We identified individuals with pituitary tumors from the UCR data collection who linked to genealogy data in the UPDB for this analysis. The use of these data resources for this study was approved by the University of Utah Institutional Review Board and by the Utah Resource for Genetic and Epidemiology Research.
To study the familial clustering of pituitary tumors, we used two different methods. First, we compared the average relatedness of all patients with pituitary tumors to the expected relatedness in this population using the genealogical index of familiality (GIF). We also estimated the relative risk (RR) for pituitary tumors in the relatives of patients with diagnosed pituitary tumors.
Genealogical index of familiality
The GIF statistic was designed to study familial aggregation of cancer within the Utah genealogy [29
] and has been used in previous studies of familiality [1
]. An analysis similar to the GIF statistic has been used with the extended Iceland genealogies [3
]. An advantage to the GIF statistic over other methods is that it takes into account all genetic relationships between all cases. The Malécot coefficient of kinship was used to measure the relatedness between all possible pairs of diagnosed pituitary tumor cases. The coefficient of kinship is the probability that randomly selected homologous genes from two individuals are identical by descent from a common ancestor [38
]. The coefficient for parent–offspring pairs is ½, for grandparent/grandchild or sibling pairs is (½)2
, for avuncular pairs is (½)3
, and for first-cousin pairs is (½)4
, and so forth. The case GIF is calculated as the mean of all coefficients of kinship between all possible pairs of cases. The case GIF is multiplied by 105
for ease of presentation.
We tested the hypothesis that there was no excess relatedness among pituitary tumor cases by comparison of the case average relatedness to the expected relatedness as observed for 1,000 sets of cohort-matched controls. For each case, a control individual was selected at random from the genealogy resource, matched on sex, 5-year birth cohort, and place of birth (in or out of Utah), resulting in a control set of the same size as the case set. The matching strategy is employed to account for potential differences in kinship based on matching characteristics. One thousand independent control sets were selected, and the GIF was measured for each set. The hypothesis of no excess relatedness of the pituitary tumor cases was tested empirically by comparing the case GIF to the distribution of the 1,000 control GIFs.
The degree of shared genetic composition between pairs of cases representing different genetic distances can be quantified with GIF analysis. When an excess of close genetic relationships is observed for cases compared with controls, it is difficult to identify whether the excess familiality is due to shared environment or shared genetic composition, or a combination of both. If, however, close relationships are ignored and a significant excess of relationships for cases compared with controls is observed, then this observed excess familiality among cases strongly supports a genetic contribution.
We compared contributions to the GIF statistic for cases and controls across close and distant relationships, measured by genetic distance between pairs of individuals. In the GIF analysis, genetic distance is approximated by path length between individuals in a pair. For example, a parent and a child are assigned a genetic distance of one, siblings are assigned a genetic distance of two, an aunt and a niece are assigned a genetic distance of three, and so forth. The empirical significance of the GIF test tells us whether overall excess familiality is observed. We also performed this same test ignoring all close relationships (genetic distance <4) to determine whether the excess familiality is also significant when ignoring all close relationships. We call this statistic the Distant GIF and also tested it empirically.
Estimation of relative risks in relatives is an alternative approach to testing the hypothesis of a genetic contribution to disease. Whereas the GIF analysis uses all relationships between all cases regardless of genetic distance, the relative risk analysis typically relies on comparisons in close relatives only. The relative risk approach compares the observed rate of disease (in this case, pituitary tumor) in relatives of probands (cases) with the expected rates of disease in relatives. All individuals in the UPDB with genealogy were used to estimate cohort-specific expected pituitary tumor rates in the UPDB.
We estimated relative risk as follows. All 2.5 million individuals in the UPDB with data linking to the original Utah genealogy were assigned to one of 132 cohorts based on birthplace (in or out of Utah), sex, and 5-year birth cohorts. For each cohort, internal cohort-specific pituitary tumor rates were calculated by summing the number of individuals in each cohort with pituitary tumor, and dividing by the number of individuals in the cohort. The expected number of pituitary tumors among first-degree relatives of patients with pituitary tumor was calculated by multiplying the number of first-degree relatives of patients with pituitary tumor (in each cohort) by the cohort-specific internal rate of pituitary tumors, and then summing over all cohorts.
The first-degree relative risk is estimated as the ratio of the number of observed pituitary tumor cases among first-degree relatives of patients with pituitary tumors to the number of expected pituitary tumor cases among first-degree relatives. The relative risk was similarly estimated for second- and third-degree relatives. The relative risk is assumed to follow a Poisson distribution, with the mean value equal to the number of expected pituitary tumor cases among relatives of probands. The Poisson distribution is an approximation to a sum of multiple binomial distributions, representing the number of expected tumors in each cohort. This Poisson approximation is appropriate for both rare and common phenotypes, being more conservative for common diseases. Probability values for one-sided tests of significance and 95% CIs for the relative risk statistic can be calculated for the Poisson distribution under the null hypothesis that the relative risk is equal to unity. While significantly elevated risks in first-degree relatives are suggestive of a genetic contribution to disease, they may also result from shared environment. Significantly elevated risks for second- or third-degree relatives, however, are strongly suggestive that a heritable component also exists.