Lyme disease exploration and research questions

Hello, my name is Tyler Cottrell and I am a third year student majoring in Applied Statistics and Global Public Health. Growing up in rural Bedford, Virginia, many peers, friends, and family members all exhibited a tick-borne illness. Lyme disease is one of the most popular diseases from ticks, that my Mom and Grandma had at one point. Therefore, I have always had an interest in zoonotics and specifically tick-borne illnesses. During my time at UVA I have developed a fondness of epidemiology, which also combines both of my majors. With all of that combined, I wanted to use this project as an opportunity to understand if Lyme disease is really that common from my hometown, or was my family insanely unlucky.

Deer Tick, Agricultural Research Service
Deer Tick, Agricultural Research Service

Research Questions:

Section 1: Is Lyme Disease Getting Worse?

For this section I want to use visualizations to determine if Lyme disease is getting worse. I will use year comparing visualizations to see if both cases and rates of Lyme disease are increasing. Determining rates at a national level is essential for my exploration because it provides evidence towards my current theory that Lyme disease is a worsening problem in the U.S.

Map of U.S. regions, The U.S. Census
Map of U.S. regions, The U.S. Census

Visualization 1: Is Lyme Disease a National problem?

My initial visualization is an exploration of Lyme disease cases by state and nationally in a horizontal bar chart. This aims to determine if cases increase or decrease on average from a few sets of years. My overall point with this graphic is to determine why this topic should be researched and why it important. I picked 2022, 2017, 2012 because they followed 5 year increments and 2008 because it was the earliest this dataset had collected information. I then subtracted 2023 data from each of those years to determine the difference between 2023 and the selected years. This helps me accomplish my task of determining if Lyme disease is increasing other the years.

Shiny applications not supported in static R Markdown documents

Based on this horizontal bar graph there is an overall increase in Lyme disease cases from each year to 2023. From 2012 to 2023, there was almost a 60,000 increase in number of Lyme disease cases nationally. Interestingly enough there was a notable drop in cases from 2008 to 2012, which could be attributed to Connecticut and Delaware having an inverse effect from the rest of the United States and actually decreasing in number of cases over time. Despite this drop in cases from 2008 to 2023 in those two states, the overall increase in cases makes this a substantial reason for further analysis on incidence rates, demographic factors, and which time of year is Lyme disease most common.

Visualization 2: How much larger are lymes incidence rates compared to cases?

For my second analysis I want to look at cases of Lyme disease compared to incidence rates. This allows us to determine if rates are really increasing or is it just State population increasing. Incidence rates are a way to look at cases accounting for the population, the way to calculate them is to divide number of cases by state population and multiply by a constant, named person-time (typically person-years). This dataset was initially multiplied by 100,000 person-time, which makes sense since a majority of the data falls in between 10 and 100 people with Lyme. This dataset combines the overall Lyme rates and incidence rates by states.

The incidence rate data only covers from 2010 - 2023, therefore, this graphic will not include rates from 2008 and 2009. I felt a plotly scatterplot was an effective measurement to determine if the people infected by state increases each year. Since the last plot looked at aggregated cases by state, I felt I could utilize the plotly visualizations to determine number of cases of each state by year could provide more evidence for this association between overall increases in rates each year. I also added a log function, because certain states have much higher incidence rates, and that allowed my graphic to have data appear in a more uniform manner.

Beginning in 2010, the incidence rates were fairly tame, with New Hampshire leading with the highest incident rate. Throughout the 13 year period, Lyme disease incidence for Northeastern states was higher than the other regions. By the time 2023 came around the incidence rates overall increased a lot. Rhode Island in particular had an incidence rate of 260, which is much larger than its 2010 rate of 17.2. Therefore this graphic confirmed my suspicion that population increases are not responsible for the increase in rates of Lyme disease, since incidence rates also increased over time.

Section 2: Are There Certain Demographic Differences in Reported Lyme Disease Cases?

As stated in my introduction I have had several family members diagnosed with Lyme disease. At the time, my Mom and Grandma both fell into categories of greater than 20 years old and female. I would like to determine if it was a fluke or do certain demographic factors effect higher frequency of Lyme disease. Additionally, I would like to examine onset months and seasons to see if climate and seasonality has an impact on Lyme cases as well.

Visualization 3: Are Age Groups Associated with Lyme Disease?

After finding that Lyme disease is indeed becoming more prevalent over time, I would like to focus in on recent years and determining if there is an association between certain demographic factors and greater prevalence of Lyme disease.

Heat maps are particularly interesting visualizations because they allow viewers to quickly identify meaningful patterns and relationships within complex datasets without having to interpret lengthy tables or aggregate function outputs. For this visualization, I decided to separate 2022 and 2023 to compare Lyme disease reported cases across geographic regions and age groups.

Within this data, The “Frequency” variable refers to the number of cases within each cohort, which allows me to compare how reported cases are distributed across geographic regions and sex categories. Each point represents a cohort’s reported case frequency rather than a single diagnosis. Therefore, I decided to find the mean of each of these cohorts in their specific region to determine the average size of these cohorts, which can help us understand how the number of cases across each U.S. region compare.

I opted for the average instead of the total, because the numbers in the heat map get really crowded and are difficult to analyze.

Additionally, separating the two years allows viewers to evaluate how these patterns change over time. Rather than looking at each year’s data independently, viewers can compare the visual distributions to identify potential increases, decreases, or persistent patterns across regions and age groups.

Examining this heat map we can see that the 20+ age group has higher average case counts of Lyme disease throughout every population. The unknown category in particular had the largest jump from 5 to 32 in 2022 and from 9 to 47 in 2023.

The Northeast region had the highest overall rates of Lyme, especially in the 2023 group. As seen with the last two graphics, the Northeastern U.S. States have higher rates of Lyme, so that was not a shock to me, however, what is perplexing is the massive jump from 0-19 to 20+. If I had to hypothesize I would say this has to do with work outside, as many outdoor workers are at a higher risk to tick encounters and therefore Lyme disease. Moving along from comparing the regions and age groups, It seems again that the overall rates of Lyme disease increased from 2022 to 2023 regardless of region and age.

One additional thing I would have liked from this dataset, is a more concise breakdown of age-groups instead of just 20+. If I were given data on 20-39, 40-59, … I could assess more grounds for association between the large difference in average Lyme disease numbers for varying ages.

Visualization 4: Is Sex Associated with Differing Rates of Lyme Disease?

Moving along from age, I would like to see if sex is associated with Lyme disease too. To accomplish this I would like to use 2 jittered scatter plots. I am trying to see across the 5 regions if sex is associated in both confirmed cases and all cases (confirmed and probable).

For this data I filtered out only the confirmed cases of Lyme to keep consistent with the other data on Lyme disease that we already have. The “Frequency” variable was utilized again in this graphic, following the same rules as for visualization 3.

The confirmed cases were much lower than the probable cases, which made me do additional research into 2022 and 2023 Lyme data. I then went back to the CDC data source to see if there was a valid explanation behind this, which there was. In 2022, the CDC altered their definition of what you need to be considered a case of Lyme disease. Now you only need laboratory evidence and no longer clinical information to be diagnosed with Lyme disease.

Although this change resulted in an increase in overall reported cases, it in return reduced the availability of onset dates. According to the CDC, illness-onset dates were available for 67.8% of cases reported during 2017-2019, compared with only 4.8% in 2022.

It seems that the results were vastly different from the graphic with both status of confirmation types versus the graphic with only confirmed cases. First, the number of cases for all cases reached above 750 for the graphic with all data, while it only reached above 60 for the filtered data. For the unfiltered data, the number of cases of Lyme disease for the Northeast region was the highest, while it was actually the lowest average number of cases for the filtered data.

Now comparing sex, it appears that there is not too much association showing that a certain sex equates to higher numbers of Lyme disease. Males tend to lead in number of cases for the most part, but there is not a definite pattern that shows males typically having higher average number of Lyme disease cases than females.

Visualization 5: Is Lyme Disease Worse During Certain Months and Seasons?

This next graphic is a shiny app that incorporates seasonal data from the CDC. I created a shiny app that allows users to have a variety of options to compare certain months or seasons of Lyme disease by year selection. I also added a heat map so users can look at the overall effects of Lyme disease by season in a slider bar of 14 years where you can select all the way from 2010 - 2023. This sandbox approach to my app allows users to explore which years was Lyme disease the highest and which onset seasons and onset months were the rates of Lyme the highest.

Onset month refers to the month that people with confirmed Lyme cases began developing symptoms of Lyme disease. The CDC uses this measurement since it is difficult to determine when patients were initially in contact with an animal (mainly ticks) with Lyme. Additionally, patients could wait several months to be officially diagnosed with Lyme, so that is why Onset Month is used in this data.

Lastly, the graphic also displays a table that corresponds to that year’s data. So if you select a different year than 2010, (the year that is the starting graphic) the table will also display data for that newly selected year.

The only time the data will not follow this exact formula, is when you are looking at varying years of the heat map, where I have programmed the map out since the actual heat map does not need the tabular information.

Shiny applications not supported in static R Markdown documents

This visualization reveals a lot of interesting information regarding this data and its patterns. The first major thing I noticed was the onset months of June and July tended to be the highest when it came to symptoms of Lyme disease beginning. The graphics for Onset month tended to follow a unimodal shape, which makes sense considering winter months tend to have the lowest onset cases of lyme disease and summer/late spring months tend to have the highest onset cases of Lyme disease.

For this analysis, I grouped the months into four three-month seasons: January through March as Winter, April through June as Spring, July through September as Summer, and October through December as Autumn. This allowed me to compare seasonal patterns in symptom onset across years. Because the dataset measures the onset of symptoms rather than the date of infection, these patterns should not be interpreted as the exact timing of tick exposure.

Something interesting I observed about my data was the 2022 and 2023 numbers were lower than I assumed they would be. Which makes sense considering the CDC altered their definition of Lyme cases.

Because this project specifically examines the seasonality of symptom onset, the reduced availability of onset dates may affect the completeness of the 2022 and 2023 observations. I kept these years to provide a more recent perspective on seasonal patterns, but interpret them cautiously. My analysis focuses on the distribution of observed onset cases across months and seasons rather than treating the counts as complete estimates of total Lyme disease cases. Therefore, the information is still usable and allows us to make adequate assumptions regarding the monthly cases and which months/seasons are the most popular.

Section 3: Do Certain Virginia Regions Have Higher Rates of Lyme Disease?

Map of VA regions, Virginia Museium of History & Culture
Map of VA regions, Virginia Museium of History & Culture

For the last exploratory portion of my project, I would like to look at the State of Virginia. I will see if specific regions or counties have higher incidence rates of Lyme disease. I would like to see if my county or surrounding counties have high rates of Lyme disease. I will use 2022 incidence rates since that is the most recent data regarding Virginia county level incidence rates. Interestingly enough the CDC reports the data in full when it comes to state-wide level data.

Visualization 6: What Does Lyme Look Like in Virginia?

For this visualization I will use a heat map for the State of Virginia and compare incidence rates based on each County. The key on the side goes from a steel blue (signifying lower rates) to a coral color (signifying higher rates). I want to identify overall patterns with the colors and determine if there is a certain region or area in Virginia with higher incidence rates of Lyme disease.

County-level incidence rates are based on the reported county of residence of individuals diagnosed with Lyme disease, rather than the county where they were exposed to ticks. Therefore, these rates represent the geographic distribution of reported cases by residence and should not be interpreted as a direct measure of local tick exposure risk.

This heat map reveals a clear pattern for which counties tend to have higher rates of Lyme disease. Counties that are considered to be a part of the Appalachian mountain range, which goes through the western counties in Virginia and through West Virginia, had the highest rates of Lyme disease. Eastern counties on the coast typically had the lowest rates. This interesting observation leads to a question of terrain and how that effects overall cases of Lyme disease.

Visualization 7: Is There an Association Between Virginia Regions and Lyme Incidence Rates?

From the last graphic I saw that counties along a vertical region tended to yield higher incidence rates of Lyme disease. To test the accuracy of this hypothesis, I created a new variable that groups counties based on their Virginia region. I then created a beeswarm plot to determine if each counties Virginia geographical region has an effect on incidence rates in Virginia.

While labeling the actual counties in this graphic would have been helpful, I decided to only name the top incidence rate counties of each region, so the data does not look super cluttered in the graphic.

Based on this beeswarm plot we can see two regions have much higher incidence rates than the other three regions. The Blue Ridge Mountain and Valley and Ridge regions both have the highest points as well as points that sit well above the highest of the other three regions. The Blue Ridge Mountains and Valley and Ridge regions both sit in the Appalachian Mountain range, with the Valley and Ridge region in particular bordering West Virginia. This helps provide association to regions being a factor in Lyme disease incidence rates in Virginia.

When it comes to Bedford county, we actually sit in the lower end of the Blue Ridge Mountains region’s incidence rates, however, many of our close western counties often exhibit high incidence rates of Lyme disease. So it is possible that my family members got bitten in those regions as opposed to in my home county, as you cannot trace back the tick bite.

Section 4: Final visualizations and Conclusion

To wrap up I would like to combine aspects of all three sections to create two density plots to see if this correlation between region and Lyme incidence rates can be applied nationally.

Visualizations 8 and 9: Are the Mountains and the East Coast Associated with Higher Rates of Lyme disease Nationally?

I initially split it up as Appalachian versus Rockies versus Non-Appalachian States, however, I also wanted to see if it was more of an East Coast thing, by incorporating a few of the states on the East Coast that were not considered a part of the Appalachia. Even though Virginia is a state in the Appalachia, I wanted to see if Mountains in general were associated with higher rates of Lyme disease, therefore I will be looking at both major mountain ranges that flow through the United States.

These density plots allow us to visualize the data in an interesting way so we can see if many States within a certain region also have an issue with Lyme disease incidence rates.

To combat the data issue with the 2022 data, I have decided to use 2019 data I subsetted from the “Lyme” dataset since that is the most recent data that has not been affected by either COVID-19 or the change of data collection. I added Pennsylvania and Vermont to both the Appalachian and East Coast data sets, despite debate on whether they should be included within those regions. I believe ecologically, they’re very similar to the other states in both columns, therefore, I decided to go with including them within the categories.

To assist with my mountains graphic, I create limits on both the x and y axis to help with analyzing the density plot. While this may remove some of the lower data points, I mainly need to see the overall density of the Appalachian and East Coast variables, and to do this I had to change the axes for these graphics.

Based on the first graphic, it seems that mountain ranges (the Appalachian or Rockies) do not equate to higher incidence of Lyme disease. The Rockies have very low incidence rates and the densities for the Appalachia incidence rates aren’t much different than the other regions.

Looking at the 2nd graphic however, I notice that there is a noticeable difference between East Coast and non-east coast densities. All of the density for the “Other Areas” category is less than incidence rates of 60 per 100,00 people. While the density for the East Coast is around the same for the whole graphic, showing that many states have higher incidence rates. This shows association between the East Coast having higher incidence rates of Lyme disease than other areas of the United States.

Conclusion:

This project was an incredible opportunity for me to explore Lyme disease data at a National and State-level. I was able to see how Lyme disease rates and incidence increased throughout years, how certain demographic factors were more prone to being diagnosed with Lyme disease, how public health policy changes alter the data we collect, and how certain geographic areas are more inclined to higher incidence of Lyme disease. Interestingly enough, Virginia and furthermore Bedford county, Virginia did not end up having alarming incidence rates of Lyme disease, while still fairly large for any disease, maybe my family was a little unlucky afterall!