The design of multi-stage tuberculin surveys: some suggestions for sampling.

Setting: Tuberculin surveys of children to estimate national or regional infection prevalences are commonly designed as multi-stage surveys. These surveys require strategies for the efficient sampling of sub-units at each stage.
Objectives: To develop guidelines for sampling in tuberculin surveys.
Design: Sampling theory was used to develop a simple and efficient sampling strategy for planning and analysing tuberculin surveys. The issue of sample sizes is considered.
Results: Formulae for the calculation of infection prevalences and their confidence intervals are developed. Sample sizes are discussed.
Conclusion: We recommend that districts be sampled using sampling proportional to size, in which larger units have a larger probability of being included in the sample. Schools are sampled next using simple random sampling, where each school within a district has the same probability of being included in the sample. In each school all eligible children are Mantoux tested. The number of children tested per district should be approximately constant. Increasing the number of selected districts is usually more efficient for increasing the precision of the estimate than increasing the number of children per district beyond several hundred to a few thousand.