Solving the cluster puzzle: clues to follow and pitfalls to avoid

Dozens of methods have been proposed for the identification of disease clusters, although only a few are used routinely in published investigations. New methods, although designed to exploit some particular aspect of the data or use some specific statistical tool, are rarely compared thoroughly in terms of power or performance. Users, when confronted with the multitude of methods available, often select methods arbitrarily, basing choices on software availability, ease of implementation, or use experience rather than considerations of statistical power, possible alternative hypotheses (that is, cluster structure) and likely confounding. In this review, we extend our typology of disease clustering methods and apply it to many of the extant methods identifying strengths, weaknesses and unique features of the methods. We conclude with recommendations for which methods should be applied to which types of situations.