The data come from the Multi-Ethnic Study of Atherosclerosis (MESA), a prospective cohort study designed to investigate the prevalence and progression of subclinical CVDs.⁸ Between 2000 and 2002, participants were recruited from six US communities in six US states (North Carolina, New York, Maryland, Minnesota, Illinois and California). At the time of enrolment, they were 45–84 years of age and free of clinical CVD. The participation rate was 60% among those eligible, and women accounted for 51% of the cohort. The original cohort included a wide range of occupations and four racial/ethnic groups, with minorities oversampled: Caucasians (38%), Chinese American (11%), African-Americans (28%) and Hispanics (23%).

The participants were asked to come to one of the six field centres in their community for a study examination. Since the baseline examination (Exam 1), four follow-up examinations were conducted (Exams 2–5) approximately every 2 years. Subclinical CVD measures were taken at Exams 1, 4 and 5. Only those who provided the data at least at two time points were included in this study (n=3441). Of those, 147 were excluded because they did not provide any occupational information, and an additional 185 were excluded because the information on the measuring location for IMT across different time points was missing. Thus this analysis used a sample of 3109 participants, which represents 66% of Exam 5 participants (mean follow-up=9.4 years, SD=0.48). Compared with those who were excluded in this analysis, included participants were on average 3.8 years younger, more likely to be Caucasian (40% vs 37%) and belong to a professional job category (29% vs 23%).

All study participants provided informed consent. The MESA study protocol was approved by the Institutional Review Board (IRB) in each of the six field centres and the National Heart, Lung and Blood Institute (NHLBI). The protocol for this analysis was approved by the IRB of the National Institute for Occupational Safety and Health (NIOSH).

We used the common carotid artery (CCA) IMT, carotid plaque score and carotid plaque shadowing as our outcome measures.⁹ B-mode ultrasound longitudinal images of the right and left CCA, bifurcation and internal carotid artery segments were recorded on a Super-VHS videotape with a Logiq 700 ultrasound system using the M12L transducer (General Electric Medical Systems, CCA frequency 13 MHz). Video images were digitised using a Medical Digital Recording device (PACSGEAR, Pleasanton, California, USA) and converted into DICOM digital records. The same ultrasound system and digitising equipment were used at each examination, but at Exam 5 the video output was directly digitised using the same recorder settings without a videotape. Trained and certified sonographers from all 6 MESA sites used preselected reference images from Exam 1 to match the scanning conditions of the initial study, including display depth, angle of approach, internal landmarks, degree of jugular venous distension and ultrasound system settings. Ultrasound images were reviewed and interpreted by the MESA Air Carotid Ultrasound Reading Center (UW AIRP, Madison, Wisconsin, USA). Digitised images were imported into syngo Ultrasound Workplace 3.5B reading stations loaded with Arterial Health Package software (Siemens Medical, Malvern, Pennsylvania, USA) for carotid IMT measurement, plaque scoring and assessment of plaque shadowing.

The distal CCA was defined as the distal 10 mm of the vessel. IMT was defined as the IMT measured as the mean of the left and right mean far wall distal CCA wall thicknesses. Carotid plaque burden was defined by the carotid plaque scored as the number of plaques (0–12) in the internal, bifurcation and common segments of both carotid arteries. Carotid plaque was defined as a discrete, focal wall thickening ≥1.5 cm or focal thickening at least 50% greater than the surrounding IMT.¹⁰ Acoustic shadowing was defined as an absence or reduction in the amplitude of ultrasound echoes caused by plaques with high beam, high attenuation.¹¹ The presence or absence of plaque acoustic shadowing was evaluated visually and recorded as a binary variable.

The intraclass correlation coefficient (ICC) for intra-reader reproducibility for mean CCA IMT was 0.99 (ie, only 1% of the variability is due to measurement error). The ICC for inter-reader CCA IMT reproducibility was 0.95. Scan–rescan reproducibility was 0.94. For carotid plaque presence and score, the intra-reader reproducibility was κ=0.83 (95% CI 0.70 to 0.96) and inter-reader reproducibility was κ=0.89 (95% CI 0.72 to 1), both values representing ‘almost perfect’ agreement.¹²

Occupational information was collected in a self-administered questionnaire at Exam 1. Four questions modelled on the US Census occupation questions were asked to determine the participant’s current occupation. In this cohort of older adults, 37% were no longer working. They reported the main job before they stopped working. Responses to open-ended questions were coded by trained personnel at NIOSH using the Census 2000 Occupation Codes. Our sample represented 354 jobs, which were then categorised into seven Census occupational categories: (1) management (48 jobs, n=561), (2) professional (96 jobs, n=900), (3) service (46 jobs, n=446), (4) sales/office and administrative support (58 jobs, n=655), (5) farming, fishing and forestry (1 job, n=1), (6) construction, extraction and maintenance (40 jobs, n=153) and (7) production, transportation and material moving (65 jobs, n=393). Since the latter three categories included a rather small number of participants in this sample, they were combined into one category of ‘blue-collar jobs.’ Occupational information was updated at each subsequent examination if the participant reported a change in the employment situation. During the study period, 8.8% of the participants reported at least one current job that was different from the one reported at baseline, but only a small fraction (3.5%) changed jobs across categories (eg, from service to management) during the study period. Therefore, we used the occupation reported at Exam 1 as a time-invariant variable in this analysis. As a sensitivity analysis, we ran the same models with only those who reported a single occupation throughout the study period.

The 354 Census 2000 Occupation Codes were also used to derive occupational exposures from the Occupational Resources Network (O*NET) V.17, a database developed by the US Department of Labor. It provides detailed descriptive information about over 900 unique jobs. The descriptions were obtained from current job holders and occupational analysts, who provided their ratings of the job on 277 questions (eg, “How often does your current job require you to work outdoors, exposed to all weather conditions?” “To what extent does this occupation allow workers to make decisions on their own?”).¹³ Owing to its comprehensive coverage of jobs and wide range of characteristics measured, O*NET has been used as a job exposure matrix.¹⁴¹⁵ We used the Census 2000 Occupation Codes to connect O*NET data to our sample. For this analysis, we focused on the following O*NET derived characteristics: physical hazard exposure, occupational physical activity, interpersonal stress, job control and job demands.

Physical hazard exposure was the mean score of 7 O*NET items addressing physical hazards traditionally studied as occupational hazards: sounds and noise levels that are distracting and uncomfortable; pollutants, gases, dusts or odors; very hot (above 90°F) or very cold (under 32°F) temperatures; extremely bright or inadequate lighting conditions; high places (eg, working on poles, scaffolding, catwalks or ladders); an environment that is not controlled (ie, without air conditioning); outdoors under cover; and outdoors exposed to all weather conditions. Cronbach’s α for the 7 items was 0.96. A Cronbach’s α >0.70 is considered as an indication of high internal consistency of the items (ie, the items together describe the characteristic, in this case the likelihood of physical hazard exposure, in a reliable manner).¹⁶

For physical activity on the job, we used 3 items: time spent sitting (reverse item), the importance of using arms and legs and moving the whole body in performing the job, and the level of general physical activities needed to perform the job. We calculated the mean of O*NET scaled means for the three variables. Cronbach’s α was 0.86.

For interpersonal stressors, we calculated the mean of 6 items: the importance of resolving conflicts and negotiating with others, frequency of conflict situations as part of the job, dealing with unpleasant, angry or discourteous people, dealing with physically aggressive people, the importance of maintaining composure and keeping emotions in check, and the importance of accepting criticisms and dealing calmly with high-stress situations. Cronbach’s α was 0.88.

For psychological job demands and job control, we used the same items used by Cifuentes et al.^14a Specifically, psychological job demands included four items addressing the ability to shift back and forth between tasks, the ability to concentrate on a task, the seriousness of the error and the importance of being accurate. We calculated the mean of the four items. Cronbach’s α was 1.68. As for job control, we used four O*NET items asking the extent to which the job makes use of workers’ abilities and allows workers to try out their own ideas, to make decisions on their own and to plan their work. Cronbach’s α was 0.97.

Employment status (ie, employed full-time, employed part-time, retired, unable to work/out of work) was also asked at each examination. While 75% of the participants reported no change in employment status over the study period, 11% retired, 4% re-entered the workforce, 4% experienced unemployment and 3% reduced their work from full-time to part-time. Thus, employment status was included in the analysis as a time-varying covariate.

Additional covariates included age at Exam 1, sex, race/ethnicity (Caucasian, African-American, Hispanic, Chinese American), nativity (born in one of the 50 states, outside of the 50 states), family history of heart attack (yes, no, do not know), socioeconomic indicators (ie, education, household income), smoking status (current, former, never), and pack-years for current and former smokers. The information was collected in the self-administered questionnaire at each examination. In addition, during the clinical examination, information was obtained on body mass index (BMI, weight (kg)/height (m²)), systolic and diastolic blood pressure, and total/high-density lipoprotein (HDL) cholesterol ratio. Regular medication was reviewed at each examination, and medication use (yes=1, no=0) for dyslipidemia and hypertension was included in the analysis. Diabetes was assessed by the fasting plasma glucose level: normal (<110 mg/dL), impaired fasting glucose (from 110 to 125 mg/dL) and untreated diabetes (>125 mg/dL).¹⁷ If participants were taking insulin or oral hypoglycaemic medication, we categorised them as ‘treated diabetes.’ All of these traditional CVD risk factors as well as the household income data were updated at each examination; thus, these variables were treated as time-varying covariates.

Because a large body of the literature on occupation and CVD has documented that men and women tend to show different associations between occupational characteristics and CVD,¹⁸ all analyses were conducted separately for men and women. Descriptive statistics by sex and occupational group were calculated for the 3109 participants included in this analysis. In order to examine the association of occupational characteristics with subclinical CVD over time, we modelled repeat measures of subclinical CVD measures as a function of time since baseline in years, occupational characteristics at baseline, and an interaction term between occupational characteristics and time. Models also included a random intercept for each person and an interaction term between baseline age (mean-centred) and time. The following covariates were treated as time-invariant: baseline age, race/ethnicity, nativity, family history, education, field centre. Time-varying covariates included current employment status, income, smoking status, pack-years, BMI, systolic and diastolic blood pressure, diabetes, total/HDL cholesterol ratio, and medication use for hypertension and dyslipidemia. For the IMT models, we also included a covariate for right versus left carotid artery. PROC MIXED was used for common carotid IMT and GLIMIXX was used for carotid plaque score (count variable, Poisson model) and plaque shadowing (binary variable with a high proportion of ‘cases’, Poisson model). Occupation was included as dummy variables (for occupational categories) or in SD units (for the O*NET job characteristic variables). Each of the occupational variables was studied separately.

For all outcome variables, we first estimated the effect of the occupational variable and its interaction with time while only age, sex, race/ethnicity, nativity and family history were included as covariates (Model 1). Then we added other SEP indicators and traditional CVD risk factors (Model 2).