Friday
Organizers:
Andrew Lawson, University of South Carolina, [email protected]
Daniel Wartenberg, Robert Wood Johnson Medical School, [email protected]
Presented under the auspices of the Special Focus on Computational and Mathematical Epidemiology.
**********************************************************
Disease clusters, defined as local excesses of disease in space, time or space and time, represent an important but vexing problem in public health. Clusters are usually identified by community residents who believe that some unusual circumstance leading to unexpected illness has befallen their families, friends or neighbors. Clusters of leukemia are reported most often, although clusters of other cancers, birth defects and other adverse health outcomes are also reported. While there are many protocols to assess whether a given cluster is etiologic, i.e., due to an identifiable cause, there is no clear consensus about how best to conduct an investigation and reach a scientifically valid conclusion. A variety of statistical issues confront investigators of clusters. For example, since cluster reports typically are based on a handful of cases, it is possible that the observed excess is simply due to random variation, particularly if the investigator fails to adjust for multiple comparisons. On the other hand, the statistical power of traditional cluster analysis methods is fairly low, likely resulting in many false negatives which might cause investigators to miss true, etiologic clusters. In addition, assumptions are made about the amount of disease expected because large data sets are generally not available at a scale that would enable investigators to determine background rates of disease, such as at the census block, census tract or zip code. Some recent methods have begun to look at approaches for conducting prospective surveillance by analyzing data collected for each time unit (e.g.,
year) it is collected. These methods offer the opportunity the overcome some of the statistical limitations of traditional cluster analyses and provide a more appropriate perspective for health officials to use in responding to community concerns. The workshop will bring together mathematicians, biostatisticians, epidemiologists and public health officials to develop an approach that, while statistically rigorous, is able to address the concerns of the public.