Data from the Korean National Health and Nutrition Examination Survey (KNHANES) 2010–2011 were used in statistical analyses. The KNHANES is a nationwide, multistage, stratified survey of a representative sample of the South Korean population conducted by the Korea Centers for Disease Control and Prevention. There were 17,476 participants in the KNHANES survey. The following categories of survey participants were excluded from the current study: males or premenopausal women (n=10,384); those without data regarding pure tone audiometry (n=2,325) or BMD (n=2,859); and those younger than 50 years old or older than 70 years old (n=426). In addition, KNHANES study participants diagnosed with external or middle ear disease (n=69), a history of brain disorders such as stroke (n=55), a history occupational or explosive noise exposure (n=261), or any subjects with asymmetric hearing results were excluded (n=88). This left a population of 1,009 KNHANES study participants who were included in the current study. Local ethical committee approval was obtained for this study (YUH-14-0512-085). Informed consent was waived since the subjects’ records and information were de-identified prior to the analysis.

Clinical and laboratory data collected from participants during the health examination included age, body mass index (BMI, kg/m²), systolic blood pressure (SBP, mmHg), diastolic blood pressure (DBP, mmHg), fasting blood glucose (FBG, mg/dL), alkaline phosphatase (ALP, IU/L), intact parathyroid hormone (i-PTH, pg/mL), vitamin D (VitD, ng/mL), triglyceride (TG, mg/dL), and high density lipoprotein (HDL) cholesterol (mg/dL) levels, alcohol consumption, smoking behavior, exposure to noise, BMD (g/cm²), and hearing thresholds.

Whole body, dual-energy X-ray absorptiometry was performed with a QDR Discovery fan beam densitometer (Hologic Inc., Bedford, MA, USA) according to the manufacturer’s recommendations. All subjects wore light clothes, and removed all items, including jewelry, that could interfere with the measurement. BMD was measured at the femoral neck (FN) and lumbar spine (LS). The case definition of OP in this study was defined as T-score below –2.5 at the FN or LS in accordance with the World Health Organization (WHO) diagnostic classification [11].

FBG, ALP, TG, and HDL cholesterol levels were measured using an enzyme method (Hitachi Automatic Analyzer 7600, Hitachi, Tokyo, Japan). The i-PTH level was measured using a chemiluminescence immunoassay (LIASON, DiaSorin Inc., Stillwater, MN, USA). The VitD level was measured from 25-hydroxy VitD using a Radioimmunoassay (1470 WIZARD gamma-Counter, PerkinElmer, Turku, Finland). Blood pressure was measured three times by trained nurses using a mercury sphygmomanometer (Baumanometer, Baum, Copiague, NY, USA), with the participant in a seated position after a 5 minute resting period. The final blood pressure value was obtained by averaging the values of the second and third blood pressure measurements. The case definition of hypertension (HTN) was SBP ≥140 mmHg, DBP ≥90 mmHg, a self-reported history of HTN, or the use of antihypertensive drugs, and diabetes mellitus (DM) was defined a self-reported history of a DM diagnosis or a fasting glucose level of ≥126 mg/dL. Smoking behaviors were classified as current smoker, ex-smoker, or nonsmoker. Alcohol intake was defined by the Korean version of standard drinking, which was based on the WHO classification [12,13]. We classified alcohol intake into three categories: abstinence (not having had an alcoholic drink within the last year); moderate drinking (women, 0.1–19.99 g pure alcohol/day and men, 0.1–39.99 g pure alcohol/day); and heavy drinking (women, ≥20 g pure alcohol/day; men, ≥40 g pure alcohol/day).

Histories of exposure to explosives or occupational noise were classified as positive or negative, as previously described [14]. Briefly, an explosive noise was defined as a sudden loud noise such as an explosion or gunshot. Exposure to occupational noise was determined according to whether the participant had worked in a location with loud machinery for ≥3 months, where loud noise was defined by whether the participant had needed to raise his or her voice to have a conversation. Participants with these histories were excluded from the present study.

Participants’ hearing thresholds were measured using an automatic audiometer at 0.5, 1, 2, 3, 4, and 6 kHz. For both ears of each participant, the pure tone averages at 0.5 were averaged to obtain the low-frequency (Low-Freq) value; those at 1, 2, and 3 kHz, to obtain the mid-frequency (Mid-Freq) value; and those at 4 and 6 kHz, to obtain the high-frequency (High-Freq) value. In the present study, the average hearing threshold (AHT) was calculated as pure tone average at 4 frequencies (0.5, 1, 2, and 3 kHz). HL was defined as an AHT >25 dB.

We performed analyses using 2 cohorts. The total cohort was defined as all 1,009 participants. To balance the baseline characteristics between participants with and without OP, we estimated a propensity score, which is a predicted probability of OP in all participants using a logistic regression model constructed to include the variables of age, BMI, DM, HTN, ALP, and VitD. Using the Greedy match algorithm, we created a propensity score matched pair without replacement (1:1 match). We decided the weight order or priority according to standardized mean difference between the 2 groups. The standardized mean difference between participants with OP and those without OP in terms of age, BMI, ALP, DM, HTN, and VitD level was 0.687, 0.296, 0.204, 0.072, 0.062, and 0.047, respectively. Before the groups were matched, the propensity standardized mean difference was 0.755. After the groups were matched, the propensity standardized mean difference was 0.103. Although the standardized mean difference in terms of DM and HTN was rather low between the groups, these comorbidities are well-known risk factors for hearing impairment. The ALP and VitD levels may be associated with pathologic changes in BMD. Therefore, these factors were included as confounding factors. The propensity score matched (PSM) cohort was defined as the cohort including participants diagnosed with OP and participants without OP.

Data were analyzed using SAS ver. 9.4 (SAS Institute Inc., Cary, NC, USA). Categorical variables were expressed as numbers and percentages, and continuous variables were expressed as mean±SD or standard error. Either the Pearson chi-square test or the Fisher exact test was used to analyze categorical variables. For continuous variables, the t-test was used to compare the means. Within the PSM cohort, we compared the difference in variables between participants with OP and each matched participant with OP by using the paired t-test. Linear regression analysis was performed to assess independent predictors of hearing thresholds. Logistic regression analyses were used to estimate the odds ratios and 95% confidence intervals (CIs), which were used to determine the relationship between the OP and HL.

Three statistical models were developed where model 1 was unadjusted, model 2 included age, and model 3 included age, BMI, alcohol intake, smoking habit, DM, HTN, HDL cholesterol, TG level, VitD, and ALP. Multivariate analyses using analyses of covariance, multiple linear regression, or multiple logistic regression were used determine the independent predictors of HL. The level of statistical significance was set at a P-value of <0.05.

There were 776 and 233 participants in the groups diagnosed without and with OP, respectively (Table 1). Age and ALP level were higher in participants diagnosed with OP when compared to participants without OP. In contrast, the BMI was lower in participants with OP than in those without OP. No significant differences in occurrence of DM, HTN, or i-PTH, VitD, HDL cholesterol, or TG levels, alcohol intake, or smoking behavior were observed between the 2 groups.

For propensity score matching, 233 pairs were selected from the 776 participants without OP. The estimated distribution of propensity scores was similar after matching, and no significant differences were identified in characteristics corresponding to the presence or absence of OP (Table 1, Fig. 1). In addition, we compared the difference between participants with OP and each matched participant without OP. Mean difference±SD was 0.16±4.65 for age (P=0.603), 0.19±3.69 for BMI (P=0.438), 1.70±98.57 for ALP level (P=0.793), 0.79±9.15 for VitD level (P=0.188), and 1.63±33.29 for i-PTH level (P=0.456). The participants with OP did not significantly differ from each matched participant without OP in terms of age, BMI, ALP, VitD, and i-PTH levels.

In the total cohort when using statistical model 1, the Low-Freq, Mid-Freq, High-Freq thresholds, and the AHT were higher in the participant group with OP than in participants without OP (Table 2). However, using statistical models 2 and 3, no significant difference in the four hearing thresholds was identified between the 2 groups. In the PSM cohort, the Low-Freq hearing threshold in the groups without OP and that with OP was 20.0±10.4 and 20.2±10.9 dB (P=0.795), respectively; Mid-Freq hearing threshold was 21.8±11.0 and 21.6±12.0 dB (P=0.896), respectively; High-Freq hearing threshold was 35.1±15.1 and 35.7±15.8 dB (P=0.686), respectively; and AHT was 21.3±10.2 and 21.3±11.2 dB (P=0.968), respectively. In the PSM cohort, statistical model 1 also yielded similar results (Fig. 2).

In the total cohort, the proportion of participants with HL in the group without and with OP was 20.4% and 30.9%, respectively (P=0.001). Logistic regression indicated that, in model 1, participants with OP had a 1.749 increased risk for HL (95% confidence interval [CI], 1.260 to 2.429; P=0.001) when compared to participants without OP. However, those with OP had a 1.164 (95% CI, 0.818 to 1.656; P=0.398) and 1.128 (95% CI, 0.835 to 1.724; P=0.323) increased risk in models 2 and 3, respectively. A significant HL risk was not observed in participants with OP.

In the PSM cohort, the proportion of participants with HL in the groups without and with OP was 30.5% and 30.9%, respectively (P=0.920). Logistic regression using statistical model 1 indicated that participants with OP had a 1.020 (95% CI, 0.688 to 1.513; P=0.920) increased risk when compared to the participants without OP.

In the total cohort, using statistical model 1, LS and FN T-scores were inversely associated with changes in the Low-Freq, Mid-Freq, High-Freq thresholds, and the AHT. However, using statistical model 2, LS T-score was inversely associated only with changes in the Low-Freq threshold (Table 3). Using statistical model 3, there were no significant associations among LS or FN T-scores and changes in the four hearing thresholds.

In the PSM cohort, the LS T-score was inversely associated with changes in the Low-Freq threshold, and the FN T-score was inversely associated with changes in the Low-Freq, Mid-Freq thresholds, and AHT. Using statistical model 2 with the PSM cohort, LS and FN T-scores were only associated with changes in the Low-Freq threshold. In contrast, using statistical model 3 with this cohort, no significant associations among LS or FN T-scores and changes in the four hearing thresholds were identified.