Figure 2

Figure 2. a. Confirmed human granulocytic ehrlichiosis (HGE) cases identified through active and passive surveillance systems, 1997–2000; b. Raw annualized incidence of confirmed HGE cases by town, 1997–2000*; c. Raw annualized incidence of...

Two hundred forty-five cases of HGE were identified through the active and passive surveillance systems in 1997–2000. A total of 136 confirmed cases of HGE were identified; 128 (94%) of these were geocoded to points in an ArcView theme (Figure 2a). Addresses that were not geocoded consisted of two incomplete ones (street names with no number) and six post office boxes from Chester, Essex, Haddam, Lyme, Madison, and Westbrook.

Annualized incidence rates for 1997–2000 were calculated by town and census block group by using 1990 census data to show the crude distribution of HGE in the 12-town area (Figure 2b,c). Rates by town ranged from 3/100,000 in East Haddam to 156/100,000 in Lyme. Rates by census block group ranged from 0/100,000 to 187/100,000 and demonstrated a high degree of random variation because of the small population size and low case numbers.

Smoothing provided a clearer picture of the areas of increased risk on a smaller scale than by town. A filter number of 10 provided the most appropriate map of smoothed incidence rates (Figure 2d). This filter number decreased random variation and showed an increased risk of contracting HGE around the mouth of the Connecticut River with risk decreasing to the north and west.

Figure 3

Figure 3. a. Single identified cluster of human granulocytic ehrlichiosis (HGE) cases within the 12-town area (maximum cluster size ≤50% total population), relative risk (RR)=1.8, p=0.001; b. Two identified clusters of HGE cases...

Using the maximum spatial cluster size of <50% of the total population, the spatial cluster analysis identified a single cluster that included all census block groups in the towns of Lyme, Old Saybrook, Chester, Essex, Deep River, and Westbrook, all but one in the town of Old Lyme, and one from the town of Clinton (Figure 3a). The identified cluster contained 46.1% of the area’s total population. The overall RR within the cluster was 1.8, with an observed number of cases of 106 compared with 59 expected cases. This elevated risk within a nonrandom pattern of disease distribution was significant (p=0.001).

To investigate the possibility of smaller clusters, the same analysis was performed with a maximum spatial cluster size of <25% of the total population. Two clusters were identified, one including Lyme and Old Lyme as well as parts of Essex, Old Saybrook, and Deep River (Figure 3b). This cluster contained 18.2% of the total population and had an overall RR of 2.6 (p=0.001), with 61 cases observed compared with an expected 23 cases. The second subcluster included areas of Deep River and Essex (Figure 3b). This cluster contained 4.2% of the total population and had an overall RR of 2.6 (p=0.16), with 14 cases observed compared with 5 cases expected. While the primary cluster identified in this analysis was significant and showed a higher overall RR, the larger cluster from the first analysis, as a result of the lack of preselection bias, better represented the areas of increased risk for infection on the basis of the spatial distribution of HGE in the 12-town area.

Cluster analyses were also performed on confirmed and probable case data. By using the <50% maximal cluster size, we included identical census block groups in the cluster (p=0.001). In the subcluster analysis that used the <25% cluster size, both subclusters identified with confirmed cases only were found to be statistically significant (p<0.005). The primary subcluster was geographically circumscribed when confirmed and probable cases were included in the analysis.

Using a GIS and spatial statistics, we investigated the spatial distribution of confirmed cases of HGE and identified areas of increased risk within an area highly endemic for tick-borne diseases. Such diseases have become recognized as serious health threats in the northeast United States in the last 20 years because of increasing prevalence and heightened detection. Areas characterized by low residential density and a landscape of recently reforested deciduous forest are strongly associated with the risk for Lyme disease (30,32,33). Areas of high Lyme disease risk have been shown to also have an increased risk for HGE (19). Our study identified spatial variations in the risk for HGE in such an area. Furthermore, the analysis demonstrated that combining thorough surveillance information with spatial analysis techniques can increase understanding of the epidemiology of HGE within a highly disease-endemic area. The next step, to investigate the underlying causes of increased risk in the identified areas, will be analysis of landscape attributes and identification of the environmental variables characteristic of high-risk areas.

The spatial statistics analyses clearly yielded a nonrandom distribution of HGE within the 12-town area. Spatial filtering (smoothing) identified areas of increased risk centered around the mouth of the Connecticut River, primarily on the eastern side of the river, in the towns of Lyme and Old Lyme. Increased likelihood of disease was seen on the western side of the river but was not as consistently high as the risk observed in Lyme and Old Lyme. Spatial cluster analysis identified a statistically significant cluster (RR=1.8, p=0.001) in the same area, around the mouth of the Connecticut River, including the towns of Chester, Deep River, Essex, Lyme, Old Lyme, Old Saybrook, and Westbrook. One census block group in southeastern Old Lyme was not included in the cluster, and one block group in Clinton was included. This cluster analysis was performed by using the default maximum spatial cluster size of <50% of the total population. Using this default method minimizes pre-selection bias of cluster size. However, to investigate the possibility of subclusters, additional cluster analysis based on a maximum spatial cluster size of <25% of the total population identified two subclusters, one significant (RR=2.6, p=0.001) and the second not significant (RR=2.6, p=0.19). The decrease in risk for HGE infection as one moves away from the coast is consistent with the results of Nicholson and Mather, who described a decreasing Lyme disease risk with increasing latitude in Rhode Island (34).

The present study analyzed the associations between human population and human disease only. Gathering and including vector population data (including population density, distribution, and infection prevalence rates) and environmental variables in the risk analysis of HGE in the 12-town area may provide a more comprehensive view of the disease risk. The relationship between Lyme disease, I. scapularis vectors, and landscape characteristics has been studied from remotely sensed and field-gathered data (35–37), but it is unknown whether these relationships can be applied to other tick-borne diseases, including HGE. Increased Lyme disease risk has been well correlated with increased tick abundance and prevalence of infected ticks (34,35,38). The spatial distribution of Lyme disease rates is correlated with widespread tick populations and pathogen prevalence (25). Environmental risk factors and landscape characteristics associated with Lyme disease have been identified (22,35–37). Using techniques similar to those used for Lyme disease, combined with the results of this study, future research will include investigating the landscape characteristics associated with HGE. Further, discernment of the aspects of the natural history of HGE that are not understood, especially pertaining to the reservoir host, may supply additional information that can be used to further refine areas of HGE risk.

While similar numbers of specimens were submitted for HGE testing to both the active and passive surveillance systems, the low rate of convalescent-phase specimen collection and the application of only one diagnostic test in passive surveillance resulted in fewer cases from passive surveillance being confirmed and included in the current analysis. Persons who live at the edges of the 12-town area may have been more likely to visit practitioners outside the active surveillance area. These case-patients would have been identified through passive surveillance but would have been less likely to be confirmed. While the practices participating in active surveillance were located throughout the 12-town area and include one practice outside that area, the lower rate of confirmed cases in the passive system may have biased the results toward the center of the surveillance area. However, the similar results obtained from the spatial statistics analysis that used both confirmed and probable cases suggest that this error may not have played a large role in the observed patterns of disease.

Because of variations in testing throughout the 4 years of surveillance, analysis for temporal clusters was not possible. Retrospective testing of banked samples from previous surveillance years or continuing accumulation of surveillance data in years to come will be needed to investigate the temporal as well as spatial spread of HGE within the 12-town area. Temporal trends, combined with time series analysis of remotely sensed land cover and land-use data, may provide indications of future areas at increased risk for HGE. Concurrent analysis of the spatial and temporal distributions of other I. scapularis–borne diseases in this area, including Lyme disease and babesiosis, may clarify the similarities and differences in risk among these common vector-borne infections.

Our study was based on the assumption that people acquire infection with the agent of HGE peridomestically, or near their homes. Falco and Fish (30) demonstrated that most cases of Lyme disease were acquired peridomestically, but no studies have investigated whether HGE infections are similarly acquired. While the life cycle similarities of these two pathogens support the assumption that HGE transmission dynamics are similar to those of Lyme disease, additional research is needed to test this hypothesis.

This spatial analysis was limited to the described 12-town study area. This area of active surveillance was identified previously by its high rates of Lyme disease, and the towns to the east were excluded because at the time the study was initiated (1997) those towns had lower rates of Lyme disease compared with the 12 study towns. However, in this analysis, understanding of the spatial distribution of HGE would be enhanced if the towns to the east of the current study area were also included in the active surveillance, given the high rates of HGE in Lyme and Old Lyme. Stemming from this analysis, the towns to the east of the original study area (Salem and East Lyme) were added to the active surveillance area in 2002. The eastern limitation in our dataset highlights another use of spatial analysis and GIS. The tools and methods described in this study can identify areas where increased surveillance is recommended.

Human behavior is a strong predictor of tick-borne disease risk, including how people move in their environment, their outdoor activities, and the individual protection they use to prevent tick bites. Reforestation in areas previously used for agriculture results in more favorable conditions for tick and reservoir hosts, while the trend towards residential preferences in well-shaded suburban and rural areas exacerbates the tick-human interactions. Change in human behavior concurrent with an ecologic transition further increases and alters tick-borne disease risk. Local weather variations and the periodicity of weather patterns also play a role in tick-borne disease risk. The combination of these factors results in a high variability of risk even within an area known to be hyperendemic for tick-borne diseases.

On the basis of data on peridomestic Lyme disease infections (30), prevention strategies are recommended that focus on persons’ risk at home. In an area in which tick-borne diseases are highly endemic, aiming prevention strategies at areas of highest risk can potentially increase the program’s effectiveness. Persons at highest risk should be informed of that risk and of the possibilities for risk reduction. Funds spent on programs might be better spent on areas where cost-effectiveness can be maximized. At this time, practical prevention advice to prevent tick-borne disease in highly disease-endemic areas is elusive.

The tools described in this article, GIS and spatial statistics, provide an opportunity to clarify and quantify the health burden from tick-borne disease within a highly endemic area and a foundation to pursue further investigation into the environmental factors resulting in increased disease burden. To implement specific and geographically appropriate risk-reduction programs, the use of such spatial analysis tools should become integral components in the epidemiologic description and risk assessment of tick-borne diseases.