Publication list via PubMed

I have a support and a research role at the Cancer Research Center. I head the Biostatistical Shared Resource, a unit tasked with providing the statistical support for the Center's research studies. The Resource works with investigators and staff to ensure that the best statistical tools available are applied to the answer research questions. I serve as a co-investigator on many epidemiology and psychosocial cancer projects. In addition, I am interested in methodological research that extends statistical techniques of relevance to our research.

My major research interest is in techniques for studying disease associations when the independent variables are measured with error. This work has particular relevance to the ongoing research efforts of the epidemiology group at the Cancer Research Center, where focus has been on diet as a disease exposure. Measurement error in exposure variables, such as that in dietary intakes, biases the disease-exposure parameter estimates and reduces the power of the study. There are several ways to deal with measurement error. In the presence of random error, using the average of repeated measurements can reduce the error in variables. However, it is generally not feasible to obtain repeated measurements from all participants in large studies, and many variables are subject to systematic as well as random error. Measurement error models provide unbiased parameter estimates that are corrected for the error in the independent variables. Most of these models require additional information about the structure of the measurement error. A substudy is often developed where the error prone variable is measured by multiple techniques. For instance, our multiethnic cohort (MEC) study included a calibration substudy where approximately 260 individuals per sex-ethnic group provided dietary information from three 24-hour dietary recalls and the food frequency questionnaire (FFQ). Assuming that the 24-hour recalls are a more accurate measure of diet, the relationship between the two dietary techniques is used to calibrate the dietary data in the entire cohort, in order to reduce bias in the parameter estimates. Recently, the OPEN study, where data from an FFQ, 24-hour recalls and biomarkers for protein and energy were compared, found that the 24-hour recalls may be a sub-optimal gold standard for comparison, leading to incomplete correction for measurement error. The agreement between the protein measures was enhanced by comparing energy-adjusted values. We also found that for the MEC that agreement was enhanced for energy-adjusted values. I have been studying the interpretation of energy-adjusted nutrients when both the nutrient and calories are measured with error. I am also studying improvements to the measurement correction methods. While few biomarkers of diet exist, we will have the opportunity to study whether incorporation of imperfect biological measurements in the calibration equations improves the correction procedure, as we are now establishing a biorepository for the MEC.

A second area of interest is in ethnic/racial classification. The state of Hawaii has a multiethnic population, and many individuals are of mixed ethnicity. Clues to the etiology of cancers can be found by comparing risks of different ethnic groups. Investigating why the cancer risk profile of one group differs from other groups can add to our understanding of carcinogenesis. However, biased risk estimates result if different racial groupings are used in numerator and denominator data of the ethnic-specific cancer rates. In order to quantify and ultimately minimize this bias, we have investigated the comparability of numerator and denominator data used for Hawaii rates, with particular attention to the patterns of assignment for those of mixed race, and determined the effect on risk estimation of using different classifications. These issues are particularly important given the changes to the race questions in the 2000 Census, where individuals can now indicate multiple categories. We have developed population estimates by sex, ethnic and age group for 1975-2000 for the state of Hawaii that will agree with the classification used for the Hawaii Tumor Registry. Population estimates by county for the year 2000 have also been created.