Evaluating cancer clusters.

We have had considerable success in identifying cancer causing agents in the workplace using epidemiologic methods. This success had made us very sensitive to the occurrence of cancer clusters among workers in the belief that identification of some common exposure could reveal the presence of a carcinogen and lead to preventive measures. This intense surveillance is both a blessing and a curse. On the one hand, it is a proven way of discovering environmental causes of cancer. On the other, it leads to false alarms or does not always lead to identification of a causal agent. It is easy to demonstrate, using tables of random number 5, how clusters can occur by chance and to demonstrate that when the number of comparisons made in identifying clusters is known there is a basis for their evaluation. Unfortunately, in most instances, when cancer clusters are detected in the workplace the number of comparisons made is unknown and the statistical significance of the cluster cannot be evaluated. Moreover, it is not usually recognized that in this situation when a study is made as a result of discovering a cluster in a particular population, the cases that make up the cluster cannot be included in a data set which tests the hypothesis that a cluster exists. This paper illustrates the above points by actual experiences.