Study design and population
This cohort study consists of workers in the Swedish construction industry that participated in health examinations between 1971 and 1993. The analysis was restricted to men with normal blood pressure at the first health examination.
The health examinations were offered freely and were performed at an interval of 2–5 years by the nationwide occupational health service up until 1993 (Toren et al.
2007). All construction workers were invited to the health examinations and at least 80% of these workers participated at least once.
There are 389,132 workers in the cohort and 19,418 of them are female. There were few women with high noise exposure (n = 99) and 30% of the women had no defined exposure level. Therefore, it is not feasible to include women in the analysis.
Workers below 15 or above 67 years of age at their first health examination were excluded. The retirement age in Sweden was 67 years until 1976 when it was decreased to 65 years.
During the health examinations between 1971–1974 and 1988–1993, the workers answered a questionnaire regarding their working conditions and their health status. Data collected from the health examinations included the region where the examination took place and occupational title. Individual data such as height, weight, tobacco use, and blood pressure were also gathered.
The BMI for all participants were calculated by dividing the weight with the square of the participant’s height. The BMI categories used were 18.5 ≤ BMI < 25 and 25 ≤ BMI < 35 kg/m2. Construction workers with BMI over 35 or under 18.5 were excluded, as were workers with missing data for calculation of BMI.
Smoking habits at the first health examination were categorised as “non-smoker”, “previous smoker”, “smoker up to 14 cigarettes per day”, and “smoker with more than 15 cigarettes per day”. If there was no information on smoking habits at the first health examination then this information was retrieved from later examinations if possible.
The blood pressure level was calculated by averaging the blood pressure level at the construction workers’ first health examination and the blood pressure levels at health examinations the following 5 years. If there was no information on blood pressure from the first health examination then the blood pressure measured at a later health examination was used. High blood pressure was defined as 140 mmHg or more in systolic blood pressure, or 90 mmHg or more in diastolic blood pressure.
A job exposure matrix was developed for the 21 different work groups in the cohort. The noise exposure levels were based on a survey of working conditions carried out during the mid-1970s by industrial hygienists. A noise exposure category was assigned for each working group in the cohort. The noise categories were graded on a 1–5 scale. Noise category level 1–3 was at that time considered acceptable noise exposure with noise exposure levels of 45–75 dB(A). At level 4 the noise exposure range was 76–85 dB(A) and for level 5 the noise exposure was above 85 dB(A). For the analysis the noise category levels were categorised as low [≤ 75 dB(A)], moderate [76–85 dB(A)], and high [> 85 dB(A)].
There were 12 different regions in Sweden where the construction workers had their first health examination. These regions were categorised into three main regions of Sweden, representing reference (Götaland), colder (Svealand), and coldest region of Sweden (Norrland). The number of days per year with a daily mean temperature (and standard deviation) below 0 °C from 1970 until 2010, for each region, were on average 67 (21), 80 (18), and 133 (13) days based on data from the Swedish Meteorological and Hydrological Institute (SMHI) (SMHI Open Data [Internet]. Swedish Meteorological and Hydrological Institute (SMHI).[cited 2016 Dec 7]. Available from:
https://www.smhi.se/en).
There was a linkage with the National Cause of Death Register to identify which individual had died (underlying cause) from ischemic heart disease (ICD-8410-412, ICD-9410-412, and ICD-10I21-I25) or cerebrovascular disease (ICD-8430-438, ICD-9430-438, and ICD-10I60-I69) using the unique personal identity number.
Statistical analysis
After excluding workers that did not fulfil the age and BMI inclusion criteria, and after excluding women, there were 359,460 men included in the analysis. Of these men, there were 194,499 with normal blood pressure. Table S1 in online supplementary material presents the exclusion of workers before the analysis.
Negative binomial regression was used to quantify any association between the exposure (noise and cold conditions) and the studied outcomes (myocardial infarction and stroke) among workers with normal blood pressure. Negative binomial regression is an alternative to using quasi-Poisson regression for over-dispersed count data and is similar to Poisson regression. Model performance was assessed using χ2-tests on the model residuals, and normal QQ-plots to spot any deviance from the normal distribution.
For the analysis of myocardial infarction and stroke, person-years were calculated from inclusion to the cohort to event or censoring. A person was censored when they died, emigrated, or turned 85 years old, or at the end of follow-up in 2010.
Potential confounders included in the analyses were age, BMI, and smoking status gathered at the workers’ first health examination. For the analysis of an association between myocardial infarction and stroke mortality with noise exposure we adjusted for cold exposure using region. When analysing the association of myocardial infarction and stroke mortality with cold exposure we used the three main regions as exposure variable and adjusted for noise exposure. To keep track of changes in the association between exposure and outcome, the initial model was unadjusted and then the potential confounders were introduced individually. The final model included all potential confounders. A possible interaction of noise and region (cold exposure) on mortality of myocardial infarction and stroke was analysed by including an interaction term in the final adjusted model.
Relative risk (RR) and 95% confidence intervals were used to determine statistical significance. For the interaction terms, Analysis of variance (ANOVA) was used to determine statistical significance, with α = 0.05. All statistical analyses were performed using R Statistical software (R version 3.4.0).