Occupational noise exposure and its association with incident hyperglycemia: a retrospective cohort study

Background: Epidemiological studies have demonstrated the association between noise exposure and diabetes, but few studies have reported the relationship between noise frequency components and fasting blood glucose. This study investigated the associations between noise levels, frequency characteristics, and the incident hyperglycemia. Methods: An industry-based cohort of 905 volunteers was enrolled and followed-up from the data of rst employment to 2012. Personal noise levels and octave-band frequencies of environmental noise were measured systematically in 2012 to classify subjects’ exposure retrospectively. Cox regression models were applied to calculate the relative risk (RR) of hyperglycemia by continuous and categorical noise-exposure and frequency-component levels, adjusting for potential confounders. Results: Subjects exposed to ≥ 80 A-weighted decibels (dBA) had an increased RR for hyperglycemia of 1.78 (95% condence interval [CI]: 1.11, 2.84) compared with those exposed to <70 dBA. The high-exposure groups at frequencies of 31.5 ( ≥ 33 decibel [dB]), 63 ( ≥ 44 dB), 125 ( ≥ 52 dB), 250 ( ≥ 59 dB), 500 ( ≥ 65 dB), 1000 ( ≥ 68 dB), and 2000 ( ≥ 68 dB) Hz had a signicantly higher risk of hyperglycemia (all p values < 0.05) than did the low-exposure groups, and those exposed at 31.5 Hz had the highest risk (Adjusted RR=1.97, 95% CI: 1.23, 3.16). Per 5-dB increase in noise frequencies at 31.5, 63, 125, 250, 500 Hz, and 1000 Hz were associated with an elevated incidence of hyperglycemia (all p < 0.05), with the highest risk of 1.27 (95% CI: 1.10, 1.47) at 31.5 Hz (p = 0.001). Conclusions: Exposure to occupational noise may an with the the association between occupational noise exposure and an increased risk of incident hyperglycemia. Positive and linear exposure–response relationships have been demonstrated at noise frequency components of 31.5, 63, 125, 250, 500, 1000, and 2000 Hz. The machinery and equipment manufacturing workers exposed to noise levels at 31.5 Hz may have the strongest risk of hyperglycemia. These ndings provide a possible link between noise exposure and cardio-metabolic disease. We recommend future studies to determine the associations between hyperglycemia and octave-band frequencies of occupational noise exposure in the different industries.

diabetes was associated with the 5-dB increase in noise exposure, mainly related to air and road tra c noise [22]. To the best of our knowledge, no studies have been conducted to investigate the relationship between occupational noise exposure and incident hyperglycemia. Furthermore, little is known about the association between noise frequency components. Therefore, this retrospective study aimed to elucidate the relationship between exposure to occupational noise and incident hyperglycemia. We also determined whether there were differences in associations between hyperglycemia and different noise frequency components.

Study population
The detailed procedures to invite cooperating companies were mentioned in a previous study [10]. Brie y, we recruited 1028 volunteers in four machinery and equipment manufacturing companies in 2012. Among them, 2 subjects with a history of diabetes before employment, and 121 subjects followed-up for less than one year were excluded. Finally, we enrolled 905 participants as study subjects in this industry-based cohort. High noise levels were identi ed among workers in the processes of metal cutting, pressing, grinding, sand blasting, polishing, and gear washing. No subjects have the shift work.
The Institutional Review Board of the School of Public Health, China Medical University reviewed and approve this study.
Written informed consent was acquired from each participant.

Hyperglycemia cases
We required all participants to fast overnight before blood sampling during the annual health examinations in 2012. Venous blood samples collected by trained nurses were used to perform blood glucose measurements by a standard glucose oxidase method. Hyperglycemia was de ned as a positive response to the following questions: "have you been diagnosed with a diabetes by a physician?" or "do you use antidiabetic drugs, and was the treatment initiated after your employment start date with the current company?", or if upon assessment in 2012 the fasting blood glucose (FBG) level was ≥100 mg/dl [23]. In addition, height, body weight, systolic blood pressure (SBP), diastolic blood pressure (DBP), total cholesterol level, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, and triglyceride levels were measured for all subjects. A trained nurse applied an automated sphygmomanometer (Ostar model P2; Ostar Meditech Corp., Taipei, Taiwan) to measure each subject's bilateral blood pressure in sitting position, and the mean of 2 measurements was represented an individual's blood pressure. We de ned hypertensive cases as he or she had one or more of the following criteria: a diagnosed hypertension by a physician; taking antihypertensive medicine; a SBP of ≥140 mmHg; a DBP of ≥90 mmHg.
Potential risk factors related to hyperglycemia or diabetes were recognized by using a self-administered questionnaire.
Demographic characteristics, lifestyle habits, a family history of diabetes, working activity, and the use of hearing-protection devices were collected and de ned speci cally to avoid information bias [10,24]. Working activity was considered each subject's time of sitting, walking, lifting heavy objects during working periods and the distance walked between the workplace the home that was further categorized into high and low levels based on the cut-off point of 10 in a scoring system [25].

Subjects' follow-up
We applied employment personnel records from the four companies to obtain the date of rst employment for each subject, and assigned this date retrospectively as the starting time to be followed up. The end of the follow-up time was established as either the date to be diagnosed as a diabetes case by a physician, the date of antidiabetic medication initiation, or the date on which blood glucose was measured in December 2012.
Occupational noise exposure evaluation and frequency component analyses The processes of noise exposure evaluation and frequency component analyses were described in detailed in a previous study [10]. Brie y, we conducted a walk-through survey and combined the workplace information to identify different numbers of similar exposure groups (SEGs) for each participating company. Workers assigned to the same SEGs showed similarities in types and frequency of tasks, agents and processes involved, and in the way of performing tasks [26].
We used a personal noise dosimeter (Logging Noise Dose Meter Type 4443, Brüel & Kjaer, Naerum, Denmark) to automatically report 5-minute continuous equivalent sound levels (Leq) with the unit of A-weighted decibel (dBA) during working periods (0800-17:00). The total amount of 96 values in 5-minute Leq (except from 12:00-13:00 pm) was used to calculate one value of 8-h time-weighted average noise levels for each SEG. Before conducting noise measurements, we calibrated this dosimeter with a sound-level calibrator (Type 4231, Brüel & Kjaer, Naerum, Denmark) and setup its determining range between 50-120 dBA for all SEGs.
In addition, an octave-band analyzer (TES-1358, TES Electronic Corp., Taipei, Taiwan) was used to record 5-minute continuous Leq with the unit of decibel (dB) at frequencies of 31.5, 63, 125, 250, 500, 1000, 2000, 4000, and 8000 Hz during the monitoring periods. We applied the total amount of 96 values in 5-minute Leq at each frequency to calculate one value of 8-h time-weighted average noise levels for a speci c frequency component. The analyzer was calibrated by a sound-level calibrator (TES-1356, TES Electronic Corp., Taipei, Taiwan) before noise measurements. The 8-h time-weighted average Leq and its octave-band frequencies were collected by two occupational hygienists to allocate speci c levels of environmental noise (dBA) and octave-band frequencies (dB) for each SEG in the four companies.
Because the regulatory workplace monitoring from these companies showed no signi cant difference in noise levels within the last 10 years (82.89.1 dBA vs. 82.08.3 dBA), we assumed personal noise levels and the frequency spectrum of occupational noise to be equal over the employment duration. Subjects were divided into high-exposure, medium-exposure, and low-exposure groups based on noise exposure assessment. The cut-off value of 80 dBA was selected to classify eld workers into the high-exposure (≥80 dBA) and medium-exposure (<80 dBA) groups because this was the median value in the distribution of personal noise exposure. O ce workers were chosen as the low-exposure group in the present study. We used the same approach to divide participants into the high-exposure ( eld workers exposed to ≥ the median) and mediumexposure ( eld workers exposed to < the median) groups to compare with the low-exposure (o ce workers) group because of the observation of large variations in median noise levels and exposure ranges at nine frequencies as shown in the Supplemental Figure S1. Additionally, we used per 5-dBA increase in personal noise exposure and per 5-dB increase in octaveband analyses of environmental noise to investigate associations with hyperglycemia.

Statistical analysis
The Shapiro-Wilk test was used to determine the normality of continuous variables. The Kruskal-Wallis test was used to perform multiple comparisons of continuous variables with non-normal distribution, among the three groups. We also applied the Chi-square test and Wilcoxon rank sum test to recognize the differences in dichotomous and continuous variables between the three groups, respectively. In addition, non-parametric Spearman correlation coe cients were estimated to investigate the correlation between individual noise exposure, environmental noise levels, and octave-band frequencies of environmental noise in the workplace.
Because this retrospective cohort study only obtained one measurement of FBG in December 2012, each participant's baseline FBG was established at that measured at the time of employment. The average length of time between baseline nonhyperglycemia FBG and follow-up FBG measurements was 7.5 years (median: 5.3 years; IQR: 7.9 years). We summed the cases of hyperglycemia identi ed by the questionnaire (n = 1) or by FBG measurements (n = 118) as the health outcome to conduct the Cox proportional hazard regression analyses. The relative risks (RRs) with 95% con dence intervals (CIs) were calculated to compare the differences in incident hyperglycemia with different groups. A basic model (Model 1) was rst established to include age and sex for biological plausibility, and two dummy variables of exposure groups. Later, we extended this model to cover 2 variables (triglyceride level and hypertension) that were signi cantly associated with the incident hyperglycemia in simple Cox regressions (Model 2). The nal model (Model 3) was set-up to include all variables in Model 2 and one confounder (the use of hearing-protection devices). We excluded the employment duration in the nal model due to its high correlation with age in the present study (Spearman's correlation coe cient was 0.648, p < 0.001). The educational level and working activity were not included because both variables were not signi cant in simple Cox regressions. The strati ed analyses were used to determine the effect modi cation of selected demographic characteristics and to test the interaction between high-exposure and low-exposure groups. We applied the SAS standard package for Windows version 9.4 (SAS Institute Incorporation, Cary, North Carolina, USA) to analyze data and set the signi cance level at 0.050 for all tests. The Bonferroni correction (i.e., the statistical signi cance level is decreased in proportion to the number of comparisons made (i.e. n = 2; p < 0.025)) was also used to avoid signi cant results in multiple comparisons purely by chance rather than real differences existing between groups.

Results
The demographic characteristics of the three study groups are presented in Table 1. Signi cant group-differences in mean age, employment duration, and SBP, and in the proportions of male sex, high educational levels, current smokers, regular exercisers, high working activity, and workers using hearing-protection devices were observed (all p < 0.05). The high-and medium-exposure groups had signi cantly higher means in SBP and higher proportions of male sex, current smokers, and high working activity, but lower means in employment duration and the proportions of high educational levels and regular exercise than those in the low-exposure group. In addition, workers in the high-exposure group were more likely to use hearingprotection devices than did the medium-and low-exposure groups (both p < 0.05).
Supplemental Table S1 shows the correlations between personal exposure, environmental levels, and octave-band frequencies of workplace noise. Personal noise levels correlated signi cantly with environmental levels and all octave-band frequencies (all p < 0.050), and higher correlations (correlation coe cients > 0.810) were observed at frequencies of 250, 500, and 1000 Hz.
Personal noise levels and octave-band analyses of environmental noise pertaining to different groups are presented in Table  2. Signi cantly higher mean levels of environmental and personal noise exposure were observed in the high-and mediumexposure groups compared with those measured in the low-exposure group (both p < 0.05). In addition, the high-and mediumexposure groups had signi cantly higher averages of noise levels at all octave-band frequencies than those measured in the low-exposure group (all p < 0.05).
The mean FBG and the RR of hyperglycemia for the three groups were shown in Table 3. A signi cant difference in FBG between groups was observed (p < 0.05). Both high-exposure and low-exposure groups had signi cantly higher mean FBG values than did the medium-exposure group (both p values < 0.05). Table 4 presents the association between occupational noise exposure and risk of incident hyperglycemia. Workers exposed to 80 dBA had an increased RR for hyperglycemia of 1.78 (95% CI: 1.11, 2.84) compared with those exposed to <70 dBA, before and after the Bonferroni correction. A direct exposure-response association was found between noise exposure and risk of hyperglycemia for all the three groups (Adjusted RR [ARR] = 1.28; 95% CI: 1.04, 1.59; p = 0.023). Sex was the only parameter found as the effect modi cation variable for comparisons between the high-exposure and low-exposure groups, as shown in Supplemental Figure S2 (p = 0.016). Women were more susceptible to an increased risk of incident hyperglycemia than were men (ARR = 4.77; 95% CI: 1.87, 12.17; p = 0.001). Figure 1 shows the risk of incident hyperglycemia according to octave-band frequencies of workplace noise exposure by group. High-exposure groups at frequencies of 31.5, 63, 125, 250, 500, 1000, and 2000 Hz had signi cantly increased risks of incident hyperglycemia compared with low-exposure groups, before the Bonferroni correction (all p < 0.050), but only those at frequencies of 31.5, 1000, and 2000 Hz had signi cant results after the Bonferroni correction (p values < 0.025). The strongest association was observed at 31.5 Hz. Participants exposed to 36.7 ± 3.1 dB at 31.5 Hz had an increased RR for hyperglycemia of 1.95 (95% CI: 1.26, 3.00) compared with those exposed to 25.4 ± 4.6 dB at the same frequency, after the Bonferroni correction (p = 0.003).
The risk of incident hyperglycemia according to a 5-dBA increase in personal noise levels and 5-dB increase at octave-band frequencies is shown in Figure 2. Five-dB increases in environmental noise levels at frequencies of 31.5, 63, 125, 250, 500, and 1000 Hz were associated with the incidence of hyperglycemia (all p < 0.05), with the highest risk of 1.27 (95% CI: 1.10, 1.47) at 31.5 Hz (p = 0.001). Only the sex variable was observed to modify the association between exposure to occupational noise and the incident hyperglycemia, as shown in Supplemental Table S2 (p = 0.025). A 5-dBA increase in occupational noise was associated with a 1.48-fold increase in risk of incident hyperglycemia among women (95% CI: 1.12, 1.96; p = 0.005). Discussion Subjects exposed to occupational noise levels 80 dBA had a signi cantly higher risk of hyperglycemia compared with those exposed to <70 dBA before and after the Bonfferroni correction. We also observed a signi cant exposure-response relationship among the high-, medium-, and low-exposure groups (ARR = 1.28; 95% CI: 1.04, 1.59; p = 0.023). These results are consistent with ndings in a population-based cohort study of 57,053 residents, which found a signi cant and increased risk of incident diabetes (ARR = 1.11; 95% CI: 1.03, 1.19) per 10-dB increase in road tra c noise [19], and with ndings in a casecrossover study from Madrid (Spain) (2001-2009), which reported a strong association between a rise of 0.5 dBA in nighttime tra c noise at a 1-day lag and a 4.6% risk (95% CI: 1.5, 7.8) of diabetic mortality [20]. The present and previous two studies all support the possibility that noise-induced cardiovascular disease may result from the activation of impaired metabolism which lead to increased blood glucose levels [11,12]. In contrast, a cross-sectional survey did not observe the signi cant association between self-reported occupational noise exposure and diabetes among 23 486 European participants [27]. The inconsistent nding may be due to the more accurate exposure assessment in measured and modeled noise levels [19,20] compared with the subjective reported ones [27]. In addition to noise-induced hearing loss [28, 29] and cardiovascular disease [1][2][3][4][5][6][7][8][9][10], the possibility of developing hyperglycemia by occupational noise exposure should be considered in future studies.
Exposure to noise levels at 31.5, 63, 125, 250, 500, 1000, and 2000 Hz was associated with the incidence of hyperglycemia, and the highest risk was found at 31.5 Hz. Signi cant exposure-response relationships were also identi ed at 31.5 Hz (ARR = 1.39; 95% CI: 1.11, 1.74; p = 0.004), 1000 Hz (ARR = 1.33; 95% CI: 1.07, 1.66; p = 0.011) and 2000 Hz (ARR = 1.35, 95% CI: 1.08, 1.68; p = 0.007). These ndings indicated that machinery and equipment manufacturing workers may be more sensitive to low and medium frequencies to elevate blood glucose. The real reason for such observations is unknown. A cross-sectional study also reported a strong association between diabetes and hearing loss at low-and medium-frequencies [30]. Annoyance and stress caused by the low-frequency noise at work may be the other reason to increase the risk of diabetes [22,31]. Therefore, we recommend conducting additional studies in the future to investigate the association between the frequency spectrum of noise exposure and speci c physiological functions.
We observed 5-dB increases at frequencies of 31.5, 63, 125, 250, 500, and 1000 Hz associated with an increased risk of hyperglycemia that may indicate the noise-induced-hyperglycemia involving multiple pathways in the biological mechanism.
Low-frequency occupational noise exposure (i.e., 31.5, 63, 125, and 250 Hz) may pose hyperglycemia indirectly through responses such as annoyance and the disturbance experienced during activities requiring selective attention or while dealing with high-load information [11,12,[32][33][34]. In contrast, middle-frequency occupational noise exposure (i.e., 500 and 1000 Hz) may cause hyperglycemia directly by repeated and prolonged stimulation of the autonomic nervous and endocrine systems [11,12,35,36]. However, more evidence is required to elucidate the reasons for the association between frequency components of noise exposure and the incidence of hypertension.
The association between occupational noise exposure and incident hyperglycemia was signi cantly in uenced by sex in the present study. Women had a higher risk of hyperglycemia compared with men. One previous study reported no signi cant effect modi cation by sex, but it also found a stronger relationship of hyperglycemia with road tra c noise among women (ARR = 1.11; 95% CI: 1.03, 1.20) compared to that among men (ARR = 1.05; 95% CI: 0.98, 1.13) [19]. Future studies are suggested to consider this modi er for investigating the association between noise exposure and hyperglycemia.
The strength of this study lies in the retrospective cohort design, which was formulated to calculate the observed person-years and distinguish the noise-induced effect after the longitudinal follow up. Therefore, the temporal association between exposure to occupational noise and the incident hyperglycemia limited in cross-sectional studies could be assessed in this study. In addition, personal exposure assessment, environmental noise measurements, and octave-band analyses of workplace noise were conducted to provide a precise and accurate evaluation of noise exposure in a real-world workplace setting.
This study has some limitations that must be mentioned. The major restriction is the retrospective design to generate a healthy-worker effect that may produce the lower proportion of hyperglycemia cases in the high-noise-exposure group or the higher incidence of hyperglycemia in the lower-noise-exposure group. The second is to underestimate the observed personyears because missing information on the noise exposure history before participants began employment for the current company or the diagnosis date of diabetes by a physician. The third is no collection of data on noise exposure out of the workplace during the employment period. Exposure to aircraft and road tra c noise have been reported with the increased risk of diabetes in a meta-analysis study [22]. The fourth is the potential recall bias in lifestyle habits that affect diabetes are only measured in 2012. Finally, participants' information about sleep disturbance and noise annoyance at home is not obtained. Noise-induced sleep disturbance and annoyance are both possible mechanisms to generate the cardio-metabolic effects [22]. This restriction may overestimate the effect of occupational noise exposure on the incident hyperglycemia.

Conclusions
Regardless of these limitations, this study observed the association between occupational noise exposure and an increased risk of incident hyperglycemia. Positive and linear exposure-response relationships have been demonstrated at noise frequency components of 31.5, 63, 125, 250, 500, 1000, and 2000 Hz. The machinery and equipment manufacturing workers exposed to noise levels at 31.5 Hz may have the strongest risk of hyperglycemia. These ndings provide a possible link between noise exposure and cardio-metabolic disease. We recommend future studies to determine the associations between hyperglycemia and octave-band frequencies of occupational noise exposure in the different industries.

Declarations
Ethical Approval and Consent to participate: All participants provided informed consent and the protocol was reviewed and approved by the Institutional Review Board of the School of Public Health, China Medical University (No. 100-03-10-4).

Consent for publication
No applicable.

Availability of data and materials
The datasets generated and analyzed during the current study are not publicly available due the con dentiality agreement with participating companies but are available from the corresponding author on reasonable request.

Competing interests
The authors declare that they have no competing interests.

Funding
The National Science Council, Taiwan (NSC 100-2221-E-039-004) provided the nancial support for this study.     Adjusted relative risk (ARR)a of incident hyperglycemia according to octave-band frequencies of occupational noise exposure for participants. ARR, adjusted relative risk; CI, con dence interval; dB, decibel; Ref, reference (i.e., o cers). a Cox regression model adjusted for age, sex, triglyceride level, hypertension, and the use of hearing-protection devices.