Study design and participants
This study was embedded in the Rotterdam Study (RS), an ongoing prospective population-based cohort study including middle-aged and older inhabitants of Ommoord, Rotterdam, the Netherlands. The RS started in 1990 and included interested inhabitants of the study area that were aged 55 years or over. The RS was extended with additional independent cohorts in 2000 and 2006 and for these cohorts, inhabitants aged 55 years or over (RS cohort II) respectively 45 years or over (RS cohort III) that were not previously invited were invited to participate. By the end of 2008, 14,926 participants aged 45 years or over had been enrolled (response rate 72%). In 2016, enrollment for a fourth cohort started aimed at inhabitants of the study area aged 40 years or over. After study entry, medical records of general practitioners (GPs) are continuously linked to the RS database and outcomes of interest are validated according to standardized guidelines by medical experts in the field. In addition, each participant is re-examined every three to six years at the research center in Ommoord and through home interviews. The RS complies with the declaration of Helsinki and has been approved by the medical ethics committee of the Erasmus University Medical Center (registration number MEC 02.1015) and by the Dutch Ministry of Health, Welfare and Sport (Population Screening Act WBO, license number 1071272-159521-PG). The RS has been entered into the Netherlands National Trial Register and into the WHO International Clinical Trials Registry Platform under shared catalogue number NTR6831. More detailed information can be found elsewhere (19).
For this study, we included participants of three independent RS cohorts (RS I-3, II-1, and III-1) with written informed consent for follow-up, available measurements of serum IgA, IgG, and/or IgM at study baseline, and information on cardiovascular outcomes of interest (n = 8,767).
Assessment of serum Igs
Blood was drawn at the research center through 1997–1999 (RS cohort I-3), 2000–2001 (RS cohort II-1), and 2006–2008 (RS cohort III-1) and the moment of blood drawing was considered the study baseline. Serum samples were subsequently stored at -80°C. Serum IgA, IgG, and IgM measurements took place between 2016–2018 through an immunoturbidimetric assay (Tina-quant® IgA/IgG/IgM Gen. 2, Roche Diagnostics GmbH, Mannheim, Germany). Recommended reference ranges according to the manufacturer’s protocol were 0.7-4.0 g/L for IgA, 7.0–16.0 g/L for IgG, and 0.4–2.3 g/L for IgM. As previously described, reference ranges based on 2.5th and 97.5th percentiles for the RS population were 0.86–4.76 g/L for IgA, 6.20–15.10 g/L for IgG, and 0.28–2.64 g/L for IgM (20).
Assessment of ACVD, cardiovascular and all-cause mortality, and CAC
ACVD was defined as myocardial infarction, revascularization (percutaneous coronary intervention or coronary artery bypass graft), or stroke in accordance with previous research and specific guidelines (21). Information on prevalent (at baseline) ACVD was retrieved through home interviews, linkage with the Nationwide Medical Registry (a national registry of all hospital discharge diagnoses of all Dutch inhabitants), and medical records of GPs. Information on incident (during follow-up) ACVD was retrieved through continuous automated linkage with medical records of GPs containing ICPC codes as diagnosed by GPs or medical specialists and was subsequently validated as described previously (22–24). Follow-up of myocardial infarction, revascularization, and the composite endpoint of ACVD was complete until January 1st 2015. Follow-up of stroke (comprising all stroke cases, i.e. ischemic, hemorrhagic, and unspecified) was complete until January 1st 2016.
Information on mortality was retrieved from medical records of GPs, hospitals, and nursing homes. Two independent research physicians classified mortality according to ICPC and ICD-10 codes. Subsequently, all coded events were reviewed by a senior physician in the field to confirm the diagnosis. Date of death was retrieved from the medical records or municipality records. Cardiovascular mortality was composed of atherosclerotic and non-atherosclerotic mortality. Atherosclerotic mortality was defined as mortality due to coronary heart disease, cerebrovascular disease, or other atherosclerotic disease (ICPC codes K75-K77, K90-K92; ICD-10 codes I21-I25, I50, I60-I70, I71.3, I71.4, I73-I74, I77-I79, Y60.5) and non-atherosclerotic mortality was defined as mortality due to other cardiovascular disease (ICPC codes K78-K87, K93-K94, K99; ICD-10 codes I05-I13, I15, I26-I28, I30-I49, I51-I52, I71[excluding I71.3 and I71.4]-I72, I80-I83, I97-I99, T82) in accordance with previous literature (22). Follow-up of cardiovascular mortality was complete until January 1st 2015. All-cause mortality comprised any (both cardiac and non-cardiac) mortality and its follow-up was complete until May 24th 2018.
CAC scores were available in a random subset of RS cohorts I-3 and II-1 (n = 1,622) and were measured by electronbeam Computed Tomography (EBT; C-150 Imatron Scanner, GE Healthcase, South San Francisco, CA). Participants had to lie still and hold their breath during the assessment. Thirty-eight images were obtained from the aorta root to the heart with 100ms scan time and 3mm slice thickness. CAC was quantified with AccuImage software (AccuImage Diagnostics Corp), displaying all pixels with a density of > 130 Hounsfield units (HU). Calcification was defined as ≥ 2 adjacent pixels of > 130 HU. CAC scores were calculated by multiplication of the area of individual calcifications in mm2 with a factor based on the peak density of the calcification. All individual CAC scores were combined to obtain a total CAC score for the entire coronary epicardial system. Assessment of CAC was performed by two independent experienced physicians (25).
Assessment of baseline covariates
Weight, height, and blood pressure were measured at the research center in Ommoord. Body mass index (BMI) was defined as weight divided by height squared (kg/m2). Blood pressure was measured twice at the right brachial artery with the participant in sitting position. The average of these two consecutive measurements was taken. Hypertension was defined as a blood pressure exceeding 140/90 mmHg or as the use of blood pressure lowering medication with the indication of hypertension. Information on baseline type 2 diabetes (DM) was collected from GPs and pharmacies and assessed through blood samples collected at the research center. DM was defined as a fasting blood glucose concentration of ≥ 7.0 mmol/L, a non-fasting blood glucose concentration of ≥ 11.1 mmol/L (when fasting samples were unavailable), or as the use of blood glucose-lowering drugs or dietary treatment for diabetes. Smoking status, alcohol consumption, and highest attained education (as proxy for socioeconomic status) were assessed through questionnaires during home interviews. Smoking status was defined as never, former, or current smoker. Alcohol consumption was reported in gram/day and categorized into none, mild (0–10 g/day), moderate (10–20 g/day), or heavy (> 20 g/day). Physical activity was assessed using validated questionnaires and expressed in standardized metabolic equivalent of task (MET) hours/week (26, 27). Serum triglycerides and total cholesterol (mmol/L) were measured using an automated enzymatic procedure. Serum CRP (mg/L) was measured with an immunoturbidimetric assay. Baseline use of medication known to potentially alter serum Ig levels (systemic corticosteroids, antiepileptic drugs, angiotensin converting enzyme inhibitors, cytostatics, immunomodulators, and/or immunosuppressants) was established during home interviews and coded based on the Anatomical Therapeutic Chemical Classification System.
Statistical analyses
Serum Igs were standardized for all analyses to facilitate comparison of results. In the longitudinal analyses, participants were followed until the first event of interest, death, or end of follow-up, whichever came first. All analyses included three models, adjusting for potential confounders based on biological plausibility and previous comparable research (21). In the first model, we adjusted for age and sex. In the second model, we additionally adjusted for smoking status, alcohol consumption, physical activity, and highest attained education. The third model included confounders that could also act as mediators and comprised BMI, DM, hypertension, serum triglycerides, CRP, and total serum cholesterol additional to the first two models. The longitudinal analyses also included RS cohort in all models to take temporal trends into account.
The association between serum Igs and incident ACVD (both composite and the individual endpoints of myocardial infarction, revascularization, and stroke) was assessed through Cox proportional hazards regression analyses after exclusion of participants with prevalent ACVD. The proportional hazards assumption was checked through the Schoenfeld test and plot and was met for all analyses. A sensitivity analysis was performed by excluding participants with serum Ig levels outside our calculated reference ranges (0.86–4.76 g/L for IgA, 6.20–15.10 g/L for IgG, and 0.28–2.64 g/L for IgM) and users of medication known to potentially alter serum Ig levels in order to limit the influence of transient outliers in serum Ig levels (e.g. due to an acute infection). With respect to stroke, we also examined the association of Igs with ischemic stroke.
The association of serum Igs with cardiovascular mortality (atherosclerotic, non-atherosclerotic, and combined) and all-cause mortality was assessed by Cox proportional hazards regression analyses. Proportional hazards were checked and confirmed for all analyses. For cardiovascular mortality, we stratified by prevalent ACVD status, age (cut-off 65 years), and sex, and performed a sensitivity analysis after exclusion of participants with serum Ig levels outside the reference range and users of medication known to potentially alter serum Ig levels. For comparison, we applied both the assay recommended and our own calculated reference ranges in this sensitivity analysis. We furthermore displayed the risk of cardiovascular mortality for participants with the highest compared to the lowest Ig reference value (4.76 vs 0.86 g/L for IgA, 15.10 vs 6.20 g/L for IgG, and 2.64 vs 0.28 g/L for IgM) while keeping all other covariates constant by testing contrasts of the Cox proportional hazards regression analyses.
For the association between serum Igs and CAC score, we categorized CAC scores into no (score = 0), mild (score 0-100), moderate (score 100–400), or severe calcification (score > 400) based on the most commonly used classification system (28). The association between serum Igs and CAC score categories was assessed through multinomial logistic regression analyses, while taking no calcification as the reference category. There was no multicollinearity and the linearity of log odds assumption was met. We performed a sensitivity analysis by excluding participants with serum Ig levels outside abovementioned calculated reference ranges and users of medication known to potentially alter serum Ig levels.
Missing values in covariates were imputed with multivariate imputation by chained equations (4 imputations, 10 iterations). Missingness was < 2% for all covariates, except for physical activity and alcohol consumption (14.1% and 20.4% respectively). Convergence was reached and the distribution of covariates before and after imputation was similar. All analyses were performed with R Statistical Software version 4.0.2. Provided that effect estimates of the second and third models were comparable, we described the fully adjusted models in the results, unless stated otherwise.