Analytical methods for evaluating reliability and validity of mobile audiometry tools.

Abstract

Statistical approaches that could be used as standardized methodology for evaluating reliability and validity of data obtained using remote audiometry are proposed. Using data from the Nurses' Health Study II (n = 31), the approaches to evaluate the reliability and validity of hearing threshold measurements obtained by a self-administered iPhone-based hearing assessment application (Decibel Therapeutics, Inc., Boston, MA) compared with measurements obtained by clinical (soundbooth) audiometry are described. These approaches use mixed-effects models to account for multilevel correlations, intraclass correlation coefficients (ICCs) of single and averaged measurements, and regression techniques with the generalized estimating equations (GEEs) to account for between-ear correlations. Threshold measurements obtained using the iPhone application were moderately reliable. The reliability was improved substantially by averaging repeated measurements; good reliability was achieved by averaging three repeated measurements. In the linear regression analyses that assessed validity, the range of intercepts (2.3-8.4) and range of slopes (0.4-0.7) indicated that the measurements from the application were likely biased from those obtained by clinical audiometry. When evaluating alternative hearing assessment tools, it is recommended to assess reliability through mixed-effects models and use ICCs to determine the number of repeated assessments needed to achieve satisfactory reliability. When evaluating validity, GEE methods are recommended to estimate regression coefficients.