Occupational Noise Exposure and Diabetes Mellitus: A 3-year Retrospective Multicenter Cohort Study

The global prevalence of diabetes has been increasing. However, occupational environmental factors inuencing it have been poorly studied. The effect of occupational noise exposure on diabetes is somewhat controversial. Thus, this study examines the relationship between occupational noise exposure ( ≥ 85 dBA) and diabetes incidence. Participants (n = 58,284) were recruited from a Common Data Model cohort of two hospitals from 2013 or 2014 and were annually followed up for three years. Drug history, clinical history of diabetes, and/or fasting glucose of 126 mg/dL or more were dened as new-onset diabetes. Multivariable time-dependent Cox proportional hazard models and Landmark analysis were implemented to estimate hazard ratios (HRs) and 95% condence intervals (CIs). Pooled HRs and 95% CIs were calculated using the weight obtained through standard error. Of the participants, 4.65% developed diabetes during the follow-up. The nal adjusted pooled HR of Cox models indicated a signicant relationship between occupational noise exposure and increased risk of diabetes (Time-dependent Cox: HR 1.35 [95% CI 1.17–1.57]; Landmark: HR 1.22 [95% CI 1.10–1.35]). There is a signicant relationship between occupational noise exposure and incidence of diabetes. Screening for diabetes, active management, and prevention may be necessary to improve the health of individuals exposed to occupational noise.


Introduction
Diabetes is a globally prevalent disease that is rapidly increasing because of aging and related lifestyle changes 1,2 . Over the past 30 years, the number of people diagnosed with diabetes has more than doubled 3 , and its prevalence was estimated to be approximately 9.3% of the entire global population, which is more than 460 million people 4 . This number may increase to approximately 700 million by 2045 4 . Many studies have elucidated that diabetes is associated with several chronic complications, including ischemic heart disease, stroke, peripheral neuropathy, chronic kidney disease, and heart failure [5][6][7] . Furthermore, diabetes can increase mortality from cardiovascular disease (CVD), which is the leading cause of death in most countries worldwide 8 . In the United States, the total estimated national medical expenditure related to diabetes increased from $188 billion to $237.3 billion between 2012 and 2017. More than 25% of the total estimated medical expenditure associated with diabetes was attributed to the working population. These costs were associated with absenteeism, presenteeism, productivity losses, work disability, and premature mortality 9 .
Given the length of time most workers spend at their workplace, the working environment may be considered a potential in uence or hindrance in the management of diabetes. There are several occupational contributors that increase the risk of developing diabetes, such as sedentary work, limited physical activity, shift work, inadequate time to rest between shifts schedules, and task stressors 10,11 .
Other occupational environmental risk factors such as exposure to noise, heavy metals, and heat are relatively less studied; however, these should be considered as risk factors in the working environment 12,13 .
Noise is an undesirable sound and is one of the most common environmental stressors present in every human activity. It can be divided into environmental and occupational noise. Occupational noise is more severe than noise in the general environment. The National Institute for Occupational Safety and Health (NIOSH) and American Conference of Governmental Industrial Hygienists (ACGIH) established a permissible noise exposure level to prevent excessive exposure in the workplace. Despite the institutions' recommendation that workers should not be exposed to noise above 85 dBA as per the time-weighted average (equal to the sum of the portion of an 8-hour work shift) 14 Observational Medical Outcomes Partnership CDM 28 . Questionnaires that were not matched to the standard vocabulary terms were de ned using KWHE-de ned coding. The distributed CDM allowed data to have the same coding and structure, allowing each institution to run the same analysis independently.
Initially, all workers with baseline health examination data were included in the study (Severance Hospital: n = 24,370, Ulsan University Hospital: n = 60,743). Subsequently, participants in companies where all workers were not exposed to occupational noise exposure (Severance Hospital: n = 2,358, Ulsan University Hospital: n = 9,858), workers who were not followed up for any health examination until 2017 (Severance Hospital: n = 4,258, Ulsan University Hospital: n = 7,023), and workers who reported that they have been diagnosed with diabetes mellitus or take medication for diabetes mellitus, or who have a fasting blood glucose of 126 or higher at the time of the baseline examination (Severance Hospital: n = 1,035, Ulsan University Hospital: n = 2,297) were excluded. After exclusion, 16,719 workers in Severance Hospital and 41,565 workers in Ulsan University Hospital were nally enrolled in the current study.

Outcomes and variables
The primary outcome of this study was to determine the incidence of diabetes among participants. Presence of diabetes was indicated if participants were diagnosed with diabetes mellitus, took medication for diabetes mellitus, or had a fasting blood glucose of 126 or higher at the time of the baseline examination. Blood glucose was measured through a blood test, not Point of Care Testing, and was performed by a trained nurse for each health examination. All participants were required to fast overnight prior to drawing blood during the health examinations.
Occupational noise exposure, which is the rst independent variable, was de ned based on the criteria of Threshold Limit Values (TLV) provided by ACGIH, depending on whether or not the worker was exposed to noise at 85 dBA or more for 8 hours during work 15 . The Korean government announced Article 125 (Working Environment Monitoring), which require government-certi cated industrial hygienists to visit the workplace and conduct a walk-through survey with workers and related o cers using interviews 29 . Noise levels in the work environment are measured in dBA, using a sound level meter conforming to the requirements of the American National Standards Institute Sound Level meters. Information about whether participants were exposed to noise or not in each health examination were extracted from the CDM exposure database.
Covariates were obtained from self-reported questionnaires and measurements of health examination.
Participants responded to a question on smoking: "Have you smoked 100 cigarettes or more in your lifetime?" Based on the responses and current smoking status, participants were grouped as nonsmokers, past-smokers, and current-smokers. According to Asian guidelines for obesity, Body Mass Index (BMI) was categorized as follows: underweight (<18.5), normal (18.5-22.9), overweight (23-24.9), and obese (≥25) 30 . Male participants who reported drinking more than seven drinks per week, and female participants who reported drinking more than ve drinks per week, were classi ed as having a history of drinking. Others were classi ed as not having a history of drinking. Hypertension among participants was de ned based on a diagnosis of hypertension, intake of medicines to control hypertension, and systolic blood pressure of 140 or higher or diastolic blood pressure of 90 or higher. Blood pressure was measured by quali ed nurses using an automated blood pressure monitor. If blood pressure was found to be too high, participants rested for 10 minutes before taking another reading. Participants were grouped based on physical activity: an exercise group that undertook high-or medium-intensity exercise more than twice a week, and a non-exercise group who did not.
In South Korea, chemical exposures-such as to carbon monoxide, nitric oxide, cyanide compounds, antimony compounds, carbon disul de, trichloroethylene, ethylene glycol dinitrate, acetonitrile, methyl chloroform, dichloro uoromethane, dichloromethane, and nitroglycerin-and physical exposure, such as vibration, high-or low-pressure, and night shift, are classi ed as risk factors for CVD by the Korean Occupational Safety and Health Act. 31 As a result, we identi ed cardiovascular-related exposure among participants with any of those factors. Further, number of cardiovascular-related exposure per participant was calculated and used as an adjusting variable. Moreover, experts that specialize in evaluating the work environment, examined all cardiovascular-related hazards.

Statistical analysis
For continuous and categorical data, independent t-tests and chi-squared tests were used to examine differences between baseline health examination data of participants with and without occupational noise exposure. As illustrated in Table 1, which includes the baseline characteristics, participants who were exposed to noise within any period of follow-up were considered as the noise exposure group. The duration from the moment of occupational noise exposure to diabetes incidence was plotted using the Kaplan-Meier method. Using a multivariable time-dependent Cox proportional hazard model to adjust for the immortal time bias, hazard ratios (HRs) with 95% con dence intervals (CIs) of diabetes incidence were estimated 32 . Each participants' health examination data and the time intervals between health examinations were used in time-dependent Cox analysis. Landmark analysis with time-xed Cox proportional hazard models, a method to reduce the immortal time bias, were further performed as a sensitivity analysis 33 . Landmark period was set to one year, which implies that participants exposed to noise within 1 year after the index date were classi ed as the noise exposure group and participants who were diagnosed with diabetes within 1 year after the index date were excluded.
The same statistical method was performed in both cohorts according to the distributed CDM method.
Pooled HRs and 95% CIs of hypertension were calculated using the weight obtained through standard error. All statistical tests were two-tailed and statistical signi cance was de ned as a p-value of less than 0.05. All statistical analyses were carried out with R version 4.0.3's "survival" packages (R Foundation for Statistical Computing, Vienna, Austria).

Ethics statement
The study protocol was approved by the Institutional Review Board of Severance Hospital (IRB: Y-2020-0011) and Ulsan University Hospital (IRB: 2020-03-043), and followed the ethical requirements of the 1975 Declaration of Helsinki. As this study is retrospective in nature, informed consent from the participants was waived by Institutional Review Board of Severance and Ulsan University Hospital.
3 Results  Table  S1. Baseline characteristics of participants at each hospital yielded a similar trend to that of the total cohort, except that exercise history was not signi cantly different between both groups in Ulsan University Hospital.
Kaplan-Meier plots of the proportion of diabetes development among participants in each hospital are shown in Figure 1a and 1b (1a: Severance Hospital, 1b: Ulsan University Hospital). Both hospitals show signi cant difference in the incidence of diabetes between noise exposure and non-exposure groups (p < 0.0001).

Discussion
This study reveals that occupational noise exposure increases the risk of diabetes incidence. This relationship was signi cant in time-dependent Cox models of two hospital cohorts, as well as the pooled cohort. All models in each group showed statistical signi cance in terms of the relationship between occupational noise exposure and diabetes. The relationship was signi cant even after adjusting for potential confounding variables. Further, a Landmark analysis, a method to reduce immortal time bias, was performed as a sensitivity analysis, and the results showed a statistically signi cant relationship between occupational noise exposure and increased risk of diabetes.
Our time-dependent Cox and Landmark analysis models adjusted for several covariates that are established risk factors of diabetes. This includes demographic factors (age and sex), lifestyle factors (BMI, smoking, exercise, and drinking alcohol), and clinical history (hypertension) [34][35][36] . Furthermore, by adjusting for factors of cardiovascular-related exposure, in addition to the existing commonly known variables, it was possible to reduce the bias caused by exposure from various work environments.
There have been several studies that elucidate the relationship between lifestyle factors and improvement of diabetes symptoms. A strategy such as the Diabetes Prevention Program (DPP) in the workplace, for prevention of type 2 diabetes mellitus was promoted to manage costs and improve population health 37 .
In 2002, the DPP research group demonstrated that a 7% body weight loss and 150 minutes of physical activity per week could reduce a three-year incidence of type 2 diabetes mellitus among people with prediabetes, by 58% 37 . The DPP lifestyle intervention has since been developed into a year-long, groupbased program that helps people lose weight in a variety of clinical settings 38,39 . However, there is insu cient research on occupational environmental factors of diabetes, compared with lifestyle factors, and intervention research is especially lacking. Intervention studies on preventing occupational noise exposure have been focused on hearing loss 40 . Considering that a lot of workers are exposed to occupational noise 16 , policies for improving the health impacts of those workers are imperative. The strong relationship between diabetes and occupational noise exposure demonstrated in this study contributes to the recognition of occupational environmental factors as an important risk factor in diabetes incidence. Future studies focused on policies and protection guidelines should be implemented.
Several studies have elucidated the association between severe occupational noise exposure and diabetes; however, their results were mostly negative, and they had limitations. Dzhambov 25 . However, in this study, noise exposure was de ned as being exposed to a very loud noise even once, which is less accurate for stratifying noise exposure. Dzhambov (2015) 27 . However, this study focused on hyperglycemiawhich includes impaired fasting glucose-and did not re ect co-exposure factors and time-varying lifestyle factors. The current study supplemented the limitations of previous studies by using timedependent Cox models and reduced bias in various ways.
In terms of potential mechanism, noise exposure could be a risk factor for diabetes by signi cantly affecting stress or sleep. Noise increases catecholamine synthesis, resulting in insulin resistance and problems with glucose homeostasis, thereby increasing stress 41,42 . Furthermore, noise exposure could result in sleep disturbances, which cause irregular blood glucose and increase in adiposity 43 . According to these theories, just as environmental noise exposure is related to diabetes, severe occupational noise exposure is also considered to have some degree of correlation.
The current study has several strengths. First, various methods were applied to overcome the immortal time bias. A multivariable time-dependent Cox proportional hazard analysis was performed and re ected the variability of lifestyle factors and BMI. The Landmark analysis was also implemented for sensitivity analysis and indicated the same trend of results. Second, using the same statistical method analysis, two hospital cohorts in different regions were included, so that the data showed diversity and could minimize bias caused by speci c regions or institutions. This distributed CDM method could result in a larger sample size by applying the same statistical syntax on different data while maintaining the security of data. Third, to reduce selection bias, the participants of companies with occupational noise exposure were enrolled and noise exposure and non-exposure groups in those companies were compared with each other. Finally, the models were adjusted with number of exposures related to cardiovascular risk, which was not considered in previous studies, along with well-known risk factors of diabetes (demographic characteristics and lifestyle factors).
However, there are some limitations due to the incompleteness of health examination data. First, the health worker effect could in uence the outcome of the current study. This implies that healthier workers survive in the company so that they could be more exposed to harmful factors. The maximum follow-up period was set at three years, which is not exceedingly long, to reduce the health worker effect. Moreover, diabetes is less severe than, for example, cancer or CVD, which implies less probability of retirement due to the disease outcome.
Second, some factors such as hearing disease history, disease treated with high-dose steroids, and the presence of personal protective equipment, which could affect diabetes, were not included in our study. Moreover, information about participants' previous occupational environment was unclear. Further welldesigned studies that overcome this limitation should be implemented.
Third, the exact date participants were diagnosed with diabetes is unclear since the de nition of diabetes outcome was based on the questionnaires and fasting blood glucose. Thus, the follow-up period may have a bias. Nonetheless, hypertension and diabetes are often found incidentally in health examinations, rather than diseases that are diagnosed by visiting a hospital due to severe symptoms. Moreover, although the time of diagnosis is not clear in the case of history of high blood pressure, diabetes, or taking medicine, the degree of duration will be spread randomly between noise exposure group and nonexposure group. Thus, this limitation can be overcome. Lastly, noise exposure was not quanti ed in this study. However, the occupational noise standard of 85 dBA lasting 8 hours or more suggested by the ACGIH was applied. Therefore, it is meaningful to evaluate the health effects of workers exposed to serious severe noise. It is also meaningful to check the dose-relationship through a quantitative noise exposure data cohort in the future.
In conclusion, there is a signi cant association between occupational noise exposure and increased risk of diabetes. Screening for diabetes, active management, and prevention are necessary to improve the health of numerous individuals exposed to occupational noise.
Declarations J.Sim, and S.Kim; Revised the manuscript: J.Lee, S.Lee and J-H.Yoon. All authors have reviewed the manuscript.

Data Availability
The datasets generated during and/or analysed during the current study are not publicly available due to the privacy of the hospital data. supplementarytables.docx