Study population

The Tromsø Study is a prospective cohort study conducted in the municipality of Tromsø, Norway.10 It was initiated in 1974 with the emphasis on epidemiology and surveillance of modifiable risk factors for cardiovascular diseases. In the fourth survey (Tromsø 4) in 1994–1995, all inhabitants 25 years or older were invited and 27 158 persons (77% of the eligible population) participated. Among them, all individuals aged 55–74 years and 5–10% random samples of the other age groups (aged 25–54 and 75–84 years) were invited for a second visit, which included extensive examinations. The selection of participants eligible for the second visit was based on economical and scientific reasons and was performed before they attended the first visit. The 6902 individuals who attended the second visit were randomly allocated based on simple randomisation with computer-generated random numbers to one of two lines of examination, one of which comprised echocardiographic examination.11 The attendance rate was 88%.12 Figure 1 shows that after exclusions, 2844 participants were included. All participants have given informed consent, the Tromsø Study complies with the Declaration of Helsinki and has been approved by the Regional Committee for Medical and Health Research Ethics, the Data Inspectorate and the Norwegian Directorate of Health.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Study population, the Tromsø Study 1994–2012.

Baseline characteristics

Questionnaire data were used to define predictor variables on diabetes (yes/no), antihypertensive medication use (current/previous/never), smoking (current/previous/never), palpitations (yes/no) and prevalent coronary heart disease (CHD) (yes/no). Prevalent CHD was defined as self-reported previous myocardial infarction. Body mass index (BMI) was calculated as weight/height² (kg/m²) and body surface area (BSA) was calculated by Du Bois formula (BSA=(Weight^0.425×Height^0.725)×0.007184). Blood pressure and heart rate were recorded three times with 1  min intervals after 2 min resting, by trained technicians using an automatic device (Dina map Vital Signs Monitor 1846, Criticon). For the analysis, the mean from the second and third readings was used. Hypertension was defined as systolic blood pressure ≥140 mm Hg or diastolic blood pressure ≥90 mm Hg or current antihypertensive medication use.

Plasma creatinine was analysed by a modified Jaffe reaction, but since creatinine-based estimation of glomerular filtration rate (GFR) is better validated for enzymatic creatinine measurements, 111 plasma samples from the 1994/1995 survey were thawed and reanalysed with an enzymatic method (Modular P/Roche). Values were fitted to a linear regression model, and recalibrated creatinine values were calculated for all participants. Estimated GFR was calculated according to the CKD-EPI formula13 and renal failure was defined as having GFR <60 mL/min/1.73 m². Non-fasting total cholesterol was analysed by enzymatic methods with commercial kits and serum high-density lipoprotein (HDL) cholesterol was measured after the precipitation of lower-density lipoprotein with heparin and manganese chloride. All blood sample analyses were performed at the Department of Clinical Chemistry, University Hospital of North Norway.

We calculated CHA₂DS₂-VASc score, which aids in identifying additional stroke risk factors for patients with AF. To calculate the score, the variables used were age (<65:0 for both sexes, 65–74: +1, ≥75: +2), sex (female ≥65: +1), history of congestive heart failure (+1), hypertension (+1), stroke/TIA/thromboembolism (+2), vascular disease (+1) and diabetes mellitus (+1).14 ,15

Echocardiography

Echocardiographic examinations were performed on all participants by two expert cardiologists using a VingMED CFM 750 (VingMED Sound A/S, Horten, Norway) with a combined 3.25 MHz mechanical and 2.5 MHz Doppler probe. The examinations were performed using the standard apical and parasternal long-axis and short-axis views. Standard two-dimensional guided M-mode registrations of anteroposterior LA size, internal dimensions of the left ventricle and wall thickness of the septum and posterior wall were made from leading edge to leading edge convention.16 In this study, LA size was indexed by BSA. Heart failure was defined as left ventricular ejection fraction <0.5.

Atrial fibrillation

All participants were followed with respect to incident clinical AF documented by an ECG before or until date of stroke. The national personal identification number of participants was linked to the diagnosis registry at the University Hospital of North Norway (outpatient clinic included), using the following diagnostic codes: ICD-9 codes 427.0–427.99 and ICD-10 codes I47 and I48. In addition, we performed manual text search for participants without an arrhythmia diagnosis but with a diagnosis of cerebrovascular or cardiovascular events. An independent committee adjudicated all events. Types of AF were categorised into paroxysmal (paroxysmal or persistent), permanent or unclassified. We combined paroxysmal and persistent AF because it was difficult to make absolute decisions about the correct type. All AF types were merged in the multivariable analysis. Participants with transient AF occurring only during an acute myocardial infarction, in connection with cardiac surgery, or in the last 7 days of life, were not classified as having AF. The adjudication of AF is described in detail previously.6 ,17

Reproducibility

A reproducibility study of the echocardiographic data was performed in a subsample of 58 participants by 2 cardiologists. All participants were examined twice with 1-week interval by both observers. Data concerning the reproducibility of the echocardiography are given elsewhere.12 This study shows that the validity of the echocardiographic measurements is good due to the use of a representative sample from a general population with a large age-span and a high attendance rate. The reliability of measurements from independently recorded heartbeats is good, and the reliability is further strengthened by 83% of the examinations being performed by one observer. There was no systematic measurement variability invalidating the data.

Classification of LA size

We indexed the LA size by BSA and categorised into two groups as normal or moderately enlarged (<2.8 cm/m²) and severely enlarged (≥2.8 cm/m²) left atria.

Classification of combined analytical groups

We combined CHA₂DS₂-VASc score, LA size and AF together and made the following groups:

• Group 1=CHA₂DS₂-VASc score=0;

• Group 2=CHA₂DS₂-VASc score ≥1 and normal or moderately enlarged LA (<2.8 cm/m²) and AF;

• Group 3=CHA₂DS₂-VASc score ≥1 and normal or moderately enlarged LA (<2.8 cm/m²) and no AF;

• Group 4=CHA₂DS₂-VASc score ≥1 and severely enlarged LA (≥2.8 cm/m²) and AF;

• Group 5=CHA₂DS₂-VASc score ≥1 and severely enlarged LA (≥2.8 cm/m²) and no AF.

End points

We followed all participants who attended Tromsø 4 for incident cases of stroke, as described previously.18 With the exclusion of subarachnoid haemorrhage, we combined ischaemic stroke, haemorrhagic stroke and unclassified stroke as one common end point. The combined stroke end point was preferred as the result for ischaemic stroke only and the combined end point were similar, but with narrower CIs for combined stroke.

Norway has a unique personal identification system that allows exact matching of population register data. The identification number of participants of the 1994–1995 survey was linked to the diagnosis registry at the University Hospital of North Norway (the outpatient clinic included), the only hospital in this area, and to the National Causes of Death Registry at Statistics Norway. Cases of possible non-fatal and fatal stroke were identified by a broad search for the following diagnosis codes of cerebrovascular disease: International Classification of Diseases (ICD) 8 and 9 codes 430–438, and ICD 10 codes I60–I69 (cerebrovascular diseases). In addition, systematic manual and electronic text searches were performed in the medical records for patients with ICD 8 and 9 diagnosis codes 410–414 and 798–799, and ICD 10 codes I20–I25 and R96, R98 and R99. Adjudication of hospitalised and out-of-hospital events was performed by an independent end point committee based on data from hospital and out-of-hospital medical records, autopsy records and death certificates. For the computation of stroke incidence (excluding subarachnoid haemorrhage), we followed 2844 participants (figure 1) until the date of first stroke or date of censoring due to death, migration or end of follow-up at 31 December 2012, whichever came first. Those who had died or emigrated from Tromsø during follow-up were identified through the Population Register of Norway.

Statistical analysis

We present the characteristics of the study population with and without stroke as means and SDs for continuous variables and proportions of group total for categorical variables. Differences between groups were assessed by t-test, χ² test and Fisher's exact test. To estimate ORs for stroke, we used both age-adjusted and multivariable logistic regression analysis adjusted for smoking, total/HDL cholesterol ratio, BMI, GFR and palpitations. The variables that were included in the final multivariable model were chosen based on known risk factors for stroke or variables significant in our univariable analysis. We have chosen logistic and not Cox proportional hazard regression analysis because the covariate AF is taken from the end point registry without knowledge of when the AF started, although both methods gave similar results. We tested for interactions between LA size and AF and sex, and the p values were 0.162 and 0.004, respectively. We calculated the c-statistic of the model to predict its clinical usefulness for distinguishing high-risk from low-risk participants. In addition, Net Reclassification Improvement (NRI) and Integrated Discrimination Improvement (IDI) were calculated to quantify improvement in model performance. A two-sided p value of <0.05 was considered statistically significant. Statistical analysis was performed using STATA V.12 (Stata, College Station, Texas, USA). The NRI and IDI were calculated with user written program by Liisa Byberg.