Association between Smoking and Noise-Induced Hearing Loss: A Meta-Analysis of Observational Studies

The purpose of this study was to synthesize the results of previously published observational studies through meta-analysis to clarify the association between smoking and noise-induced hearing loss (NIHL). We searched several databases as of October 2019. Based on the results of heterogeneity analysis (Q statistic and I2 statistic), a fixed effect model (for no heterogeneity; Q test P > 0.1 and I2 ≤ 50%) or a random effects model (for heterogeneity) was used to calculate the pooled odds ratios (ORs). We explored the potential dose-response relationship between smoking and NIHL as well. In total, 27 studies involving 30,465 participants were included. Compared with non-smokers, the pooled OR of current smokers was 2.05 (95% Confidence interval (CI): 1.71–2.46), and of former smokers was 1.11 (95% CI: 1.05–1.18). We found a curve linear association between an increasing number of pack-years (packages/day × smoking years) and risk of NIHL. The dose-response meta-analysis suggested that when the number of pack-years was less than fifteen, the risk of NIHL was increasing, and the highest combined OR was 5.25 (95% CI: 2.30–11.96) for pack-years of fifteen. After fifteen pack-years, the pooled OR had a slow decline. Our study indicated that smoking is a risk factor for NIHL. Current smokers have a higher risk than former smokers, and there is a positive dose-response relationship between smoking and NIHL.


Introduction
Noise-induced hearing loss (NIHL) is chronic and irreversible sensorineural hearing loss resulting from long-term exposure to noise. It affects the daily social life of patients and brings a huge burden on society and economy. According to the WHO, around 466 million people worldwide suffer from disabling hearing loss, and it is estimated that unaddressed hearing loss poses an annual global cost of US$750 billion [1]. About 16% of the world's disabling hearing loss is caused by noise exposure in the workplace [2], and NIHL has become one of the most important work-related diseases around the world. In addition, it is estimated that nearly 600 million workers worldwide have a history of occupational noise exposure [3].
NIHL is related to multiple factors. In addition to occupational noise, other factors (such as organic solvent [4], high-temperature [5], no use of hearing protection device [6], alcohol [7], genes [8], comorbidity [9], etc.) may be independent factors or have a synergistic effect with noise to increase the risk of NIHL. Smoking is a risk factor for many illnesses, and many published studies [10][11][12][13] have suggested that it may also be associated with NIHL. Some toxic and harmful substances like nicotine from tobacco burning may affect hearing. However, smokers are widely distributed all over the world, especially in China, with an estimated more than 300 million people (one third of the total number of smokers worldwide) [14]. As a lifestyle that is one of the leading preventable causes of death but difficult to quit, the association between smoking and NIHL has drawn increasing attention. Although a meta-analysis [15] has concluded that smoking is associated with hearing loss, prior analyses focused on people without occupational noise exposure, and the pathogenesis of NIHL is different from other hearing loss. Therefore, we conducted a meta-analysis of observational studies to assess the relationship between smoking and NIHL in noise exposed workers.

Literature Search Strategy
The meta-analysis was performed in accordance with the PRISMA guidelines [16]. We conducted a literature search in Pubmed, Embase, Web of Science, Scopus, Wanfang, and CNKI databases for studies published in English and Chinese up to October 2019. The search terms were NIHL and smoking with their synonyms (noise induced hearing loss or noise induced deafness or noise deafness) AND (smoke or smoker or smoking or cigarette or tobacco or cigar). We also reviewed the reference lists of retrieved articles for other pertinent papers.

Inclusion and Exclusion Criteria
The inclusion criteria were as follows: (1) the study design was a cohort, case-control or cross-sectional; (2) study population had a history of occupational noise exposure; (3) NIHL was clearly defined as the outcome; (4) study provided odds ratio (OR) or relative risk (RR) with the corresponding 95% confidence interval (CIs). If a study was published in multiple papers, we included only one with sufficient information. Review, conference, or experimental articles were excluded.

Data Extraction and Quality Assessment
Two investigators (X.L. and X.R.) independently extracted the following information from eligible studies: first author, year of publication, country, source of participants, study design, sample size, age, gender, diagnostic of NIHL, smoking information, adjusted OR/RR with 95% CIs, and adjusted or matched variables. Disagreements were solved through discussion.
We used the Newcastle-Ottawa Scale [17] to assess the quality of cohort or case-control studies. The judgement was based on three areas: selection of participants, comparability of groups, and exposure/outcome ascertainment. Scores ranging from 0 to 9 reflect an improvement quality of studies. For cross-sectional studies, an 11 items checklist recommended by Agency for Healthcare Research and Quality (AHRQ) [18] was applied. Articles scoring 0-3 points, 4-7 points, and 8-11 points were classified as low, moderate, and high-quality studies.

Statistical Analysis
OR was used as a measure of the association between smoking and NIHL. Due to the low incidence of NIHL, the reported RR was approximately considered as OR. When smoking status was just divided into smokers and non-smokers, smokers were defined as current smokers. In addition, when OR was reported separately at different smoking levels, we extracted the highest level of results. Two articles [19,20] separately estimated OR and 95% CI in two levels of noise exposure, and they were treated as different studies in the analysis.
Before calculating the overall pooled OR, we used Q test and I 2 statistic to quantify the heterogeneity of studies. If P value for Q test was more than 0.10 and I 2 value was less than 50%, we used a fixed effect model. Otherwise, a random effects model was applied [21]. And we did subgroup analyses according to study design (cohort vs. case-control vs. cross-sectional), gender (both vs. male vs. female), mean age (<40 vs. ≥40), race (Mongoloid vs. Caucasian vs. others), quality of studies (high quality vs. moderate quality), number of adjusting variables (0 vs. ≥1) and publication year (<2010 vs. ≥2010). The race was roughly classified on the basis of the country reported in the study. People living in China, Japan, Malaysia, Indonesia, Brunei, and Nepal were seen as Mongoloid; Caucasians were from the United States (USA), Britain, Italy, Iran, Denmark, Switzerland, and Germany; Brazilian was classified separately.
In addition, a dose-response analysis of pack-years and NIHL was estimated. Pack-years is a measure of the amount of cigarettes a person has smoked over an extended period, which is equal to the number of packages (/20 cigarettes) per day multiplied by the smoking time (/years). For example, smoking one package every day for two years equals to two pack-years. Articles that provided at least three quantitative categories were included in this calculation. Since all studies reported dose in groups, we assigned the midpoint of the group range as the dose value, and for the highest open-ended group, multiplied the lower limit by 1.5 times. We evaluated the potential curve liner relationship between the number of pack-years and NIHL by using restricted cubic splines with three knots (10%, 50%, and 90%) of the distribution [22,23]. We tested whether the coefficient of the second spline is equal to zero to determine whether the relationship is linear or non-linear. A coefficient not equal to zero for the second spline indicates a non-linear relationship [24].
A sensitivity analysis was also conducted to see the influence on the overall result by omitting each study. In addition, we recalculated the pooled OR after omitting those articles with extremely high ORs (>10). We used Begg's rank correlation test and Egger's linear regression test to assess potential publication bias [25,26]. If publication bias was indicated, further trim-and-fill method [27] was used to recalculate the pooled OR. All statistical analyses were performed using R version 3.6.0 (R Foundation for Statistical Computing, Vienna, Austria). For all statistical tests other than Q tests, which have different statistical significance criteria as described above, a two-sided P-value < 0.05 was considered statistically significant.

Literature Search
Based on the database and search terms, we obtained 1541 articles to screen. Through browsing titles and abstracts of papers, we excluded 1467 records. After retrieving and reviewing 74 full articles, we excluded 47 records. Seven studies were excluded because participants in the research didn't have a history of occupational noise exposure. A further 30 studies were excluded because they didn't provide effect size estimate and 95% CIs to calculate the pooled OR. 4 studies were excluded because they used auditory threshold as outcome. Two studies were excluded because results were repeatedly reported in other articles. Four studies not published in Chinese or English were also excluded. Our meta-analysis includes 27 studies [7,[10][11][12][13]19,20,[28][29][30][31][32][33][34][35][36][37][38][39][40][41][42][43][44][45][46][47] in this meta-analysis. Selection details are shown in Figure 1. There were four cohort studies, two case-control studies, and 21 cross-sectional studies.  Table 1 shows the main characteristics of 27 studies included in our meta-analysis. Studies were published between 1987 and 2018. And a total of 30,465 workers were included in our review. Among all included studies, 18 studies were from Asia; 5 studies were from Europe; 3 studies were from North America, and 1 study was from South America. The diagnostic criteria showed differences among these studies; 16 studies were based on speech frequency, and 11 studies were based on high frequency. In addition, the quality assessment scores for the 4 cohort studies were in the range of 6-8, and the average score was 7.25 points. For two case-control studies, the scores were 7 and 8 points. There were 21 cross-sectional studies with the scores ranging from 6 to 10, and the average score was 8.14 points.

Association between Current Smokers and Risk of NIHL
Of the 27 included articles, 29 studies (2 articles reported OR separately for different noise exposure history) assessed the association between current smokers and risk of NIHL. Among the 29 studies, 20 reported a positive relationship between current and risk of NIHL, while nine found no association. Figure 2 shows the results of pooled OR by a random effects model. The pooled OR of NIHL for current smokers was 2.05 (95% CI: 1.71-2.46) with a significant heterogeneity across studies (Q test P < 0.001, I 2 = 87%).

Association between Former Smokers and Risk of NIHL
Six studies (1 articles reported OR separately for different noise exposure history) provided information on the association between former smokers and risk of NIHL. Supplementary Figure S1 shows the results of pooled OR from a fixed effect model. Of the six included studies, only one showed a positive relationship, while others suggested no statistical significance. The pooled OR of NIHL for former smokers was 1.11 (95% CI: 1.05-1.18). No heterogeneity was detected (Q test P = 0.394, I 2 = 4%). Table 2 shows the results of subgroup analysis for current smokers and NIHL risk. Study design, gender, mean age, race, quality of studies, number of adjusting variables and publication year were conducted in the subgroup analysis. Overall, the results for most subgroups indicate a positive relationship between current smokers and risk of NIHL. According to the study design, the main heterogeneity came from cross-sectional studies (Q test P < 0.001, I 2 = 89%), and for cohort studies and case-control studies, there was no evidence of heterogeneity (Q test P = 0.504, I 2 = 0%; Q test P = 0.418, I 2 = 0%). According to gender, the pooled OR was 3.05 (95% CI: 1.90-4.89) for male, 1.50 (95% CI: 1. 28

Sensitivity Analysis and Publication Bias
The results were not significantly different after omitting two studies [37,42] with extremely high ORs. The sensitivity analysis for current smokers and NIHL risk showed that the result was not significantly affected by removal of any one study. Nevertheless, among studies for former smokers, sensitivity analysis hinted that, omitting the study by Dement [11], the pooled OR would change to 1.18 (95% CI: 1.00-1.39) (Supplementary Figure S2).
Supplementary Figure S3 shows an asymmetric funnel plot of studies researching the relationship between current smokers and NIHL. It indicated a potential publication bias. In addition, the Begg's rank correlation test and the Egger's linear regression test both confirmed potential publication bias (P = 0.007; P < 0.001). In view of this, we used trim-and-fill method to recalculate the pooled OR. The result was 1.34 (95% CI: 1.10-1.64), which still indicated the same positive association. Publication bias about former smokers and NIHL was not found by either the Begg's test or the Egger's test (P = 0.091; P = 0.173).

Discussion
In this meta-analysis, we confirmed the hypothesis that smoking is associated with increased risk of NIHL. Both current smokers and former smokers had a higher risk of NIHL than non-smokers. And we found a dose-response relationship between smoking and NIHL.
However, the specific mechanism of smoking and NIHL is unclear. Nicotine and other substances in tobacco may have ototoxicity, damaging cochlear hair cells by increasing carbon monoxide hemoglobin or reducing the volume of cochlear blood flow [48,49]. In addition, experimental studies have found nicotine-like receptors in hair cell, suggesting that nicotine has a direct ototoxic effect on hair cell function [50]. Smoking may be an independent risk factor for NIHL. Some studies [37] found that the combined effect of smoking and occupational noise was comparable to the sum of the independent effects of each factor. However, other studies [47] indicated that smoking and noise might have a synergistic effect on NIHL.
Our study showed that the pooled OR for current smokers was higher than OR for former smokers, indicating that quitting smoking could reduce the risk of NIHL. It may be associated with the dose-response relationship between smoking and the risk of NIHL risk; former smokers have lower exposure to smoke. We performed a dose-response analysis based on eight eligible studies. It showed a non-linear relationship between cigarette intake pack-years and NIHL. The OR and its 95%CI of each dose are always greater than one. The general trend also showed that the risk of NIHL would increase with the smoking dose increasing. The slow decline in the fitting model may be due to the fact that the dose concentration of the included studies was mainly between 0 and 15 pack-years. In the total of 19 dose points, only three points were larger than 15, which made the latter trend less accurate. In addition, the studies included in the dose-response analysis were mainly cross-sectional studies. There might be a healthy worker effect in the population. Therefore, a smoking cessation program is important for workers in a noise exposure environment.
In the subgroup analysis, we found that heterogeneity mainly came from cross-sectional studies. It is known that the results of cohort and case-control studies are more reliable. But there were no differences in results between the three types of studies. The subgroup analysis by gender showed that the positive association between smoking and NIHL was stronger in male than female. Because the smoking group of women was small, we only included one study focusing on female participants. And in general, men smoke more than women. Further, in the working environment, men generally have a higher noise exposure dose and a longer duration. Some studies suggested that men were more likely to suffer from NIHL as well [51][52][53]. According to subgroup analysis of race, the OR of current smoking was the highest in the Caucasian population. It may be attributed to genetics, and the previous study indicated that white people were more susceptible to NIHL [54]. Since some cited countries are multi-ethnic, and there are some country-level factors (such as country-specific industry standards for allowable noise levels), the method of simply classifying races based on the country is rough, so the results have some limitations. With regard to number of adjusting variables, group 0 didn't show a positive association. It might attribute to those unadjusted variables (such as age) overwhelming the certain effect of smoking.
Egger's linear regression test and Begg's rank correlation test for the study of current smokers and NIHL were both suggestive of publication bias, which may be related to the inclusion of articles included published in Chinese and English only. In the process of paper screening, four studies published in other languages were eliminated. There was no search for unpublished grey documents. To address the potential of publication bias, we conducted the trim-and-fill method to adjust the influence of publication bias. There was no substantial change, suggesting that the result was not affected by this bias.
Despite previous meta-analysis [15] suggesting that smoking may increase the risk of hearing loss, this study was the first meta-analysis study to explore the relationship between smoking and NIHL. In our study, we focused exclusively on workers with a noise exposure history. Although there was high heterogeneity, subgroup analysis and sensitivity analysis suggested that our results were robust. And a dose-response analysis of pack-years and NIHL was also carried out to assess the dose-response relationship between them.
There are still some limitations in the present study. First, due to the lack of cohort and case-control studies, we added cross-sectional studies. However, cross-sectional studies have a selection bias due to the defects of its design, so that the results are not as reliable as the other two study designs. Secondly, the diagnostic criteria of NIHL were entirely different in each study. Some chose high-frequency (3 kHz) hearing loss as the standard, and some were based on speech frequency (−2 kHz) hearing loss. Moreover, the frequency selections of hearing thresholds were varied as well, which might lead to the heterogeneity. Third, there are also significant differences in the correction of factors that may affect the relationship between smoking and NIHL. The previous studies may not have adjusted all of the confounding factors similarly, such as heredity, with a great effect on NIHL. Fourth, in the dose-response analysis, few of included studies had high dose results, which affected the stability of the rear part of the curve.
According to the shortcomings of our research, we hope to have more long-term follow-up prospective cohort studies to explore the relationship between smoking and NIHL to further confirm our conclusions. They also need uniform diagnostic criteria. In addition, the synergistic effect between smoking and noise interests us. We look forward to seeing a subgroup analysis of diverse occupational noise exposure history. Based on the available information, we have reason to believe that smoking is a risk factor for many diseases, and it can affect workers' hearing health. Relevant departments can provide some smoking cessation programs in the occupational noise environment, especially for men. The Government can also reduce tobacco consumption by increasing tobacco taxes.

Conclusions
In summary, our study indicated that smoking is a risk factor for NIHL. Quitting smoking can reduce the risk of NIHL. There is a non-linear dose-response relationship between the number of smoking pack-years and NIHL. When the dose is less than 15, the risk will add over the increase of pack-years.
Author Contributions: Review conception and design, X.L. and X.R.; Acquisition of data, X.L. and X.R.; Data-interpretation, X.L., X.R., Z.W. and A.L.; Data analysis, X.L.; Drafting of manuscript, X.L.; Critical revision, X.L., X.R., Z.W. and A.L. All authors have read and agreed to the published version of the manuscript.

Conflicts of Interest:
The authors declare no conflict of interest.