A Space-Time Permutation Scan Statistic for Evaluating County-Level Tickborne Disease Clusters in Indiana,2009-2016

Introduction

Over the past 2 decades, spatial distribution of tickborne diseases has expanded in the United States.^1,2 For example, the geographic range of the lone star tick responsible for transmitting ehrlichiosis, tularemia, and other pathogens has extended to the northern and midwestern states. Additionally, the number of counties reporting the blacklegged tick—responsible for transmitting Lyme disease, babesiosis, and anaplasmosis—has doubled, and the number of counties now considered to be of high incidence for Lyme disease diagnosis has increased by 300% in the northeastern United States. ³ Clusters of elevated incidence of tickborne disease diagnosis and related deaths have also been reported across parts of the United States.^1,4 These increases have led to calls for public health officials and scientists to develop a comprehensive understanding of the drivers of tickborne disease, with a view to reducing disease risk. ⁵ In Indiana, tickborne diseases have increased tremendously as indicated by confirmed and probable cases of Lyme disease, Rocky Mountain spotted fever, ehrlichiosis, and anaplasmosis. ⁶ Previous studies have demonstrated differences in the occurrence of tickborne disease by age group, gender, and rural/urban areas.^7-9 In a previous study, we identified clusters of Indiana counties with a significant high risk of self-reported tickborne disease, suggesting the need for continued surveillance. ⁷ Although various tick species and tickborne diseases have been reported across counties in Indiana,^10,11 the drivers of tickborne disease, including counties with higher than expected distribution, are currently unknown. It is expected that climate change, urbanization, population growth, and human encroachment into tick habitats will lead to increased diagnosis of tickborne diseases. ¹² Increased surveillance of populations at greatest risk is critical.

The ability to detect space-time clusters has important public health implications. This is because the identification of some common elevated risk factor of limited geographical and/or temporal extension will naturally warrant further investigation and, thus, lead to identification of those risk factors. Particularly in situations where resources are limited, the ability to map areas with higher rates of tickborne disease is important for prioritizing tick exposure and tickborne disease control programs for prompt community interventions designed to modify risk exposure. Timely implementation of disease prevention programs expectedly will reduce morbidity and mortality for the most at-risk counties. Using retrospective data for the state of Indiana, the purpose of the current study is to determine the age, gender, rural-urban, and space-time distributions of tickborne disease diagnoses in Indiana. Study outcomes may have important implications for tickborne disease surveillance and risk-reduction interventions.

Methods

We obtained the number of retrospective probable and confirmed tickborne disease cases in Indiana at the county level between 2009 and 2016 from the Epidemiology Resource Center at the Indiana Department of Health. Data from 5 tickborne diseases were available: Lyme disease, Rocky Mountain spotted fever, ehrlichiosis, tularemia, and typhus/rickettsia. For the purposes of this analysis, we considered both probable and confirmed tickborne disease as cases. This was based on the understanding that both probable and confirmed cases are reported to the National Notifiable Diseases Surveillance System. ²

We analyzed the data separately for each of the 5 tickborne diseases reported for the period under investigation. We also analyzed all 5 diseases, collapsed by county (summated tickborne diseases). Seven tickborne disease cases did not have a county of diagnosis and were removed from further analysis. We used frequency tabulation to determine the percentage of each tickborne disease reported from 2009 to 2016. Furthermore, we used chi-square cross tabulation to test whether there were statistically significant (P ≤ .05) differences in reported tickborne disease cases by gender (male vs. female), median age (≤ 44 and >44), and county classification (rural vs. rural-mixed vs. urban). The age cutoffs are based on previous reports that suggest certain tickborne diseases were more prevalent in males older than 40 years of age. ⁸ County classification was based on the modified classification method of rural, rural-mixed, and urban by Ayres et al ¹³ (ie, population of less than 40,000 = rural; 40,000 to 99,999 = rural-mixed; and 100,000 or more = urban). To test the hypothesis that county classifications (ie, rural, rural-mixed, and urban) of summated tickborne diseases are identical populations, we used the nonparametric Kruskal-Wallis test. As opposed to 1-way analysis of variance (ANOVA), the Kruskal-Wallis test is more appropriate for discrete data.

Finally, we used retrospective space-time permutation models in SaTScan to test the hypothesis of complete spatiotemporal randomness of tickborne disease diagnosis. A space-time permutation test looks for space-time interaction clusters, automatically adjusting for purely spatial and purely temporal clusters/variation. ¹⁴ SaTScan is a global clustering test that applies a likelihood ratio test to evaluate and detect both the location of clusters and their statistical significance, while adjusting for multiple testing. ¹⁵ SaTScan has been used by several scholars for space-time cluster detection. For example, Kulldorff et al ¹⁶ used it on data spanning multiple years to evaluate spatiotemporal clustering of brain cancer, while Zhao et al ¹⁷ used space-time clustering in SaTScan to analyze county-level tuberculosis data in China from 2005 to 2011. The space-time permutation version of SaTScan checks to see if the observed spatial clusters are also clustered in time and vice versa.

In this study, clusters represented Indiana counties with persistent tickborne disease diagnoses during periods of 1 year or more. Because the data were recorded by year, our time aggregation unit was yearly, with a time aggregation length of 1. Using the 2010-2015 population estimates for Indiana counties ¹⁸ as the maximum circle size to represent the underlying population at risk, we set the maximum cluster size to 25% of this population. Hence, clusters were evaluated only if the expected number of tickborne disease diagnoses cases were less than 25% of the total number of tickborne disease diagnosis cases in the county. This decision to set the maximum cluster size to 25% (as opposed to the default of 50%) was based on the understanding that from a local outbreak detection/interventionist perspective, only small clusters are of public health importance. ¹⁹ To increase the statistical power of the analysis, the number of Monte Carlo replications was set at 9,999 random permutations. A county boundary map of Indiana was used as the background for all map projections (using the World Geodetic System [WGS 1984] as the Global Positioning System reference coordinate system). Frequencies, chi-square cross tabulation, and the Kruskal-Wallis test were conducted using RStudio version 1.1.414 (RStudio, Boston, MA). Retrospective space-time permutation analysis was conducted in SaTScan version 9.6 (Martin Kulldorff with Information Management Services Inc., Boston, MA), and shapefile output from SaTScan cluster detection were analyzed in ArcGIS software version 10.6 (Environmental Systems Research Institute, Redlands, CA), to produce the maps with the detected clusters.

Results

A total of 1,329 tickborne diseases were diagnosed in Indiana from 2009 to 2016—of these 1,322 had complete county information and were thus analyzed for this study. Of the 1,322 cases, Lyme disease accounted for 64% (n = 851), Rocky Mountain spotted fever 17% (n = 225), ehrlichiosis 16.7% (n = 221), tularemia 1.4% (n = 19), and typhus/rickettsia 0.5% (n = 6). Fifty-nine percent (n = 784) of subjects were male and 40% (n = 530) were female, and approximately 1% (n = 8) were gender deidentified. The mean age of subjects was 41.4 years (standard deviation [SD] = 22, median = 44 years, minimum = 22 years, maximum = 91 years). Age groups were divided evenly, with 50% (n = 660) older than 44 years and 50% (n = 662) younger than 44 years. Using county classification, we observed that 53% (n = 709) of tickborne disease cases were in urban counties, 30% (n = 398) in rural counties, and 17% (n = 229) in rural-mixed counties. Tickborne disease data were missing for 16% (n = 14) of Indiana counties. The counties in this category include Blackford, Decatur, Jay, Ohio, Randolph, Rush, Sullivan, Switzerland, Tipton, Union, Vermillion, Wayne, Wells, and Whitley counties.

Tickborne disease diagnosis differed significantly between the 2 age groups (chi-square [χ ² ] = 82.03, degrees of freedom [df] = 4, 1322, P < .001). Specifically, more Lyme disease diagnoses were reported among the younger age group (≤ 44 years of age), compared with the older age group (> 44 years) (Table 1). Conversely, more cases of ehrlichiosis, Rocky Mountain spotted fever, and tularemia were reported among older Indiana residents (> 44 years). We also found statistically significant differences in tickborne disease diagnosis based on county classification (χ ²  = 184.68, df = 10, 1322, P < .001). Urban counties had the highest percentage of diagnosis for Lyme disease, Rocky Mountain spotted fever, tularemia, and typhus/rickettsia, whereas the highest percentage of ehrlichiosis diagnosis was found in rural counties. No statistically significant differences (P ≤ .05) were observed for tickborne disease between male and female subjects (χ ²  = 14.02, df = 8, 1322, P = .081). The Kruskal-Wallis test was conducted to examine differences in summated tickborne disease by county classification. We found significant differences (χ ²  = 22.71, df = 2, P < .001) between rural, rural-mixed, and urban counties. Based on a Conover post hoc test combined with the Bonferroni option to adjust for multiple comparisons, there was a significant difference (P < .05) in summated tickborne disease between urban and rural counties, and between urban and rural-mixed counties.^20,21 No other differences were observed.

Results of SaTScan space-time permutation for tickborne disease are presented in Table 2. Rows in the table contain data for each detected disease cluster. For each cluster, we reported the radius (miles), location ID indicating the number of counties in the cluster, cluster period corresponding to the time frame (years), expected number of cases, observed number of cases, test statistic, and corresponding P value. Location IDs are listed in the Appendix (www.liebertpub.com/doi/suppl/10.1089/hs.2019.0159).

A total of 6 clusters were detected for summated tickborne disease, with the cluster radius ranging from 0 to 64 miles. Two of these clusters had significant P values (P < .05), suggesting the presence of more tickborne disease for the specified time period in counties within these clusters than would be expected if the distribution were completely spatiotemporally random. Cluster 1 (P = .001, test statistic = 9.576) had a radius of 24 miles and consisted of 5 counties in 2014. Cluster 2 (P = .04, test statistic = 6.761) had a radius of 22 miles, consisted of 4 counties, and was observed from 2010 to 2011. No significant clusters of each type of tickborne disease diagnosis were observed.

Figure 1 summarizes the geographic distribution of space-time clusters of tickborne disease in Indiana from 2009 to 2016, by county classification (rural, rural-mixed, urban). As stated earlier, 2 significant clusters of summated tickborne disease diagnosis (Figure 1A) were observed. The first significant cluster consisted of 5 counties and was observed in south-central Indiana. The second significant cluster consisted of 4 counties and was located in western Indiana. The counties involved in the significant clusters of summated tickborne disease were predominantly rural. No significant clusters of each type of tickborne disease diagnosis were detected: Rocky Mountain spotted fever (Figure 1B), Lyme disease (Figure 1C), ehrlichiosis (Figure 1D), tularemia (Figure 1E), and typhus/rickettsial diseases (Figure 1F).

OPEN IN VIEWER

Figure 1.

Space-time permutation clusters of tickborne disease diagnoses in Indiana (2009-2016). Counties within significant clusters have a malachite green cylindrical window; A—summated tickborne diseases diagnosis; B—Rocky Mountain spotted fever; C—Lyme disease; D—ehrlichiosis; E—tularemia; F—typhus/rickettsia. Detailed information for clusters with a radius of 0 miles are not displayed on the map. Abbreviations: RMSF, Rocky Mountain spotted fever; TBD, tickborne disease.

Discussion

In this study, we sought to determine in the study population (1) the prevalence of various tickborne diseases; (2) whether significant differences in tickborne diseases diagnoses existed by gender, age group, and county classification; and (3) the hypothesis of complete spatiotemporal randomness of tickborne disease diagnoses, using data obtained from the Epidemiology Resource Center at the Indiana Department of Health. Our hypotheses were informed by evidence in the scientific literature that suggests a geographic expansion in tickborne disease prevalence by gender, age group, and rural-urban differences and the importance of spatiotemporal cluster detection in disease surveillance in regard to implementation of risk-reduction interventions.

Compared with other tickborne diseases, we found that Lyme disease was the most prevalent diagnosis in Indiana during the study period. Rocky Mountain spotted fever and ehrlichiosis diagnosis, respectively, were next. It is possible that some cases reported as ehrlichiosis in the dataset analyzed were actually anaplasmosis, because information obtained from the Indiana Department of Health indicated that anaplasmosis and ehrlichiosis were both reported under the condition “ehrlichiosis” during the study period. These results are in agreement with those from other states, suggesting that overall, Lyme disease is the most prevalent tickborne disease in Indiana. ²²

Our results also revealed differences in tickborne disease by age group. More Lyme disease diagnosis was reported among Indiana residents younger than 44 years of age, but we found more ehrlichiosis, Rocky Mountain spotted fever, and tularemia among Indiana residents who were older than 44 years of age. Some previous studies have shown that Lyme disease is more prevalent among children and older adults in the United States, while other studies report mixed results.^7,8,22 It is not clear what is responsible for the differences between age groups in diagnosis of tickborne disease found in our study.

We know that the risk of acquiring any of these diseases is present in Indiana because the 3 common tick vectors—Ixodes scapularis, Amblyomma americanum, and Dermacentor variabilis—found in various parts of Indiana are capable of transmitting the diseases reported in this study. ²³ Additionally, several of these tick vectors are capable of transmitting multiple diseases. ²⁴ It is possible that exposure to these tick vectors at locations where people live, work/farm, learn, and play may account for the differences by age group found in this study. Furthermore, we found differences in the distribution of tickborne disease diagnosis by county classification. Specifically, Lyme disease diagnosis was highest in urban counties, compared with rural and rural-mixed counties. Conversely, ehrlichiosis diagnosis was highest in rural counties, compared with urban and rural-mixed counties. Previous studies have shown that the highest risk factors for tick exposure and tickborne disease occur in rural areas. ⁹ We would therefore expect rural counties to report the highest number of all types of tickborne disease diagnosis, including Lyme. However, urban counties have more clinicians with adequate training and experience, awareness, and resources to diagnose Lyme disease, compared with rural counties. Whether the higher Lyme disease diagnosis reported from urban counties in this study is due to higher incidence, higher awareness for the disease, or presence of more resources is unclear. The finding of more ehrlichiosis diagnosis in rural counties of Indiana is supported by other studies, which indicates that the primary disease vector (Amblyomma americanum) is predominant in rural counties of Indiana.^25,26

Although previous studies have found significant gender-based differences in tickborne disease diagnosis,^7,8 the current study found no such differences. With a .05 significance level, based on the Kruskal-Wallis chi-squared test and subsequent Conover post hoc test, we can conclude that the count of summated tickborne disease diagnoses in Indiana is significantly higher in urban counties, compared with rural and rural-mixed counties.

Based on the summary score of summated tickborne disease diagnoses, we found 2 significant spatiotemporal clusters. The first significant spatiotemporal cluster was observed in south-central Indiana in 2014, and the second significant spatiotemporal cluster was located in western Indiana between 2010 and 2011. In a previous study, we reported 2 spatial clusters of counties with significantly higher relative risk of self-reported tickborne disease diagnosis in southeastern and southwestern Indiana. ⁷ One of the clusters reported in the current study overlapped with the cluster from our previous study. These findings support the hypothesis that county-level variations in the distribution of summated tickborne disease diagnoses exist in Indiana. Our map layers also show that the significant clusters of all tickborne disease diagnosis are predominantly in rural and rural-mixed counties. While the current exploratory study does not investigate the relative distribution of risk factors in association with spatiotemporal clusters, previous studies have reported risk factors associated with increased tickborne disease incidence and prevalence include rural dwelling, male gender, and outdoor occupational exposure to ticks.^7,9,27 Although we detected significant spatiotemporal clusters when summated tickborne disease was used in the analysis, no significant spatiotemporal clusters were detected when each type of tickborne disease (ie, Rocky Mountain spotted fever, Lyme disease, ehrlichiosis, tularemia, and typhus/rickettsial diseases, respectively) was used as standalone variable in our analysis. It remains to be seen whether this spatiotemporal clustering of summated tickborne diseases diagnoses demonstrates a common infection source, or if other factors are at play.

Certain limitations of our study findings are worth pointing out. First, we are unable to determine if the higher prevalence of Lyme disease diagnosis observed in this study, compared with other tickborne diseases, is due to a higher Lyme disease incidence. While the diagnoses of most tickborne diseases depend on laboratory criteria, the occurrence of erythema migrans has been used in the probable diagnosis of Lyme disease, which may inflate the prevalence of Lyme disease reporting. ²⁸ This is particularly true in South Central United States where patients have been known to develop erythema migrans-like skin lesions, but serological tests for antibodies to Borrelia burgdorferi are almost always negative. ²⁹ Secondly, because spatial scan statistics provide only an approximate location and size of the clusters, the exact borders of the clusters are uncertain. This means that for any given cluster, there may be areas within the cluster that do not have an excess risk of tickborne disease and areas outside the cluster that could have an excess risk. Thirdly, in the absence of a detailed travel history to verify whether exposure and subsequent bite of infected ticks occurred in the county of diagnosis, the results of tickborne disease diagnosis may not reflect actual location of risk for tick exposure. This is because a time lag exists between the bite of an infected tick and symptom development among exposed individuals. ²⁸ If there are significant differences between the county of exposure and county of diagnosis, the results of cluster detection could be misleading in relation to the actual risk. Other limitations are generally associated with surveillance data procurement, including underreporting and misclassification, and changes in case definition status, particularly for Lyme disease. ³⁰

Conclusion

This study shows that retrospective tickborne disease diagnosis data provide important clues for identifying geographic foci of tickborne disease diagnoses. The study also shows that all tickborne diseases diagnosis counts were significantly higher in urban countries compared with rural and rural-mixed counties. Assuming the counties identified in this study represent actual locations of high tickborne disease risk, our current findings can inform more efficient targeted surveillance strategies, monitoring of the possible geographic expansion of tickborne diseases, and an improved understanding of the spatial epidemiology of tickborne disease diagnoses in Indiana. Nonetheless, there is a need to further investigate the apparent discrepancy between an observance of higher tickborne disease diagnoses in urban counties, and the appearance of higher risk factors for tick exposure in rural and rural-mixed counties. Further, the detection of 2 clusters of summated tickborne diseases diagnoses suggests a need for continued surveillance of the counties observed within these clusters. Future studies should be designed to test the association between tick abundance, tick infection rate, self-reported tick exposures, and tickborne disease diagnosis. Furthermore, future studies should identify risk factors for tickborne disease diagnosis in counties of Indian where clusters have been detected, compared with counties without significant cluster detection. ⁹