This research is for the development of new approaches to the analysis of data from large cohort studies, either epidemiologic or clinical trials, with many qualitatively different variables observed over several time points, and with many possibly correlated outcomes of interest.
The aim i s to develop methods for discovering patterns of higher order variable interactions that suggest unusually high risk for some outcomes, that are not readily found by more traditional methods. The methods proposed are an attempt to develop tools that merge so-called data mining approaches with more traditional biostatistical methods, and which have the ability to generate hypotheses which can then be further examined by classical parametric statistical methods, or by modern multivariate semi-parametric model building methods, such as Smoothing Spline Analysis of Variance, (SS-ANOVA), which have been developed under this research program and elsewhere. An additional goal is to incorporate family structure information for a subset of study participants in parallel with the search for high order interactions among variables to uncover patterns that may be related to family structure, and to examine the tradeoff between family related and other information in predicting, or estimating the probability of various outcomes. Data from the Wisconsin Epidemiological Study of Diabetic Retinopathy and the Beaver Dam Eye Study will be used to examine the models under study for their reasonableness and for their ability to answer questions meaningful to the study scientists. The results will have broad applicability to other large epidemiological studies as well as to clinical trials.
Epidemiological and clinical studies have much responsibility for the dramatic improvement in public health and longevity in the last fifty years or so. Better understanting of the effect of lifestyle factors, treatment opportunities, and genetic factors have come about as the result of straightforward as well as sophisticated analysis of the data gleaned from these studies. With extensive data collection and complex data structures, as well as improved computational and software resources, there are opportunities to further develop and extend modern data analysis methods to better capture complex relations between variables that affect outcomes of important personal and public health interest. It is proposed to exploit these opportunities.
Showing the most recent 10 out of 14 publications
