Development of audiometric parameters throughout the lifespan. II: Relationships between parameters

Aging is accompanied by a gradual decline of hearing functions. Manifestations of this physiological process become significant after ca. 60 years of age (presbycusis), however, they may be observed already at the age of 30 (Borchgrevink et al., 2005; Helfer et al., 2020; Paping et al., 2023). The most common symptom, which is usually clinically examined, is a progressive elevation of the pure-tone hearing thresholds (Gates and Cooper, 1991), corresponding to deterioration of the auditory periphery. During the years, it has become apparent that presbycusis, or more generally age-related change of hearing functions, has also a central component which may be to some extent independent on the peripheral function (Humes et al., 2012; Mazelová et al., 2003; Profant et al., 2019; Sanchez-Lopez et al., 2020). This central component is usually more difficult to detect, partly because it may encompass several different factors (in particular, alterations in the supracochlear parts of the auditory system, and cognitive aspects), partly because its manifestations may be mixed with, or even masked by the influence of the peripheral decline. It has been proposed that impairments of the central auditory functions affect particularly processing of temporal acoustic features (Grose and Mamo, 2010), as well as perception of auditory patterns (Musiek and Pinheiro, 1987) and, most notably, speech comprehension (Pichora-Fuller and Singh, 2006).

To characterize the age-related decline in auditory functions, we studied a large pool of volunteers with age-appropriate normal hearing aged from 18 to 87 years using a complex set of auditory tests, including tone detection thresholds under various conditions, detection of gaps in noise, detection of temporal modulation, binaural interactions, perception of auditory patterns, and intelligibility of noisy and interrupted speech. One of the aims of the study was to determine how the mean values and variability of the test outputs change during the lifespan, also in dependence on sex; these data are provided in our companion paper (Čapková et al., 2025), in which the groups means and variability of the acquired auditory parameters are presented in 10-year age categories and also as regression fits with prediction bands. However, because the individual auditory parameters are known to be interrelated, the current paper takes the next step, aiming to identify possible relationships between the measured variables and to explore if and how these relationships change with age or hearing status.

The acquired psychophysical variables form a multidimensional space. While some of the variables might provide similar information, others may be independent. In order to explore this field, we perform basic correlational analysis among the variables. Such analysis proved useful in our several recent works (Bureš et al., 2019; Profant et al., 2017; Sommerhalder et al., 2023) and was also used in previous attempts to put different auditory measures into relation (Divenyi and Haupt, 1997a; Glasberg and Moore, 1989; Strelcyk and Dau, 2009; Summers et al., 2013; Van Esch and Dreschler, 2015). However, simple correlations do not reflect the dimensionality of the problem; for this reason, we cluster the variables within the multidimensional space using a method based on principal component analysis. A similar approach has been recently tested, for example, in the works by Sanchez-Lopez and colleagues (Sanchez Lopez et al., 2018).

An important issue that we also take into account is the relation between the elementary perceptual measures (such as mean hearing threshold, detection threshold of a tone in noise, or sensitivity to short acoustic events and modulation) and speech comprehension. The results of previous works suggested that elementary psychophysical measures might serve as a predictor of understanding speech, particularly in noisy conditions (Füllgrabe et al., 2014; Glasberg and Moore, 1989; Houtgast and Festen, 2008; Mazelová et al., 2003). As speech understanding is an important factor influencing human quality of life, we analyze how well the measured variables predict speech intelligibility in our data sample. We focus on two types of speech distortion: sentences embedded in a multitalker babble noise, and utterances that are periodically interrupted and thus the conveyed information is incomplete. The predictions of speech understanding are calculated using two different methods: 1) statistically oriented exploration, which assesses the relative importance of each variable in predicting the target variable, and 2) gradient boosting, which iteratively constructs decision trees and minimizes the Mean Square Error function.

The above mentioned analyses are performed separately in several specific subject subgroups, for example in aged subjects with excellent hearing thresholds, in subjects with poor tone-detection thresholds and concurrent good speech comprehension ability, etc. We hypothesize that different subgroups of subjects will exhibit different relationships between the variables, and that their speech comprehension ability will rely on different audiometric parameters. By using several methods of exploring relationships among audiometric variables, we aimed to provide a more complex view on the topic: while classical statistical methods are well known, their sole utilization might result in misleading interpretation of the subject matter; it is thus appropriate to verify the results also using multidimensional non-linear methods.

Comments (0)

No login
gif