Objectives: To describe our experience using satellite image-based sampling to conduct a health survey of children in an urban area of Lusaka, Zambia, as an approach to sampling when the population is poorly characterized by existing census data or maps.
Methods: Using a publicly available Quickbird image of several townships, we created digital records of structures within the residential urban study area using ArcGIS 9.2. Boundaries were drawn to create geographic subdivisions based on natural and man-made barriers (e.g. roads). Survey teams of biomedical research students and local community health workers followed a standard protocol to enroll children within the selected structure, or to move to the neighbouring structure if the selected structure was ineligible or refused enrollment. Spatial clustering was assessed using the K-difference function.
Results: Digital records of 16 105 structures within the study area were created. Of the 750 randomly selected structures, six (1%) were not found by the survey teams. A total of 1247 structures were assessed for eligibility, of which 691 eligible households were enrolled. The majority of enrolled households were the initially selected structures (51%) or the first selected neighbour (42%). Households that refused enrollment tended to cluster more than those which enrolled.
Conclusions: Sampling from a satellite image was feasible in this urban African setting. Satellite images may be useful for public health surveillance in populations with inaccurate census data or maps and allow for spatial analyses such as identification of clustering among refusing households.