Study population

The Primary Prevention Study (PPS) is a population-based cohort study from Gothenburg, Sweden.13 The initial study population was a random sample of 10 000 men born between 1915 and 1925. Of those, 7494 men (75% of the sample) participated in screening examinations between January 1970 and March 1973.9 ,14 Three years later, in 1974–1977, a follow-up investigation involving 7133 men was carried out. In the present study we used data from the follow-up investigation, because there were complete occupational data from this follow-up. All participants gave their informed consent to participate in the study and it was approved by the Ethics Committee for Medical Research at Gothenburg University.

Occupation at baseline (1974–1977) was classified at three-digit level according to the Nordic Classification of Occupations, NYK-74.15 For assessing psychosocial workplace exposure, we used a previously published job-exposure matrix (JEM).16 ,17 This JEM was developed in the late 1970s based on information from large Swedish population surveys where around 12 000 randomly selected subjects aged 25–74 years were classified according to psychological demands and decision latitude. The JEM provides separate estimates of demand and control for 261 occupations separated into gender and age (25–44 and 45–74).

Psychological job demands were explored with four items and decision latitude was explored with seven items. All items were scored using a scale,1–10 with a score of 10 indicating high psychological demands or high decision latitude, respectively. Each subject was assigned a score based on occupation and age. The scores were then dichotomised into high and low, using the median of the distribution as cut-off. Combining demand and control with the median cut-offs divides the participants into four categorical quadrants: high strain (high demand–low control), active (high demand–high control), passive (low demand–low control) and low strain (low demand–high control).

At baseline, information was collected about age, country of birth (Sweden/other), body mass index (BMI), systolic blood pressure, diastolic blood pressure, use of antihypertensive medication (yes/no), history of diabetes (yes/no), coronary heart disease (yes/no), atrial fibrillation (yes/no), stroke (yes/no), smoking and socioeconomic status (white collar or blue collar worker) as previously described.14 ,18 Hypertension was defined as systolic blood pressure >140 mm Hg or diastolic blood pressure >90 mm Hg or receiving antihypertensive treatment. An ECG was registered.

Based on unique personal identification numbers, the subjects were matched with the Swedish national register on cause of death and the Swedish hospital discharge register. The hospital discharge register has operated on a nationwide basis since 1987, while all discharges from Gothenburg hospitals have been entered in the national register since 1970 (except for 1976, because of a legislative change for that year). The International Classification of Disease (ICD) codes listed in the registries were used to identify events with atrial fibrillation (both non-fatal and fatal events) during the entire follow-up period. The eighth version of the ICD code was used until 1986, ICD-9 was used from 1987 to 1996, and ICD-10 was used from 1997 onwards. Atrial fibrillation was defined as ICD codes 427.92, 427D or I48.

Subjects with a history of coronary heart disease or stroke at baseline were excluded, as were men with a previous hospitalisation discharge code of atrial fibrillation or an ECG showing atrial fibrillation, resulting in a study population of 6035 subjects (table 1).

Statistical analysis

Descriptive statistics are presented as percentages or mean values with SDs. The SAS statistical package (V.9.3) was used for most of the analyses. The data were analysed with Cox regression models using SAS (PROC PHREG). The proportional hazards assumptions were investigated using tests based on cumulative sums of martingale residuals,19 and were found reasonable for all models. Hospital care or death (whichever came first) from atrial fibrillation were events and time was measured as months since baseline. Subjects dying of other causes were censored at the time of death. The observation period stopped at the age of 75, but a separate analyses was performed for age <65 years. In the full model, HRs were calculated using high strain based on the job demand–control model with adjustments for age, smoking, BMI, hypertension, diabetes and socioeconomic status at baseline. There was also one model with 5 years of latency time, that is, excluding subject with any event occurring during the first 5 years during follow-up. Smoking was categorised as never-smoking, former smoking, starter during follow-up, current smoking and unknown smoking habits. There was also one model using all four categories of the job-strain model using low strain as the reference category, as well as models with demand and control as continuous variables. In the model with the continuous variables, the interaction between demand and control was tested using a multiplicative term included as a covariate.20