Introduction

In the aging population, hearing loss is a major concern1. The importance of hearing loss has been increasing as it is considered to be a critical modifiable risk factor for dementia2. The Lancet Commission stated that hearing loss is the leading factor responsible for dementia in middle age3. Therefore, identifying risk factors of hearing loss is important for the early detection and rehabilitation of hearing loss.

The apolipoprotein E (APOE) ε4 allele is the most important genetic risk factor for Alzheimer’s disease3,4. APOE exhibits an important role in peripheral cholesterol metabolism and cholesterol transport in the central nervous system. Three APOE isoforms (ε2, ε3 and ε4) have been identified in human, and the e4 allele is associated with Alzheimer disease5.

Given that both hearing loss and the APOE ε4 allele are common risk factors for Alzheimer's disease, their potential association has been investigated. In an animal study using APOE knockout mice, stenosis of the spiral modiolar artery caused by atherosclerosis induced hair cell loss and high frequency hearing loss6. However, several previous cohort studies have reported contradictory results. Some studies suggested that the APOE ε4 allele had a negative effect on hearing levels7,8, other studies reported no association9,10,11,12, and one study even showed a protective effect of the APOE ε4 allele on hearing13.

In this retrospective multicenter cohort study, we aimed to investigate the relationship between APOE genotypes and sensorineural hearing loss.

Methods

Clinical data warehouse

Clinical data was extracted from a clinical data warehouse (CDW) that includes data from seven affiliated Catholic Medical Center (CMC) hospitals (Seoul St. Mary’s Hospital, Yeouido St. Mary’s Hospital, Uijeongbu St. Mary’s Hospital, Bucheon St. Mary’s Hospital, Eunpyeong St. Mary's Hospital, St. Vincent’s Hospital and Incheon St. Mary’s Hospital). The CMC CDW includes a fully anonymized database of approximately 15 million electronic medical records; researchers can extract data based on required inclusion and exclusion criteria14.

This study was approved by the Ethics Committee of Seoul St. Mary’s Hospital (No. KC21WISI0924) and followed the tenets of the Declaration of Helsinki. Since this study is a retrospective cohort study using anonymized clinical data in CDW, the requirement for informed consent was waived by Ethics Committee of Seoul St. Mary’s Hospital.

Study population

We enrolled 2,458 subjects aged 20–100 years with APOE genotype information from seven affiliated hospitals. The follow-up period was from January 2006 to December 2021. Patients without audiometric and cognitive function testing were removed. Only participants with records of APOE genotypes, audiometric, and cognitive function testing were included. Additionally, cases indicating conductive or mixed-type hearing loss, where the average air-conduction threshold and bone-conduction threshold in pure-tone audiometry (PTA) were ≥ 10 dB HL, were excluded. Thus, the final dataset included 1,162 patients.

Clinical outcomes

APOE genotyping

DNA was extracted from blood samples using the QIAmp Blood DNA Maxi Kit protocol (Qiagen, Valencia, CA). TaqMan SNP genotyping assays (Applied Biosystems, Foster City, CA) were used to determine the genotypes for two APOE SNPs, rs429358 (ε4) and rs7412 (ε2). The participants were classified into two groups based on APOE ε4 variations: ε4 negative group (ε2/ε2 or ε2/ε3 or ε3/ε3) and ε4 positive group (ε2/ε4 or ε3/ε4 or ε4/ε4). The ε4 (-) group served as the comparison group for all analyses.

Hearing tests

Audiometric tests were conducted in a soundproof booth using the GSI 61TM audiometer (Grason-Stadler, Inc., St. Eden Prairie, MN), calibrated to meet American National Standards Institute standards (S3.6-1996). The PTA thresholds were reported in decibels of hearing level, and the PTA averages were calculated using the hearing thresholds at 0.5, 1, 2, and 4 kHz. In cases of asymmetric hearing thresholds, the better-hearing ear threshold was used for analysis. Hearing loss was defined in accordance with the World Health Organization’s definition of impairment as a four-frequency average of PTA exceeding 25 dB HL15. The speech discrimination score (SDS) test was performed using the standardized Korean version of phonetically balanced monosyllabic words and reported as a percentage16.

Cognitive function

Cognitive function was evaluated using the Mini-Mental State Examination (MMSE), a brief screening tool developed to assess the overall cognitive function of elderly individuals. The MMSE comprises orientation, attention, memory, language, and visual-spatial skill tests. Scores range from 0 to 30, with lower scores indicating poorer cognitive function. Cognitive impairment was defined as a MMSE score lower than 24. The MMSE test results conducted on the closest date were extracted and employed, in accordance with the baseline audiometric test.

Other covariates

In the reviewed prior research, several potential confounding variables that are connected with hearing loss and cognitive impairment have been discovered. Data on the sex of the patient, the patient’s age when the audiometry test was performed, and lipid tests were collected. The study used ICD-10 diagnostic codes to filter hearing loss due to diabetes (E11) and hypertension (I10). Standard procedures were used to determine total cholesterol (TC), high-density lipoprotein (HDL-C), low-density lipoprotein (LDL-C), and triglycerides (TG) levels. For the summary statistics, laboratory findings obtained 50 days before and 50 days after the date of the hearing test were used.

Statistical analysis

Means, standard deviations, and percentages were used to characterize demographic and clinical variables. The student's t-test was used to analyze the continuous variables in the summary statistics. The Chi-square test was used to analyze the categorical variables. Participants' APOE allele frequency distributions (ε4 +) were compared to those of the established general population (ε4 −).

In this study, we aimed to investigate the effect of the APOE ε4 genotype on PTA thresholds. The dataset included various demographic and clinical variables such as MMSE, low-density lipoprotein (LDL), triglycerides (TG), high-density lipoprotein (HDL), total cholesterol (TC), presence of diabetes and hypertension. The primary outcome variable was PTA measure in dB HL, and the primary predictor of interest was the presence of the APOE ε4 allele. To comprehensively assess the characteristics of the data, we conducted univariate analyses considering the presence of the E4 allele, MMSE scores, four lipid tests (HDL-C, LDL-C, TC, TG), as well as the presence of diabetes and hypertension. However, due to the correlations among the covariates, we aimed to control the model by including only a minimal set of covariates in the multivariate analysis. Therefore, only the presence of hypertension and diabetes were included in the multivariate analysis.

We fitted a Generalized Linear Mixed Model (GLMM) using a Gamma distribution with a log link to investigate the relationship between PTA and several predictors, including age, APOE ε4 allele status, sex, diabetes, and hypertension status. The model also accounted for random effects at both the individual level and the frequencies of PTA. To address potential overfitting issues in the non-parametric regression analysis, we performed a tenfold cross-validation. As shown in the equation below, we included age, sex, and the presence of diabetes and hypertension as covariates to estimate hearing thresholds. Additionally, the presence of the APOE ε4 genotype and the interaction term between genotype and age were included as fixed effects to estimate hearing loss. The GLMM allowed us to model the correlation within clusters and to include random intercepts for each subject (\({{\varvec{u}}}_{{\varvec{i}}}\)) and each frequency of PTA (\({{\varvec{u}}}_{{\varvec{j}}}\)), reflecting repeated measures for each patient. In this model \(\upmu {\text{ij}}\) represents the expected PTA score for individual i at frequency j. To analyse PTA across all frequencies simultaneously, we transformed the independent hearing thresholds at each frequency into a stacked data format. The effects at each frequency were then compared by examining the size of the random effects. To standardize the levels of the variables, numerical variables were transformed using min–max normalization. The p-values for coefficients less than 0.05 indicated statistically significant relationships.

$$log\left(\upmu {\text{ij}}\right)={a}_{0}+{\beta }_{1}age+ {\beta }_{2} APOE \varepsilon 4+{\beta }_{3}\left(age \times APOE \varepsilon 4\right)+ {\beta }_{4}sex+ {\beta }_{5} diabetes+ {\beta }_{6}hypertension+{u}_{i}+ {u}_{j}+{\varepsilon }_{ij}$$

Based on the results of our GLMM analysis, we discovered that the APOE ε4 genotype plays a moderating role in age-related hearing loss. As shown in Fig. 1A, there is a noticeable reversal in the average estimated hearing threshold at a specific age. Consequently, to closely observe the critical age at which the average hearing levels between groups with and with the APOE ε4 genotype is reversed, we conducted ROC (Receiver Operating Characteristic) analysis and Kaplan–Meier analysis. The area under the ROC curve (AUC) is a graph that compares the true positive rate (sensitivity) against the false positive rate (1 − specificity). The positive likelihood ratio, also known as LR + (sensitivity/1 − specificity), expresses the degree to which positive outcomes are significantly more common among individuals who have hearing impairment compared with those who do not17 A higher LR + number has an excellent diagnostic informative value; hence, the optimal cut-off was determined based on the highest LR + . We utilized Kaplan–Meier plots to visually observe changes in the proportion of individuals with hearing loss across different age groups and APOE genotype groups. Hearing loss was classified as a PTA threshold of 25 dB HL or higher, as previously described.

Fig. 1
figure 1

Relationship between age and pure tone audiometry (PTA) thresholds by apolipoprotein E genotypes and sexes. A, The ε4 (+) group shows lower PTA thresholds at younger ages compared to the ε4 (−) group, but the thresholds increase rapidly with age. B, Women generally exhibit lower PTA thresholds, and the rate of hearing deterioration with age does not differ between sexes.

Full size image

Results

Demographic statistics according to APOE ε4 status

Table 1 shows descriptive statistics by APOE ε4 status. Among the 1,162 individuals, 280 were ε4 allele carriers and the remaining 882 did not have the ε4 allele. There was no significant variation in the average age of the population (75.7 years) based on genotype. The incidence rate of cognitive impairment (MMSE < 24) was higher in the ε4 (+) group, while the incidence rate of hearing loss based on PTA (> 25 dB HL) was higher in the ε4 (−) group.

Table 1 Descriptive statistics by Apolipoprotein E ε4 status.
Full size table

Generalized linear mixed model

The GLMM results provided insights into the relationship between various covariates and PTA thresholds. The model included fixed effects for age, sex, hypertension, diabetes, and the interaction between age and the presence of the APOE ε4 allele. Random effects were included to account for repeated measures within individuals and across frequencies. As shown in Table 2, age was identified as a significant factor influencing hearing ability, with an estimate of 1.784 (SE = 0.096, p < 0.001). This means that each additional year of age is associated with an increase of 1.741 in the PTA threshold, indicating a decline in hearing ability as individuals grow older. Additionally, sex differences were observed, with males having higher PTA thresholds than females. The estimate for males was 0.136 (SE = 0.026, p < 0.001), suggesting that males generally experience worse hearing than females.

Table 2 The Result of Generalized Linear mixed regression analysis.
Full size table

Results for individuals with the APOE ε4 allele indicated an estimate of − 0.363 (SE = 0.126, p = 0.005), suggesting that individuals carrying this allele tend to have lower PTA thresholds. However, as seen in Fig. 1A, the estimated PTA values by age intersect at a certain point depending on the presence of the allele. This suggests that the APOE ε4 allele acts as a moderating variable in age-related hearing loss. The interaction term for age and APOE ε4 has a coefficient of 0.557 (SE = 0.214, p = 0.006), suggesting that the APOE ε4 gene may accelerate the decline in hearing ability with age. On the other hand, as observed in Fig. 1B, there was no moderating effect of sex on the relationship between age and PTA thresholds.

Kaplan–Meier plot

In the Kaplan–Meier plot (Fig. 2), the probability of hearing loss-free survival between the ε4 (+) group and the ε4 (−) group was different depending on age at the time of the test. The probability of hearing loss–free survival was higher in the ε4 (+) group at a younger age, while the ε4 (−) group had a higher probability of hearing loss–free survival at an older age.

Fig. 2
figure 2

Kaplan–Meier plot for hearing loss-free survival. Before the age of 75, the ε4 (+) group had a higher probability of hearing loss-free survival compared with the ε4 (−) group. The ε4 (−) group exhibited a higher likelihood of hearing loss-free survival after the age of 75. PTA, pure-tone audiometry.

Full size image

ROC analysis for determining cut-off value

ROC analysis was used to derive the optimal age cut-off value for subgroup analysis. The appropriate hearing impairment age threshold was determined using data from the prediction model. The AUC was 0.77, with a 95% confidence range of 0.74–0.79. The optimal cut-off value of 75.5 was determined based on predictive sensitivity of 73.1% and a specificity of 69.3% (Fig. 3).

Fig. 3
figure 3

Receiver operating characteristic (ROC) curve to investigate the optimal age cut-off value. The optimal cut-off value was 75.5-year-old based on predictive sensitivity of 73.1% and a specificity of 69.3%. AUC, area under the ROC curve.

Full size image

Subgroup analysis

In the subgroup analysis, the ε4 (+) group demonstrated significantly better hearing levels in both mean PTA and SDS among individuals younger than 75 years old (P < 0.01, T-test). In the older subgroup (≥ 75 years), the mean age of the ε4 (+) group was significantly lower than the ε4 (−) group and audiologic tests indicated no significant difference between the two groups (Table 3).

Table 3 Summary statistics for the group of patients < 75- and ≥ 75-year-old.
Full size table

Discussion

In the context of the ongoing discussion regarding the potential correlation between APOE genotype and hearing loss, which has yielded conflicting outcomes to date, our study has produced the following conclusions. Firstly, ε4 allele carriers show that hearing deteriorates more rapidly with increasing age compared to non-carriers, with no differences regarding sex. Secondly, among ε4 carriers, hearing is better before the age of 75, but this difference diminishes after the age of 75 due to the rapid deterioration of hearing with age.

To the best of our knowledge, seven studies have been published investigating the correlation between APOE genotype and hearing loss, with conflicting results reported (Table 4). Our findings provide insight into the discrepancies among previous studies, which may be attributed to variations in the age range of the cohorts examined. A potential explanation for the inconsistent results of prior investigations could be the age of the cohorts studied. Specifically, five of the previous studies evaluated hearing loss using PTA in cohorts with an average age ranging from 64 to 72 years and failed to identify a significant correlation between the ε4 genotype and hearing loss or reported a positive relationship with good hearing9,10,11,12,13. In contrast, Kurniawan et al. identified a significant association between the ε4 genotype and hearing loss in the oldest cohort examined, with a mean age of 85 years8. Our investigation found that the ε4 variant had a protective effect before the age of 75.5, but a detrimental effect thereafter. This likely explains the negative impact of the ε4 allele variant observed in the older age group. Kim et al. reported a higher risk of hearing loss in male ε4 carriers in a relatively young cohort; however, this study relied on self-reporting rather than audiometry, which may be subject to age and sex-related biases7. Given that subjective hearing impairment can be influenced by age and sex, as previously noted18, we suggest that our results, which show no significant differences in analysis by sex, are more reliable.

Table 4 Summary of previous studies and current results.
Full size table

Our findings suggest a biphasic effect where individuals with APOE ε4 allele variants initially exhibit better hearing at a younger age, followed by a rapid deterioration in hearing with advancing age. Additionally, our results indicate a reversal of the earlier hearing protective effect of the APOE ε4 allele around the age of 75.5. Considering that the incidences of Alzheimer’s disease regarding APOE genotypes significantly differ from about 70 years of age, it is thought that the onset of hearing loss appears at a later age of about 5 years than Alzheimer’s disease19,20. The reason why the ε4 allele increases Alzheimer’s disease prevalence is thought to be because of amyloid-β accumulation and tau-mediated neurodegeneration21. As with APOE genotypes, there exists a debate concerning the link between amyloid-β and tau protein and their correlation with hearing loss22,23,24,25. A study has reported an association between higher amyloid-β and tau burden and age-related hearing loss22. Furthermore, in an animal study using transgenic mice expressing amyloid-β, hair cell loss in the cochlear basal turn due to amyloid-β has been observed, and tau protein has been demonstrated to synergistically enhance amyloid-β-induced hearing loss23. Nonetheless, the underlying mechanism by which the ε4 allele impacts hearing loss remains inadequately investigated. A previous investigation using APOE knock-out mice suggested that hyperlipidemia and atherosclerosis may damage the cochlea APOE plays an important role in lipoprotein metabolism and can increase LDL-cholesterol levels in ε4 allele carriers26. Several studies have reported a potential negative impact of elevated LDL-cholesterol levels on hearing loss27. However, the effect of APOE on hearing cannot be fully explained by its role in lipid metabolism alone. First, lipid profile results did not significantly differ in the present study. Second, the biphasic effect observed in our study cannot be explained by lipid metabolism alone, which changes monotonically. Thirdly, there are differing findings on the correlation between blood lipid levels and hearing loss, with some studies reporting weak or no correlation28.

Most of the studies on APOE genotypes have been focused on the negative effects of the APOE ε4 allele; however, recent investigations provided some indicators for positive effects. Several studies have shown that innate immune biomarkers are reduced in ε4 carriers, implying that ε4 allele carriers may be more protected from inflammatory burden through lower innate immune sensing and faster clearance following the resolution of an acute inflammatory spike29,30,31,32. In the study of Garcia et al.31, C-reactive protein (CRP) levels were 30% lower in ε4 carriers compared with homozygous ε3/ε3 individuals, and ε4 carriers exhibited a lower eosinophil to lymphocyte ratio and lower total leukocytes, suggesting that ε4 allele may play a beneficial role in the immune response, despite the fact that ε4 carriers have higher blood lipid levels. These Janus-like characteristics of the ε4 allele provide clues for understanding the results of our study. We hypothesized that the positive effect of ε4 allele on the immune response may be a reason for the hearing protective effect at a younger age, while the negative effects of blood lipids accelerate the hearing loss at an older age. Nevertheless, this remains a hypothesis since our study was not designed to reveal the pathophysiology of APOE genotypes and hearing levels. Therefore, further research in this area is necessary.

The present study demonstrates methodological strengths compared with previous research. The audiologic data analyzed in this study was obtained in a soundproof chamber using calibrated equipment, rendering it reliable. Additionally, the sample size was sufficiently large to produce trustworthy findings. Above all, our study has significant meaning in that it is the first report of a biphasic effect of the ε4 allele on hearing loss, with an age-related cut-off value determined through statistical analysis.

However, this study has several limitations. First, the cohort used in this study was limited to participants with Korean ethnicity and patients who visited hospitals. Therefore, the MMSE values of our study cohort exhibited lower results compared to the general population. Second, it was not feasible to analyze the effect of APOE ε4 dosage as the number of ε4 (+/+) individuals in the cohort was insufficient. Third, there is a limitation that the timing of the conducted assessments may not be uniform. Unlike the APOE genotype test, the MMSE test, audiometric assessment, and blood tests can vary based on the timing of the examination. Therefore, while it would be most ideal for all assessments to be conducted at the same time for analysis, it should be noted that this study acknowledges the presence of variations in the timing of assessments due to the retrospective nature of the research. Finally, due to limitations in the data available from medical records, we were unable to include various covariates. For instance, several risk factors for hearing loss, such as a history of noise exposure, smoking, familial history of hearing loss, race, and the use of ototoxic medications, as well as other audiological symptoms like tinnitus and hyperacusis, were not included in the analysis33,34. Additionally, genomic principal components, which could control for the polygenic risk of hearing loss and non-genetic confounders affecting the relationship between APOE and hearing loss, were not included as covariates.

The correlation between APOE genotype and hearing loss has yielded contradictory results in prior studies. Our study indicated that individuals with the ε4 allele experience a hearing protective effect at a younger age but undergo a more rapid deterioration in hearing as they age. This biphasic effect of the APOE genotype provides a clue to understanding the inconsistent findings of previous investigations. Further studies investigating the underlying mechanisms of this relationship are needed.