Existing statistical methods are insufficient for quality control or analysis of audiometrically-assessed hearing measurements. Statistical analysis of audiometrically-assessed hearing measurements is challenging due to the complex correlation structure of the data, and because the definition of hearing loss is typically based on the pure tone average which is typically the average of measurements at three or four specified frequencies, leading to a potential loss of useful information available in the individual frequency data. In addition, it is now possible to perform audiometric testing outside of the clinic with inexpensive electronic equipment in large- scale epidemiologic hearing studies. In doing so, statistical analysis must account for measurement error in the hearing tests to prevent bias in the estimates of associations and causal effects. We will develop entirely novel methods for quality control of hearing data, and for validly and efficiently assessing the exposure-hearing loss associations and their causal relationships, while accounting for the multiple layers of correlation and multivariate outcomes from multiple frequencies in audiometrically-assessed hearing data. We will develop methods to correct for measurement error-induced bias in the estimated associations and causal effects for studies where hearing outcomes are measured in non-clinical settings. We will apply the hearing data analysis methods to the Conservation of Hearing Study based in the Nurses? Health Study II. 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 expert theoretical and applied statisticians, epidemiologists and audiologists, and we expect to be well equipped to solve the challenging problems that have been identified.
Existing statistical methods are insufficient for quality control or analysis of audiometrically-assessed hearing measurements; we will develop entirely novel analytical methods for quality control of hearing data, and validly and efficiently quantifying not only the exposure-hearing loss associations but also their causal relationships, while handling the various layers of correlation and multivariate outcomes arising from the measurement of multiple frequencies in audiometrically-assessed hearing data. In addition, we will correct for measurement error-induced bias in the estimates of the associations and causal effects in studies where hearing outcomes are evaluated in non-clinical settings. We will apply the hearing data analysis methods to the Conservation of Hearing Study based within the Nurses' Health Study II, and user-friendly software developed to implement all new methods will be made publicly available.
