To determine the degree that concordance occurred for a particular organ or for other clinical outcomes, we computed к statistics for affected sibling pairs using a technique developed by Fleiss. While this technique was initially developed for measuring agreement between raters, it has also been applied to sibling analysis in multiple sclerosis and rheumatoid arthritis. Each variable tested for concordance was dichotomized by the presence or absence of the condition. Proportions of siblings with the condition present were computed to determine if the concordance differed from what would be expected by chance. The resulting к statistic ranged from + 1 (indicating perfect concordance) to – 1 (indicating perfect discordance). A separate analysis was performed for full siblings only.
Because к is unreliable in cases of extreme lack of variation in the distribution of a response variable, concordance results were not computed for organ systems in which there was almost universal involvement (97.6% of the affected siblings had lung involved) or rare involvement. Fewer than 10% of the affected siblings had neurologic, cardiac, spleen, calcium metabolism, renal, parotid/salivary gland, muscle, or bone marrow involved. Observed and expected numbers of sibling pairs concordant or discordant for a specific phenotype (or clinical outcome) were computed. Observed and expected numbers counted all permutations of sibling pairs from each family. For continuous variables, intraclass correlations coefficients (ICCs) were used as measures of sibling concordance. www.medicine-against-diabetes.net
ORs and 95% CIs from general estimating equation logistic regression models were used to estimate the risk of the condition in the sibling with a later date of diagnosis. The presence of the condition in the sibling with an earlier date of diagnosis was used as the predictor in the models. All permutations of sibling pairs within a family were included in the analysis. Each general estimating equation model was adjusted for full-sibling and half-sibling pair status and accounted for the clustering of observations within a family (tendency of members within the same family to be more alike than members between two different families). The primary study hypothesis in the SAGA study concerned detecting one or more disease predisposing genetic loci in a full genome scan, and the study sample size was configured for that purpose. The present phenotype analysis deals with secondary hypotheses and was considered exploratory in nature. Therefore, no formal corrections of p values for multiple comparisons were made. Results were considered significant if the p < 0.05 level was attained.