Analysis of community cancer mortality rates
Author(s)Vatland, Janice A. (Janice Audrey)
Massachusetts Institute of Technology. Division of Bioengineering and Environmental Health.
William G. Thilly.
MetadataShow full item record
Residents of small communities or neighborhoods often observe what appears to them to be an excessively high number of cancer cases. Some of these observations are brought to the attention of public health officials in the hope that an environmental cause within these communities may be discovered and subsequently eliminated. However, what appears to be a cluster of cancer cases locally may be a chance occurrence when viewed from the perspective of the entire state or region. The primary aim of this work was to obtain and analyze a data set of sufficient magnitude to provide a means to discover if distributions of cancer mortality rates among communities within a particular state, for any of the most common forms of cancer, were compatible with chance. To this end cancer mortality data for six of the largest states in the United States (California, Florida, Massachusetts, New York, Pennsylvania and Texas) were collected, converted into mortality rates and analyzed to discover if the variations among communities could be accounted for by chance alone. These data comprised one third of all recorded deaths in the period of approximately 1969 to 1998. The 21 most common forms of adult and 6 most common forms of pediatric cancers were organized to permit analyses within each of the 6 states with regard to age (0-19, 65-84 and >/= 85 years), gender and ethnicity (European American and Non-European American descent). Key to this work was the mode of statistical analysis. For each community an expected mortality rate and its expected distribution was defined by the average mortality rate for all communities within each state and the binomial distribution, respectively. These expected distributions were summed for all communities to define the expected chance distribution of community mortality rates for each state, cancer, gender, age cohort and ethnicity. This produced nearly 800 separate "chance" distributions. Each of these was compared to the corresponding observed distribution using the Kolmogorov-Smimov statistical test. This test was designed to discover statistically significant differences between any two distributions. Here it was used to determine which of the nearly 800 observed distributions could not be accounted for by chance alone. Of these comparisons, 16 were found to have observed distributions significantly different from the expected by chance distributions. All 16 had distributions that exhibited greater dispersion than expected by chance, and none had distributions with less dispersion than the chance expectation.
Thesis (Ph. D.)--Massachusetts Institute of Technology, Division of Bioengineering and Environmental Health, 2001.Includes bibliographical references (leaves 186-193).
DepartmentMassachusetts Institute of Technology. Division of Bioengineering and Environmental Health.
Massachusetts Institute of Technology
Division of Bioengineering and Environmental Health.