This project aims to develop methods for analyzing complex interactions in longitudinal data collected from biomedical studies that contain a strong social and behavioral component. Of special interest are data collected on elderly populations, within which interaction of various forms-between and within multiple risk factors, co-morbidities, concurrent interventions, and with time-are commonplace. Added to this complexity is the potential of this interaction to affect a single or multiple outcomes. A starting point to operationalize the study of interaction concerns the effect of a factor (e.g., intervention) on the variation of an outcome over time, or the so-called """"""""intervention by time"""""""" interaction. For clinical trials, the most important research question often concerns an """"""""intervention by time"""""""" interaction. A statistically significant intervention by time interaction would signal that the average response patterns of the intervention group and the control group are different over time, and thus would point to a significant effect of the intervention. Epidemiological studies that compare 2 groups of subjects by a risk condition such as a psychosocial stressor of depression, also hinge on the investigation of the """"""""risk condition by time"""""""" interaction. Motivated by the ubiquity of rather complex interaction effects of risk factors in the study of elderly populations and the lack of specialized methods for analyzing such interactions, this proposal focus on the further development of methods that go beyond analyzing the simple and often linear """"""""intervention by time"""""""" interaction effects.
The specific aims of the proposal include the further development and extension of a of 2 relatively new methods--hidden Markov models for analyzing multiple outcomes and transition between health states, and a broad survival marginal models framework for analyzing longitudinal multivariate outcome data The first subproject is especially relevant to the modeling of age-related disablement, while the second is suitable for analyzing mixed modes of multivariate outcomes, including recurrent-recurrent events data. Correspondingly 3 significant data sets will be analyzed using these methods. These 3 data sets are collected, respectively, from the Health, Aging, Body Composition (Health ABC), the Cardiovascular Health Study (CHS), and Lifestyle Intervention and Independence for Elders (LIFE). A model integrating the hidden Markov model and survival marginal model will also be explored. Another specific aim of the project is to disseminate research results and user-friendly computer programs on a common platform. The proposed interdisciplinary program of study is also designed to promote cross-fertilization of scientific ideas and solutions among team members-both through method development efforts, and across applications--and to fully realize the quality and scientific power of longitudinal data collected in the biomedical and social sciences.
Complex interactions between multiple clinical risk conditions, psychosocial stressors, co-morbidities, and concurrent medical treatments over time are commonplace in elderly populations. Using an interdisciplinary team of clinical, social and methodology scientists, this project develops a selection of advanced specialized analytic tools that could help clinicians better understand the mechanisms of how intertwining factors contribute to human health and well-being, especially that of the elderly.
