Project Summary / Abstract Traditional methods for estimating exposure-hearing relations, which typically use pure-tone average hearing thresholds to define the outcome, may not comprehensively capture useful information encompassed by the shape of the audiogram across the range of frequencies. Since existing statistical methods are insufficient for analysis of the audiogram shape data, built on our previous work (R01DC017717) on dealing with multi-layers of correlation structures in hearing data, we will develop rigorous and robust analytical methods for estimating exposure-hearing relations when the outcomes are hearing phenotypes characterized by audiogram shapes, either from supervised learning methods with pre-specified phenotype categories or unsupervised learning methods with phenotype categories determined solely by the data at hands. In addition, current analytical methods rely on either pure-tone audiometry data or self-reported hearing data as the outcome, but not both; these two outcome measures are in fact complementary, and we will develop statistical methods for integrating the pure-tone audiometric data and self-reported data. We will apply the hearing data analysis methods to the Conservation of Hearing Study within the Nurses’ Health Study II, the Medical University of South Carolina Hearing Study, and the National Health and Nutrition Examination Survey (NHANES). User-friendly publicly available software development will be a central feature accompanying all new methods to be developed. We have formed an interdisciplinary team of theoretical and applied statisticians, epidemiologists and hearing experts, and we expect to be well equipped to solve the challenging problems that have been identified.