BACKGROUND: Surveillance and prevention programs often struggle to utilize spatial cluster and hot spot detection tools that can be routinely employed to examine patterns, trends, and magnify health disparities within widely-accessible health data systems including vital records (birth and death data), hospital inpatient discharge data and emergency department visit data. With an increased lens on health equity, it is advantageous and important for public health professionals working with data to embrace a general understanding of the definitions and purpose of cluster and hot spot detection methods and be able to routinely translate or apply their results to increase our understanding of health inequities, surveillance, and other health promotion activities.
METHODS: A five-year data set comprised of pre-term birth rates, infant mortality rates, heart-disease death rates, asthma hospital inpatient discharge rates, and diabetes emergency department visit rates was compiled using indicators from several primary health data sets in Colorado: Colorado Birth Dataset, Colorado Death Dataset, Colorado Hospital Discharge Dataset and the Colorado Emergency Department Visit Dataset. Data from these systems was aggregated to the census tract geography and rates were age-adjusted based on five year population estimates. Three software packages (two are free of charge) containing cluster and hot spot detection tools were installed and the following four statistics (depicting location and magnitude of clusters and hotspots) were calculated for each of the five health outcomes: Kulldorff Spatial Scan Statistic, Anselin Local Moran's I, K-means clustering, and the Getis-Ord local G.
RESULTS: Results from the cluster and hotspot statistics include defined hot spots, census tracts where high or low values are clustered, census tracts that are considered outliers, in addition to information on the most likely cluster, the relative risk of each cluster, and the observed/expected ratio of each cluster, for each of the five health outcomes. In order to make sense of the complex and similar output calculated from each of the four statistical methods, maps and tables detailing each method’s output and findings were developed with simplicity in mind so that program staff with little to no experience working in spatial statistics could interpret the findings from these methods and provide additional perspective as to the distribution of health disparities across a region or state.
CONCLUSIONS: Programs that assess health status or identify health disparities based on well established health data sets should take advantage of spatial statistics tools and routinely implement a strategy for determining and visualizing clusters or hot spots.