Abstract

In this issue of the Journal, Heath (1) presents a detailed account of investigations into eight clusters of childhood leukemia and lymphoma carried out by the US Centers for Disease Control and Prevention (CDC) during the 1960s and 1970s. The clusters, chosen for characteristics suggestive of an infectious etiology, were selected from 50 such investigations undertaken during this time period. Heath concludes that indicators of interpersonal contact suggest that infectious disease underlies the etiology of childhood leukemia and lymphoma. Do clusters really provide such evidence?

A cluster is an excess incidence of related health events occurring at the same place, the same time, or (more usually) both. The difficulties involved in identifying clusters of chronic disease have been discussed extensively (e.g., see Rothman (2)). Heath states that the CDC considered post-hoc investigation of a cluster with formal statistical evaluation irrelevant, which is in line with the position of many researchers (e.g., see Alexander (3)). Indeed, one might argue that, were statistical testing required to prove the existence of a cluster, that cluster would be less than useful for causal inference. There are more objective arguments that tests of statistical significance are not appropriate. Principal among these are the definitions of closeness between cases in space, time, and diagnosis employed. These assumptions motivated Grufferman (4) to compare cluster identification to the method of the “Texas sharp-shooter,” who shoots at a wall and then draws a target around the bullet holes: Spatial, temporal, and diagnostic boundaries may be engineered to provide the biggest possible cluster. Heath did not make clear how the CDC investigations of these clusters drew their boundaries, but at least some of the clusters were brought to the CDC's attention through cases living in very close proximity.

https://academic.oup.com/aje/article/162/9/817/58326/Community-Clusters-of-Childhood-Leukemia-and