Statistical Methods for Medical Studies

Prentice, Ross

Abstract

The aim is the development and evaluation of improved statistical methods for a range of medical and biomedical studies. The proposal incorporates ongoing research efforts in the development of statistical methods for epidemiologic cohort and case-control studies, and in the development of statistical methods for therapeutic studies, along with new research emphasis on disease prevention/risk factor intervention trials and on certain types of animal studies, to form a program project proposal. Specific research topics in the disease prevention/risk factor intervention area include the use of a case-cohort design to avoid expensive assembly of covariate histories on the entire trial cohort, the development of data analysis procedures to accommodate correlated endpoint data as may arise, for example, if subjects are randomized in groups, and the development of designs and procedures to avoid control group contamination. Topics in the analysis of epidemiologic studies will include the further development of relative risk regression and odds ratio regression methods with emphasis on such sub-topics as linear relative risk estimation, interaction testing with sparse data, and evaluation of the impact of covariate measurement errors. Methodologic issues in genetic epidemiology and in the analysis of ordered categorical data will also be considered. The work directed toward therapeutic trials will emphasize study design issues, including randomization strategies, and sequential designs; exploratory data analysis issues, including diagnostics, graphical displays and median survival curve estimation; and data analytic issues, including study of the use of multistate and multivariate models. The statistical methods for animal carcinogenicity testing project will examine the use of parametric models to compare tumor incidence rates based on death and sacrifice data, will evaluate testing procedures currently in use and will propose alternate procedures that may, for example, incorporate historical control information.