J Gaudart1, R Giorgi, B Poudiougou, O Touré, S Ranque, O Doumbo, J Demongeot. 1. Equipe biomathématiques et informatique médicale, LIF-UMR 6166-CNRS, faculté de médecine de Marseille, université Aix-Marseille, Marseille cedex 05, France. jean.gaudart@univmed.fr
Abstract
BACKGROUND: Various statistical methods have been developed to describe spatial heterogeneity, in terms of high risk zones. If no source can be determined, this heterogeneity can be globally or locally described. Global methods test a statistic estimated over the whole studied geographical area, whereas local methods estimate a statistic on each spatial unit (or regrouping unit). This paper aimed to present, and to compare results of an epidemiological application, of five methods of spatial cluster detection. METHODS: The two global detection methods were: 1) Moran's coefficient, a classically used autocorrelation coefficient; 2) Tango's statistic, a spatial generalization of the Chi(2) statistic. The three local methods were: 1) the local application of Moran's coefficient, proposed by Anselin; 2) the scan statistic, which searches for grouping of spatial units; 3) the oblique regression tree, which splits the studied zone into sub-zones of different risks. These five methods were applied to the description of the spatial heterogeneity of the malaria risk over a hyperendemic village, in Mali. RESULTS: All the methods highlighted a significant spatial heterogeneity. Both global methods (Moran's coefficient and Tango's statistic) showed weak spatial correlations. Local Moran's coefficient (with Bonferronis' adjustment) highlighted five spatial units. The scan statistic identified a single high risk cluster. The regression oblique tree split the study area into six sub-zones; the sub-zone with the higher risk was consistent with the cluster identified by the scan statistic. CONCLUSION: These presented methods do not require any previous knowledge of a source. They allow evaluating spatial risk heterogeneity over the entire geographical area under study. It is noteworthy that shape, size, and spatial heterogeneity characteristics (either global or local) of the study area, as well as the definition of the proximity, significantly influence the spatial risk analysis' outcome. Although their results should be cautiously interpreted, these methods are useful for preliminary field studies or epidemiological surveys.
BACKGROUND: Various statistical methods have been developed to describe spatial heterogeneity, in terms of high risk zones. If no source can be determined, this heterogeneity can be globally or locally described. Global methods test a statistic estimated over the whole studied geographical area, whereas local methods estimate a statistic on each spatial unit (or regrouping unit). This paper aimed to present, and to compare results of an epidemiological application, of five methods of spatial cluster detection. METHODS: The two global detection methods were: 1) Moran's coefficient, a classically used autocorrelation coefficient; 2) Tango's statistic, a spatial generalization of the Chi(2) statistic. The three local methods were: 1) the local application of Moran's coefficient, proposed by Anselin; 2) the scan statistic, which searches for grouping of spatial units; 3) the oblique regression tree, which splits the studied zone into sub-zones of different risks. These five methods were applied to the description of the spatial heterogeneity of the malaria risk over a hyperendemic village, in Mali. RESULTS: All the methods highlighted a significant spatial heterogeneity. Both global methods (Moran's coefficient and Tango's statistic) showed weak spatial correlations. Local Moran's coefficient (with Bonferronis' adjustment) highlighted five spatial units. The scan statistic identified a single high risk cluster. The regression oblique tree split the study area into six sub-zones; the sub-zone with the higher risk was consistent with the cluster identified by the scan statistic. CONCLUSION: These presented methods do not require any previous knowledge of a source. They allow evaluating spatial risk heterogeneity over the entire geographical area under study. It is noteworthy that shape, size, and spatial heterogeneity characteristics (either global or local) of the study area, as well as the definition of the proximity, significantly influence the spatial risk analysis' outcome. Although their results should be cautiously interpreted, these methods are useful for preliminary field studies or epidemiological surveys.
Authors: Jordi Landier; Jean Gaudart; Kevin Carolan; Danny Lo Seen; Jean-François Guégan; Sara Eyangoh; Arnaud Fontanet; Gaëtan Texier Journal: PLoS Negl Trop Dis Date: 2014-09-04