Parasitology

SECTION 1 ADVOCACY AND DEFINING AREAS IN NEED OF CONTROL

Bayesian geostatistical modelling for mapping schistosomiasis transmission

P. VOUNATSOUa1 c1, G. RASOa2a3, M. TANNERa1, E. K. N'GORANa4a5 and J. UTZINGERa1

a1 Department of Public Health and Epidemiology, Swiss Tropical Institute, P.O. Box, CH-4002 Basel, Switzerland

a2 Division of Epidemiology and Social Medicine, School of Population Health, The University of Queensland, Public Health Building, Herston Road, Brisbane, Queensland 4006, Australia

a3 Molecular Parasitology Laboratory, Queensland Institute of Medical Research, 300 Herston Road, Brisbane, Queensland 4006, Australia

a4 UFR Biosciences, Université de Cocody-Abidjan, 22 BP 582, Abidjan 22, Côte d'Ivoire

a5 Centre Suisse de Recherches Scientifiques, 01 BP 1303, Abidjan 01, Côte d'Ivoire

SUMMARY

Progress has been made in mapping and predicting the risk of schistosomiasis using Bayesian geostatistical inference. Applications primarily focused on risk profiling of prevalence rather than infection intensity, although the latter is particularly important for morbidity control. In this review, the underlying assumptions used in a study mapping Schistosoma mansoni infection intensity in East Africa are examined. We argue that the assumption of stationarity needs to be relaxed, and that the negative binomial assumption might result in misleading inference because of a high number of excess zeros (individuals without an infection). We developed a Bayesian geostatistical zero-inflated (ZI) regression model that assumes a non-stationary spatial process. Our model is validated with a high-quality georeferenced database from western Côte d'Ivoire, consisting of demographic, environmental, parasitological and socio-economic data. Nearly 40% of the 3818 participating schoolchildren were infected with S. mansoni, and the mean egg count among infected children was 162 eggs per gram of stool (EPG), ranging between 24 and 6768 EPG. Compared to a negative binomial and ZI Poisson and negative binomial models, the Bayesian non-stationary ZI negative binomial model showed a better fit to the data. We conclude that geostatistical ZI models produce more accurate maps of helminth infection intensity than the spatial negative binomial ones.

(Received October 24 2008)

(Revised February 24 2009)

(Accepted February 27 2009)

(Online publication June 02 2009)

Correspondence:

c1 Corresponding author: Penelope Vounatsou, Department of Public Health and Epidemiology, Swiss Tropical Institute, P.O. Box, CH-4002 Basel, Switzerland. Tel: +41 61 284-8109. Fax: +41 61 284-8105. E-mail: penelope.vounatsou@unibas.ch

Metrics