To our knowledge, this is the first study to examine the impact of diabetes on hearing using the criteria of the World Report on hearing [18]. In summary, there was a higher hearing threshold in those participants who had diabetes no matter sex sex-specific or age-specific it is. We found that diabetes, longer duration and higher FBG were independently associated with hearing loss for speech-frequency and high-frequency HL, and the associations were stronger at high-frequency HL (Fig. 3). At low/mid-frequency HL, only > 5 years duration of duration and higher FBG were associated with HL independently. Furthermore, the “healthier population” (no hypertension, no dyslipidemia and younger age) has stronger associations and higher risks.

The associations of diabetes and duration with speech-, low/mid- and high-frequency hearing loss. OR (95% CI) were from logistic regression models, which were adjusted for age(< 65/ ≥ 65 years), education level, smoking (current, former, never), alcohol consumption (current, former, never), tea (current, former, never), coffee (no/yes), BMI (≤ 18.5, 18.5 to < 24, ≥ 24 kg/m2), use of earphone (no/yes), tinnitus (no/yes), electric otoscopy (normal/abnormal), occupational noise exposure (no/yes), use of ototoxic medication (no/yes), hypertension (no/yes) and dyslipidemia (no/yes). OR indicates odds ratios; CI indicates confidence intervals

The potential reasonable biological mechanisms of diabetes-related hearing loss are mainly clustered to two aspects, neurological etiology and microvascular etiology. Diabetic neuropathy is mainly caused by diabetes-related metabolism imbalance and an imbalance in the mitochondrial redox state [9]. These imbalances collectively culminate in negative effects on mitochondrial function and gene expression along with inflammation and oxidative stress [9, 10], further leading to hearing loss. Atrophy of the spiral ganglion and demyelination of the eighth cranial nerve among autopsied diabetic patients suggest a neurological etiology to diabetes-related hearing impairment [22]. Microvascular changes in the cochlea occur at the early stage of diabetes, and studies have found thickened basilar membranes and capillaries of the stria vascularis and atherosclerotic narrowing of the internal auditory artery among autopsied people who had diabetes, but not in people without diabetes [22]. These microvascular changes reduce blood supply in the cochlea and further lead to hypoxia and insufficient energy supply, which in turn affects inner ear function. These negative changes involving abnormality of metabolism, blood supply and the cochlea structure would lead to hearing loss gradually. More terribly, mitochondrial dysfunction facilitates further mitochondrial ROS production in a positive feedback loop and affects metabolism and energy supply, ultimately resulting in the activation of cells' apoptotic pathways in cochlear and subsequent hearing loss, namely a vicious circle.

In our study, the effect size (OR) was small relatively, which may result from the difference in age range, inconsistent definitions of HL, different categorized criteria of speech-, low/mid- and high- frequency and unadjusted important potential confounders (e.g., BMI, ototoxic drug and hypertension) among studies. Nonetheless, the results were consistent with previous studies overall. A meta-analysis, in which data were obtained from cross-sectional studies, suggested a higher risk of HL in those participants with diabetes compared with those without diabetes (OR (95%CI): 2.15 (1.72–2.68)) [23]. Dalton et.al also found there was an association between NIDDM and hearing loss when adjusting for potential confounders (OR(95% CI): 1.41 (1.05–1.88))[15] at speech-frequency HL, although they did not finely adjust for BMI and ototoxic drugs. The evidence from National Health and Nutrition Examination Surveys (NHANES), which enrolled younger participants than ours, showed OR (95%CI) of 1.82 (1.27, 2.60) and 2.16 (1.47, 3.18) without adjusting chronic diseases (e.g., hypertension and dyslipidemia) for the low/mid-frequency (500, 1000, 2000 Hz) and high frequency (3000, 4000, 6000, 8000 Hz) HL(PTA > 25 dB HL), respectively [8]. In addition, we also found that the magnitude of the associations between diabetes and HL is the strongest compared with those of other cardiometabolic diseases and HL, which suggested that the associations of diabetes and hearing loss are really disease-dependent and not a reflex of the aging process itself.

The impact of diabetes duration on HL was inconsistent in present studies may be due to inconsistency of method and population. The results from NHS I and II [12] found that a longer duration (for ≥ 8 years) of diabetes was associated with a higher risk of moderate self-reported HL. Mitchell et.al. also found an association between diabetes and hearing loss for diabetes duration ≥ 10 years (OR (95% CI): 2.08 (1.10–3.94)) at speech-frequency HL (PTA > 25 dB HL) [24]. It was observed that the degree of high-frequency HL (> 2000 Hz) was related to the duration(< 1 year, 1–5 years, 6–10 years and > 10 years) [25] proportionally in a clinical study. Likewise, HAl-Sofiani et.al found that high-frequency hearing loss was significantly and positively correlated with age and duration of type 1 diabetes [26]. By contrast, Dalton et.al did not find associations between the duration of diabetes (1-year change) and hearing loss (OR (95% CI) 0.99 (0.96–1.02)) [15] at speech-frequency HL, which could result from designed short term of duration group. Also, the cross-sectional study from Brazil showed there were no significant differences in mean hearing threshold (duration ≥ 9 years vs duration < 9 years) after adjusting for age, gender, and hypertension at octave frequencies from 250 to 8000 Hz, although the mean hearing threshold was higher in duration ≥ 9 years group [27]. In our study, we found longer diabetes duration impaired hearing, which has a mild dose–response relationship. We have no idea to identify the reason for the inconsistent observations and more studies are needed to explore the relationship. However, the observed associations suggested diabetes-mediated changes in hearing occurred over the long term, namely negative cumulative effect.

A higher FGB level is associated with a higher risk of HL in all frequencies, which is similar to previous studies. The fifth Korea National Health and Nutrition Examination Survey, where HL was defined as the average hearing threshold at 500, 1000, 2000, 3000, 4000 and 6000 Hz exceeding 25 dB HL, suggested increased fasting glucose was independently associated with HL (OR (95% CI): 1.4 (1.1–1.8)) in participants with metabolic components [28]. Similarly, data from the Dongfeng‑Tongji cohort study in China found significant dose-dependent relationships between increasing high-frequency ((4000 and 8000 Hz) hearing loss levels [29]. Cochlea metabolic disturbance and the death of hair cells induced by long-term exposure to pathoglycemia may lead to HL.

Several specific findings were also observed. At low/mid-frequency HL, there was no association between diabetes and HL while a higher risk in those with diabetes for > 5 years duration. Also, there was a significant association between diabetes duration ≤ 5 years and HL compared with those without diabetes at high-frequency HL only. In addition, the impacts of diabetes-related risks on hearing were strongest at high-frequency HL. Many epidemiology studies have shown higher frequency threshold differences between diabetes and those without diabetes were larger than lower frequency [8, 24, 30, 31]. It suggested that the main frequency affected by diabetes will gradually progress from high frequency to low frequency with onset. In other words, the first damaged structures in the inner ear are those located at the cochlear basal turn, which is responsible for the high-frequency threshold. The mitochondrial damage increased with time and the damage in the cochlea basal turn was the most severe [32]. The guinea pig model has shown 5 h-room temperature survival rate of OHCs from the base was less than one-third of that from the apex, which suggested that basal OHCs may be more vulnerable to free-radical damage than apical OHCs [33]. Also, outer hair cells (OHCs) play a crucial role in cochlear amplification, which is responsible for the exquisite sensitivity and frequency selectivity.

The stronger associations between HL and diabetes, longer duration and higher FBG in those “healthier populations” (no hypertension, no dyslipidemia and younger age) are, to our knowledge, uncommon findings, while the associations were stronger in men rather than in women may due to the protective effect (otoprotection and neuroprotection) of estrogen on hearing function [34]. There are several studies shown similar results. The study from America, which investigated the association between hearing loss and CVD comorbidities in 80 years and above older individuals, showed the effect value became no significance obviously in the multiple linear regression model for both high- and low-frequency pure-tone average (β (95% CI): 3.24 (− 5.84 to 12.33) and 3.69 (− 4.82 to 12.20)) [35]. Similarly, the study from the Canadian Longitudinal Study on Aging (CLSA) found stronger associations between PTA and poor health status in the younger age group rather than the older age group in the female [36]. There are several plausible interpretations underlying with these findings. It’s not consistent between chronological age and biological age which may result from poor health effects and age acceleration [37]. Thus, adjusting for chronological age is not completely equal to adjusting for biological age of hearing function. Diabetes has been associated with an increased risk of many negative outcomes, such as CVD [38], obesity, etc., which also affect the hearing effect. In older samples, the difference of the associations arising from other important causes of age-related HL (competing risk factors) may mask the contribution of diabetes to HL if other risks are imperfectly measured and thus not completely controlled in statistical models [31]. The potential competing effect was also implied by the stronger associations between diabetes and HL in those who without hypertension and dyslipidemia. Namely, in those who with other cardiometabolic diseases, the associations between diabetes-related risks and HL in participants with higher risks may be affected by competing risk factors with more probability. And it suggested that we should pay more attention to hearing function in the younger age with diabetes so that we can make earlier findings, perform earlier treatment and lower the burden of HL. We also found one more interact item exists when calculating the risk of HL associated with a continuous variable, FBG, at low/mid- and high-frequency HL. It may be because categorized variables may lose the original information [39] and lead to one of the categories sample small although it makes analysis and interpretation easier and mimics medical practice, and continuous variables use the information more sufficiently, and have more statistical power [39, 40].

Our study has several strengths. First, we examined octave frequencies from 500 to 8000 Hz, from low frequency to high frequency, so that we can explore the association between diabetes-related risk factors and different frequency HL collectively. Second, we categorized HL using the latest classification of HL in World Report on hearing [18]. In previous studies, the definition or classification of HL was not consistent, which weakens the conviction of the impact of diabetes on HL to an extent and leads to inconsistent conclusions and hard interpretations. Furthermore, we gave evidence to the relationship of diabetes duration with all frequency HL while the few related studies have different results.

There are also several limitations. First and most, FBG reflects levels at a single point in time and it does not indicate how glucose varied over time. So, the association between the hearing and FBG is likely stronger than we have found. The significant association between FBG and HL suggested that it is sufficiently accurate to be used. In further studies, we could measure FBG over time longitudinally or glycated hemoglobin to more precisely contrast the decline in hearing associated with diabetes-related risk factors, and to determine whether effective treatment of diabetes may delay the onset and progression of hearing loss. Secondly, we did not distinguish type 2 diabetes and type 2 diabetes. The potential mechanisms linked type 1 diabetes to HL and linked type 2 diabetes to HL may be different. In further studies, we should definite the type of diabetes for more clarity. Thirdly, our study was performed in Jiangsu province in China, and the findings may not apply to region or race/ethnicity groups. Studies in multiple regions and races are needed to examine the relationship between diabetes and HL and to explore the differences in varying populations. Finally, residual confounding cannot be excluded due to the observational design, although we have controlled important potential confounders (e.g., ototoxic drugs, exposure to occupational noise, etc.). And we have robust results.