A case-cohort study was designed within the follow-up of the REGICOR (Registre GIroní del COR) population cohort study.14 In REGICOR, a 6352-participant population cohort was recruited in 2005, with a response rate > 70%. Of these participants, 5404 were aged between 35 and 74 years and were free of cardiovascular disease at the time of inclusion. All participants were followed up between 2009 and 2013.

In line with the case-cohort study design, we included all participants who had a coronary event (n = 117) together with a random subcohort of participants included in the baseline evaluation at an approximate ratio of 1:6 (n = 667). Of these participants, we selected for analysis those with an available biological sample (105 case participants and 651 subcohort participants). The incidence of the event of interest in the original 5404-participant sample was 2.7%; assuming a type I error rate of 5%, the case-cohort sample size (n = 756) provides a statistical power of 80% for the statistically significant identification of a relative risk ≥ 1.58 per standard deviation increase in the biomarker. Statistical power was calculated using the ccsize function in the gap package in R. We also included 43 competing noncoronary cardiovascular events.

All participants were informed of the aims of the study and gave written consent. The study was approved by the local clinical research ethics committee.

Participants provided biological samples after a 10 to 14 hour fast. Once the blood was centrifuged, and after coagulation in the case of serum, samples of serum and EDTA (ethylenediaminetetraacetic acid) plasma were aliquoted and stored at –80°C until analysis. Glucose, total cholesterol, and triglycerides were measured in serum by enzymatic methods, whereas high-density lipoprotein (HDL) cholesterol was measured directly (Roche Diagnostics; Basel, Switzerland) in a Cobas Mira Plus autoanalyzer (Roche Diagnostics). If the triglyceride concentration was < 300 mg/dL, low-density lipoprotein cholesterol was estimated with the Friedewald equation.

Vitamin D was measured in serum by competitive ELISA for 25-hydroxy vitamin D (Immunodiagnostic Systems; Boldon, United Kingdom). Plasma albumin concentration was assayed by bromocresol green dye binding (ABX Diagnostics; Montpellier, France) and plasma apoA1 and apoB concentrations by immunoturbidimetric assay (ABX Diagnostics) in a Cobas Mira Plus autoanalyzer.

The variables analyzed in the Cobas Mira Plus autoanalyzer (albumin, apoA1, and apoB) were checked against external quality assurance standards from the Spanish Society of Clinical Chemistry (Sociedad Española de Química Clínica). Interassay variation coefficients for albumin, apopA1, apoB, and 25-OH-vitamin D were 3.19%, 3.37%, 3.21%, and 5.72%, respectively.

Various information sources were used during follow-up. Nonfatal events of interest were identfied through regular structured telephone interviews and a re-evaluation visit, and fatal events were identified from official death records. Events of interest were the appearnace of angina, fatal or nonfatal myocardial infarction, and coronary death.

Nonfatal events were validated by reviewing the medical records of participants reporting angina pectoris or a myocardial infarction. Deaths due to coronary disease were identified according to ICD (International Classification of Diseases) codes ICD-9 410-414 and ICD-10 I20-I22, I24, I25. A committee categorized all events according to standard criteria. Angina pectoris was defined as the presence of symptoms accompanied by an objective demonstration of ischemia in the electrocardiogram or > 50% coronary stenosis by coronary angiography. Myocardial infarction was defined according to international criteria.15

Recorded competing cardiovascular events were noncoronary cardiovascular death and nonfatal cerebrovascular disease.

Questionnaires and standard methods were used to determine conventional cardiovascular risk parameters14: systolic and diastolic blood pressure, total cholesterol, HDL cholesterol, smoking, and diet. Smokers were defined as participants who were active smokers or who had stopped smoking within the previous year. Diabetes was defined as baseline blood glucose ≥ 126 mg/dL or treatment with oral antidiabetic drugs or insulin. Cardiovascular risk was estimated with the REGICOR score validated for the Spanish population.3

Biomarkers of interest showing a nonnormal distribution were logarithmically transformed. Vitamin D was stratified into 3 bands widely used in the medical literature: < 30 nmol/L, 30-49 nmol/L, and ≥ 50 nmol/L.12 Differences between the case population and subcohort in sociodemographic variables, risk factors, and biomarkers were determined using the method of Lin and Ying for case-cohort studies.16 The relative risk of coronary disease associated with a 1 standard deviation increase in each biomarker of interest was estimated using the method proposed by Pintilie et al. to accommodate the existence of competing risks.17 We developed 2 multivariate models, the first adjusted for age and sex and the second additionally adjusted for systolic and diastolic blood pressure, total cholesterol, HDL cholesterol, diabetes, and smoking status. Furthermore, each biomarker was analyzed separately; all biomarkers were subsequently included in the multivariate model, and only those showing a statistically significant association with the event of interest were maintained. All variables except for diabetes fulfilled the assumption of risk proportionality. We confirmed that stratification for diabetes did not significantly modify the magnitude of association between the biomarkers of interest and coronary events.

The net contribution of biomarkers to the predictive capacity of conventional risk factors was evaluated from the improvement in discrimination, determined by analyzing the difference in Harrell's C statistic adapted to case-cohort studies.18 Proper reclassification was verified by estimating the categorical and continuous net reclassification improvement (NRI). NRI categories were defined using the standard cut-points for the 10-year risk of a coronary event in the Spanish population (5% and 10%).3 The number of events at 10 years in each risk category was obtained by preparing event-free survival curves. The NRI confidence intervals to accommodate uncertainty in the estimates were calculated using bootstrap resampling techniques.19 For the estimation of NRI in the intermediate risk group, we used the method proposed by Paynter and Cook to correct overestimation due to the use of a single risk category.20

Statistical analyses were conducted in R version 3.0.1 (R: a language and environment for statistical computing; the R Foundation for Statistical Computing, Viena, Austria). We used the function developed by Pintilie et al. to accommodate the existence of competing risks in case-cohort studies17 and wrote R code functions to calculate the C statistic for case-cohort studies.

The findings presented here reinforce the results of several studies analyzing the influence of albumin on coronary and cardiovascular risk and total mortality. A 1998 meta-analysis of data from 8 studies already confirmed the association between low albumin concentrations and an elevated risk of coronary disease.11 This association was also later reported in large epidemiological studies, such as ARIC21 and Framingham.22 A recent study concluded that low albumin concentration explains 41% of the attributable population risk for a first acute myocardial infarction.23 Low albumin concentration is also associated with cardiovascular diseases such as stroke24 and with total cardiovascular risk.25 Moreover, as far back as 1989, the British Regional Heart Study reported higher cardiovascular mortality in men with low systemic albumin concentrations. 26 These results have been confirmed in male participants in the Paris Health Study27 and the National Health and Nutrition Examination Survey I,28 but not in other studies.29

There is no consensus on whether albumin concentration is simply a nonspecific variable or is causally involved in cardiovascular disease.30 However, the relationship could be explained by several biological properties of albumin: a) Albumin has antioxidant and anti-inflammatory properties and accounts for more than 50% of total plasma antioxidant activity due to its free–radical-binding sulfhydryl groups.31 Moreover, albumin-bound bilirubin inhibits lipid peroxidation,32 and albumin interacts with factors that regulate the intensity of arterial inflammation.33b) Albumin transports free fatty acids and free cholesterol and participates in reverse cholesterol transport.34c) Albumin inhibits endothelial apoptosis and improves endothelial function.35 Nevertheless, the association between albumin and cardiovascular risk appears not to be very specific, since low albumin concentrations are also associated with noncardiovascular death and cancer.36

In this study, we have provided the first evaluation of albumin as a predictive risk biomarker. We found that inclusion of plasma albumin concentration in the REGICOR score improved the estimation of coronary risk and appropriately reclassified a significant proportion of the study population. The magnitude of the reclassification was higher than that observed with other biochemical biomarkers7 or with genetic biomarkers,6 and was similar to that obtained with intracoronary calcium.9 These results support the usefulness of albumin as a predictive biomarker of coronary risk. Nevertheless, confirmation is needed in other cohort studies, which should also analyze the predictive capacity of simultaneously considering different types of biomarkers.

The second biomarker analyzed in this study was the apoA1:apoB ratio. Several prospective studies have reported an association between the occurrence of coronary events and this ratio, sometimes expressed inversely as the apoB:apoA1 ratio.10,37 In the INTERHEART study,38 the aopB:apoA1 ratio was the biomarker associated with the highest population attributable risk, and apoB was identified as a more precise cardiovascular risk biomarker than non–HDL-cholesterol.39 In the present study, the apoA1:apoB ratio showed an inverse association with coronary risk, but this association disappeared after adjustment for conentional risk factors (total cholesterol and HDL cholesterol).

Given the lack of association with coronary risk, we did not analyze the influence of apoA1:apoB ratio on the predictive capacity of a conventional risk score. A previous study reported that inclusion of the apoA1:apoB ratio slightly improved the discriminatory capacity and reclassification of risk scores containing total cholesterol and HDL cholesterol.10

Vitamin D is much more than a simple regulator of calcium and phosphorus metabolism, and is linked to many cardiovascular risk factors, including dyslipidemia, hypertension, and diabetes.40–42 Nevertheless, the relationship between vitamin D concentration and coronary risk is highly controversial; although epidemiological studies indicate an association with lower cardiovascular risk,43 vitamin D supplementation in experimental studies has not demonstrated a cardioprotective effect.44 In the present study, the association of vitamin D concentration with reduced coronary risk disappeared after adjustment for conventional risk factors. Given the lack of an association with coronary events, we did not analyze the contribution of vitamin D concentration to the predictive capacity of a conventional risk score; however, where an association with cardiovascular risk has been observed, inclusion of vitamin D concentration did not improve risk prediction.43

The strengths of this study include its prospective design and its population base, together with the evaluation of several biomarkers for their predictive association with coronary events.

Study limitations include the small number of coronary events, which limits the statistical power of the study. It should also be noted that vitamin D is subject to seasonal variations that may have influenced the results obtained.45 However, re-evaluations took place at all times of the year, so a seasonal effect would seem unlikely.