PhD

dissertationThis dissertation describes the application of a new analytical technique to the epidemiologial analysis of observations on large, computerized genealogies. Statistical comparison of distributions of kinship (genetic distances) among cancer cases and controls is made in contrast to the classical methodologies of genetic epidemiology that rely upon comparison of incidences of a disease under investigation among cases and controls. This research represents the first analysis of a population completely ascertained for human cancers. Kinship analysis of a population completely ascertained for a disease (the inclusion of both high and low incidence genealogies) introduces new methodological difficulties since the classical techniques of genetic epidemiology are intended for the analysis of only partially ascertained populations. Several of these difficulties are of statistical import, foremost among them a lack of statistical independence among the related pairs of individuals that comprise the kinship distributions. The lack of independence among kinship observations members of a genealogy complicates the use of traditional statistical tools that rely on independence. A novel analytic approach to epidemi1ogical observations designed and developed to meet the demands posed by the unusual nature of the data and the selected strategy of analysis is described. A new and potentially powerful analytical feature of large computerized genealogies is described in reference to the problem of ascertainment bias: the appearance of seemingly nonrandom configurations of common traits in large, intensively screened populations by purely random means. The probability of observing random configurations that mimic nonrandom mechanisms of disease etiology increases with both population size and disease frequency. The influence of ascertainment biases upon uncorrected epidemiolgical studies of common diseases in populations may not be fully appreciated by current researchers who lack the means to adjust for ascertainment. Large computerized genealogies offer the potential of estimating an ascertainment density function for a set of observations on the genealogy to any desired degree of accuracy through reiterative sampling of matched controls. A statistic describing the degree of familiality of a trait and incorporating a population based ascertainment correction is introduced in the form of an empirical significance index and applied to a survey of cancer among the Utah Mormons. A comparative ranking by degree of familiality of major cancer sites in the Utah Mormon population is offered. Prostate cancers are found to be more familial by far than any other major site, followed by breast, colorectal and other reproductive and digestive cancer sites. An overall pattern of excess kinship among major cancer sites in comparison to the kinship expected under an hypothesis of random distribution of the disease with respect to genetic distances is reported. A survey of the tendencies of pairs of cancer sites to occur together in families is also presented: by and large the cancers that have the greatest tendency to aggregate in families also coaggregate to a similar excessive extent. This result is interpreted to suggest that cancer syndrome families are more frequent in the Utah Mormons than are single cancer fami lies. A critique of epidemiological techniques and interpretation of the reported results in terms of current concepts of cancer genetics is offered with reference to problems of ascertainment biases