Statistical methodology is being applied and developed for longitudinal studies and other studies of aging. The research program focuses on several types of statistical models: 1) longitudinal mixed-effects regressi models which consider both within- and between-subject variation in analyzing the repeated measurements for all individuals in the study population, 2) survival analysis for studying risk factors in prospective studies, 3) multiple comparisons for testing group differences in experimental or observational designs, 4) mixture models for describing age changes in distributions of biological markers, and 5) experimental design. Other techniques used include Bayesian, maximum likelihood and numerical computing methods. A major emphasis of the research program is the development of methods which yield cogent yet easily understood results when applied to data. The impact of the normality assumption for random effects on their estimates in the linear mixed-effects model has been investigated by allowing for differing numbers of subpopulations to describe the random effects. It is shown that if the distribution of random effects is a finit mixture of normal distributions, then the random effects may be badly estimated if normality is assumed, and the current methods for inspecting the appropriateness of the model assumptions are not sound. Further, it is argued that a better way to detect the components of the mixture is to build this assumption in the model and then """"""""compare"""""""" the fitted model with the Guassian model. This so-called mixture model approach has been illustrated using data from the prostate study. The resulting model provides a classification of patients into one of four diagnostic groups (normal, BPH, local regional cancer, and metastatic cancer) with 81% of patients correctly classified. In addition, mixed-effects regression model have been used to examine measurement error bias and its relationship with regression dilution bias in risk factor studies. Results from a simulation study show that baseline risk factor values represented by mixed-effects predicted values provide a more accurate measure of the association between a follow-up study's outcome and a risk factor value. This research develops and applies new approaches toward the use of longitudinal data in epidemiological and biomedical studies of aging and associated disease states.
