The GCKD study involved 5 217 participants of European ancestry aged 18 to 74 years with an eGFR of 30 to 60 mL·min–1 per 1.73 m2 (corresponding to CKD stage G3, A1-3) or an eGFR ≥60 mL·min–1 per 1.73 m2 in the presence of severely increased albuminuria (i.e., >300 mg·g–1 creatinine) (corresponding to CKD stages G1-2, A3).45 The GCKD study was approved by the ethics committees of all participating centers (Friedrich Alexander University of Erlangen-Nuremberg, University of Freiburg, Ludwig-Maximilians University of Munich, University of Hannover, Charité—Universitätsmedizin Berlin, University of Würzburg, RWTH Aachen, University of Jena, and Heidelberg University) and was registered in a national database for clinical studies (Deutsches Register für Klinische Studien (DRKS) 00003971). The main exclusion criteria included solid organ or stem cell transplantation, active malignancy within 24 months prior to screening, non-European ancestry, and severe heart failure (New York Heart Association (NYHA) Stage IV). Each study participant provided written informed consent.

OPG, 25-OH vitamin D, P1NP, CTX1, iPTH, calcium, phosphate, FGF23, and BAP were measured in baseline serum samples from the GCKD study cohort that had been transported on dry ice and were stored at –80 °C. Measurements were performed at the Institute of Clinical Chemistry and Laboratory Medicine, Greifswald, Germany. OPG was measured by sandwich enzyme-linked immunosorbent assay (ELISA, Biomedica Immunoassays, Vienna, Austria, Cat# BI-20403) with a limit of detection (LOD) of 5.4 pg·mL–1 and an intra-assay variation ≤3%. C-terminal FGF23 was measured by ELISA (Biomedica Immunoassays; Cat# BI-20702) with an LOD of 0 pmol·L–1 + 3 standard deviations (SD) of 0.07 pmol·L–1 and an intra-assay variation ≤8%. 25-OH Vitamin D (LOD of 2.4 ng·mL–1 and an intra-assay variation of 4.6%–6.2%), P1NP (LOD < 1.0 ng·mL–1, intra-assay variation 2.6%–3.0%), iPTH (LOD 2.5 pg·mL–1, intra-assay variation 1.1%–6.3%), CTX1 (LOD 0.023 ng·mL–1, intra-assay variation 2.1%–4.9%), and BAP (LOD 0.4 μg·L–1, intra-assay variation 1.4%–2.0%) were measured by chemiluminescence assay on an IDS-iSYS platform (Immunodiagnostic Systems, Frankfurt, Germany). Reagents and standards were used as recommended by the manufacturers.

In addition, a standardized set of biomarkers was measured in a central certified laboratory using standardized protocols (Synlab, Germany).46 Calcium and phosphate were measured on a clinical chemistry platform (Cobas, Roche Diagnostics, Rotkreuz, Switzerland). Baseline serum and urinary creatinine were quantified using an IDMS traceable methodology (Creatinine plus, Roche Diagnostics, Rotkreuz, Switzerland). GFR was then estimated using the 2009 creatinine-based CKD Epidemiology Collaboration formula. Baseline urinary and serum albumin as well as low-density lipoprotein (LDL) cholesterol, high-density lipoprotein (HDL) cholesterol, and high-sensitive C-reactive protein (hsCRP) were quantified using a turbidimetric method (Tina-quant, on Roche/Hitachi MODULAR P platform (Roche Diagnostics, Rotkreuz, Switzerland). The urine albumin creatinine ratio (UACR) was calculated from the measured urinary albumin and urinary creatinine (mg·g–1).

During follow-up, patients were subjected to yearly interviews by trained personnel in alternating phone visits and face-to-face interactions. In this structured interview setting, data on any hospitalizations or clinical events were recorded. The collected data were verified by discharge and outpatient letters authored by the respective physician(s) responsible for the patients’ treatments. These reports were further subjected to extraction of outcomes according to a prespecified endpoint catalog by independent physicians (endpoint adjudication committee). Eventual deaths of study participants were confirmed by obtaining death certificates from civil registry offices whenever possible. The outcomes analyzed in this report include non-CV death, CV death (including death due to myocardial infarction, coronary artery disease, death due to a cerebrovascular event, sudden cardiac death, and death due to other cardiac causes, including death due to pulmonary embolism, decompensated heart failure, cardiac valve disease, or pulmonary embolism), nonfatal major adverse cardiac events (MACE, i.e., nonfatal myocardial infarction and nonfatal ischemic or hemorrhagic stroke), and hospitalization for CHF.

Participants who had missing values for bone and mineral biomarkers were excluded from the analysis (n = 917). Thus, the analysis data included 4 246 participants. The participants included in our analysis had characteristics comparable to those of the total GCKD study population (Fig. S2). We used mean values and standard deviations for normally distributed variables and median values and interquartile ranges for nonnormally distributed variables to describe the overall population and across the calcium, phosphate, P1NP, CTX1, OPG, FGF23, BAP, 25-OH vitamin D, and iPTH quintiles. Values for categorical variables are presented as frequency distributions with percentages. If patients failed to complete the 6.5-year follow-up period, censoring was performed at the time of the last follow-up, e.g., when participants left the study due to loss to follow-up or refused to further participate in the study. Patients were censored at death if it was not part of the outcome of interest; thus, all hazard estimates obtained from our models were cause specific. Data extraction from the main GCKD study database was performed in February 2021. At this time, 188 (2.3%) participants had been lost to follow-up, and 311 (5.9%) participants had prematurely left the study. All data collected during their active participation were used for this analysis.

We used Cox proportional hazard models to analyze the associations of OPG, FGF23, BAP, calcium, phosphate, P1NP, CTX1, iPTH, and 25-OH vitamin D with non-CV death, CV death, MACE, and hospitalization due to CHF during follow-up. We assessed each outcome with three statistical models. First, we applied a univariate model to OPG, FGF23, BAP, calcium, phosphate, P1NP, CTX1, iPTH and 25-OH vitamin D alone (Model 1). The subsequent multivariate model (Model 2) was additionally adjusted for traditional risk factors for CVD, including age, gender, body mass index (BMI), systolic blood pressure, presence of diabetes mellitus, history of smoking, preexisting CVD, laboratory parameters including LDL cholesterol, eGFR, UACR, hsCRP, serum albumin, and relevant medication, including use of statins, renin-angiotensin system inhibitors (RASi), anti-platelet aggregation agents, beta blockers, vitamin D and analogs supplementation, and mineralocorticoid receptor antagonists. The selection of variables adjusted for in Model 2 was also based on the differences in the clinical characteristics between the quintiles of the different biomarkers provided below. In multivariate Model 3, we additionally corrected for all other bone and mineral biomarkers included in this study. The resulting hazard ratios (HRs) are presented with 95% confidence intervals (CIs). Statistical analysis was performed with SAS version 9.4 (SAS Institute Inc., Cay, NC, USA). Data were visualized with SAS, R version 4.2.2, and GraphPad Prism version 9 (GraphPad Software Inc., San Diego, CA, USA).