A total of 134 volunteers (≥65 years), with a Mini-Mental State Examination (MMSE) score > 24, were prospectively involved (between 2016 and 2018) in the Auditory and Dementia study (ANDES) cohort. They were all Chileans, belonging to the Recoleta and Independencia districts from Santiago, Chile, and spoke Spanish as their native language. Thirteen subjects did not complete the audiological evaluation, and two patients had missing data. In addition, following the recommendations of the American Academy of Audiology (2010) guidelines for the evaluation of auditory processing (AP), three subjects that had the best ear pure-tone average (PTA) greater than 50 dB HL were not considered for AP evaluations. The exclusion criteria were previous ear disease, previous use of hearing aids, asymmetrical hearing loss (defined as a difference greater than 15 dB HL in at least two contiguous frequencies), conductive hearing loss (defined as a PTA air-bone gap greater than 10 dB HL), stroke, dementia, and major psychiatric and neurological disorders. All volunteers gave written informed consent in accordance with the Declaration of Helsinki. All procedures were approved by the Ethics Committee of the Clinical Hospital of the University of Chile, permission number: OAIC752/15.

Cognitive performance was assessed by an experienced psychologist in cognitive tests that were blind to auditory evaluations. Instructions were given orally and visually using a desktop computer. The battery included the Trail Making Test Part A (TMT-A) for processing speed and Part B (TMT-B) for executive functions (Arango-Lasprilla et al., 2015), Wechsler Digit Symbol for processing speed (Peña-Casanova et al., 2009), the Boston Naming Test for language in an abbreviated version of 30 items (Kaplan et al., 1978), the Rey-Osterrieth Complex Figure Test for visuospatial abilities (Rey, 1959), the Forward and Backward Digit Span for verbal working memory and attention (Peña-Casanova et al., 2009), and the total recall of the Free and Cued Selective Reminding Test (FCSRT) to explore verbal episodic memory (Grober et al., 1988).

For descriptive analysis, we used the median and interquartile range (IQR) of demographic variables. We also explored gender differences using the Mann-Whitney U test and ear differences for auditory processes using the Wilcoxon rank-sum test.

We performed exploratory factor analyses for hearing loss, auditory brain stem, and cognitive domains, allowing us to build composite scores that represented the main aspects of the auditory processes analyzed. To define how many scores were produced by each of these elements, we used the parallel analysis method (Hayton et al., 2004). Factors were extracted with principal axis factorization rotated using oblimin (Costello and Osborne, 2005), and each factor was submitted to Cronbach’s alpha internal consistency analysis (Cronbach, 1951) using only the variables with loadings greater than 0.3. Composite scores were estimated using the regression method based on exploratory factor analyses results. The following variables were used for each factor analysis:

Once hearing loss, suprathreshold ABR, and cognitive domain scores were defined and estimated, we evaluated the differential contribution of these elements on auditory processes using generalized multiple linear regression models. Each category of AP was explored independently and not combined in a composite score. We chose this theoretical-driven approach rather than a data-driven approach to avoid artifactual results (Costello and Osborne, 2005). As dependent variables, we included the performance of each ear in SIN, dichotic digits, SSW, and FP tests. All our dependent variables, except for the GIN task, were performance percentages with bounded scores from individual answers, which could be right or wrong. These kinds of variables are better modeled using a binomial modeling approach (Crawley, 2013). In the case of obtaining overdispersion, we used a quasibinomial approach. In the case of the GIN task, variables that respond to waiting time events, and are zero bounded, are expected to have gamma distributions (Nobel and Tijms, 2006). For this reason, we modeled the GIN task assuming gamma distribution. As independent variables, we included the composite scores derived from the exploratory factor analyses (EFA), as well as sex, years of education, and age. Since many of these evaluations required not only a proper cochlear function but also comprehension of the task, we included an interaction between cognition and hearing loss factors. Non-significant regressors (including interaction) were removed from the models using a backward method followed by a forward method. Both results and procedures were manually compared. When the solution was not convergent, we kept the solution presenting the highest pseudo-R-squared value. To estimate pseudo-R-squared values, we used a variance function-based method (Zhang, 2020). The p-values were corrected using Bonferroni’s method. A model was significant if the pseudo-R-squared values were bigger than 0.15 (Schober and Schwarte, 2018). All statistical analyses were performed with the R project, with an alpha value of 0.05.

We used two tests for the evaluation of dichotic speech: dichotic digits and SSW. For dichotic digits, 18 and 12% of the variability in performance for the right and left ears correspondently was explained by cognition. These pseudo-R-squared values were near the cut-off value of 0.15 that we used, showing that other variables not included in our study could be important. Previous reports have described a correlation between dichotic digits and cognition. Gates evaluated a group of 313 adults aged between 71 and 96 years (Gates et al., 2010). The cognitive assessment included TMT-A, TMT-B, and working memory. A composite cognitive score, similar to the one we achieved with the EFA, explained 16% of the variance in dichotic digits (Gates et al., 2010). Fischer studied a group of 3,655 adults, aged between 21 and 100 years, with a free recall of three digits. A cognitive evaluation was done using MSSE. Five factors, including age, sex, education, hearing loss, and MMSE, accounted for 22.7% of the variance in dichotic digits (Fischer et al., 2017). In the study by Gates et al. (2010), 12% of the subjects were users of hearing aids, and neuroplasticity could be a non-controlled variable affecting the results. Nevertheless, the variance attributable to cognition had a similar value to our results. The study by Fischer et al. (2017) used a cognitively more demanding challenge (three digits), and this could explain why cognition had a higher load in the model.

Hearing and cognition were significant variables for SSW and explained 17% of the variability in the right ear and 27% of the variability in the left ear. Since SSW results were expressed as values corrected by word discrimination score, it is not surprising that in addition to cognition, hearing loss was also a significant predictor.

Even though we identified significant models for SSW and dichotic digits, these were able to explain less than 30% of the variance in the results. Other variables could be considered in the assessment of dichotic speech in elderly subjects. One of these could be interhemispheric communication. The dichotic listening paradigm makes the asymmetry between cerebral hemispheres evident (Lazard et al., 2012). With the exposure of verbal input to both ears simultaneously, the information from the right ear predominates over the left ear. Kimura described this observation in the early 60s as the right ear advantage (REA) (Kimura, 1967; Jerger and Martin, 2004). The information from the right ear ascends mainly to the contralateral cerebral hemisphere directly to the language center in the left hemisphere. Conversely, the information from the left ear ascends mainly to the right hemisphere and must cross through the corpus callosum to the language center in the left hemisphere. Information from functional and structural studies supports this theory (Hugdahl and Westerhausen, 2016). The REA persists in aged subjects (Jerger et al., 1994; Westerhausen et al., 2015; Cañete et al., 2020), and this can have a clinical relevance since, in some cases, the use of bilateral hearing aids could result in interference rather than improvement in hearing perception. Our results on dichotic digits and SSW confirmed the presence of REA (Table 2).

Two tests explored temporal resolution of the auditory signal: GIN detection and FP. Cognition was a significant variable in the model of FP for both ears, while hearing had no role. Similar results have been published before (Sheft et al., 2015; Murphy et al., 2018). Two additional variables emerged as significant in the models of FP: gender for the right ear and midbrain for the left ear. Gender differences in the performance of FP in adults or elderly subjects have not been reported earlier (Sanchez et al., 2008; Majak et al., 2015). Other studies do not explore gender differences (Murphy et al., 2018). Hearing levels do not account for the difference observed in our participants, since the variable “hearing” had no significance in the model. Further research is needed to clarify this issue.

The second test we used for exploring the temporal processing was GIN detection across the channel. The test was not easy to explain to our subjects, and a percentage of them were not able to execute it (8%). The model including cognition had a pseudo-R of 0.08 and was considered not significant. O’Brien et al. (2021) assessed a group of 213 subjects aged > 50 years and found no correlation between auditory gap detection measured by using the Adaptive Test of Temporal Resolution and cognitive domains. Even though the test we chose to explore GIN discrimination was cognitively challenging, our results did not confirm a correlation between across channel GIN detection and cognition.

Our volunteers were mostly women. An effort was made to control gender bias by including gender as a variable in our models. Inclusion criteria required normal hearing or mild to moderate presbycusis participants that could execute tests for AP, thus our results cannot be extended to severe hearing loss. We selected AP tests that were available in Spanish and previously validated in our population (Fuente and McPherson, 2006; Cañete et al., 2020). With this selection, we compromised other aspects, for example, the SIN test we used was easy for our subjects and had a ceiling effect.