BACKGROUND: Spatial analysis is increasingly being utilized by public health, particularly in disease cluster detection. Cluster detection can be used to identify geographic areas with high-risk population so that targeted public health prevention strategies can be put in place. This method may also be useful in inferring geographic directionality of disease patterns in low incidence states, particularly in counties which share a border with other states. With increasing outward expansion of reported Lyme disease (LD) cases from the Midwest and Northeast, it is important to determine if state border effects contribute to clustering of LD cases in low incidence states due to their proximity to high incidence states. Previous studies have employed spatial statistics for targeted surveillance efforts, yet a significant gap in knowledge exists on methods to assess border bias potentially influencing case reporting near state lines.
METHODS: Yearly county-level case report numbers for probable and confirmed LD cases report obtained for low incidence states of Ohio and West Virginia, and high incidence states of Maryland and Pennsylvania (2008-2014). Yearly county-level population estimates were obtained for each state from the U.S Census Bureau. Data were joined in ArcMap 10.1 to a U.S. counties by state shapefile. Local indicators of spatial autocorrelation (LISA) and spatial empirical Bayesian smoothed rates were calculated in GeoDa 1.6.
RESULTS: During 2008-2014, 14,726 human Lyme cases were reported with 3.25% (n=479) in Ohio; 6.53% (n=962), West Virginia; 12.6% (n=1,859), Pennsylvania; and 77.6% (n=11,426) Maryland. Cluster analysis was performed on low incidence states initially, and found significant clustering (Z = 0.65, P = 0.05) of high human LD reporting in West Virginia’s Northeastern Panhandle. However, once high incidence bordering states (Pennsylvania and Maryland) were included in the spatial analysis no significant clustering occurred within West Virginia or Ohio. Of particular interest, smoothed rates indicated elevated human risk in West Virginia, particularly in those counties which border CDC recognized high incidence states
CONCLUSIONS: Potential limitations for the analysis are linked to data quality issues associated with changes in LD case definitions during study period, potential misdiagnosis, inaccurate case reporting, and availability of county-level data for surrounding parts of the region. Despite these limitations, a need for revised LD surveillance strategies in low incidence states may be warranted. This pilot study fill gaps in knowledge and proposes a low cost methodology for targeted surveillance and determination of Lyme progression within low incidence states.