Methods
The MESA population consists of 6814 men and women who, at recruitment in 2000–2002, were aged 45–84 years and without clinical atherosclerotic CVD, heart failure or atrial fibrillation, and were not undergoing active treatment for cancer.15 Participants were of four races/ethnicities including white, black, Hispanic and Chinese, and recruited from six communities in the USA including Los Angeles County, California; Chicago, Illinois; Baltimore/Baltimore County, Maryland; Saint Paul, Minnesota; Forsyth County, North California and Northern Manhattan/Bronx, New York. After baseline (2000–2002), participants attended up to five additional clinic visits at exam 2 (2002–2004), exam 3 (2004–2005), exam 4 (2005–2007), exam 5 (2010–2012) and exam 6 (2016–2018). Additionally, they took part in regular phone calls which queried recent hospitalisations. The design of the MESA study has been previously published.15
After exclusions for missing baseline RHR (n=48), missing covariates in our primary model (n=258) or missing VTE follow-up data (n=29), a total of 6479 participants were included for our primary analysis for incident VTE (online supplementary figure 1). In a supplemental analysis conducted to help establish potential mechanisms, we also examined the cross-sectional association of RHR with several plasma haemostatic factors and endothelial markers (further described below in this section). Some of these biomarkers were measured in the entire cohort, while others were measured in a random sample of a 1000 MESA participants.16 The baseline characteristics of these ‘MESA 1000’ participants did not differ from that of the entire MESA cohort.
For the primary analysis, we used data regarding RHR and clinical and laboratory covariates that were obtained at the MESA baseline examination. Participants were instructed to fast for at least 12 hours and avoid heavy exercise before the exam. A 12-lead ECG obtained at rest was used to record the RHR of participants at baseline.
Surveys and questionnaires were used to gather information on age, sex, race/ethnicity, education and smoking status. Information on physical activity was obtained by asking questions on time and frequency of physical activity during the week using a survey instrument adapted from the Cross-Cultural Activity Participation Study,17 and total minutes of moderate and/or vigorous exercise per week were calculated. Body mass index (BMI) was measured as the weight (kilograms)/height (kg/m2). Diabetes was based on self-report, medication history as well as a fasting glucose level of ≥126 mg/dL. A medication inventory approach18 was used to determine atrioventricular (AV)-nodal blocker use (defined as participants who were on either beta-blockers, verapamil or diltiazem), aspirin and oral anticoagulants, as well as use of antihypertensive and lipid-lowering therapies. The Chronic Kidney Disease Epidemiology equation was used to calculate the estimated glomerular filtration rate (eGFR).19
Serum concentration of high-sensitivity C reactive protein (hsCRP) and fibrinogen were measured using the Dade Behring Nephelometer II Analyzer System (BNII) (Deerfield, Illinois, USA).11 Interleukin-6 (IL-6) serum concentration was measured using ultrasensitive ELISA (Quantikine HS Human IL-6 Immunoassay, R&D Systems).11 D-dimer was measured using Sta-R analyzer (Liatest D-DI; Diagnostica Stago, Parsippany, New Jersey, USA).20
As previously described, plasma haemostatic factors and endothelial markers were measured in various samples of the MESA cohort.16 At baseline, in the entire cohort, factor VIII and plasmin–antiplasmin (PAP) were measured, whereas plasminogen activator inhibitor-1 (PAI-1), von Willebrand factor (VWF), soluble thrombomodulin (STM) and E-selectin were measured only in the ‘MESA 1000’ random sample.16 Intercellular adhesion molecule 1 (ICAM-1) was measured from the ‘MESA 1000’ sample and also from all participants enrolled before February 2003.16 Due to missing data and exclusions, the numbers included slightly varied across these biomarkers.
Participants were followed from study baseline (2000–2002) to 2015. They were contacted by telephone every 9–12 months to obtain information regarding self-reported hospitalisation, and the medical records from these hospitalisations were obtained for review. Incident VTE events were identified via these hospitalisation records and also from death certificates by using International Classification of Diseases (ICD)-9 and ICD-10 codes. These codes were predetermined by a panel of VTE experts in the MESA coordinating centre. They were chosen to be consistent with the ICD codes used for the Longitudinal Investigation of Thromboembolism (LITE) study21 whose goal was to determine the incidence of VTEs in two community-based cohorts: the Atherosclerosis Risk in Communities (ARIC)22 and the Cardiovascular Health Study cohorts23 (see online supplementary table 1 for the specific ICD-9 and ICD-10 codes used).
We stratified baseline characteristics by the presence/absence of incident VTE. We present continuous variables as mean±SD or median (25–75th percentile) for normally distributed and skewed variables, respectively. We present categorical variables as frequency (percentage). We used two samples t-test, Mann-Whitney test or X2 test to describe differences between groups as appropriate. We used multivariable-adjusted Cox proportional hazards regression models to determine the HRs and 95% CIs of incident VTE by RHR groups categorised by previously established cut points (<60, 60–69, 70–79 and ≥80 bpm).10 24 25 We tested for trends across RHR categories by using an ordinal variable for each RHR category and modelling this as a continuous variable for association with VTE. We also examined RHR as a continuous variable per 10 bpm increment. We tested non-violation of the Cox proportional assumption using Schoenfeld residuals.
We used progressively adjusted models. In model 1, we adjusted for demographic variables of age (continuous), sex (men; women) and race/ ethnicity (white; black; Hispanic; Chinese). Model 2 (our primary analytical model) was further adjusted for socioeconomic and lifestyle factors such as education (<high school; high school or vocational school; college, graduate or professional school), BMI (continuous) and physical activity level (metabolic equivalent of task (MET) minutes /week of moderate or vigorous activity; log-transformed continuous). Model 3 was additionally adjusted for cigarette smoking (current; former; never), diabetes (yes; no), eGFR (continuous), use of AV nodal blocking agents (yes; no), aspirin (yes; no) and anticoagulants (yes; no). Note that the use of oral anticoagulants was very low in MESA (0.4% of our analytic sample), as MESA excluded those with clinical CVD and atrial fibrillation at baseline. We did not adjust for participants with oral contraceptive or hormone therapy use as data were only limited to women and was not associated with VTE in this cohort (not shown). In a final model, model 4, we further adjusted for the inflammatory and coagulation markers (all log-transformed, continuous) of hsCRP, IL-6, fibrinogen and D-dimer, which may play an intermediary role in the relationship between RHR and VTE.
To further characterise the relationship between RHR and incident VTE, we used restricted cubic splines centred at the median RHR (62 bpm) with five knots placed at the 5th, 27.5th, 50th, 72.5th and 95th percentiles adjusted for variables in our main analytic model (model 2).
We performed several sensitivity analyses. First, we explored if the association between RHR and VTE differed by sex based on a priori hypotheses.10 We also examined interactions by age groups (per 10-year increments). Second, we excluded participants on AV-nodal blocker medications. Third, we modelled RHR and some covariates in our models as time-varying, with variables updated at MESA exams 2, 3, 4 and/or 5, as available (of note, since VTE events were recorded through 2015, data from MESA exam 6 which began in 2016 was not included). Lastly, to evaluate for potential mechanisms, we examined the partial correlations of RHR with various inflammatory and coagulation factors in MESA, using Spearman’s correlation (r), after adjusting for age, sex, and race/ethnicity.16 We considered a p<0.05 to be statistically significant, and performed our analyses using Stata V.15.