Survival analysis of familial data: A study of the aggregation of length of life

Epidemiology is being increasingly faced with the problem of correlated data (e.g., familial data, clusters of individuals exposed to a risk factor). Such data requires analysis by random effects models. Recent theoretical advances have provided models for the inclusion of random effects into semi-parametric survival models. Using these developments, this research (1) tested survival models specific for dependent data, (2) described mortality patterns in an Old Order Amish genealogy, and (3) tested for familial aggregation of length of life in the Old Order Amish genealogy.; Three different survival models based on the conventional Cox proportional hazard model, but allowing for clustered data, were tested using simulation techniques. The models were compared to the standard Cox model for robustness of the estimated regression coefficient and Type I error levels. The Cox model and two models for clustered data demonstrated consistant estimation of the regression coefficient. While the Cox model showed elevated Type I errors, the Type I errors were appropriate when analyzed by two of the random effects models. Analysis of these simulated data by a third model was impeded by the simulation conditions.; An Old Order Amish genealogy was selected to test for aggregation of length of life. This demographic study focused on factors which affect length of life independent of family membership. Gender, year of birth and inbreeding level were found to significantly affect life span in this population.; The random effects survival models were applied to selected sibships from the Old Order Amish genealogy to test for aggregation of length of life after adjustment for these covariates. There was significant clustering in age of death after univariate adjustment, accounting for a 10-20% increase in hazard. There was evidence that the clustering may be due, in part, to temporal patterns of mortality not fully addressed by including year of birth in the model. The development of random effects survival models and the unique characteristics of this population combine to provide an excellent opportunity to determine the level of familial aggregation of length of life.