774 adult patients who presented at our outpatient clinic for bone diseases between August 2021 and February 2022 were retrospectively evaluated. As part of routine diagnostics, patients undergo DXA and HR-pQCT imaging as well as laboratory analysis. All patients with detectable lower leg arterial calcifications (LLAC) in the scan volume were included into the LLAC group, whereas patients without detected LLAC were assigned to the control group. As outlined by Paccou et al., arterial calcifications of the lower leg are defined as hyperdensities of linear or tubular character with a circular or semi-circular shape [6]. All HR-pQCT scans were meticulously examined by a single observer, who manually assessed each image to identify LLAC based on anatomical positions, allowing for a distinction between anterior and posterior vessels, as illustrated in Fig. 1A. Patients with genetic diseases, CKD grade ≥ 3B (< 45 ml/min) according to the KDIGO (Kidney Disease: Improving Global Outcomes CKD-MBD Work Group Guidelines [18]) or active tumors were excluded. Calcium supplementation was assessed based on a medical history questionnaire. This study was conducted in accordance with local guidelines and the Declaration of Helsinki with informed consent of participating individuals.

Dual-energy X-ray absorptiometry (Lunar iDXA, GE Healthcare, Madison, WI, USA) scans were performed at the lumbar spine (L1–4) and both proximal femora (femoral neck and total hip). Subsequently, areal bone mineral density (aBMD), T-score and Z-score were determined. For further analysis, the lowest T-score of the lumbar spine and the proximal femora with corresponding aBMD and Z-score were used. Daily calibration scans were performed with the dedicated phantom according to the manufacturer’s recommendations including precision tests following the recommendations of the International Society for Clinical Densitometry (ISCD) [19].

Blood and Urine samples were analyzed on a routine clinical base at the local laboratory, as described elsewhere [20]. Levels of serum calcium, phosphate, creatinine, alkaline phosphatase (ALP), creatinine, bone-specific alkaline phosphatase (bALP), osteocalcin, parathyroid hormone (PTH) and 25-hydroxyvitamin D (25-OH-D) as well as urinary deoxypyridinoline (DPD) per urinary creatinine were assessed. Glomerular filtration rate (GFR) was estimated by CKD-EPI [21].

All patients were examined with second-generation HR-pQCT (XtremeCT II, Scanco Medical AG, Brüttisellen, Switzerland) at the distal tibia. Scans were carried out according to the manufacturer’s standard in vivo scan protocol (68 kVp, 1470 μA, 43 ms integration time, 60.7 μm voxel size).

Scans at the tibia were taken at a defined offset of 22 mm proximal to the reference line positioned at the inflection point of the endplate of the tibial plafond in accordance with Whittier et al. [22]. The entire scanning volume extended over 168 slices (10.2 mm). Microarchitectural parameters followed the standardized nomenclature of the IOF-ASBMR-ECTS working group and included bone volume to total volume ratio (BV/TV), trabecular number (Tb.N, mm−1), trabecular thickness (Tb.Th, mm), trabecular separation (Tb.Sp, mm), cortical thickness (Ct.Th, mm), cortical pore diameter (Ct.Po.Dm, mm) and cortical porosity (Ct.Po). Volumetric bone mineral density was expressed as total BMD (Tt.BMD, mg HA/cm3), cortical BMD (Ct.BMD, mg HA/cm3), and trabecular BMD (Tb.BMD, mg HA/cm3). Geometric values include total bone area (Tt.Ar, mm2), trabecular bone area (Tb.Ar, mm2), cortical bone area (Ct.Ar, mm2), and cortical perimeter (Ct.Pm, mm). All HR-pQCT images were manually examined for motion artifacts [23]. If motion grade 4 or 5 was detected, patients were excluded [22], resulting in the exclusion of 7 patients in total (6.14%).

Previous trauma or microtrauma may influence the presence of calcification and the bone microstructure. Therefore, we manually excluded calcifications that were not associated with vessels, such as heterotopic calcifications, which was the case for 7 patients (6.14%).

For quantification of the vessel calcification, XamFlow 1.8.0.0 (Lucid Concepts AG, Zürich, Switzerland) was used. Image stacks were batch loaded and processed via a defined workflow to detect and quantify LLAC in the soft tissue surrounding the bones.

Therefore, first the image stacks were filtered with a gaussian filter (sigma = 3, support = 3) to remove noise. To isolate the full human tissue area from background, a threshold tissue was applied with a lower threshold at −189.8 mg HA/cm3 and an upper threshold of 1656.8 mg HA/cm3. Subsequently, mineralized tissue was masked using a specified threshold mineral with the lower value set to 169.6 mg HA/cm3 and the higher threshold value to 1868.8 mg HA/cm3. To address partial volume effects in the atherosclerotic tissue, we chose a relatively high threshold with respect to the density of the atherosclerosis. Next, the bone volume was subtracted from the full tissue volume. Therefore, the largest two elements (fibula and tibia) of the mineralized tissue volumes were chosen to be subtracted from the full tissue volume. The remaining mineral volumes were kept referring to potential arterial calcification.

A despeckling was subsequently added to the workflow to remove small noise artifacts below a volume of 15 voxels. For any mineralized volumes that did not correspond to the region of arteries or if the shape was clearly not arterial wall calcification, we applied a manual correction to exclude these volumes. Mineralized areas (excluding the bones) were than evaluated with respect to density and volume.

Statistical analysis was performed using SPSS software version 28.0.1 (IBM, Armonk, NY, USA), GraphPad Prism 9.5.0 (GraphPad Software, San Diego, CA, USA) and G*Power software version 3.1.9.6 (Heinrich-Heine-University Düsseldorf, Düsseldorf, NRW, Germany). The results are presented as absolute values or as means ± standard deviations. To evaluate normal distribution, the Shapiro–Wilk test was performed. All categorical variables were compared using the Fisher's exact test. For differences between two subgroups, the unpaired two-tailed students t-test was used for normally distributed data and the Mann–Whitney-U test was performed for non-normally distributed data. Correlation analyses were performed to evaluate the relationships between vascular calcification variables (density and volume) and HR-pQCT bone microstructure and density parameters. For non-normally distributed data Spearmen correlation was used. Additionally, due to interactions among many parameters, a multiple linear regression model (enter method) was used to evaluate the independent predictive value of the various variables on plaque density (mg HA/cm3) and plaque volume (# voxels) in lower limb vascular calcifications.

Parameters that showed relevance in previous analyses, such as correlation analysis and (sub-)group comparison were included. In addition to the characteristics of the overall model (R2, adjusted R2, F, and p-value), individual regression coefficients (B, ß, and p-value) were calculated. Variance inflation factors (VIF) were calculated to check for multicollinearities, with no multilinearity indicated if the VIF values were between 1 and 5. We decided to perform propensity score matching in mitigating confounding effects and improving the reliability of our tests [24]. We thereby minimized other influencing factors such as age and BMI between the different groups, enhancing the overall power. However, due to the matching, sample set size and thus calculated power were reduced.