Answering Cluster Investigation Requests: The Value of Simple Simulations and Statistical Tools

Abstract

Cluster investigations remain an important public health issue as the number of reported clusters and public concern increase. This study shows how statistical considerations and a simulation tool may be helpful in providing communities with proper answers to the questions usually raised in such situations: How surprising is an observed childhood cancer excess? What could be learned from a statistical test? What could be learned from a case-control study? Using real demographic and incidence-rate data together with simulations based on the hypothesis that incidence rates are homogeneous, the probabilities of observing given situations were estimated. A number of real situations have been used as examples. The results of the simulation study showed, in detail, that no reliable information on the reality of an observed excess could be obtained a posteriori from a statistical test. A cluster of the same size may or may not be surprising, depending on the spatial area and time window to which the cases are related (i.e., the expected number of cases), and depending on the size of the referential territory to which this area is associated. The lack of power of a case-control study if no particular unusual exposure is present is also addressed. The approach described in this paper can easily be reproduced and adapted to many situations. It may be of assistance to health departments conducting cluster investigations and communicating with the public.

Spatial and space–time clustering of childhood acute leukaemia in France from 1990 to 2000: a nationwide study

“…Over the last few years, the number of localised childhood cancer clusters reported to the health authorities has shown a tendency to increase in several countries, among which France. However, hardly ever has an environmental factor been hold responsible for the observed excess of cases; such clusters were probably due to chance alone (Bellec et al, 2005). Likewise, no excess in the incidence of childhood leukaemia was evidenced over 1990 -1998 around the 29 French nuclear installations (White-Koning et al, 2004).…”

Détection de clusters spatiaux sans point source prédéfini: utilisation de cinq méthodes et comparaison de leurs résultats

“…Plus τ sera grand, plus le test sera sensible aux larges clusters et inversement. La déter-mination optimale de la matrice de proximité est un des problè-mes les plus débattus [18,25,57,[65][66][67]. Un des problèmes de la matrice de contiguïté est l'exclusion des US n'ayant pas d'unité directement voisine (contiguë) [68].…”

Cluster Detection Tests in Spatial Epidemiology: A Global Indicator for Performance Assessment

“…For development of new statistical methods, simulation studies are obviously essential. In field investigation, they provide useful knowledge for interpretation of real data and decision making [1]. However, from a methodological point of view, there is still no commonly accepted protocol for simulation studies in spatial epidemiology.…”