An Exact Test for Overdispersion in Screening Mammography Assessments

点击翻译按钮获取中文摘要

DESCRIPTION (provided by applicant): Early detection tests for breast cancer save many thousands of lives each year, and many more lives could be saved if even more women and their health care providers took advantage of these tests. Early detection of breast cancer depends in part on the accuracy of screening mammogram interpretation. The accuracy of human interpreters in detecting the presence or absence of breast cancer can be evaluated by estimating indices like sensitivity and specificity. Efficient and precise estimation of both of these accuracy parameters have important consequences for public health planning and for patients. When multiple observations and statistical modeling approaches are employed for assessing the accuracy indices, the standard assumption of the binomial distribution for the data needs confirmation. Several tests based on large sample theory are widely available. However, the data generated in biomedical studies frequently involve small samples, and thus the accuracy of the asymptotic approach for analyzing such data is in question. The goal of this research proposal is to develop a method of exact testing for overdispersion in small samples. The project will involve both theoretical and empirical work, drawing on existing data sources and collaborative opportunities provided through the investigator's collaboration at H. Lee Moffitt Cancer Center & Research Institute. The specific aims for this proposal are described as: Specific Aim 1: Develop an exact test for overdispersion; Specific Aim 2: Develop a computational algorithm and release a publicly available implementation; Specific Aim 3: Apply and evaluate the proposed method using a real life dataset.