In the occupational setting, noise is primarily regarded as a hazard to hearing sensitivity. However there is increasing evidence of “non-auditory” health effects including cardiovascular disease, and particularly hypertension.1 It is hypothesised that the pathological mechanism involves a dysfunctional stress response: the auditory apparatus (which is “hard-wired” to the sympathetic nervous and neuroendocrine systems) responds to sound stimuli through reflexes that are manifested by cardiovascular responses and by changing levels of catecholamines, adrenaline, noradrenaline and corticosteroids. When the pathways are continually excited by chronic or repetitive noise stimuli these otherwise normal, transient, responses may become pathogenic—for example, leading to a permanent upward resetting of baroreceptors (“vascular autoregulation”) and thence to hypertension.2 3

Epidemiological methods have been used to test various aspects of the biological model. Several studies have demonstrated elevated levels of “stress hormones”, demonstrating that catecholamines and cortisol were higher in exposed populations compared with non-exposed individuals.4 Other studies focused on the sympatho-adrenal medullary axis and shown increased heart rate and elevated blood pressure.5 The authors recently reported on a longitudinal study of heart disease mortality in a cohort of 27 464 sawmill workers and reported elevated risk of acute myocardial infarction (for which hypertension is a major risk factor) among those who were chronically exposed to high levels of noise.6 This finding was replicated in another large longitudinal study the following year.7

Despite a large number of studies conducted over the past two decades, evidence linking noise exposure and hypertension has been inconsistent.3 Several factors might have contributed to this inconsistency: small sample size8; errors in blood pressure (BP) monitoring9; differing degrees of ability to control for potential confounders, especially when the study has a small sample size3 10; differences in outcome measures (BP change versus diagnosed hypertension); and differences between exposure assessment strategies such as subjective rating versus personal noise measurement.

In this study we examined the hypothesis in a novel way, examining hypertension incidence among a cohort of sawmill workers for whom we had quantitative exposure noise exposure data and examining health using hospital discharge and physician visit records.

METHODS

Participants

The study population consisted of 10 681 male subjects who were employed in production (for example, sawyers, graders, forklift drivers) and maintenance jobs (for example, welders, millwrights, electricians) for at least one year between 1950 and 1995 in one of 14 British Columbia sawmills, and who were alive on 1 April 1991 (the date electronic health records began).11 Health outcome data were obtained by linking the cohort to British Columbia provincial health administration data through the British Columbia Linked Heath Database. It includes data on hospital discharges and physician (outpatient) visits. It has excellent coverage, because of British Columbia’s publicly funded provincial healthcare system, allowing near complete follow-up post-employment.12 Our analyses were restricted to males as the number of females was small (n = 151) and they were not represented in the highly exposed categories.

Cases were defined by: (1) death or (2) hospital admission or (3) having had three doctor visits within 70 days, all using ICD-9 codes 401–405. Criterion 3 was based on British Columbia clinical practice guidelines.13 For subjects who had more than one hypertension “event” only the first diagnosis was used.

Follow-up

Follow-up began 1 April 1991 or one year after first employment in a cohort sawmill for subjects who started employment after this date. End of follow-up was defined as either the date when a subject became a case or for non-cases, 31 July 1998 if they were still employed. Those lost to follow-up were handled in two ways. Subjects who had ever linked to the British Columbia Linked Health Database, because of its near complete (98%) coverage of the British Columbia population, were assumed to be alive until 31 July 1998. For those who had never linked, end of follow-up was set to their date of last employment, an approach that maximised follow-up without making unverifiable assumptions.

Exposure assessment

Noise exposure was retrospectively assessed, based on a predictive model created using 1900 personal noise dosimetry measurements from cohort sawmills that were gathered from several sources (regulatory agencies, research and industry data) between 1970 and 1997.6 The model predicted noise exposure for each unique mill/job/time combination (n = 3700) utilising an exposure data matrix.14

Cumulative exposure above 85 dBA was estimated using time-weighted average exposure (in units of dBA×year). Cumulative exposure was determined using logarithmic addition of noise intensity (from L_eq) and duration of employment (T) in a given job j, for jobs 1 to k as follows:

The resulting unit for cumulative exposure metric has a logarithmic scale, consequently a doubling of either sound level or time results in an increase of 3 dBA×year. Since the relation between noise and hypertension was not known, we investigated three “duration of exposure above threshold” metrics, defined for the number of years worked in jobs in which the noise levels exceed thresholds of 85 dBA, 90 dBA and 95 dBA. A reference category of less than 3 years was selected to ensure adequate number of cases in the reference group.

Few subjects in the cohort had exposure levels below 85 dBA for an 8-hour shift average, and using this as lower cut-off level also excluded most non-occupational exposures, thereby removing this source of potential misclassification.

Ethical approval

The study protocol was approved by the behavioural research ethics board of the University of British Columbia.

Statistical methods

Person-days at risk were stratified using life-table analyses software (PCLTAS, NIOSH, Cincinnati, OH, USA). Internal analyses were conducted using Poisson maximum likelihood regression in Stata (Stata Corporation, College Station, TX, USA).

Age and calendar year were entered in the model in 5-year categories and collapsed if cells contained fewer than 10 cases. We also examined ethnicity; South Asian and Chinese subjects were distinguished from “other” (mostly of European descent) using the last name as an indicator.6

RESULTS

The cohort had a mean cumulative exposure of 111.1 dBA×year. Our case definition resulted in 828 cases. Table 1 describes the main characteristics of the study population both at entry in the cohort and at the time a subject became a case. Of the 828 cases, 818 cases originated from outpatient (that is, physician) visits and 10 from hospital visits. Sixty-three subjects eventually died of hypertension, but all had had previous outpatient or hospital contact for hypertension

Cumulative exposure was a strong predictor of risk of hypertension (fig 1A), with the relative risk in the highest exposed population (>115 dBA×year) 32% higher than for the baseline (<95 dBA×year). Cumulative exposure showed a relatively monotonic increasing risk with exposure (p = 0.006). Figure 1B shows hypertension morbidity by duration of exposure above 85 dBA, 90 dBA and 95 dBA thresholds. The p values for exposure-response trends were significant in all three analyses (p values 0.002, 0.006, and 0.036 respectively).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1 Number of cases (above bar), relative risk estimates and 95% confidence intervals for hypertension morbidity. (A) Using cumulative exposure metric; reference group those who are exposed for less than 95 dBA×years. (B) Using duration above thresholds metrics; reference group those who are exposed less than 3 years at threshold level. All analyses adjusted for age, calendar year, both entered in 5-year categories, and ethnicity.

As expected, age was a statistically significant predictor. Ethnic group was also a predictor. Being South Asian (mostly Sikhs in this cohort) had double the risk (RR = 2.1 CI: 1.7 to 2.5) and being Chinese a 30% increase (RR = 1.3 CI: 0.8 to 2.0) compared to “other” ethnicities (mostly white race).

DISCUSSION

We found an elevated risk of hypertension for subjects exposed to high levels of noise over long periods in their working environment. Following adjustment for age, race, and calendar year, we found a monotonically increasing risk of hypertension for cumulative exposure to noise, and to increasing duration of exposure above thresholds of 85 dBA and higher. The relative risks we report are of a similar magnitude to those described in earlier studies.10 However, there is reason to believe that our results may be underestimating true risk. Because we were unable to adjust our exposure estimates for use of hearing protection there is probably a systematic overestimation of exposure. In a previous analysis of noise and acute myocardial infarction in the same cohort, exposure-response associations increased when the cohort was restricted to those who never used hearing protection devices, presumably because non-differential misclassification had been reduced.6

Despite the fact that many earlier studies have shown a positive association, epidemiological evidence is still inconsistent.10 Our study addressed several of the limitations that may have contributed to this inconsistency. This was a longitudinal study, whereas most others have been cross-sectional in design; it was very large, with more than 10 000 subjects; it utilised a quantitative exposure assessment based on personal noise dosimetry; and it utilised objective health outcome measures (hospitalisation or physician visits for hypertension).

As is the case with most retrospective cohort studies, we lacked information on individual risk factors (for example, smoking) and personal attributes, which might act as confounders. However, all participants in the cohort were blue collar, being production and maintenance workers; salaried office staff were excluded from the cohort. Given this homogeneous cohort, with respect to socioeconomic attributes, the likelihood of significant bias due to confounding would have been reduced. Also, to be a confounder, the risk factor must be correlated with exposure as well as outcome, and we had no reason to believe that this was occurring. For example, with regard to smoking, we were able to look at smoking rates in a small, randomly selected subset of the cohort (6%) interviewed in 1997, and their smoking rates were similar across all noise-exposure groups. Furthermore, when we examined risk for lung cancer with increasing noise exposure we found no relation; a relation would be predicted if smoking was confounding the noise-cardiovascular disease association. We did not have exposure data for work outside of cohort mills, which may have resulted in exposure misclassification. However the workforce was fairly stable in these mills, and few jobs outside the sawmill would be as noisy as sawmill work where the average noise level is about 91 dBA14; thus assuming them “unexposed” is not unreasonable, given our threshold of 85 dBA for “exposed”. Other pertinent occupational exposures were not considered to be potential confounders because they were unlikely to be correlated with noise (for example, shiftwork) or, with respect to dust, because the particle size distribution was much larger than the ultrafine range associated with cardiovascular disease in other studies.

We were concerned that our case definition may have been too specific; we therefore performed sensitivity analyses comparing two more relaxed definitions: three outpatient visits in a one-year period (for example, a patient who presents borderline blood pressure); and (as a midpoint) where subjects had had three outpatient visits in six months. While the number of cases increased slightly as the criteria were relaxed (828, 1117 and 1237 cases for 70-day, 6-month and 12-month inclusion, respectively) the models behaved identically.

We found that risks varied substantially by ethnicity. Our results showed that the Sikhs were at double the risk for hypertension compared with the “other” (mostly European) subjects, our reference group. Sikhs have been previously shown to be at increased risk of cardiovascular disease.15

The results of this study suggest that chronic noise exposure, even at levels deemed safe by many jurisdictions (85 dBA), may be an important risk factor for hypertension, a leading cause of cardiovascular diseases. Therefore, the health impact of noise not only extends beyond the traditionally recognised noise-induced hearing loss, but may be significant as the vast majority of industrial workplaces have a noise environment above the regulated cut-off level.16