Carotid artery volumetric measures associate with clinical ten-year cardiovascular (CV) risk scores and individual traditional CV risk factors in rheumatoid arthritis; a carotid-MRI feasibility study

Demographic data, traditional CV risk factors and for patients with RA, disease phenotype including 3-variable (3v) Disease Activity Score in 28 joints (DAS28) based on C-reactive protein (CRP) protein [13] and Health Assessment Questionnaire Disease Index [14] were recorded on clinical evaluation. Fasting lipid profile and glucose were measured, and rheumatoid factor, anti-cyclic citrullinated peptide antibody, CRP and erythrocyte sedimentation rate were measured in those with RA. Two 10-year cardiovascular risk scores were calculated, given the pilot nature of this work; Framingham (commonly used and therefore understood by a wide audience) and Joint British Societies 2 (in line with our cardiology partnership/network practice), and both were adjusted as per 2010 [15] and 2017 [34] European League Against Rheumatism (EULAR) guidelines. Additional samples were processed and stored for later measurement of NT-proBNP and homeostasis model assessment-estimated insulin resistance (HOMA-IR, fasting insulin (μU/ml) × fasting glucose (mg/dl)/405) [16].

Non-contrast carotid MRI was performed at 3.0 T (Philips Achieva, Philips, Best, The Netherlands) using a small 10-cm phased-array receiver coil (Philips dStream Flex, Best, The Netherlands). Survey images were used to locate the bifurcation of the common carotid artery into the internal and external carotid artery. These images were used to plan perpendicular, non-breath-held, time-of-flight (ToF) proton density-weighted (PD), T1 and T2 weighted acquisitions of the common carotid artery (10 slices in total, 5 above and 5 below the carotid bifurcation). Voxel sizes for each of these acquisitions were as follows: ToF 0.89 × 0.89 × 3 mm, PD 0.7 × 0.7 × 2 mm, T1 0.7 × 0.7 × 2 mm, T2 0.7 × 0.7 × 2 mm.

Using methodology similar to previously published studies in the general population [17‐19] as described subsequently, a cardiology fellow (GF) reviewed and reported the carotid-MR images blinded to all patient details and disease status using post-processing software (QMASS MR 7.5, Medis, Leiden, The Netherlands). An expert cardiovascular MRI-cardiologist (JPG) reviewed a subset of images/contours to ensure all were assessed as expected. A second junior then repeated contouring/measurements for a subset of images to allow for calculation of inter-observer variability.

Cross-sectional T1-weighted images of the right carotid artery were assessed in short-axis views. Slices 2, 4, 6 and 8 mm below the bifurcation of the common carotid artery were identified, and the carotid artery endothelial wall manually contoured on each slice. The outer (adventitial) wall of the artery was also contoured for each slice. Each cross-sectional image was then divided into 6 segments, and the post-processing software derived the minimum and maximum wall thickness, along with MWT of all 24 segments of the carotid wall. Figure 1 illustrates the images acquired and contours drawn. The software then used the contours to calculate the carotid wall mass (wall volume x the density of the tissue (1.05 mg/ul) [20]) for all four slices, allowing calculation of carotid wall volume. Total carotid wall volume was normalised for vessel size (carotid wall volume index) by calculating carotid wall volume/(carotid wall volume + luminal volume).

If the right carotid artery images were of poor quality or did not provide enough slices, the left carotid artery was used for measurement.

Within the RA group, Spearman’s correlation was used to measure associations between carotid-MRI measures and 10-year CV risk scores, and Pearson’s correlation and univariable linear regression analyses (UVA) were used to measure associations between each of the baseline variables and caroti- MRI measures, using log-transformed values when appropriate. Any variables reported in the literature as associated with carotid-MRI measures or found to be strongly correlated in the preliminary analyses (Pearson’s correlation coefficient > 0.3) were then included in a multivariable linear regression model. When the heteroskedasticity of regression model residuals was not improved by logarithmic transformation of the data, robust standard errors were employed.

Intra and inter-observer variability was tested (between GF and LAB blinded to one another) for assessment of carotid MWT to demonstrate an acceptable coefficient of variance.

In the event of missing serology results the most recent value preceding the visit was carried forward into the data, excluding CRP due to its capacity to vary, or lipid/glucose profile, as a fasting state could not be verified.

Of 69 patients with RA and 25 healthy controls that underwent a carotid-MRI scan, 64 with RA and 24 controls had images available to analyse (using left carotid images in 12 and 4 individuals, respectively); 4 data sets were of poor quality and 2 were missing due to image processing errors. Of the patients with RA, 60 patients had data on carotid arterial wall volume measurements; data were missing in 4 patients as the carotid artery bifurcated too low to allow measurement up to 8 mm below the bifurcation.

Table 1 outlines the demographic, CV risk profile and soluble CV biomarkers, and Table 2 describes the RA disease-specific features of those who had carotid artery images available. The mean age (standard deviation) of patients with RA was 59.6 (9.4) years, 70% were female and 94% white. Median (interquartile range) disease duration was 17.3 (10.7, 25.7) years, 84% were seropositive for rheumatoid factor (RF) or anti-citrullinated protein antibodies (ACPA), 82% had erosive disease and 67% were taking biological disease-modifying anti-rheumatic drugs (DMARDs). Patients overall were in remission; median (interquartile range) 3v DAS28 was 2.38 (1.15, 3.25). A significant proportion had CV risk factors, including 33% with known hypertension. The control group were younger (mean (standard deviation) age 51.8 (11.7) years) with fewer (54%) women and fewer CV risk factors. There was little difference in glucose and lipid profile, HOMA-IR or NT-proBNP between patients with RA and controls.

There were no significant differences between patients with RA and controls in carotid arterial wall measurements, including MWT, maximum wall thickness, wall volume, luminal volume, wall volume index and wall volume indexed/body surface area (BSA) (see Table 3).

In patients with RA, carotid wall volume and luminal volume correlated well with 10-year CV risk scores (adjusted as per 2010 [15] and 2017 [34] EULAR guidelines). The 2010 adjustments appeared to perform better; the carotid wall and luminal volume Spearman correlation coefficients for the 2010 adjusted Joint British Societies 2 risk scores were 0.37 (p = 0.005) and 0.43 (p = 0.001), respectively, compared to the 2017 adjusted Joint British Societies 2 risk scores (rho = 0.33 (p = 0.012) and rho = 0.35 (p = 0.008), respectively) (see Table 4).

No variables (traditional CV risk factors or RA disease-specific factors) were associated with carotid MWT (see Additional file 1: Table S1). A history of smoking and lower BMI correlated with higher maximum wall thickness (Pearson’s correlation coefficient r = 0.25 p = 0.049, r = − 0.25 p = 0.026) (see Additional file 1: Table S2); given r < 0.3, only age and gender were added to the multivariable analysis model (MVA) and no independent association was revealed.

In unadjusted analyses, increasing age (r = 0.35, p = 0.007) and a history of smoking (r = 0.33, p = 0.010) were associated with carotid wall volume, although only age remained independently associated in the MVA (B 1.13 (95% CI 0.32, 1.93), p = 0.007, R² = 0.245 (see Table 5). Increasing age (r = 0.37, p = 0.004), HOMA-IR (r = 0.30, p = 0.024) and RA disease duration (r = 0.34, p = 0.007) were associated with carotid luminal volume, although the association with HOMA-IR was lost with the exclusion of the high outlier. Age and RA disease duration remained independently associated in the MVA (age, B 3.69 (95% CI 0.55, 6.83) p = 0.022, RA disease duration; B 3.88 (95% CI 0.80, 6.97) p = 0.015, R² = 0.257 (see Additional file 1: Table S3).

A history of previous joint surgery and increasing HOMA-IR was associated with lower carotid wall volume index (r = − 0.26 p = 0.043 and r = − 0.28 p = 0.035, respectively), again the association with HOMA-IR was lost with the exclusion of the outlier. Only age and gender were included in the MVA and no independent association was determined (see Additional file 1: Table S4).

Cardiology fellow GF re-analysed 10 carotid images. The mean difference for intra-observer variability was 0.021 mm (95% CI − 0.012, 0.054) and the limits of agreement were − 0.072 to 0.113 mm. The coefficient of variance was 3.66%. The mean difference for inter-observer assessments (between fellow GF and LAB, using 10 images) was 0.032 mm (95% CI − 0.075, 0.010) and the limits of agreement were − 0.088 to 0.152 mm. The coefficient of variance was 4.93%.

There are many limitations associated with this study as outlined below. No widely recognised protocol for carotid-MRI reporting is currently available, and although our methodology mirrored previously published methods, this is an area in need of further study, including concordance between right and left carotid artery measurements. In addition, given the small differences in carotid wall thickness, our inter-observer and intra-observer variability limits of agreement are relatively wide and any future larger study would need to be appropriately powered to be meaningful. The variability in measurement may ultimately prove too wide for carotid-MRI to be a valid outcome measure. This cross-sectional study was also unable to quantify the burden of inflammation to which patients with RA were exposed over their disease duration. The cohort had established severe disease (lengthy disease duration, high proportion of seropositivity/erosions); therefore, although they were in remission, we feel the total burden of inflammation was likely significant as many had required multiple (including biological) disease-modifying anti-rheumatic drugs and orthopaedic surgery. However, the lack of association between carotid imaging measurements and RA disease activity/inflammation may be due to the smaller representation of those with high disease activity in our study population. The effect of acute active RA disease compared to chronic active RA, and those with previous high disease activity but now in remission needs evaluating further with larger studies to make any firm conclusions. We acknowledge that any further study would also require matching for age/sex/CV risk factors to have the greatest chance of detecting any real difference, if present at all, between those with RA and those without. Comparing carotid-MRI measures to US-measured CIMT would also help determine the role/applicability of carotid-MRI in this field both as a research tool and for clinical use in CV screening. Its role in screening is especially important given the relative lack of access to, and greater cost/expertise required, for carotid-MRI and the conflicting evidence supporting inclusion of carotid artery structural measures in CV risk scores in the general population [29, 30].

In addition to the needs already discussed, larger cross-sectional studies and prospective longitudinal studies are required to validate the utility of carotid-MRI in RA. The use of contrast to characterise carotid plaque further, differentiating unstable from stable plaques, lipid-rich necrotic cores from intra-plaque haemorrhage [31], combined with positron emission tomography imaging [32] could potentially help provide information on the pathophysiology of CVD in RA and on identifying those at risk of CVD. Our study also provides supportive evidence for strategic management of modifiable traditional CV risk factors in reducing CV morbidity and mortality in RA; an important message given that the management of such patients remains suboptimal [33].