Free Access
Issue
Vet. Res.
Volume 38, Number 4, July-August 2007
Page(s) 585 - 596
DOI https://doi.org/10.1051/vetres:2007018
Published online 30 May 2007
How to cite this article Vet. Res. (2007) 585-596
Vet. Res. 38 (2007) 585-596
DOI: 10.1051/vetres:2007018

A spatial hazard model for cluster detection on continuous indicators of disease: application to somatic cell score

Emilie Gaya, b, Rachid Senoussia and Jacques Barnouinb

a  INRA, UR546, Biostatistique et processus spatiaux, Avignon, 84000, France
b  INRA, UR346, Épidémiologie animale, Saint-Genès-Champanelle, 63122, France

(Received 28 August 2006; accepted 12 February 2007 ; published online 30 May 2007)

Abstract - Methods for spatial cluster detection dealing with diseases quantified by continuous variables are few, whereas several diseases are better approached by continuous indicators. For example, subclinical mastitis of the dairy cow is evaluated using a continuous marker of udder inflammation, the somatic cell score (SCS). Consequently, this study proposed to analyze spatialized risk and cluster components of herd SCS through a new method based on a spatial hazard model. The dataset included annual SCS for 34 142 French dairy herds for the year 2000, and important SCS risk factors: mean parity, percentage of winter and spring calvings, and herd size. The model allowed the simultaneous estimation of the effects of known risk factors and of potential spatial clusters on SCS, and the mapping of the estimated clusters and their range. Mean parity and winter and spring calvings were significantly associated with subclinical mastitis risk. The model with the presence of 3 clusters was highly significant, and the 3 clusters were attractive, i.e. closeness to cluster center increased the occurrence of high SCS. The three localizations were the following: close to the city of Troyes in the northeast of France; around the city of Limoges in the center-west; and in the southwest close to the city of Tarbes. The semi-parametric method based on spatial hazard modeling applies to continuous variables, and takes account of both risk factors and potential heterogeneity of the background population. This tool allows a quantitative detection but assumes a spatially specified form for clusters.


Key words: spatial epidemiology / cluster detection / hazard function / mastitis / dairy herd

Corresponding author: emilie.gay@u707.jussieu.fr

© INRA, EDP Sciences 2007