This study used data from a health screening dataset created by Asan Medical Center from March to December 2019. During this period, data for 1,341 subjects aged ≥60 years was obtained. Of the 1,341 participants, 970 cases were included for which brain MRI data could be obtained. Here we excluded cases with download failures (n = 3), analysis errors (n = 7), missing T1 images or images with a thickness other than 1 mm (n = 45), and cases with enhanced images (n = 316) since these were not suitable for FreeSurfer analysis. We also excluded cases in which there was an inter-ear difference in hearing of more than 15 dB (n = 225; Figure 1). Finally, demographic features of patients, including blood lipid levels (i.e., LDL, HDL, triglycerides, and total cholesterol), hypertension (HTN), diabetes (DM), glucose, and HbA1C levels, were examined. In addition, a survey on smoking and alcohol consumption habits was conducted.

MR images were obtained using a 3.0-T MRI scanner (Achieva 3T TX MRI System, Philips HealthCare, Best, Netherlands). T1-weighted imaging was performed in the sagittal orientation with the following parameters: TR/TE = 500/10; FA = 90°; FOV = 240; matrix = 512 × 512; thickness = 1 mm. Cortical surface reconstructions and regional GM volume estimations were performed using FreeSurfer version 7.2.0 (Figure 2).

In this process, raw Digital Imaging and Communications in Medicine (DICOM) files were first converted to NIfTI (Neuroimaging Informatics Technology Initiative) format, followed by registration to the MNI305 template. Initial image preprocessing steps such as bias field correction, skull stripping, and tissue segmentation were applied. The hemispheres were then separated, and white and pial surfaces were reconstructed along intensity gradients to enable accurate cortical modeling. The procedural details of these analyses have been documented in prior publications (11–14). For cortical regions, FreeSurfer uses the Desikan–Killiany atlas to parcellate regions of interest based on cortical gyri (15, 16). In addition, subcortical structures were automatically segmented and labeled, following the methodology presented by Fischl et al. (14). The analyzed regions of interest included the total GM, limbic system, including the hippocampus, thalamus, amygdala, accumbens, as well as the white matter and cortex of the cerebellum for both hemispheres.

Audiological evaluation was performed using a diagnostic audiometer (SA-203; Entomed, Malmö, Sweden) at 0.5, 1, 2, and 4 kHz in a soundproof booth. The average thresholds for each of the four frequencies for each ear were used. Moreover, cases with a history of sudden hearing loss, Meniere's disease, or other inner ear disorders were excluded. We included patients with bilateral symmetric hearing thresholds (within a threshold difference of 15 dB), which is consistent with the definition of ARHL. Patients with no significant abnormalities found in the tympanic membrane examination and no history of otitis media or other related conditions were included in this study. The analysis was performed by categorizing the hearing levels of both ears into three groups: < 20 dB, 20–40 dB, and above 40 dB.

The study protocol for data analysis was approved by the Institutional Review Board (IRB 2020-0646). The studies were conducted in accordance with the local legislation and institutional requirements. Written informed consent for participation was not required from the participants or the participants' legal guardians/next of kin in accordance with the national legislation and institutional requirements.

We used Kruskal–Wallis tests and Fisher's exact tests to compare continuous- and categorically-measured demographic characteristics among the subgroups, respectively. Next, we adopted an approach employed in a previous study such that each specific regional volume was adjusted by dividing it by the estimated total intracranial volume then multiplying this result by the average estimated total intracranial volume for all subjects (to accommodate variation in head size among individuals) (17). The resulting estimates were represented in cubic millimeters (mm³).

To ensure the statistical robustness of our findings, a priori power analysis was conservatively performed assuming a small effect size (Cohen's f² = 0.10), a stringent significance level (α = 0.01), and a high power threshold (1–β = 0.99). Under these criteria, the minimum required sample size for a multiple regression model with six predictors (Age, Sex, HTN, DM, Hyperlipidemia, Hearing group) was estimated to be 203. Since the total sample size in this study far exceeded this threshold, the analysis was sufficiently powered even under these conservative conditions.

Multivariable linear regression analyses were used to confirm the association between hearing impairment and volumetric changes in the brain. To control for potential confounding factors, covariates including age (continuous), sex (binary), and the presence of HTN, DM, and hyperlipidemia (all coded as binary variables: 0 = absence, 1 = presence) were included in the model. All variables were entered using stepwise method. To assess multicollinearity among the independent variables, the independence of residuals was evaluated using the Durbin-Watson statistics. In all models, tolerance values were >0.1, variance inflation factors were below 5, and condition indices were < 15, indicating no substantial multicollinearity. Residual normality was assessed using the Shapiro–Wilk test and Q–Q plots. The results indicated no significant deviation from normality (p > 0.05). For all analyses, a p-value < 0.05 was considered to be statistically significant. All statistical analyses were performed using SPSS version 29.0 (IBM SPSS Inc.).

The baseline characteristics of study participants according to hearing level are shown in Supplementary Table 1. A total of 745 patients were included. Our sample showed a median age of 65 years and 405 (54.36%) of all subjects were male. When divided into three groups based on hearing level, we observed that the severity of hearing loss was associated with older patients, as well as a higher proportion of males (p < 0.05). When examining the relationship between average PTA and patient age via linear regression, it was observed that as age increased, the hearing threshold significantly increased (B = 0.94, r² = 0.22, p < 0.000; Figure 3).

Patients in moderate-to-severe hearing loss showed significantly higher prevalence in HTN, DM, and showed higher level of glucose and HbA1c than patients in normal hearing and mild hearing loss (Supplementary Table 1). Contrary to our expectations, the total cholesterol and LDL levels were significantly lower in the worse hearing group, despite their higher average age. There was a slight difference in smoking history, with a higher percentage of current smokers in the moderate-to-severe hearing loss group, but this difference was not statistically significant.

Next, we compared cerebellar white matter, cerebellar cortex, limbic system components, and total GM volume among the three subgroups (Supplementary Table 2, Figure 4). The more severe the degree of hearing loss, the greater the tendency for most areas to show a significant decrease in volume (p < 0.01). There was a significant decrease in the volume of the bilateral cerebellar white matter, cerebellum cortex, thalamus, hippocampus, amygdala, accumbens, and total GM in the moderate-to-severe hearing loss group compared to normal hearing. However, there were no significant differences among groups with respect to the size of the caudate and pallidum.

We examine the relationship between hearing loss severity and brain volume through multiple regression analysis. After adjusting for factors such as age, sex, and medical history of HTN, DM, and hyperlipidemia, multiple regression analysis was conducted (Table 1). In the areas analyzed, cortical volume significantly decreased with age (p < 0.001). A significant volume reduction in both hippocampi was observed in the moderate-to-severe hearing loss group (p < 0.05), although there was no significant difference between the normal hearing and mild hearing loss group. Total GM volume reduction was observed in the mild hearing loss group, and this trend was even more pronounced in the moderate-to-severe hearing loss group. However, no meaningful volume changes were observed in the white matter of the cerebellum, the cerebellar cortex, and other areas of the limbic system (i.e., thalamus, amygdala and accumbens) among the groups.

Multiple linear regression analyses were performed separately for 13 cortical regions, with hearing groups as the main predictor. To correct multiple comparisons across the 13 regions, the significance of the hearing group factor was assessed using Type III sum of squares within the general linear model framework. The resulting p-values were adjusted using the false discovery rate (FDR) method based on the Benjamini–Hochberg procedure. Although uncorrected p-values ranged from 0.011 to 0.902, none reached the predefined significance threshold after FDR correction (Supplementary Table 3).