FREE ACCESS

Spatial epidemiological analysis of Lyme disease in southern Ontario utilizing Google Trends searches

Publication: Environmental Health Review22 February 2022https://doi.org/10.5864/d2021-025

Abstract

Lyme disease is of growing concern in Ontario with endemic areas increasing in size. Differential diagnosis of Lyme disease patients should include their exposure status assuming knowledge of high-risk areas. The goal of this study was a spatial analysis of Lyme disease in southern Ontario for the years 2015–2019 with a focus on the association between Lyme disease prevalence and Internet search frequencies recorded by Google Trends. A choropleth map visualized the raw prevalence of Lyme disease across the 28 public health units of southern Ontario. A disease cluster comprising five public health units was identified in eastern Ontario using the flexible scan statistic (standard morbidity ratio = 4.9, p = 0.01). Poisson regression modeling revealed an association between Lyme disease prevalence and the search term “Lyme disease” in Google Trends (p = 0.032). Lyme disease prevalence was correlated with Google Trend searches, with an increase in relative risk by a factor of 1.19 (CI95%: 1.03, 1.39) for every 1% increase in search activity. Knowledge of the existence and location of high-risk or exposure areas for Lyme disease is important to properly diagnose patients. Exploiting the association between Lyme disease and Internet search activity by the population at risk can also further disease surveillance.

Introduction

Lyme disease is an emerging vector-borne disease transmitted by the blacklegged tick (Ixodes scapularis), which occurs in temperate zones across the northern hemisphere (Marques, 2010; Ogden et al., 2009). The number of annually reported cases in Canada has increased from 144 to 1,487 between 2009 and 2018 (GOC, 2021). The Canadian provinces of Ontario, Quebec, and Nova Scotia make up a noteworthy amount of these cases, indicating that blacklegged tick populations have been established there for years and will continue to increase in abundance (GOC, 2021).
Lyme disease is the most reported vector-borne disease in the United States and is Canada’s most common tick-borne illness (Lindsay, 2016; Wormser et al., 2006). Lyme disease has been emerging across the United States and Canada since the 1970s and is spreading further geographically in recent years (Marques, 2010). Endemic tick areas have been gradually moving towards central and eastern Canada (Ogden et al., 2009). With endemic areas growing, it has caused Lyme disease cases to increase exponentially over the last several years (Lindsay, 2016). Therefore, tick surveillance and the notification of Public Health Units (PHUs) of endemic areas is considered essential to ensure patients are properly diagnosed and treated (Lindsay, 2016).
Lyme disease is caused by infection with the spiral shaped bacterium known as Borrelia burgdorferi and is transmitted in Canada by the bite of the blacklegged tick (Marques, 2010). Lyme disease can be identified primarily by skin lesions and erythema migrans, which can grow larger than 5 cm in diameter from the site of the bite (Lantos et al., 2021; Ogden et al., 2009). Early diagnosis is critical in preventing severe Lyme disease. If noticed early enough the disease can be treated with antibiotics (Lantos et al., 2021; Ogden et al., 2009). The Lyme diagnosis begins by examining the bite spot (Lantos et al., 2021; Ogden et al., 2009). If an erythema migrans spot is recognized, immediate treatment with antibiotics is advocated to halt further spread of the bacterium and avoid any possible health complications. Additional tests are conducted for patients not showing lesions but experiencing specific neurologic manifestations or advanced atrioventricular heart block (Wormser et al., 2006). Further examination entails a two-tiered test provided by Canada’s Public Health laboratories, which consist of an enzyme-linked immunosorbent assay (ELISA) paired with a Western blot (Lantos et al., 2021; Ogden et al., 2009). The first assay is known to have a low sensitivity to early Lyme, thus the Western blot allows for increased sensitivity (Ogden et al., 2009). Interpretation of the diagnostic tests is further informed by the patient’s exposure status to any endemic areas (Lantos et al., 2021; Ogden et al., 2009). Any suspicions of Lyme should be taken seriously. When the lesions begin to disappear, the bacterium begins to spread hematogenously throughout the body, potentially causing serious issues and (or) damage to the patient’s nervous system, heart, and joints (Lantos et al., 2021; Marques, 2010; Ogden et al., 2009). The consequences that could arise are Lyme meningitis, Lyme carditis, Lyme arthritis, and other neurologically debilitating and potentially life-threatening diseases that require long-term care (Wormser et al., 2006).
With Internet technology becoming a greater asset, scientists have become more inclined to use it to their advantage. Google’s search engine has become a popular gateway for the general public seeking medical information (Seifter et al., 2010). Google Trends is a feature that allows individuals to acquire data and graphs that recall search frequencies for various search terms and/or topics (Seifter et al., 2010). This can be used alongside disease frequencies to aid in surveillance and determining whether Lyme disease is spreading or becoming more prevalent in an area on the basis of an increase in search traffic.
With increasing cases of Lyme disease, it remains unclear how endemic areas are expanding or shifting. Identifying these hot spots and new endemic areas can reduce patient misdiagnosis as well as allow there to be a greater awareness brought towards spatial changes. The general goal of this project was to study the geographic distribution of Lyme disease in southern Ontario between the years of 2015 and 2019. The specific objectives were to develop a disease map, identify clusters and respective high-risk areas, and to analyze the relationship of Lyme prevalence with Google Trends.

Methods

Counts of prevalent cases of Lyme disease per year and the corresponding population sizes for each PHU in southern Ontario through the years of 2015 to 2019 were retrieved from Public Health Ontario (PHO, 2020). These cases included all ages and sexes in Ontario. Moreover, the annual counts collected for each PHU were aggregated over the five years of study to present a total number of cases and population size for each region. The boundary map file used as the base for this spatial analysis of southern Ontario was retrieved from Statistics Canada (SC, 2020). The map used to identify certain regions with associating PHUs was also retrieved from Statistics Canada (SC, 2015). The search frequency data of the terms “tick bite,” “Lyme disease,” and “tick” for the years 2015–2019 in southern Ontario was retrieved from Google Trends (Google, 2015–2019).

Disease mapping

The number of Lyme disease cases and population sizes were aggregated into annual totals for the year 2015 to 2019 to report prevalence per 100,000 individuals for each PHU. A choropleth map was then created (Berke, 2001). For the choropleth map of Lyme prevalence, the PHU’s prevalence values were specifically divided into quintiles and colour coded, assigning darker grey shades to represent increasing prevalence (Brewer & Pickle, 2002).

Disease cluster identification

Disease clusters or hot spots of Lyme disease were investigated using the flexible scan statistic (Tango & Takahashi, 2012). This flexible scan statistic is accurate in finding circular as well as noncircular clusters (Tango & Takahashi, 2012). The scan statistic was applied using the R package smerc (French, 2018). Case clusters identified by the flexible scan statistic were highlighted on the choropleth map. Any identified disease cluster was reported with its location, observed and expected case count, standard morbidity ratio (SMR) with its 95% confidence interval, and the Monte Carlo p value from the flexible scan test.

Lyme prevalence with Google Trends

Google Trends data were acquired from the Google Trends website (Google, 2015–2019). Google Trends reports search frequencies as a standardized percentage specific to the search terms between 2015 and 2019, as well as the geographic region being investigated. The Google Trends data were reported for cities within the study area (southern Ontario) and the city-specific data were converted and allocated to its respective PHU. A Quasi-Poisson regression model was used to determine any log-linear relationship between the search term frequencies for “tick,” “tick bite,” and “Lyme disease,” and the observed prevalence of Lyme disease across the PHUs. Quasi-Poisson regression models (Roback & Legler, 2021) adjust for overdispersion, i.e., for situations where standard regression modeling assumptions such as independence and variance homogeneity do not hold, but more elaborate models, i.e., spatial regression models are not adequate due to small sample sizes.
All disease mapping and data manipulations were done using R statistical software (R Core Team, 2020; RStudio Team, 2020). Disease mapping employed the R packages rgdal, maptools, tmap, and rgeos. A significance level of α = 0.05 was applied where appropriate.

Results

There were 28 PHUs within southern Ontario that were investigated for the purpose of this study (see Table 1). Total reported Lyme disease cases per PHU ranged from 7 to 557 cases, with prevalence ranging from 0.64 to 64.62 per 100,000 individuals. The regional Lyme disease prevalence had a mean of 7.86 and a standard deviation of 15.27 per 100,000 individuals.
Table 1: 
Table 1: Lyme disease data from the 28 public health units (PHUs) of southern Ontario from 2015 to 2019.
Note: Names of PHUs forming part of the disease cluster are in bold. NA, not available.

Disease mapping

A choropleth map visualizing the geographic distribution of the average annual Lyme disease prevalence at PHU level in southern Ontario during 2015 to 2019 is presented in Figure 1. A clear pattern demonstrates high prevalence in the eastern portion of southern Ontario.
Figure 1: 
Figure 1: Choropleth map of the 5-year Lyme disease prevalence per 100,000 population at risk across the 28 public health units of southern Ontario during 2015–2019. The boundaries of 5 PHUs marked in black indicate a disease cluster with SMR = 4.9. The 3 letter labels refer to the short hand names of the PHUs; see Table 1 for the full names.

Disease cluster identification

A single Lyme disease cluster was identified by the flexible scan statistic that displayed an increased prevalence of Lyme (p = 0.001). The identified Lyme disease cluster consists of five PHUs that are outlined on the choropleth map by thick black boundary lines (Figure 1) and have the labels HPE, KFL, LGL, OTT, and EOH, respectively (see Table 1 for details). Within the cluster a total of 1,952 cases were observed, whereas the expected number of cases was about 398. The estimated SMR for this cluster was SMR = 4.91 (95% CI: 4.69–5.13), meaning that the populations within this cluster was at about 5 times higher risk of developing Lyme disease than the total population in southern Ontario.

Lyme disease prevalence in relation to Google Trend searches

Google Trends provided trend search data for 24 cities, only covering 16 of the 28 PHUs. The reported search frequencies as reported by Google Trends are included in Table 1. The data are standardized values that can range between 0 and 100. The values for “Lyme disease” ranged from 12 to 22, for “tick bite” from 2 to 7, and for “tick” from 19.5 to 39.
Quasi-Poisson rate regression modeling of the Lyme disease counts for 16 of 28 PHUs over 2015–2019 and, using the regional population at risk per 100,000 as an offset, it revealed the Google Trends search term “Lyme disease” as the only predictor (p = 0.032). The two other search terms (“tick bite” and “tick”) were not found to be associated with Lyme disease prevalence. The exponentiated regression parameter is interpreted as usual, i.e., as the relative risk (RR) of Lyme disease for a 1 unit increase in Google search activity using the search term “Lyme disease”: RR = exp(β = 0.1797) = 1.19 and CI95%(RR) = (1.03, 1.39).

Discussion

Lyme disease prevalence was not evenly distributed across southern Ontario. The majority of Lyme disease occurred towards the east, in the Peterborough area (Figure 1). It is believed that the map presents itself in this manner due to numerous bodies of water distributed throughout the area. Along with the deciduous and mixed forests in these areas, it creates a favourable climate for tick species to thrive and reproduce (Nelder et al., 2018).
The disease cluster located in this area consists of five PHUs (HPE, KFL, LGL, OTT, EOH) where people are about 5 times more at risk of contracting Lyme disease because their estimated SMR is 4.91. Quasi-Poisson regression modeling adjusted for overdispersion, i.e., the presence of disease clusters, and identified “Lyme disease” as the only relevant search term. This presents some evidence for a relationship between the search term “Lyme disease” and Lyme disease prevalence at PHU level. The prevalence of Lyme disease was associated to search frequencies at a RR of 1.19. With respect to the confidence interval of the RR, a 1% increase in search activity of the term “Lyme disease” was associated to an increase in prevalence of Lyme disease by a factor of 1.03 to 1.39. Meaning, an increasing search frequency by 1 unit translates to an average 19% increase in prevalence, but this can also jump to an 39% increase (for the upper limit of the confidence interval).
These findings align with literature that supports the argument that Google Trends search frequencies can be useful in identifying the presence of Lyme disease (Seifter et al., 2010). Being able to identify endemic areas and predict outbreaks of Lyme disease can aid in providing a framework for a risk-based surveillance system and reduce the chance of misdiagnosis. Studies show that individuals who are commonly misdiagnosed tend to turn to the Internet for a diagnosis and search for alternative help from commercial labs (Kim et al., 2020).
One of the largest limitations to any study regarding humans is the unpredictability of human behaviour. In this study it is unknown where individuals have travelled before they have contracted Lyme. The information publicized only includes case numbers, as PHUs privately hold patients travel information for Lyme Disease Enhanced Surveillance (GOC, 2021). Because this information was not available for the study, it creates uncertainty in terms of not knowing the entire story. Additionally, there is also the issue of underreporting, as some cases of early Lyme may not be reported due to asymptomatic disease.
Another study limitation relates to the reliability of Google Trends searches. It is not clear why people search the Internet and if the search is linked to illness or triggered by individuals who become curious about Lyme disease. For example, Lyme disease research at the University of Guelph might inflate search activity of the general public. A reason, other than a spike in cases, that can cause Lyme disease to trend on the Internet and in the news is the chance of a celebrity being infected by Lyme or a public health information push in a region.
Moreover, a choropleth map can change in appearance depending on the prevalence range cut-offs as well as the colour coding selected. To minimize visual bias in this study, only grey scales were selected, and their classification was based on quintiles of the empirical distribution of the map variable to minimize visual bias (Brewer & Pickle, 2002).
This study is the first step into determining whether there is a spatial (or geographical) association between Google Trends search frequencies and Lyme disease prevalence in Ontario. Even though the Google Trends data used for this study is only from 16 PHUs over the years 2015 to 2019, it presents evidence that past endemic Lyme disease areas were linked to current endemic Lyme disease areas as presented on Public Health Ontario’s Lyme Disease Risk Map 2020 (OAHPP, 2020).

Conclusion and future research

People from across Ontario were affected by Lyme disease (Figure 1), although exposure to the vector (the blacklegged tick) is concentrated to few areas. This study detected one hot spot in eastern Ontario and estimated the SMR for the respective population to be 4.9. The study also determined a link between Google Trends and Lyme disease (p = 0.032) that associates an increase in Internet searches for “Lyme disease” with an increase of risk by the factor 1.19. Further research should be dedicated to confirming this link as it can aid disease surveillance in determining whether Lyme disease is spreading or become more prevalent. Additionally, other factors that could influence the propagation of ticks, such as changing climates and urbanization should also be investigated.

References

Berke O.2001Choropleth mapping of regional count data of Echinococcus multilocularis among red foxes in Lower Saxony, GermanyPrev Vet Med522119131
Brewer C. A. and Pickle L.2002Evaluation of methods for classifying epidemiological data on choropleth maps in seriesAnn Assoc Am Geogr924662681
French, J. 2018. smerc: Statistical Methods for Regional Counts. R package version 0.4.5. Available at: https://CRAN.R-project.org/package=smerc [accessed 28 November 2021].
Google. 2015–2019. Google Trend files: Southern Ontario 2015–2019 [Internet]. Available at: https://trends.google.com/trends [accessed 26 March 2021].
Government of Canada (GOC). 2021. Surveillance of Lyme disease. In: Diseases and conditions [Internet]. Available at: https://www.canada.ca/en/public-health/services/diseases/lyme-disease/surveillance-lyme-disease.html [accessed 18 March 2021].
Kim D., Maxwell S., and Le Q.2020Spatial and temporal comparison of perceived risks and confirmed cases of Lyme disease: An exploratory study of Google TrendsFront Public Health8395
Lantos P. M., Rumbaugh J., Bockenstedt L. K., Falck-Ytter Y. T., Aguero-Rosenfeld M. E., Auwaerter, et al.2021Clinical Practice Guidelines by the Infectious Diseases Society of America, American Academy of Neurology, and American College of Rheumatology: 2020 Guidelines for the Prevention, Diagnosis, and Treatment of Lyme DiseaseNeurology966262273
Lindsay L. R.2016Present state of common vector-borne diseases in CanadaCan Comm Dis Rep42200201
Marques A. R.2010Lyme disease: A reviewCurr Allergy Asthm R101320
Nelder M. P., Wijayasri S., Russell C. B., Johnson K. O., Marchand-Austin A., Cronin K., et al.2018The continued rise of Lyme disease in Ontario, Canada: 2017Can Commun Dis Rep4410231236
Ogden N. H., Lindsay L. R., Morshed M., Sockett P. N., and Artsob H.2009The emergence of Lyme disease in CanadaCMAJ1801212211224
Ontario Agency for Health Protection and Promotion (Public Health Ontario) (OAHPP). 2020. Ontario Lyme disease map 2020: Estimated risk areas [Internet]. Toronto, ON: Queen’s Printer for Ontario.
Public Health Ontario (PHO). 2020. Infectious Disease Trends in Ontario 2015–2019 [Internet]. Available at: https://www.publichealthontario.ca/en/data-and-analysis/infectious-disease/reportable-disease-trends-annually#/34 [accessed 18 March 2021].
Statistics Canada (SC). 2015. Ontario, Health Units – Southern Ontario Health Regions [Internet]. Available at: https://www150.statcan.gc.ca/n1/pub/82-402-x/2015002/maps-cartes/rm-cr07-eng.htm [accessed 24 March 2021].
Statistics Canada (SC). 2020. Health regions: Boundaries and correspondence with census geography files 2018 [Internet]. Available at: https://www150.statcan.gc.ca/n1/pub/82-402-x/2018001/hrbf-flrs-eng.htm [accessed 24 March 2021].
R Core Team. 2020. R: A language and environment for statistical computing. Vienna, Austria: R Foundtion for Statistical Computing.
Roback, P., & Legeler, J. 2021. Beyond multiple linear regression: Applied generalized linear models and multilevel models in R. Boca Raton, FL: Chapman and Hall/CR.
RStudio Team. 2020. RStudio: Integrated Development for R. Boston, MA: RStudio, PBC.
Seifter A., Schwarzwalder A., Geis K., and Aucott J.2010The utility of “Google Trends” for epidemiological research: Lyme disease as an exampleGeospatial Health42135137
Tango T. and Takahashi K.2012A flexible spatial scan statistic with a restricted likelihood ratio for detecting disease clustersStat Med313042074218
Wormser G. P., Dattwyler R. J., Shapiro E. D., Halperin J. J., Steere A. C., Klempner M. S., et al.2006The clinical assessment, treatment, and prevention of Lyme disease, human granulocytic anaplasmosis, and babesiosis: Clinical Practice Guidelines by the Infectious Diseases Society of AmericaCID4310891134

Information & Authors

Information

Published In

Environmental Health Review cover image
Environmental Health Review
Volume 64Number 4December 2021
Pages: 105 - 110

History

Published online: 22 February 2022

Permissions

Request permissions for this article.

Key Words

  1. Lyme disease
  2. Borrelia burgdorferi
  3. Google Trends
  4. spatial epidemiology
  5. disease mapping

Authors

Affiliations

Maria Kutera
Department of Population Medicine, University of Guelph, Guelph, ON, Canada
Department of Population Medicine, University of Guelph, Guelph, ON, Canada
Kurtis Sobkowich
Department of Population Medicine, University of Guelph, Guelph, ON, Canada

Metrics & Citations

Metrics

Other Metrics

Citations

Cite As

Export Citations

If you have the appropriate software installed, you can download article citation data to the citation manager of your choice. Simply select your manager software from the list below and click Download.

There are no citations for this item

View Options

View options

Full Text

Open Full Text

PDF

Download PDF

Get Access

Media

Media

Other

Tables

Share Options

Share

Share the article link

Share with email

Share on social media