Synergistic effect of smoking on age-related hearing loss in patients with diabetes

This study investigated the synergistic effects of risk factors on age-related hearing loss (ARHL) using nationwide cross-sectional data of 33,552 individuals from the 2010‒2013 Korea National Health and Nutrition Examination Survey. Patients with ARHL were selected based on their pure-tone audiometry results. Previously reported risk factors for ARHL were analyzed using logistic regression and propensity score-matching, and synergistic effects between risk factors were analyzed using propensity score-matching. Of the 12,570 individuals aged 40–79 years, 2002 (15.9%) met the criteria for ARHL. Male sex, exposure to occupational noise, and diabetes showed a significant relationship with ARHL (p < 0.05) in both the logistic regression and propensity score-matching analyses. Smoking and diabetes showed the strongest significant synergistic effect on ARHL (odds ratio [OR] 1.963, 95% confidence interval [CI] 1.285‒2.998; p = 0.002). In the subgroup analysis based on smoking status, current smokers with diabetes had a significant relationship with ARHL (OR 1.883, CI 1.191‒2.975; p = 0.009), whereas ex-smokers with diabetes did not (OR 1.250; CI 0.880‒1.775; p = 0.246). This implies that current smokers with diabetes may benefit from the cessation of smoking. In conclusion, patients with diabetes should strictly avoid or cease smoking to prevent the progression of ARHL.

included the 10 proposed risk factors as independent variables and ARHL as a dependent variable in the regression model. The six variables (age, occupational noise exposure, stroke, male sex, current smoking, and dyslipidemia) included in the final regression model showed 84.8% accuracy of classification and a Nagelkerke's R 2 of 0.306 (i.e., 30.6% of events could be explained by this model) ( Table 1). The p value, according to the Hosmer-Lemeshow test, was 0.596, which indicated that this regression model was valid. In the regression model, occupational noise exposure showed the highest correlation with ARHL. The well-known cumulative effect of age on noise-induced hearing loss was consistent with this result 19  In the PSM analysis, male sex, HTN, diabetes mellitus, and noise exposure showed significantly increased ORs (Table 2). Among these, noise exposure history had the highest OR (OR 1.779, CI 1.461-2.167), which was consistent with the results of the logistic regression analysis. HTN (OR 1.157, CI 1.004-1.333) showed a significantly increased OR in the PSM but not in the logistic regression analysis. The ARHL risk factors common to both statistic models were male sex, noise exposure, and diabetes. and obesity) and current smoking. Individuals who had both these targeted risk factors were compared to those who had a single or no targeted risk factor. Ten combinations of the five modifiable risk factors were investigated by PSM (Table 3). Notably, the combination of diabetes and current smoking was associated with the most significant increase in the prevalence of ARHL (OR 1.963, 95% CI 1.285-2.998; p = 0.002), after completely controlling for the other risk factors. The combination of diabetes and HTN was also statistically significant, but the OR (OR 1.388, 95% CI 1.100 − 1.750; p = 0.007) was smaller than that for the combination of diabetes and smoking. Other combinations showed no significant difference in terms of the prevalence of ARHL according to the presence of risk factors. Next, we performed a subgroup analysis of the prevalence of ARHL according to smoking status. The OR of ARHL did not differ between the ex-smokers and the never-smoked group (OR 1.008, 95% CI 0.789-1.287; p = 1.000) (Supplementary Table S2 online). Additionally, there was no synergistic effect between ex-smoking status and diabetes in terms of the OR of ARHL (OR 1.250; 95% CI, 0.880-1.775, p = 0.246) ( Table 4). However, in the analysis of the current smokers and the never-smoked subgroups, a synergistic effect on the odds of developing ARHL was observed between current smoking status and diabetes (OR 1.883, 95% CI 1.191-2.975; p = 0.009) ( Table 4). The combined effect of current smoking status and diabetes on the hearing function is clearly presented in Fig. 1. Unlike current smoking status, the ex-smoking status did not increase the OR of ARHL in patients with diabetes.

Discussion
In this study, we performed a PSM analysis and observed that the combination of smoking history and diabetes contributed most significantly to ARHL. Moreover, we observed that diabetic ex-smokers showed a reduced OR compared to current diabetic smokers. In the PSM analysis, the OR of the combination of diabetes and current smoking (1.963) was higher than the ORs of current smoking (1.171), diabetes (1.285), and HTN (1.157) alone as well as the combination of diabetes and HTN (1.388).Considering that both diabetes and smoking may affect cardiovascular circulation, the progression of ARHL may be attributed to previously suggested risk factors, such as HTN, dyslipidemia, stroke, obesity, and CVD 9,14,20 .
ARHL is a slowly progressive disease that develops over decades and has an extremely complex etiology. Although a definite causative gene contributing to the development of ARHL in the population is unknown, numerous genes with common polymorphisms or rare variants have been suggested as a cause for ARHL in some patients; for instance, KCNQ4 is a candidate gene associated with hearing loss that progresses with age 16,[21][22][23][24][25][26][27][28][29] . However, the genetic information of the subjects was not available in this study. The analysis of genetic polymorphisms associated with ARHL would be more helpful in substantiating the causative relations of the risk factors for ARHL in the future. Further, several environmental contributors to ARHL have been suggested, including Table 3. Odds ratios of age-related hearing loss for each combination of two risk factors in the propensity score-matched groups. *Compared to the group with one or no risk factors; CI confidence interval; HTN hypertension; OR odds ratio; # p value < 0.05. www.nature.com/scientificreports/ ototoxic medication, alcohol intake, tobacco use, and head trauma 17 . Because ARHL has complex etiologies, the effects of some risk factors, such as diabetes and dyslipidemia, remain controversial 4,14,30 . In this study, dyslipidemia showed a negative correlation and no correlation with ARHL in the logistic regression and PSM analysis, respectively. Interestingly, dyslipidemia seems to have a lower association with ARHL, speculation of which has been supported by recent studies 31,32 . We also observed a discrepancy in the effects of some risk factors (HTN, smoking) on ARHL between logistic regression and PSM analyses. Additionally, other medical conditions, such as CVD, stroke, obesity, and dyslipidemia, did not contribute to the difference in the OR for ARHL in logistic regression and PSM analyses, which was inconsistent with the findings of previous literature [9][10][11][12] . This may indicate that several medical conditions have mutual effects, and no single effect from each risk factor can explain the pathology of ARHL. In this regard, we consider that multifactorial analyses may be more helpful in explaining the pathophysiology of ARHL. Considering that several risk factors, such as HTN, diabetes, CVD, dyslipidemia, obesity, and smoking, are associated with vascular disease and are interrelated, controlling for confounding factors is important when evaluating a single risk factor for ARHL 9,14,20 . Meanwhile, the logistic regression analysis revealed that older age, noise exposure, male sex, current smoking, and diabetes were significantly associated with ARHL, whereas the PSM revealed that noise exposure, male sex, diabetes, and HTN significantly increase the OR of ARHL. Because the PSM and logistic regression analyses are complementary, it is reasonable to adopt the results from both methods. In particular, we used PSM analysis to confirm the hierarchical association of the known risk factors for ARHL without the interactions between covariates. PSM analysis was also preferably chosen for determining the synergistic effects of each risk factor, since PSM can substantially minimize the collinearity of variables, thereby accurately predicting the synergistic effects. Additionally, PSM can convincingly define an association hierarchy among varying combinations by perfectly matching the other co-variables between groups. Although the interaction regression model using the cross-product variable can be applied to evaluate the interactions of covariates, it is less appropriate in this study because too many medical conditions should be hypothesized as interacting conditions.
Smoking is a modifiable risk factor in preventing the progression of ARHL in patients with diabetes. In particular, we observed that ex-smoking status did not have a significant effect on the OR for ARHL in diabetic patients, indicating that ARHL may be alleviated if a patient quit smoking. The synergistic mechanism of smoking and diabetes in ARHL remains unclear, and the duration of smoking cessation that is required to obtain a reduced OR of ARHL is not known. In general, smoking and diabetes have synergistic harmful effects on blood circulation; smoking increases the progression of nephropathy and the risk of CVD in patients with diabetes [33][34][35] . It has also been reported that C-reactive protein, a systemic inflammation marker, is increased in current smokers compared to ex-smokers and those who have never smoked 36,37 . Moreover, a recent study has suggested that inflammatory food consumption is closely related to the incidence of ARHL 38 . Given that chronic inflammation increases the risk of cardiovascular events, as supported by the results of the JUPITER trial, smoking and diabetes may cause vascular pathologies, such as atherosclerosis 39 . This synergism may eventually destroy the vascular supply to the auditory end organ, the cochlea, and cause progressive hearing deterioration. Another possible mechanism may be direct cellular damage to the sensory hair cells in the cochlea or auditory neural pathway. In a mouse model, a high glucose level affects auditory midbrain function, loss of spiral ganglion cells, decreased cochlear blood flow, and increased susceptibility to noise exposure 40,41 . Smoking also increases susceptibility to noise exposure and decreases spiral ganglion cells due to increased oxidative stress in the mouse model 42,43 . According to the data in our study, cochlear damage due to chronic smoking may be reversed through cessation of smoking. Therefore, strict and timely control of smoking is advised, particularly in patients with diabetes.
A limitation of this study is that it did not provide specific information on the duration of smoking cessation required to observe its preventive effect on ARHL in diabetic patients. Prospective observational studies may be www.nature.com/scientificreports/ needed to clarify details about the prevention of ARHL. Further, the PSM analysis has an innate limitation; in an ideal PSM, it is hypothesized that all possible covariates are included. In this study, 10 validated risk factors from previously published studies were included. However, there may be additional latent covariates affecting the development of ARHL (i.e., genetic polymorphisms) that were not considered in our analyses. Future studies should reflect novel discovered risk factors for ARHL. Additionally, the causal relationship between ARHL and several risk factors could not be clearly elucidated in this study. This is because the KNHANES is a cross-sectional survey, which may have led to a reverse causality bias in this study. However, considering the pathophysiologic mechanism of the medical risk factors associated with ARHL, it is unlikely that ARHL may have increased the risk of medical diseases such as HTN, diabetes, and dyslipidemia. Further, the reduced OR in diabetic ex-smokers compared to current diabetic smokers seems to imply that the cessation of smoking can ameliorate the risk of ARHL; however, this result should not be overestimated since a cross-sectional study cannot fully elucidate the direction of causality. Therefore, further prospective studies may be required in the future to verify the ameliorating effects of smoking cessation. Lastly, the World Health Organization recently defined ARHL as a hearing threshold of > 40 dB HL in 0.5, 1, 2, and 4 kHz frequencies.
Although the cut-off threshold of 30 dB HL used in this study has been applied in several publications, the results, especially the OR values, may slightly vary if the cut-off threshold is set at 40 dB HL.
In conclusion, several environmental and medical factors may affect the risk of ARHL. Among them, noise exposure, male sex, and diabetes seem to be important factors, based on both the logistic regression and PSM analyses. In addition to the individual effect of each risk factor, a combinatorial effect of these factors should be considered when attempting to prevent ARHL. As the most modifiable factor, cessation of smoking should be strongly prescribed in high-risk subjects, such as diabetic patients.

Participants.
A total of 12,570 individuals (all Korean) included in the 2010 − 2013 KNHANES 18 participated in this study. The inclusion criteria were normal findings on otoscopic examination, presence of pure-tone audiometry data, and an age of 40-79 years. We excluded individuals who showed a difference of > 30 dB HL in the bilateral pure-tone threshold since ARHL is typically associated with symmetric hearing impairment 4,44 . Individuals with missing values regarding any candidate risk factor [age, sex, HTN, diabetes, dyslipidemia, stroke, CVD, occupational noise exposure, smoking history, and body mass index (BMI)] were also excluded (see Supplementary Figure S1  Korea National Health and Nutrition Examination Survey. The KNHANES is a cross-sectional survey that has been conducted annually in South Korea since 1998. The fifth KNHANES was conducted from 2010 to 2012, and the sixth KNHANES was conducted from 2013 to 2015. The sample design of the KNHANES was extracted using a two-stage stratified cluster (region and household) sampling. The rate of participation in the fifth and sixth KNHANES was 80.8% and 78.3%, involving 25,533 and 22,948 people, respectively. The KNHANES consisted of three categories: the medical survey, the nutrition survey, and the health examination. Specifically, we only collected datasets that included both noise exposure history and a pure-tone audiogram. Although pure-tone audiometry was conducted from 2008 to 2013, the data from the fourth KNHANES (2007-2009) was not included in this study as it did not include a noise exposure history survey. Since noise exposure history-taking and the pure-tone audiometry test were conducted only from 2010 to 2013, the data from the fifth KNHANES (2010-2012) and the sixth KNHANES (for 2013) were merged vertically to increase the statistical validity. Medical history and smoking information were obtained from the medical survey. BMI and air-conduction pure-tone threshold data were obtained from the health examination.
Hearing assessment. The subjects underwent pure-tone audiometry in an ordinary soundproof booth that was installed in the vehicle. The test included pure-tone air-conduction threshold measurements at 0.5, 1, 2, 3, 4, and 6 kHz. Individuals with an average pure-tone threshold of ≥ 30 dB HL in the "better" ear at 0.5, 1, 2, and 3 kHz (according to the guidelines of the American Academy of Otolaryngology-Head and Neck surgery foundation 45 ) were included in the ARHL group 46,47 . An otorhinolaryngologist examined and reported the tympanic membrane findings in each subject. A well-visualized tympanic membrane without pathologic findings such as retraction, perforation, or middle ear effusion and no cerumen impaction in the middle ear and external auditory canal were defined as normal findings.
Risk factors. We selected 10 candidate risk factors: age, male sex, HTN, diabetes, dyslipidemia, stroke, CVD, occupational noise exposure, smoking, and obesity. Physician-diagnosed diseases (HTN, diabetes, dyslipidemia, stroke, and CVD) were identified through a self-reported survey. Smoking history was categorized into three groups: never-smoked, ex-smoker, and current smoker. The ex-smoker category indicated that a subject had smoked daily during their lifetime but was not smoking at the time of the survey. BMI was converted into a dichotomous variable termed "obesity"; a BMI > 25 kg/m 2 was considered obese 48  Statistical analysis. The 10 candidate risk factors were treated as interacting covariates, and logistic regression and PSM analyses were performed with the same sample and covariates. In order to classify the most associated determinants, the logistic regression analysis was performed using the stepwise backward selection (cut-off p value = 0.05) method with the Hosmer-Lemeshow test (valid if p value > 0.05). The accuracy of classification and Nagelkerke's R 2 were calculated to evaluate the explanatory power. ARHL was included as a dependent variable, and the 10 risk factors were included as independent variables in the regression model. Propensity scores were calculated using logistic regression analysis, which included the target risk factor as a dependent variable and other risk factors as independent variables. We performed 1:1 matching between individuals with the same propensity score to obtain two exact propensity score-matched groups. From the matched pairs, ORs and 95% confidence intervals (CIs) were calculated from a 2 × 2 table, including ARHL (presence or absence) and the target risk factor (presence or absence). We defined two groups to evaluate the ORs of the combined risk factors: one was the combined risk group with both risk factors, and the other was a control group with one or no risk factors. After classification, PSM and OR calculation was performed in the aforementioned manner.
The McNemar's test was performed to evaluate the statistical significance of the difference in ARHL risk factors between the propensity score-matched groups. The pure-tone threshold of each frequency was compared using two-way analysis of variance with Sidak's post-hoc. A p value < 0.05 was considered statistically significant. The statistical analyses were conducted using Statistical Package for the Social Sciences version 25.0 (IBM; Armonk, NY).

Data availability
The data are available from the corresponding author upon reasonable request. www.nature.com/scientificreports/ Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creat iveco mmons .org/licen ses/by/4.0/.