Synergy among data analysis techniques: An analysis of episodic memory function.

摘要

The purpose of this dissertation was to investigate whether stronger conclusions about the associations among variables predictive of memory decline could be drawn by combining the results of multiple analytic techniques, each providing a different perspective on the data. Substantial decline in episodic memory, even over short intervals, is hypothesized to be a marker for memory difficulty as other cognitive systems (associational memory, processing speed) become less robust with age.;Data from the Vietnam Era Twin Study of Aging were used. Many other phenotypes were collected (including neuropsychological tests, physiological data, and psycho-social information) on a sample of monozygotic and dizygotic males who served in the military during the Vietnam era. Episodic memory was assessed using standard neuropsychological tests: the Wechsler Memory Scale (WMS) Logical Memory, and the WMS Visual Reproduction. Both tests included immediate and delayed recall conditions.;Several exploratory and confirmatory statistical techniques were applied to the data. Correlation matrices and regression analyses demonstrated that there was considerable inter-subject variability in episodic memory, initial and delayed recall performance. The recall performance of the vast majority of subjects declined significantly during the test-retest interval, but subjects declined at different rates, irrespective of their relative initial performance levels. Regression analysis demonstrated strong relationships between episodic memory and other variables, particularly general cognitive ability. Multivariate models represented as directed acyclic graphs differed substantially depending on a priori assumptions regarding causality; however adding or subtracting variables from any given model did not affect the results as significantly as the causality assumption. Biometrical modeling demonstrated a close correspondence between the genes responsible for immediate and delayed recall. Finally, the use of Structural Equation models demonstrated that good fitting latent variable models were achievable and less sensitive to extraneous influences than observed variable models. Although each modeling approach was found to have limitations, exploratory data analysis may proceed efficiently by using several models in combination. Therefore, using neuropsychological tests and other demographic information, a phenotype predictive of poor memory function may be identified. This phenotype would be characterized by life-long memory problems, lower-than-average cognitive ability, and a genetic predisposition, as well.