovirus, varicella, influenza, and HIV [15], [19],[17], [40], [56]. First described in 1952, the topic of dry eyedisease has gained significant popularity. DED is one of themost frequent ophthalmic morbidities in the United States,with 25% of patients in ophthalmology clinics complainingof dry eye [14], [36]. Recent estimates show that about 30million people suffer from dry eye disease in the US [11], [39].The tremendous incidence of DED signifies the increasing importance of monitoring and treating patients who sufferfrom dry eyes. Dry eye disease is a chronic multifactorialocular surface disease in which the surface of the eye isinadequately lubricated due to pathophysiology involving acontinuum of evaporative dry eye and poor tear quantity orquality [12], [8], [49], [31][14,15,22,32]. Risk factors includeold age, female sex, environment (low humidity, extremesof temperature, air pollution), cigarette smoking, screen timeor blink rate, refractive surgery, contact lens use, severalmedications (such as anti-depressants), and several conditions(auto-immune disease, inflammatory disease, and aqueous ormucinous abnormalities) [14]. The presenting symptoms ofDED vary from a burning sensation to blurry vision, ocularpain, or the sensation of a foreign body in the eye [31].Treatments include ocular lubricants (such as artificial tears),various anti-inflammatory medications, punctal plugs, modifi-cation of local environment, dietary modifications (includingoral essential fatty acid supplementation), lid hygiene, andwarm compresses [13], [8]. Despite these therapies, patientsoften will continue to have several DED symptoms. Dry eyedisease plays a major socioeconomic impact on our society. Inthe US, DED patients visit the doctor an average of 6 times peryear, costing a total of $800 USD; this totals to a national costof $4 billion USD per year [9], [58]. Furthermore, consideringthe loss of productivity,estimations of the annual financialburden of DED in the United States exceed $55 billion USD.The true costs of DED on society are greater when consideringthe effects of DED on quality of life, vision, and productivity,as well as the psychological and physical impact of pain[8]. This overall socioeconomic burden motivates the need tobetter identify geographic regions with DED and relieve dryeye symptoms at a low cost. This study aims to address this ig. 6.
Scatter plot for comparison of Google Trends data for dry eye andtemperature or humidity in the United States:
A linear model was used toanalyze the data, with the equation of the line of best fit and the coefficientof determination, R2, listed. A: Dry eye search intent versus temperature. B:Dry eye search intent versus humidity. problem by examining the potential to use of readily availableinternet data, such as Google Trends, to geographically mapDED-related search intent in the US. Google search intent fordry eye disease in the US population has grown by 157% from2004 to 2019. The dry eye disease epidemic will continue toexpand, with forecasts predicting that DED search intent willgrow an additional 57% by 2025 and 441% by 2040. Thisrise reflects the epidemiological trends reported from nationalsurvey data, with an increase in US DED prevalence from4.34% in 2009 to 6.8% in 2017 [44], [12]. This demonstratesthe concept that as the prevalence of overall DED rises in theUnited states, so does the volume of dry eye information-seeking behavior on Google Trends. Furthermore, dry eyesearch intent showed a significant seasonal pattern with thehighest search volume in Spring and lowest in Fall, whichconfirms previous findings by Kumar et al. in 2015 [21].When compared across the four US census regions, our resultsdemonstrated no significant difference in DED search intentacross the regions, which matches traditional reports of dryeye regional prevalence in 2017 [12]. Nevertheless, regional
Fig. 7.
Box and whisker diagrams for comparison of (A) dry eye searchintent, (B) temperature, and (C) humidity across US census regions:
A:There were 2 outliers for dry eye search intent in the West, California (100a.u.) and Wyoming (11 a.u.), and 1 outlier in the South, Florida (94 a.u.). B,C: There were no outliers for temperature or humidity. ifferences in dry eye prevalence have been observed invarious other countries, including Korea, Taiwan, and India[18], [16], [55], [43]. Thus, it comes as a surprise that thereis no regional DED prevalence differences in the US despitevast differences in climate and environment. The United Statesis comprised of a wide variety of natural ecosystems, rangingfrom desserts to wetlands [32], [42]. Perhaps many of theseecosystems are present in all 4 US census regions, andtherefore the grouping itself may be hiding unique geographicdifferences in DED search intent. When looking at individualstates, localized patches of hot spots exist, such as Californiaand Florida, mostly along the coastline. This may be explainedby the high speed coastal winds, as high wind speed hasbeen shown to be significantly correlated with dry eye disease[34], [53]. Understanding the unique nature of these regionsand how they play a role on the geographic trends in DEDadds an important perspective to the investigation of DEDin this nation. Regarding environmental factors, our resultsdemonstrated that temperature and coastal status were moreimportant predictors of DED search intent than humidity,sunshine duration, air pollution, or geographic region. A studyconducted in South Korea found that meteorological factorssuch as higher temperature, lower humidity, higher windspeed, longer sunshine duration, and higher air pollution wereall significantly correlated with DED [53]. Human studies havedemonstrated that humidity was associated with measures suchas tear evaporative rate, blink rate, and tear volume [53],[52], [50], [22], [1]. In the current study, the relationshipbetween DED and humidity may be drowned out by otherstronger relationships. Furthermore, the observed geospatialpatterns of DED might be due to regional differences in amyriad of factors spanning beyond just the environmentalcomponents studied. These factors might include regionaldifferences in lifestyle including screen habits, demographics(age, sex, ethnicity, etc.), socioeconomic factors, awareness,advertising and media exposure, or various other types ofnoise e.g. internet bots or users tampering with Google toalter market size for financial reasons. Further investigationis needed for the association between population DED searchintent and the various underlying factors contributing to thetrends. While mining the web for epidemiology in real timeis a fascinating perspective, in the present day this type ofinternet data should not be assumed to replace any effortsof public health organizations or clinicians [5]. As we movetoward the future of reporting internet epidemiologic data inhealth care, greater transparency in the methodology throughwhich we gather our internet data can improve its reliabilityas a research tool and bring us closer to real time digitalepidemiology [35]. On the topic of dry eye disease, futurestudies should expand our understanding of DED populationsby examining the relationship between DED search intentand various other regional factors e.g. meteorological, demo-graphic, socioeconomic, lifestyle. This data in the US canthen be compared to trends across the world. In the future,hospitals may be able to monitor diseases like DED in realtime based on the unique regional characteristics of their own community via publicly available internet data. This datacan then be used to power hospital logistics according tothe changing incidence of a limitless variety of illnesses indifferent months of the year. Beyond the study of disease atthe macro level with population digital epidemiology, futurework should investigate diseases, such as dry eye disease, atthe micro level for individual patients. With lifestyle patterntracking in todays world, patients can collect personal dataincluding daily habits, climate, geospatial data, and screentime [41], [27], [37], [38]. In the future, this type of datacan one day be incorporated to monitor and advise on anindividuals personal eye health state [30], [25], [28]. Thisindividual monitoring allows for personal health navigationto control the health state of the eye to the best potential fora certain individual [26], [29], [25]. Several limitations arepresent in our study. First, interpretation of population trendsin US DED via Google Trends is challenging without theclinical information provided by traditional surveys such asmedical comorbidities and symptom severity. Second, thereis a possibility of internet data being influenced by variousunknown factors, like media exposure, or altered by users orbots. Third, the data sets comparing DED search intent totemperature and humidity were consolidated to average data-points per state based on the climate data available, limitingthe sample size. At the same time, this sample of fifty statesmay have been skewed by outliers e.g. California or Wyoming.Additionally, the search terms utilized might reflect otherconditions of the ocular surface e.g. allergic conjunctivitis,rather than being specific to DED. On the same note, Googleusers may have entered synonyms of dry eye symptoms ortreatments that were missed in our collection.V. CONCLUSIONSOur study is the first to demonstrate that internet searchintent such as Google Trends can be used as a novel digital epi-demiologic approach for geographically mapping populationdry eye disease in relation to the unique regional characteris-tics of the United States. This type of mapping allows for easyvisualization of localized hot spots in dry eye search intent,which were mostly located along the coastline. The interest indry eye in the United States grew tremendously since 2004,with an overall upward trend and a seasonal variation pattern.Meteorological factors varied in their relation to dry eye, withtemperature and coastal status being more important predictorsof DED search intent than humidity, sunshine duration, air pol-lution, or geographic region. This paper creates an avenue forthe future exploration of geographic information systems forlocating dry eye and other diseases through online populationdisease metrics. Further investigation is needed to determinewhat other regional factors are contributing to the differencesin DED search intent across the nation, how these resultscompare to dry eye trends across the world, and how thisinformation can be applied at the micro level to monitor anindividuals personal eye health state [30], [25]. Continuousaccess to estimating this personal eye health state allows forthe foundation by which individual’s may recieve automateduidance on how to best ensure their corneal health stays well[26]. R
EFERENCES[1] Anastasia Alex, Austin Edwards, J. Daniel Hays, Michelle Kerkstra,Amanda Shih, Cintia S. de Paiva, and Stephen C. Pflugfelder. Factorspredicting the ocular surface response to desiccating environmentalstress.
Investigative Ophthalmology and Visual Science , 54(5):3325–3332, 5 2013.[2] Alphabet. Google Trends.
Web , 2006.[3] Berta Andr´as, Tth Molnr Edit, and Csutak Adrienne. New internationalconsensus statement about the definition, classification, ethiology, di-agnostics and therapy of dry eye (TFOS DEWS II).
Orvosi Hetilap ,159(20):775–785, 5 2018.[4] Julia Silvestre De Castro, Iara Borin Selegatto, Rosane Silvestre De Cas-tro, Eliana C.M. Miranda, Jos Paulo Cabral De Vasconcelos, Keila Mon-teiro De Carvalho, Carlos Eduardo Leite Arieta, and Monica Alves.Prevalence and Risk Factors of self-reported dry eye in Brazil using ashort symptom questionnaire.
Scientific Reports , 8(1), 12 2018.[5] Gianfranco Cervellin, Ivan Comelli, and Giuseppe Lippi. Is GoogleTrends a reliable tool for digital epidemiology? Insights from differentclinical settings.
Journal of Epidemiology and Global Health , 7(3):185–189, 9 2017.[6] William G. Christen, J. Michael Gaziano, and Charles H. Hennekens.Design of Physicians’ Health Study II - A randomized trial of beta-carotene, vitamins E and C, and multivitamins, in prevention of cancer,cardiovascular disease, and eye disease, and review of results of com-pleted trials.
Annals of Epidemiology , 10(2):125–134, 2 2000.[7] William G. Christen, Joann E. Manson, Robert J. Glynn, Umed A.Ajani, Debra A. Schaumberg, Robert D. Sperduto, Julie E. Buring,and Charles H. Hennekens. Low-dose aspirin and risk of cataractand subtypes in a randomized trial of U.S. physicians.
OphthalmicEpidemiology , 5(3):133–142, 1998.[8] Jennifer P. Craig, J. Daniel Nelson, Dimitri T. Azar, Carlos Belmonte,Anthony J. Bron, Sunil K. Chauhan, Cintia S. de Paiva, Jos A.P.Gomes, Katherine M. Hammitt, Lyndon Jones, Jason J. Nichols, Kelly K.Nichols, Gary D. Novack, Fiona J. Stapleton, Mark D.P. Willcox,James S. Wolffsohn, and David A. Sullivan. TFOS DEWS II ReportExecutive Summary, 10 2017.[9] Victor J. Drew, Ching Li Tseng, Jerard Seghatchian, and ThierryBurnouf. Reflections on dry eye syndrome treatment: Therapeutic roleof blood products, 2 2018.[10] EPA. Air Data: Air Quality Data Collected at Outdoor Monitors Acrossthe US, 2019.[11] Marjan Farid. Dry Eye Disease: Let’s Start Thinking Outside of theArtificial Tear Box, 11 2017.[12] Kimberly F. Farrand, Moshe Fridman, Ipek zer Stillman, and Debra A.Schaumberg. Prevalence of Diagnosed Dry Eye Disease in theUnitedStates Among Adults Aged 18 Years and Older.
American Journal ofOphthalmology , 182:90–98, 10 2017.[13] Anat Galor, Elizabeth R. Felix, William Feuer, Nabeel Shalabi, Eden R.Martin, Todd P. Margolis, Constantine D. Sarantopoulos, and Roy C.Levitt. Dry eye symptoms align more closely to non-ocular condi-tions than to tear film parameters.
British Journal of Ophthalmology ,99(8):1126–1129, 8 2015.[14] Johnny L. Gayton. Etiology, prevalence, and treatment of dry eyedisease, 2009.[15] Jeremy Ginsberg, Matthew H. Mohebbi, Rajan S. Patel, LynnetteBrammer, Mark S. Smolinski, and Larry Brilliant. Detecting influenzaepidemics using search engine query data.
Nature , 457(7232):1012–1014, 2 2009.[16] Sang Beom Han, Joon Young Hyon, Se Joon Woo, Jung Jae Lee,Tae Hui Kim, and Ki Woong Kim. Prevalence of dry eye disease inan elderly Korean population.
Archives of Ophthalmology , 129(5):633–638, 5 2011.[17] Anupam B. Jena, Pinar Karaca-Mandic, Lesley Weaver, and Seth A.Seabury. Predicting new diagnoses of HIV infection using internet searchengine data.
Clinical Infectious Diseases , 56(9):1352–1353, 2013.[18] Y. Jie, L. Xu, Y. Y. Wu, and J. B. Jonas. Prevalence of dry eye amongadult Chinese in the Beijing Eye Study.
Eye , 23(3):688–693, 2009.[19] Amy K. Johnson and Supriya D. Mehta. A comparison of internet searchtrends and sexually transmitted infection rates using google trends.
Sexually Transmitted Diseases , 41(1):61–63, 1 2014. [20] Seung Pyo Jun, Hyoung Sun Yoo, and San Choi. Ten years ofresearch change using Google Trends: From the perspective of bigdata utilizations and applications.
Technological Forecasting and SocialChange , 130:69–87, 5 2018.[21] Naresh Kumar, William Feuer, Nicole L. Lanza, and Anat Galor.Seasonal Variation in Dry Eye.
Ophthalmology , 122(8):1727–1729, 82015.[22] Alberto L´opez-Miguel, Marisa Tes´on, Vicente Mart´ın-Monta˜nez, AmaliaEnr´ıquez-De-Salamanca, Michael E. Stern, Margarita Calonge, andMara J. Gonz´alez-Garc´ıa. Dry eye exacerbation in patients exposed todesiccating stress under controlled environmental conditions.
AmericanJournal of Ophthalmology , 157(4):788–798, 4 2014.[23] S. E. Moss, R. Klein, and B. E.K. Klein. Prevalance of and risk factorsfor dry eye syndrome.
Archives of Ophthalmology , 118(9):1264–1268,9 2000.[24] Beatriz Mu˜noz, Sheila K. West, Gary S. Rubin, Oliver D. Schein,Harry A. Quigley, Susan B. Dressler, and Karen Bandeen-Roche. Causesof blindness and visual impairment in a population of older Ameri-cans: The Salisbury eye evaluation study.
Archives of Ophthalmology ,118(6):819–825, 2000.[25] Nitish Nag.
Health State Estimation . PhD thesis, University ofCalifornia, Irvine, 3 2020.[26] Nitish Nag and Ramesh Jain. A Navigational Approach to Health:Actionable Guidance for Improved Quality of Life.
Computer , 52(4):12–20, 4 2019.[27] Nitish Nag, Vaibhav Pandey, and Ramesh Jain. Health multimedia:Lifestyle recommendations based on diverse observations. In
ICMR2017 - Proceedings of the 2017 ACM International Conference onMultimedia Retrieval , pages 99–106, New York, New York, USA, 62017. Association for Computing Machinery, Inc.[28] Nitish Nag, Vaibhav Pandey, Likhita Navali, Prateek Mohan, andRamesh Jain. Synchronizing Geospatial Information for PersonalizedHealth Monitoring.
Arxiv , 7 2019.[29] Nitish Nag, Vaibhav Pandey, Hyungik Oh, and Ramesh Jain. CyberneticHealth.
Arxiv , 2017.[30] Nitish Nag, Vaibhav Pandey, Preston J Putzel, Hari Bhimaraju, SrikanthKrishnan, and Ramesh Jain. Cross-modal health state estimation. In
MM2018 - Proceedings of the 2018 ACM Multimedia Conference , pages1993–2002, 2018.[31] National Eye Institute. Hyperopia — National Eye Institute, 2020.[32] National Geographic. Asia: Physical Geography — National GeographicSociety, 2020.[33] National Oceanic and Atmospheric Administration. Data Access —National Centers for Environmental Information (NCEI) formerly knownas National Climatic Data Center (NCDC), 2019.[34] National Weather Service. Coastal and Frontal Winds, 2020.[35] Sudhakar V. Nuti, Brian Wayda, Isuru Ranasinghe, Sisi Wang, Rachel P.Dreyer, Serene I. Chen, and Karthik Murugiah. The use of google trendsin health care research: A systematic review, 10 2014.[36] Paul D. O’Brien and Louis M.T. Collum. Dry eye: Diagnosis and currenttreatment strategies, 2004.[37] Vaibhav Pandey, Nitish Nag, and Ramesh Jain. Ubiquitous event miningto enhance personal health. In
UbiComp/ISWC 2018 - Adjunct Proceed-ings of the 2018 ACM International Joint Conference on Pervasive andUbiquitous Computing and Proceedings of the 2018 ACM InternationalSymposium on Wearable Computers , pages 676–679, New York, NewYork, USA, 10 2018. Association for Computing Machinery, Inc.[38] Vaibhav Pandey, Nitish Nag, and Ramesh Jain. Continuous HealthInterface Event Retrieval. In
Proceedings of the 2020 InternationalConference on Multimedia Retrieval , pages 486–494, New York, NY,USA, 6 2020. Association for Computing Machinery (ACM).[39] Adam J. Paulsen, Karen J. Cruickshanks, Mary E. Fischer, Guan HuaHuang, Barbara E.K. Klein, Ronald Klein, and Dayna S. Dalton. Dry eyein the beaver dam offspring study: Prevalence, risk factors, and health-related quality of life.
American Journal of Ophthalmology , 157(4):799–806, 2014.[40] Camille Pelat, Clment Turbelin, Avner Bar-Hen, Antoine Flahault, andAlain Jacques Valleron. More diseases tracked by using google trends,2009.[41] Byron Reeves, Nilam Ram, Thomas N. Robinson, James J. Cummings,C. Lee Giles, Jennifer Pan, Agnese Chiatti, Mj Cho, Katie Roehrick,Xiao Yang, Anupriya Gagneja, Miriam Brinberg, Daniel Muise, YingdanLu, Mufan Luo, Andrew Fitzgerald, and Leo Yeykelis. Screenomics:A Framework to Capture and Analyze Personal Life Experiences andhe Ways that Technology Shapes Them.
Human-Computer Interaction ,2019.[42] Ramroop T. Boudreau T. McDaniel M. Santani T. Sprout E. Costa H.Hun Jeff. Jeff Hun Rutledge, K. Ecosystem — National GeographicSociety, 2011.[43] Anshu Sahai and Pankaj Malik. Dry eye: Prevalence and attributable riskfactors in a hospital-based population.
Indian Journal of Ophthalmology ,53(2):87–91, 6 2005.[44] Debra A. Schaumberg, Reza Dana, Julie E. Buring, and David A.Sullivan. Prevalence of dry eye disease among US men: Estimates fromthe physicians’ health studies.
Archives of Ophthalmology , 127(6):763–768, 6 2009.[45] Debra A. Schaumberg, David A. Sullivan, Julie E. Buring, and M. RezaDana. Prevalence of dry eye syndrome among US women.
AmericanJournal of Ophthalmology , 136(2):318–326, 8 2003.[46] O. D. Schein, B. Munuz, J. M. Tielsch, K. Bandeen-Roche, andS. West. Prevalence of dry eye among the elderly.
American Journal ofOphthalmology , 124(6):723–728, 1997.[47] Oliver D. Schein, Marc C. Hochberg, Beatriz Mu˜noz, James M. Tielsch,Karen Bandeen-Roche, Thomas Provost, Grant J. Anhalt, and SheilaWest. Dry eye and dry mouth in the elderly: A population-basedassessment.
Archives of Internal Medicine , 159(12):1359–1363, 6 1999.[48] Joshua D. Stein, David M. Childers, Bin Nan, and Shahzad I. Mian.Gauging interest of the general public in laser-assisted in situ ker-atomileusis eye surgery.
Cornea , 32(7):1015–1018, 7 2013.[49] Tear Film and Ocular Surface Society. Introduction to the report of theinternational Dry Eye WorkShop (2007), 2007.[50] Marisa Tes´on, Mara J. Gonz´alez-Garc´ıa, Alberto L´opez-Miguel, AmaliaEnr´ıquez-de Salamanca, Vicente Mart´ın-Monta˜nez, Mara Jess Benito,Mara Eugenia Mateo, Michael E. Stern, and Margarita Calonge. Influ-ence of a controlled environment simulating an in-flight airplane cabinon dry eye disease.
Investigative Ophthalmology and Visual Science ,54(3):2093–2099, 3 2013.[51] Nataliya Tkachenko, Sarunkorn Chotvijit, Neha Gupta, Emma Bradley,Charlotte Gilks, Weisi Guo, Henry Crosby, Eliot Shore, Malkiat Thiarai,Rob Procter, and Stephen Jarvis. Google Trends can improve surveil-lance of Type 2 diabetes.
Scientific Reports , 7(1):1–10, 12 2017.[52] Eduardo Uchiyama, Joel D. Aronowicz, Igor A. Butovich, and James P.McCulley. Increased Evaporative Rates in Laboratory Testing ConditionsSimulating Airplane Cabin Relative Humidity: An Important Factor forDry Eye Syndrome.
Eye & Contact Lens: Science & Clinical Practice ,33(4):174–176, 7 2007.[53] Sun Bi Um, Na H. Kim, Hyung K. Lee, Jong S. Song, and Hyeon C.Kim. Spatial epidemiology of dry eye disease: Findings from SouthKorea.
International Journal of Health Geographics , 13(1), 8 2014.[54] US News. State Pollution Rankings — US News Best States, 2020.[55] Tsung Jen Wang, I. Jong Wang, Chao Chien Hu, and Herng Ching Lin.Comorbidities of dry eye disease: A nationwide population-based study.
Acta Ophthalmologica , 90(7):663–668, 11 2012.[56] K. Wilson and J. S. Brownstein. Early detection of disease outbreaksusing the Internet.
Canadian Medical Association Journal , 180(8):829–831, 4 2009.[57] World Media Group. U.S. Average Wind Speed State Rank, 2020.[58] Junhua Yu, Carl V. Asche, and Carol J. Fairchild. The economic burdenof dry eye disease in the United States: A decision tree analysis.