Survival Analysis. 1) I am interested in the problem of analyzing relative survival, i.e., survival relative to a standard population, subsequent to a diagnosis, using the Life Table Probability function. This extends mere comparison with "expected survival" to a regression framework, with particular interest in the dependence on age, sex, and calendar year. 2) I am using the results of this modeling approach, along with modeling incidence of a disease (as a function of age, sex, and calendar year), to create a model for prevalence of the disease, not in Olmsted County, but in an artificial closed system where immigration and emigration do not exist. Such a prevalence model would then be more generalizable to a large population that is less affected by immigration/emigration. 3) I am developing methods to represent the results of standard survival models (e.g., Cox models or Kaplan-Meier) in terms of ROC curves where the "disease state" to be predicted is "dead within 5 years," for example.

Systolic Blood Pressure in the Elderly. I am developing methods for analyzing both cross-sectional and longitudinal data on blood pressure control and medication usage, in order to understand the age-related decline in blood pressure control. 1) For the cross-sectional data, I am thinking of these data as the result of a game between medicine and nature. The goal of medicine is to get to target blood pressure, while the goal of nature is to keep the blood pressure out of control. This leads to use of survival methodology but applied to intensity of drug use as the "time" variable. 2) For the longitudinal data, each encounter of the patient with the doctor is regarded as an experiment in measuring the practice patterns, i.e., the propensity to increase the medication regimen. We then model this propensity using logistic modeling controlling for patient age, sex, and also current BP levels, and current intensity.

SNP Array Analysis. I am working with Dr. Steve Turner and his colleagues, as well as the Mayo genetics group, in the analysis of SNP arrays of 100 K and, soon, 500 K. I am especially interested in the use of principle components on the allele count variables to reduce the dimensionality of the X-space. I am interested in the comparison between this principle components approach and a genetic haplotype model approach that is implemented here, but is hard to scale up to the huge dimension involved here. Optimal divisions of the array into analysis chunks, missing data, and many other interesting analytic issues arise.