[Estimation of the dental status of 3- to 5-year-old children by means of cluster samples]

Abstract

Objectives: Surveys on the dental status of preschool children in selected regions or at specific ages often take place in randomly assigned nursery centres (cluster sample). The mean dmft value in the population serves here as an indicator for the average caries decay. It is evaluated from the individual dmft values of the children on the basis of statistical procedures. The study at hand assesses the impact of cluster structure in the data on the estimated dmft mean and its variance.

Methods: We defined 170 nursery centres with 7 578 children aged 3-5 years old as population and drew 100 times 10%-cluster samples (with replacement) and by this means we simulated 100 surveys. Estimation of the population mean and the variance was established each time with and without respecting the cluster structure.

Results: If the cluster structure is not obeyed in the course of calculating, the resulting value of the variance for the dmft mean estimator is too small; and hence confidence intervals are too narrow or tests exceed their type I error. If, in contrast, the cluster structure is taken into account, the variance is increased in all 100 cases (variance inflation), with a magnitude that, however, varies randomly. For surveys with cluster samples, this effect can have an impact on sample size calculations. A bias of the dmft average itself is possible for samples of a moderate extent once the mean occupancy of the nursery centres in the sample deviates markedly from that in the population.

Conclusions: If a cluster design is chosen for a caries epidemiology survey in nursery centres, the sample size calculation and the analysis should take that into account in order to avoid misleading results.