Statistical power of disease cluster and clustering tests for rare diseases: a simulation study of point sources

Abstract

Two recent epidemiological studies on clustering of childhood leukemia showed different results on the statistical power of disease cluster and clustering tests, possibly an effect of spatial data aggregation. Eight different leukemia cluster scenarios were simulated using individual addresses of all 1,009,332 children living in Denmark in 2006. For each scenario, a number of point sources were defined with an increased risk ratio at centroid, decreasing linearly to 1.0 at the edge; aggregation levels were administrative units of Danish municipalities and squares of 5, 12.5 and 25 km(2). Six statistical methods were compared. Generally, statistical power decreased with increasing size of aggregated units. In our scenarios, statistical tests based on individual data usually had lower statistical power than the best test based on aggregated data. In conclusion, spatial aggregation does not necessarily blur a clustering effect; this depends on the nature of clustering and the aggregated units.