We published methods showing how to evaluate agreement between two exposure measures when data are categorized into percentile groupings, such as quartiles. Such analyses are common in nutritional epidemiology and other fields where precise exposure measurements are difficult. Standard methods of analysis are inappropriate. We showed that inference on variance components based on the Wald statistic can be misleading, and illustrated this point with data on hormone assay reliability. We published data showing that case-control designs to detect interactions between environmental exposures and genetic factors require very large sample sizes, particularly when genes or exposures are rare or when exposures are subject to measurement error. The efficiency of the community intervention design used to study interventions to promote smoking cessation was improved by matching the communities on demographic factors. The efficiency of chemoprevention trials can increase, in principle, by using genetic markers, particularly if such markers can be used to define subpopulations that benefit especially from the intervention. Computer algorithms were developed to evaluate cancer rates and trends to facilitate graphical displays, smoothing methods, and age-period-cohort analyses. A computer program has been developed to enable one to select a parsimonious subset of dietary questions to estimate nutrient intake from questionnaire data. A cluster-analysis demonstrated that groupings of lymphocytes based on binding profiles corresponded well to classifications based on other functional assays. A random effects model was developed to perform a meta-analysis on cohorts of underground miners exposed to radon. We presented analyses to refute claims based on ecologic studies that radon exposure is not associated with lung cancer risk. We published an interview with Nathan Mantel that explored his many contributions to methods for analyzing epidemiologic data.
Showing the most recent 10 out of 16 publications
