Study populations

This study included two cohorts, participants from the Peripheral Artery Disease in Västmanland study (PADVa), and a population-based sample recruited as control subjects for the Västmanland Myocardial Infarction Study (VaMIS); ClinicalTrials.gov number, NCT01452178). Individuals with a verified PAD diagnosis in the population-based cohort were excluded.

The PADVa cohort

Consecutive patients referred to the Vascular Ultrasound Laboratory of the Department of Vascular Surgery, Västmanland County Hospital, Västerås, Sweden, were evaluated for inclusion between April 2006 and February 2011. At least one of the following criteria were required to be enrolled in the study:

(i)

symptoms of claudication intermittent (described as pain from lower limb reproduced by exercise and relieved within 10 min’ rest) in combination with ankle-brachial index (ABI) ≤ 0.9;

(ii)

symptoms of claudication intermittent with signs of occlusive arterial disease on ultrasound examination in the ipsilateral extremity;

(iii)

ultrasound verified stenosis or occlusion of the internal carotid artery (ICA).

Among 614 patients fulfilling the criteria for participation, 162 (26.4%) rejected inclusion, and 15 (2.4%) were excluded due to any missing biomarker values, leaving 437 individuals with carotid or lower limb artery disease included in the study.

The population-based cohort

In the VaMIS study, consecutive patients hospitalized for acute myocardial infarction from November 2005 to May 2011 were included [18]. Using the Swedish Population register, a control subject was recruited from the general population for each included VaMIS patient. A random individual with the closest date of birth, same sex, and same municipality as an included VaMIS patient was invited to participate. The patients from the PADVa cohort and controls in the VaMIS study underwent the same study protocol [15]. From the 855 control subjects in the VaMIS study, 175 individuals (20.5%) were excluded as they fulfilled the criteria for PAD specified above. Further, 37 (4.3%) with missing biomarker values were excluded, leaving 643 subjects for analysis in the population-based cohort.

Study protocol

The study protocol for both cohorts included medical history, current medication and smoking habits obtained from self-administered questionnaires. Diagnoses of previous CV events and diabetes mellitus and were confirmed from medical records. Hypertension was defined if diagnosed by a physician and prescribed antihypertensive medications. The ABI was calculated as the highest of the systolic blood pressures in the dorsalis pedis artery and the posterior tibial artery divided by the systolic blood pressure in the arm. ABI was defined as abnormal if ≤ 0.90 or 1.40 in any leg. Stenosis in the internal carotid artery (ICA) was considered if the ultrasonography investigation detected a localized protrusion of the vessel wall into the lumen in combination with turbulent color Doppler flow, spectral broadening in the spectral Doppler flow, and a peak systolic flow velocity of ≥ 1.2 m/s [19]. Left ventricular ejection fraction (LVEF) was assessed using the biplane Simpson’s formula [20]. In subjects for whom it was impossible to obtain the Simpson’s LVEF, a visual estimation of LVEF was made.

Blood samples

All the blood samples were taken after a night fasting by trained staff and immediately sent to the accredited Laboratory of Clinical Chemistry, Västmanland County Hospital, Västerås, Sweden, for analyses or freezing. Blood samples aimed for freezing were obtained in 5 ml lithium heparin-coated vacuum tubes. The tubes were centrifuged at 2000 g for 10 min (Becton Dickinson and Co., Franklin Lakes, New Jersey, USA) or 2200 g for 10 min (Vacuette, Greiner Bio-One GmbH, Austria). The plasma was reallocated to 5 ml plastic tubes, frozen and stored at − 70 °C within 2 h.

Blood analyses at baseline

Plasma levels of NT-proBNP were determined at baseline by a sandwich immunoassay using monoclonal antibodies and biotin-streptavidin separation (Elecys 1010 and Cobas e411 instruments, Roche Diagnostics, Germany) [21]. The analytical range was 5–35,000 ng/L and total coefficient of variation (CVa) of 5.4% and 4.4% at 101 ng/L and 908 ng/L, respectively. Serum creatinine was measured by the Jaffé method standardized against isotope dilution mass spectrometry (Synchron LX 20 and UniCel DxC instruments, Beckman Coulter, USA) with a total CV of 5.3% and 2.4% at 90 µkat/L and 379 µkat/L, respectively. Total cholesterol was measured by enzymatic conversion of cholesterolesterase by a time-endpoint method (Synchron LX 20 and UniCel DxC instruments, Beckman Coulter, USA) with a total CV of 1.5% and 1.6% at 3.18 mmol/L and 7.7 mmol/L, respectively. HbA1c was measured using high-performance liquid chromatography using cation exchange separation and calibrated against the Swedish Mono S method (TOSOH automated Glycohemoglobin Analyzer G7, Tosoh, Japan). HbA1c Mono S was converted to IFCC units by the equation \(HbA1c\left(IFCC\right)=HbA1c\left(Mono S\right)\times 10.45-10.62\). This formula differs slightly from the IFCC master equation \(HbA1c\left(IFCC\right)=HbA1c\left(Mono S\right)\times 10.11-8.94\) and is due to a recalculation in 2004 [22, 23].

Analyses on thawed blood samples

Before analysis, the samples were thawed at room temperature, mixed and centrifuged at 3470 g at 4 °C for 15 min and aliquoted into a microtiter plate using a pipetting robot, the Tecan Freedom Evolyze. The analyses were performed at the Clinical Biomarkers Facility, Science for Life Laboratory, Uppsala University, Uppsala, Sweden.

Plasma levels of GDF-15 were analysed in 2017 by a an immunoassay based on the specific Elecsys electrochemiluminescence (ECL) detection system (Roche Diagnostics) [24]. The analytical range was 400–20,000 ng/L with a total CVa of 4.9%.

The Proseek Multiplex CVD I 96 × 96 panel (Olink Bioscience, Uppsala, Sweden) was used to analyze 92 proteins previously associated with CVD or inflammation, including NT-proBNP and GDF-15 [16]. The content of the proteins in each plasma specimen was measured simultaneously by the binding of paired oligonucleotide-labeled antibodies to the target proteins. The subsequent formation of new polymerase chain reaction (PCR) amplification targets was detected and quantified by high-throughput real-time PCR. The measures are specified and presented in relative units on a binary logarithmic scale, the Normalized Protein Expression (NPX).

Validation of the assay including 90 proteins and seven samples analysed in nine separate runs showed a mean intra-assay coefficient of variation of 8% (range 4–13%) and an inter-assay coefficient of variation of 15% (range 11–39%).

Follow-up and outcomes

The participants in both cohorts were followed from the index examination until an endpoint or, at the latest, December 31, 2013. The endpoint was a CV event defined as CV death (International Statistical Classification of Diseases and Related Health Problems, 10th revision code I00-I99) or hospitalization for myocardial infarction (MI) (I21), heart failure (I50, I11.0, or I25.5), or ischemic stroke (I63). Follow-up data were collected from the Swedish National Inpatient Register and the Swedish National Cause of Death Register.

Statistical analyses

We present continuous variables as mean and standard deviation (SD) or median and interquartile range, and categorical variables as frequency and percentage.

Comparisons between the plasma values of NT-proBNP and GDF-15 measured by the conventional immunoassays and the PEA were performed using Spearman’s rank correlation coefficients (rho) with corresponding 95% confidence intervals (CI) and visualized in scatter plots with local polynomial regression lines (LOESS). Due to highly positively skewed plasma levels in the conventional analyses, the values were binary logarithmized (to the same scale as the PEA NPX values). To adjust for technical variation with internal controls in the PEA panels, the PEA and conventional plasma values were standardized to a mean of 0 and an SD of 1, grouped by the PEA panel.

Unadjusted Cox proportional hazard regression was used to evaluate the ability of the proteins to predict future CV events. The predictive values of the plasma levels of NT-proBNP and GDF-15 obtained with the PEA were compared with prediction based of plasma levels of the conventional assays in both cohorts. The unadjusted Cox regression results are presented as hazard ratios, with corresponding 95% CI. The discriminatory accuracy is presented as the time dependent areas under the receiver operating characteristics curves (tdAUC) [25]. The AUC values were calculated at 5 years of follow-up and demonstrated in plots.

Two-sided p-values < 0.05 were considered statistically significant. R (R Foundation for Statistical Computing; 2016, Vienna, Austria; https://www.r-project.org) were used for the statistical analyses.