Fracture risk prediction using bone mineral density and biochemical markers of bone and mineral metabolism in dialysis and non-dialysis CKD patients Authors: Jin Woo Choi, Ji Eun Kim Published: 2026 ABSTRACT BackgroundBone mineral density (BMD) is a recognized predictor of fracture risk in the general population, but its utility in chronic kidney disease (CKD) remains uncertain. This study evaluated BMD and biochemical markers of bone and mineral metabolism (BMMs)—parathyroid hormone (PTH) and bone-specific alkaline phosphatase (bone ALP)—as fracture predictors in CKD patients, both with and without dialysis.MethodsWe retrospectively analyzed 918 adult patients with CKD who underwent BMD and BMM assessments between 2010 and 2024. Patients were stratified by dialysis status (RRT: n = 514; non-RRT: n = 404); non-RRT patients who initiated RRT during follow-up were censored at the date of RRT initiation. Cox proportional hazards models (Firth correction) and Fine–Gray competing-risk models were applied, with adjustment for age, sex, BMI, and diabetes mellitus. Time-dependent receiver operating characteristic (ROC) curves were used to compare predictive performance for 5-year fracture risk.ResultsOver a median follow-up of 1,694 days (IQR 751–2,314), 83 fracture events occurred. Fracture risk did not significantly differ across tertiles of PTH or bone ALP in either group. In contrast, BMD was significantly associated with fracture risk in both groups: patients with osteoporosis (T-score ≤ − 2.5) had directionally elevated fracture risk in both groups, although the extremely wide confidence intervals around the categorical hazard ratios preclude precise interpretation of effect magnitude. More stable estimates from Fine–Gray competing-risk models showed that each one-unit increase in BMD T-score was associated with an approximately 45% lower fracture risk in both groups (RRT: SHR 0.54, 95% CI 0.38–0.75; non-RRT: SHR 0.57, 95% CI 0.41–0.79). Time-dependent ROC analysis confirmed BMD significantly outperformed BMMs for 5-year fracture risk prediction, with AUCs of 0.712 (RRT) and 0.658 (non-RRT).ConclusionBMD was strongly associated with fracture risk in CKD patients, regardless of dialysis status, and significantly outperformed BMMs in predicting fracture risk. In contrast, BMMs were not associated with fracture risk. These results support the role of BMD assessment in fracture risk evaluation across the full spectrum of CKD.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1186/s12882-026-05097-4. Introduction Chronic kidney disease-mineral bone disorder (CKD-MBD) is a systemic disorder observed in patients with CKD, involving abnormalities in mineral metabolism, vascular calcification, and alterations in bone structure [1]. These pathological changes lead to decreased bone density, disrupted micro-architecture, and compromised bone quality. These alterations increase fracture risk in CKD patients compared to the general population, with patients in CKD stages G3a-G5D having a 2.36 times increased risk of hip fracture [2]. In addition, fractures in hemodialysis patients are associated with 3.7 times higher mortality rate compared with the general hemodialysis population [3]. Bone mineral density (BMD) reliably predicts fractures in non-CKD populations. However, its application in CKD is limited due to heterogeneous bone turnover patterns, which range from high-turnover osteitis fibrosa to low-turnover adynamic bone disease [1]. Biochemical markers of bone and mineral metabolism (BMMs) such as bone-specific alkaline phosphatase (bone ALP) and parathyroid hormone (PTH) reflect dynamic bone remodeling. However, their predictive value in CKD remains inconsistent [4]. Previous studies have demonstrated that BMD serves as a critical predictor of fracture risk in CKD patients, particularly in non-dialysis cohorts. In contrast, BMMs have shown limited predictive value in dialysis populations [5]. In light of these findings, the present study aimed to evaluate the utility of BMD and BMMs in predicting fracture risk among both dialysis and non-dialysis CKD patients. Materials and methods Study population This study was approved by the Institutional Review Board of Korea University Guro Hospital (IRB No. 2023GR0442). The data were retrieved for research purposes on September 4, 2024. As a retrospective study using anonymized electronic medical records, the requirement for informed consent was waived. The authors had no access to identifiable personal information of the participants during or after data collection. We retrospectively analyzed data from adult patients with CKD who underwent BMMs and BMD assessments between 2010 and 2024. CKD was defined as the presence of estimated glomerular filtration rate (eGFR) < 60 mL/min/1.73 m² or dipstick albuminuria (1 + or greater on spot urinalysis), confirmed within 6 months of the index date. Eligible patients had at least one measurement of bone ALP during this period. Patients with a history of parathyroidectomy, prior fractures before the index date, or who were under 18 years of age were excluded. Patients who had undergone kidney transplantation were also excluded. For patients in the non-RRT group who initiated RRT during follow-up, follow-up was censored at the date of RRT initiation to avoid exposure misclassification while preserving information up to the change in exposure status. Patients were stratified based on renal replacement therapy (RRT) status into dialysis (RRT) and non-dialysis (non-RRT) groups. The final analytic cohort comprised 514 patients receiving RRT and 404 patients not on RRT at the index date, of whom 65 initiated RRT during follow-up and were censored at that time. A detailed flow diagram of patient selection is presented in Fig. 1. Fig. 1Flow diagram of patient selection. Of 1,256 patients with at least one bone ALP measurement between 2010 and 2024, patients with a history of parathyroidectomy or age under 18 years, prior fractures before the index date, and fractures occurring within 30 days of the index date were sequentially excluded. The final analytic cohort comprised 918 patients, of whom 514 were in the RRT group and 404 in the non-RRT group at the index date. Sixty-five patients in the non-RRT group initiated RRT during follow-up and were censored at the date of RRT initiation. RRT = renal replacement therapy Flow diagram of patient selection. Of 1,256 patients with at least one bone ALP measurement between 2010 and 2024, patients with a history of parathyroidectomy or age under 18 years, prior fractures before the index date, and fractures occurring within 30 days of the index date were sequentially excluded. The final analytic cohort comprised 918 patients, of whom 514 were in the RRT group and 404 in the non-RRT group at the index date. Sixty-five patients in the non-RRT group initiated RRT during follow-up and were censored at the date of RRT initiation. RRT = renal replacement therapy Data collection and variables The index date was defined as the first measurement date of bone ALP. Demographic and clinical variables including age, sex, body mass index (BMI), hypertension (HTN), and diabetes mellitus (DM) were recorded. eGFR was calculated from serum creatinine measured at the index date using the CKD-EPI equation and is reported only for the non-RRT group, as creatinine-based eGFR estimates are not physiologically interpretable in patients on maintenance dialysis. Dipstick albuminuria was defined as a spot urinalysis result of 1 + or greater within 6 months of the index date; patients without available urinalysis data were treated as missing. BMMs—PTH and bone ALP—were extracted with dates. Data on PTH and BMD were collected if measured within ± 6 months of the index date. Intact PTH was measured using an immunoradiometric assay (IRMA; Beckman Coulter, Brea, CA, USA). Bone ALP was measured using chemiluminescence immunoassay (CIA; Beckman coulter, Brea, CA, USA). BMD was assessed via dual-energy X-ray absorptiometry at the lumbar spine (L1–L4), femoral neck, and total hip. T-scores were recorded for each site, and the lowest T-score among all measured sites was used for osteoporosis classification, in accordance with WHO diagnostic criteria. Site-specific T-scores were not systematically extracted from the database; therefore, only the lowest T-score per patient was available for analysis. Fracture events were identified using ICD diagnostic codes indicative of osteoporotic fractures: S22, S32, S42, S52, S72, T08, T10, T12, or M48. Only fracture events occurring more than 30 days after the index date were included, to minimize reverse causation, as bone ALP testing may have been performed due to a recent fracture. For patients with multiple fracture events, only the first fracture was included as the primary outcome. Study outcomes The primary outcome was the risk of incident fracture according to levels of each BMM and BMD. Patients without a fracture were censored at the last follow-up date or on December 31, 2024, or — for patients in the non-RRT group who initiated RRT during follow-up — at the date of RRT initiation, whichever occurred first. As secondary analyses, we evaluated the associations between BMMs and BMD, and the predictive performance of BMMs and BMD for fracture risk. Statistical analysis Baseline characteristics were compared by RRT status using t-tests or Wilcoxon rank-sum tests for continuous variables and chi-square tests for categorical variables. BMM distributions and inter-marker correlations were analyzed separately for RRT and non-RRT groups. Differences in marker distributions between RRT and non-RRT groups were tested using the Kolmogorov–Smirnov test. Fracture risk was assessed using Cox proportional hazards models with tertile categorization of each BMM. Restricted cubic spline models were constructed to assess associations between BMMs and fracture risk. BMD-fracture associations were also evaluated using Cox models with Firth correction to address sparse events. To account for mortality as a competing event, Fine–Gray subdistribution hazard models were used for continuous BMD analyses. Models were adjusted using a hierarchical framework as detailed below. Given a total of 83 fracture events distributed across the two strata (RRT and non-RRT), we adopted a hierarchical modeling framework to balance confounding adjustment with model stability and avoid overfitting. Model 1 was minimally adjusted for age and sex. Model 2 (primary parsimonious model) additionally adjusted for BMI and diabetes mellitus, selected a priori based on established clinical relevance to fracture risk in CKD; this maintained an events-per-variable ratio above the conventional threshold of 10 within each stratum. Model 3 added hypertension, serum calcium, phosphorus, and albumin in both groups, with dipstick albuminuria additionally included in non-RRT analyses only, given its limited interpretability in dialysis patients. In BMD–fracture analyses (Table 3), Model 3 also included PTH and bone-specific alkaline phosphatase as covariates. In BMM–fracture analyses (Table 2), the BMM being tested as the primary exposure was not included as a covariate; the other BMM was included (e.g., for PTH tertile analyses, bone ALP was a Model 3 covariate, and vice versa). Sensitivity analyses were performed by restricting RRT analyses to hemodialysis patients only (excluding 62 peritoneal dialysis patients). ROC curve analysis for 5-year fracture prediction was performed using the time-dependent ROC method (timeROC package), with area under the curve (AUC) and 95% confidence intervals (CI) calculated. DeLong tests were used to compare AUCs between each BMM and BMD. All statistical analyses were conducted in R (version 4.4.3). Data availability The data generated in this study are available upon request from the corresponding author (permitted only for data types for which a community-recognized, structured repository does not exist). Results Baseline characteristics and biochemical markers of bone and mineral metabolism in RRT and non-RRT groups Among 918 patients included in the final analysis, 514 (56.0%) were undergoing dialysis. 404 (44.0%) were not on RRT, of whom 65 initiated RRT during follow-up and were censored at the date of RRT initiation. Compared to non-RRT patients, those in the RRT group were younger, more likely to be male, and had a higher prevalence of hypertension and diabetes. Levels of PTH and bone ALP were markedly elevated in the RRT group. Notably, the median BMD T-score was also slightly higher (less negative) in RRT patients, with a lower proportion classified as osteoporosis. The full baseline characteristics are summarized in Table 1. Table 1Baseline characteristics of the study populationVariablesMissing nNon-RRT (n = 404)RRT (n = 514)p-valueAge, years064.2 ± 14.557.4 ± 12.8< 0.001Sex0< 0.001 Female263 (65.1%)232 (45.1%) Male141 (34.9%)282 (54.9%)BMI, kg/m²3223.4 ± 4.123.0 ± 3.90.054Dialysis modality0- Hemodialysis-452 (87.9%) Peritoneal dialysis-62 (12.1%)Creatinine, mg/dL02.3 ± 3.027.06 ± 3.8< 0.001eGFR, ml/min/1.73m2062.0 ± 37.4N/A-Dipstick albuminuria, %70107 (26.4%)150 (29.2%)< 0.001HTN, %0319 (79.0%)490 (95.3%)< 0.001DM, %0208 (48.5%)303 (58.9%)0.028Ca, mg/dL519.3 ± 0.89.3 ± 1.00.611P, mg/dL693.7 ± 1.14.7 ± 1.4< 0.001Albumin, g/dL524.0 ± 0.44.0 ± 0.40.001PTH, pg/mL7046.8 [27.3–99.5]174.0 [73.9-395.3]< 0.001bone ALP, µg/L012.2 [9.1–18.3]15.5 [11.2–27.6]< 0.001BMD, T-score185-2.4 [-3.1- -1.7]-2.0 [-2.9- -1.3]< 0.001BMD category, %1850.004 Normal42 (12.4%)68 (17.2%) Osteopenia135 (39.9%)186 (47.1%) Osteoporosis161 (47.6%)141 (35.7%)Data are presented as mean ± standard deviation, median (interquartile range), or number (%). PTH, bone ALP, and BMD are presented as median (IQR) given their skewed distributions. BMD T-score represents the lowest T-score among the lumbar spine (L1–L4), femoral neck, and total hip. eGFR was not calculated for the RRT group, as creatinine-based eGFR estimates are not physiologically interpretable in patients on maintenance dialysisBMI = body mass index, HTN = hypertension, DM = diabetes mellitus, PTH = parathyroid hormone, bone ALP = bone-specific alkaline phosphatase, BMD = bone mineral density Table 1Baseline characteristics of the study populationVariablesMissing nNon-RRT (n = 404)RRT (n = 514)p-valueAge, years064.2 ± 14.557.4 ± 12.8< 0.001Sex0< 0.001 Female263 (65.1%)232 (45.1%) Male141 (34.9%)282 (54.9%)BMI, kg/m²3223.4 ± 4.123.0 ± 3.90.054Dialysis modality0- Hemodialysis-452 (87.9%) Peritoneal dialysis-62 (12.1%)Creatinine, mg/dL02.3 ± 3.027.06 ± 3.8< 0.001eGFR, ml/min/1.73m2062.0 ± 37.4N/A-Dipstick albuminuria, %70107 (26.4%)150 (29.2%)< 0.001HTN, %0319 (79.0%)490 (95.3%)< 0.001DM, %0208 (48.5%)303 (58.9%)0.028Ca, mg/dL519.3 ± 0.89.3 ± 1.00.611P, mg/dL693.7 ± 1.14.7 ± 1.4< 0.001Albumin, g/dL524.0 ± 0.44.0 ± 0.40.001PTH, pg/mL7046.8 [27.3–99.5]174.0 [73.9-395.3]< 0.001bone ALP, µg/L012.2 [9.1–18.3]15.5 [11.2–27.6]< 0.001BMD, T-score185-2.4 [-3.1- -1.7]-2.0 [-2.9- -1.3]< 0.001BMD category, %1850.004 Normal42 (12.4%)68 (17.2%) Osteopenia135 (39.9%)186 (47.1%) Osteoporosis161 (47.6%)141 (35.7%)Data are presented as mean ± standard deviation, median (interquartile range), or number (%). PTH, bone ALP, and BMD are presented as median (IQR) given their skewed distributions. BMD T-score represents the lowest T-score among the lumbar spine (L1–L4), femoral neck, and total hip. eGFR was not calculated for the RRT group, as creatinine-based eGFR estimates are not physiologically interpretable in patients on maintenance dialysisBMI = body mass index, HTN = hypertension, DM = diabetes mellitus, PTH = parathyroid hormone, bone ALP = bone-specific alkaline phosphatase, BMD = bone mineral density Baseline characteristics of the study population Data are presented as mean ± standard deviation, median (interquartile range), or number (%). PTH, bone ALP, and BMD are presented as median (IQR) given their skewed distributions. BMD T-score represents the lowest T-score among the lumbar spine (L1–L4), femoral neck, and total hip. eGFR was not calculated for the RRT group, as creatinine-based eGFR estimates are not physiologically interpretable in patients on maintenance dialysis BMI = body mass index, HTN = hypertension, DM = diabetes mellitus, PTH = parathyroid hormone, bone ALP = bone-specific alkaline phosphatase, BMD = bone mineral density Fracture risk by BMM tertiles in RRT and non-RRT groups In the overall cohort, 83 fracture events occurred during follow-up. The median follow-up duration was 1,694 days (IQR, 751–2,314). The incidence rate was 1.98 per 100 person-years. When stratified by RRT status, 46 fractures occurred in the non-RRT group and 37 in the RRT group, corresponding to incidence rates of 2.46 and 1.59 per 100 person-years, respectively. The median follow-up durations were 1,536 days (IQR, 702–2,416 days) and 1,758 days (IQR, 809–2,150 days), respectively. To further assess the prognostic relevance of BMMs, we evaluated their associations with fracture risk using tertile-based analyses stratified by RRT status. Tertiles were calculated separately within each group. In Cox models, no statistically significant associations were observed between BMM tertiles and fracture risk in either dialysis or non-dialysis groups across all model specifications (Table 2). When using tertile 1 as the reference, hazard ratios for tertiles 2 and 3 did not show a consistent increase in risk for either PTH or bone ALP, regardless of dialysis status. The null findings were consistent in the parsimonious model (Model 2: adjusted for age, sex, BMI, and diabetes) and in the extended model (Model 3: additionally adjusted for hypertension, serum calcium, phosphorus, and albumin, in both groups, the alternate BMM as a covariate (e.g., bone ALP in PTH analyses, and vice versa), and dipstick albuminuria in non-RRT analyses only), supporting the robustness of these results despite the limited number of fracture events. Table 2Comparison of fracture risk by BMM tertiles in RRT and non-RRT groupsGroupMarkerComparisonUnadjusted HR (95% CI)Model 1 h(95% CI)Model 2 h (95% CI)*Model 3 h (95% CI)†RRTPTHT2 vs. T10.88 (0.43–1.80)1.13 (0.53–2.40)1.07 (0.49–2.31)1.05 (0.48–2.30)T3 vs. T10.38 (0.15–0.99)0.54 (0.20–1.48)0.51 (0.18–1.41)0.62 (0.21–1.78)Bone ALPT2 vs. T10.86 (0.39–1.90)1.04 (0.46–2.34)0.94 (0.41–2.15)1.20 (0.50–2.88)T3 vs. T10.80 (0.35–1.80)1.14 (0.47–2.72)1.10 (0.46–2.66)1.27 (0.52–3.12)Non-RRTPTHT2 vs. T10.93 (0.42–2.07)1.09 (0.49–2.43)0.97 (0.41–2.32)0.81 (0.30–2.21)T3 vs. T10.66 (0.26–1.69)1.17 (0.44–3.11)1.20 (0.44–3.27)0.42 (0.09–1.93)Bone ALPT2 vs. T10.68 (0.31–1.46)0.78 (0.36–1.69)0.80 (0.35–1.82)0.84 (0.29–2.49)T3 vs. T11.15 (0.59–2.25)1.66 (0.84–3.30)1.97 (0.96–4.07)1.72 (0.62–4.74)Data are presented as hazard ratios (95% confidence intervals)* Model 1: adjusted for age and sex† Model 2 (primary parsimonious): adjusted for age, sex, BMI, and diabetes mellitus‡ Model 3 (extended): additionally adjusted for hypertension, serum calcium, phosphorus, and albumin in both groups, with dipstick albuminuria included in non-RRT analyses only. In each BMM tertile analysis, the other BMM was also included as a covariate (i.e., bone ALP was a covariate in PTH tertile analyses, and PTH was a covariate in bone ALP tertile analyses)BMM = biochemical markers of bone and mineral metabolism, RRT = renal replacement therapy, PTH = parathyroid hormone, bone ALP = bone-specific alkaline phosphatase Table 2Comparison of fracture risk by BMM tertiles in RRT and non-RRT groupsGroupMarkerComparisonUnadjusted HR (95% CI)Model 1 h(95% CI)Model 2 h (95% CI)*Model 3 h (95% CI)†RRTPTHT2 vs. T10.88 (0.43–1.80)1.13 (0.53–2.40)1.07 (0.49–2.31)1.05 (0.48–2.30)T3 vs. T10.38 (0.15–0.99)0.54 (0.20–1.48)0.51 (0.18–1.41)0.62 (0.21–1.78)Bone ALPT2 vs. T10.86 (0.39–1.90)1.04 (0.46–2.34)0.94 (0.41–2.15)1.20 (0.50–2.88)T3 vs. T10.80 (0.35–1.80)1.14 (0.47–2.72)1.10 (0.46–2.66)1.27 (0.52–3.12)Non-RRTPTHT2 vs. T10.93 (0.42–2.07)1.09 (0.49–2.43)0.97 (0.41–2.32)0.81 (0.30–2.21)T3 vs. T10.66 (0.26–1.69)1.17 (0.44–3.11)1.20 (0.44–3.27)0.42 (0.09–1.93)Bone ALPT2 vs. T10.68 (0.31–1.46)0.78 (0.36–1.69)0.80 (0.35–1.82)0.84 (0.29–2.49)T3 vs. T11.15 (0.59–2.25)1.66 (0.84–3.30)1.97 (0.96–4.07)1.72 (0.62–4.74)Data are presented as hazard ratios (95% confidence intervals)* Model 1: adjusted for age and sex† Model 2 (primary parsimonious): adjusted for age, sex, BMI, and diabetes mellitus‡ Model 3 (extended): additionally adjusted for hypertension, serum calcium, phosphorus, and albumin in both groups, with dipstick albuminuria included in non-RRT analyses only. In each BMM tertile analysis, the other BMM was also included as a covariate (i.e., bone ALP was a covariate in PTH tertile analyses, and PTH was a covariate in bone ALP tertile analyses)BMM = biochemical markers of bone and mineral metabolism, RRT = renal replacement therapy, PTH = parathyroid hormone, bone ALP = bone-specific alkaline phosphatase Comparison of fracture risk by BMM tertiles in RRT and non-RRT groups Data are presented as hazard ratios (95% confidence intervals) * Model 1: adjusted for age and sex † Model 2 (primary parsimonious): adjusted for age, sex, BMI, and diabetes mellitus ‡ Model 3 (extended): additionally adjusted for hypertension, serum calcium, phosphorus, and albumin in both groups, with dipstick albuminuria included in non-RRT analyses only. In each BMM tertile analysis, the other BMM was also included as a covariate (i.e., bone ALP was a covariate in PTH tertile analyses, and PTH was a covariate in bone ALP tertile analyses) BMM = biochemical markers of bone and mineral metabolism, RRT = renal replacement therapy, PTH = parathyroid hormone, bone ALP = bone-specific alkaline phosphatase These findings were further supported by restricted cubic spline analyses, which demonstrated no significant nonlinear associations between continuous marker levels and fracture risk across the full observed range in either group (Fig. 2). Fig. 2Hazard ratios for fracture risk across continuous BMM levels in RRT and non-RRT groups. Hazard ratios for fracture risk are plotted across continuous levels of bone-specific alkaline phosphatase (bone ALP, left panels, green) and parathyroid hormone (PTH, right panels, orange) in the RRT group (upper panels) and the non-RRT group (lower panels). Solid lines represent point estimates and shaded areas represent 95% confidence intervals. The dashed horizontal line indicates a hazard ratio of 1. Models were adjusted using the parsimonious primary model (Model 2: age, sex, BMI, and diabetes mellitus). BMM = biochemical markers of bone and mineral metabolism, RRT = renal replacement therapy, Bone ALP = bone-specific alkaline phosphatase, PTH = parathyroid hormone Hazard ratios for fracture risk across continuous BMM levels in RRT and non-RRT groups. Hazard ratios for fracture risk are plotted across continuous levels of bone-specific alkaline phosphatase (bone ALP, left panels, green) and parathyroid hormone (PTH, right panels, orange) in the RRT group (upper panels) and the non-RRT group (lower panels). Solid lines represent point estimates and shaded areas represent 95% confidence intervals. The dashed horizontal line indicates a hazard ratio of 1. Models were adjusted using the parsimonious primary model (Model 2: age, sex, BMI, and diabetes mellitus). BMM = biochemical markers of bone and mineral metabolism, RRT = renal replacement therapy, Bone ALP = bone-specific alkaline phosphatase, PTH = parathyroid hormone Fracture risk according to BMD category in RRT and non-RRT groups BMD measurements were available in 733 patients. Among them, 110 (15.0%) were classified as normal, 321 (43.8%) as osteopenia, and 302 (41.2%) as osteoporosis. In the non-RRT group, 42 (12.4%), 135 (39.9%), and 161 (47.6%) patients were classified as normal, osteopenia, and osteoporosis, respectively. In the RRT group, the corresponding numbers were 68 (17.2%), 186 (47.1%), and 141 (35.7%). Fracture events increased progressively with worsening BMD category in both groups. In the non-RRT group, fracture rates were 0%, 9.6%, and 18.6% in patients with normal BMD, osteopenia, and osteoporosis, respectively. A similar pattern was observed in the RRT group (0%, 5.4%, and 14.2%, respectively). In contrast to BMMs, BMD was significantly associated with fracture risk in both RRT and non-RRT groups. Kaplan–Meier curves demonstrated lower fracture-free survival with decreasing BMD in both groups (Fig. 3). In hierarchical Cox models with Firth correction, osteoporosis was directionally associated with increased fracture risk in both groups across all model specifications, with no fractures observed in the normal BMD category. The categorical hazard ratios consistently indicated an elevated fracture risk among patients with osteoporosis (e.g., Model 2 h 17.36 in the RRT group, HR 16.52 in the non-RRT group), but the extremely wide confidence intervals (lower bound ~ 2, upper bound > 2000)—a consequence of sparse-event instability and the absence of fractures in the normal BMD category—preclude precise interpretation of the effect magnitude. We therefore emphasize the directionality of these associations rather than the point estimates themselves. Osteopenia also showed a similar directional pattern but did not reach statistical significance in the RRT group. In competing-risk analyses using Fine–Gray models with continuous BMD, each one-unit increase in BMD T-score was associated with an approximately 45% lower fracture risk in both groups (RRT: SHR 0.54, 95% CI 0.38–0.75; non-RRT: SHR 0.57, 95% CI 0.41–0.79). Results were consistent across minimally adjusted (Model 1), parsimonious (Model 2), and extended (Model 3) models (Table 3), supporting the robustness of the BMD–fracture association. Table 3Fracture risk according to BMD in RRT and non-RRT groupsUnadjustedModel 1‡Model 2§Model 3¶ Non-RRT Categorical HR (95% CI)* HR (95% CI)* HR (95% CI)* HR (95% CI)* Normal1 (ref)1 (ref)1 (ref)1 (ref) Osteopenia9.27 (1.23–1187)8.74 (1.14–1121)8.32 (1.07–1073)4.71 (0.51–684) Osteoporosis20.95 (2.96–2655)17.79 (2.34–2282)16.52 (2.12–2131)12.34 (1.39–1807) Continuous SHR (95% CI)† SHR (95% CI)† SHR (95% CI)† SHR (95% CI)† BMD (per + 1 T-score)0.56 (0.43–0.72)0.59 (0.44–0.79)0.57 (0.41–0.79)0.45 (0.26–0.77) RRT Categorical HR (95% CI)* HR (95% CI)* HR (95% CI)* HR (95% CI)* Normal1 (ref)1 (ref)1 (ref)1 (ref) Osteopenia6.91 (0.89–889)6.38 (0.81–822)6.08 (0.77–784)4.53 (0.57–586) Osteoporosis17.73 (2.43–2257)15.80 (2.08–2025)17.36 (2.22–2239)15.27 (1.90–1979) Continuous SHR (95% CI)† SHR (95% CI)† SHR (95% CI)† SHR (95% CI)† BMD (per + 1 T-score)0.57 (0.46–0.72)0.57 (0.43–0.75)0.54 (0.38–0.75)0.53 (0.37–0.77)* Categorical BMD groups (Normal, Osteopenia, Osteoporosis) were analyzed using Cox proportional hazards models with Firth correction† Continuous BMD (T-score) was analyzed using Fine–Gray subdistribution hazard models, with mortality treated as a competing event. SHR represents the change in fracture risk per one-unit increase in BMD T-score‡ Model 1: adjusted for age and sex§ Model 2 (primary parsimonious model): adjusted for age, sex, BMI, and diabetes mellitus¶ Model 3 (extended sensitivity model): additionally adjusted for hypertension, serum calcium, phosphorus, albumin, PTH, and bone-specific alkaline phosphatase in both groups, with dipstick albuminuria included in non-RRT analyses onlyBMD = bone mineral density; RRT = renal replacement therapy; HR = hazard ratio; SHR = subdistribution hazard ratio; CI = confidence interval; PTH = parathyroid hormone Table 3Fracture risk according to BMD in RRT and non-RRT groupsUnadjustedModel 1‡Model 2§Model 3¶ Non-RRT Categorical HR (95% CI)* HR (95% CI)* HR (95% CI)* HR (95% CI)* Normal1 (ref)1 (ref)1 (ref)1 (ref) Osteopenia9.27 (1.23–1187)8.74 (1.14–1121)8.32 (1.07–1073)4.71 (0.51–684) Osteoporosis20.95 (2.96–2655)17.79 (2.34–2282)16.52 (2.12–2131)12.34 (1.39–1807) Continuous SHR (95% CI)† SHR (95% CI)† SHR (95% CI)† SHR (95% CI)† BMD (per + 1 T-score)0.56 (0.43–0.72)0.59 (0.44–0.79)0.57 (0.41–0.79)0.45 (0.26–0.77) RRT Categorical HR (95% CI)* HR (95% CI)* HR (95% CI)* HR (95% CI)* Normal1 (ref)1 (ref)1 (ref)1 (ref) Osteopenia6.91 (0.89–889)6.38 (0.81–822)6.08 (0.77–784)4.53 (0.57–586) Osteoporosis17.73 (2.43–2257)15.80 (2.08–2025)17.36 (2.22–2239)15.27 (1.90–1979) Continuous SHR (95% CI)† SHR (95% CI)† SHR (95% CI)† SHR (95% CI)† BMD (per + 1 T-score)0.57 (0.46–0.72)0.57 (0.43–0.75)0.54 (0.38–0.75)0.53 (0.37–0.77)* Categorical BMD groups (Normal, Osteopenia, Osteoporosis) were analyzed using Cox proportional hazards models with Firth correction† Continuous BMD (T-score) was analyzed using Fine–Gray subdistribution hazard models, with mortality treated as a competing event. SHR represents the change in fracture risk per one-unit increase in BMD T-score‡ Model 1: adjusted for age and sex§ Model 2 (primary parsimonious model): adjusted for age, sex, BMI, and diabetes mellitus¶ Model 3 (extended sensitivity model): additionally adjusted for hypertension, serum calcium, phosphorus, albumin, PTH, and bone-specific alkaline phosphatase in both groups, with dipstick albuminuria included in non-RRT analyses onlyBMD = bone mineral density; RRT = renal replacement therapy; HR = hazard ratio; SHR = subdistribution hazard ratio; CI = confidence interval; PTH = parathyroid hormone Fracture risk according to BMD in RRT and non-RRT groups * Categorical BMD groups (Normal, Osteopenia, Osteoporosis) were analyzed using Cox proportional hazards models with Firth correction † Continuous BMD (T-score) was analyzed using Fine–Gray subdistribution hazard models, with mortality treated as a competing event. SHR represents the change in fracture risk per one-unit increase in BMD T-score ‡ Model 1: adjusted for age and sex § Model 2 (primary parsimonious model): adjusted for age, sex, BMI, and diabetes mellitus ¶ Model 3 (extended sensitivity model): additionally adjusted for hypertension, serum calcium, phosphorus, albumin, PTH, and bone-specific alkaline phosphatase in both groups, with dipstick albuminuria included in non-RRT analyses only BMD = bone mineral density; RRT = renal replacement therapy; HR = hazard ratio; SHR = subdistribution hazard ratio; CI = confidence interval; PTH = parathyroid hormone Fig. 3Fracture-free survival according to bone mineral density categories in RRT (a) and non-RRT (b) patients. Kaplan–Meier curves showing fracture-free survival probability according to BMD category (normal, green; osteopenia, orange; osteoporosis, red) in RRT (p = 0.0012) and non-RRT (p < 0.001) groups. Tick marks indicate censored observations. Numbers at risk are shown below each panel. RRT = renal replacement therapy, BMD = bone mineral density Fracture-free survival according to bone mineral density categories in RRT (a) and non-RRT (b) patients. Kaplan–Meier curves showing fracture-free survival probability according to BMD category (normal, green; osteopenia, orange; osteoporosis, red) in RRT (p = 0.0012) and non-RRT (p < 0.001) groups. Tick marks indicate censored observations. Numbers at risk are shown below each panel. RRT = renal replacement therapy, BMD = bone mineral density BMM associations with BMD Correlation analysis showed inverse relationships between PTH and BMD in both RRT and non-RRT groups. In the RRT group, both PTH (R = − 0.18, p < 0.001) and bone ALP (R = − 0.28, p < 0.001) were negatively correlated with BMD. In contrast, in the non-RRT group, PTH showed a non-significant trend toward an inverse correlation with BMD (R = − 0.11, p = 0.098), and bone ALP was not correlated with BMD (R = − 0.02, p = 0.68) (Fig. 4). Fig. 4Correlation analysis between BMD and BMMs in RRT and non-RRT patients. Scatter plots with linear regression lines show the correlation between BMD (T-score) and bone ALP (left panels, green) or PTH (right panels, orange) in the RRT group (upper panels) and the non-RRT group (lower panels). Shaded areas represent 95% confidence intervals of the regression line. Pearson correlation coefficients (R) and p-values are shown within each panel. BMD = bone mineral density, BMM = biochemical markers of bone and mineral metabolism, PTH = parathyroid hormone, Bone ALP = bone-specific alkaline phosphatase Correlation analysis between BMD and BMMs in RRT and non-RRT patients. Scatter plots with linear regression lines show the correlation between BMD (T-score) and bone ALP (left panels, green) or PTH (right panels, orange) in the RRT group (upper panels) and the non-RRT group (lower panels). Shaded areas represent 95% confidence intervals of the regression line. Pearson correlation coefficients (R) and p-values are shown within each panel. BMD = bone mineral density, BMM = biochemical markers of bone and mineral metabolism, PTH = parathyroid hormone, Bone ALP = bone-specific alkaline phosphatase Fracture risk prediction: ROC analysis Time-dependent ROC analysis demonstrated that BMD consistently showed the highest performance for 5-year fracture prediction in both RRT (AUC 0.712, 95% CI 0.602–0.822) and non-RRT (AUC 0.658, 95% CI 0.546–0.769) groups (Fig. 5). DeLong tests confirmed that BMD significantly outperformed bone ALP in both groups (RRT: p < 0.001; non-RRT: p = 0.014) and outperformed PTH in the RRT group (p = 0.034), with a borderline difference in the non-RRT group (p = 0.054) (Additional file 1). Fig. 5Time-dependent ROC Curves for 5-year fracture risk prediction in RRT (left panel) and non-RRT (right panel) patients. Curves compare the predictive performance of BMD (green), bone ALP (red), and PTH (blue). Area under the curve (AUC) values with 95% confidence intervals are shown within each panel. RRT = renal replacement therapy, BMD = bone mineral density, Bone ALP = bone-specific alkaline phosphatase, PTH = parathyroid hormone Time-dependent ROC Curves for 5-year fracture risk prediction in RRT (left panel) and non-RRT (right panel) patients. Curves compare the predictive performance of BMD (green), bone ALP (red), and PTH (blue). Area under the curve (AUC) values with 95% confidence intervals are shown within each panel. RRT = renal replacement therapy, BMD = bone mineral density, Bone ALP = bone-specific alkaline phosphatase, PTH = parathyroid hormone Sensitivity analysis: Hemodialysis-only subset To address potential heterogeneity introduced by including peritoneal dialysis patients in the RRT group, we performed sensitivity analyses restricted to hemodialysis patients (n = 452, 31 fracture events). The findings were consistent with the main RRT analyses across all major outcomes. In Cox models with Firth correction, osteoporosis remained directionally associated with increased fracture risk, with similarly wide confidence intervals (Model 2 h 13.68, 95% CI 1.70–1775; p = 0.008). In Fine–Gray competing-risk analyses, each one-unit increase in BMD T-score remained associated with lower fracture risk (SHR 0.57, 95% CI 0.40–0.81; p = 0.002). The 5-year time-dependent AUC for BMD in the hemodialysis subset (0.700, 95% CI 0.580–0.820) was comparable to that of the full RRT group (0.712), and BMD remained the most discriminative predictor among the three markers. These results indicate that the BMD–fracture associations and the superior predictive performance of BMD over BMMs in the RRT group were not driven by inclusion of peritoneal dialysis patients (Additional file 2). Discussion This study demonstrates that BMD is strongly associated with fracture risk in patients with CKD, across both dialysis and non-dialysis populations, whereas BMMs—including PTH and bone ALP—did not show significant associations with future fracture risk. In both groups, osteoporosis-defined BMD thresholds (T-score ≤ − 2.5) were directionally associated with increased fracture risk. Competing-risk analyses using Fine–Gray models, accounting for mortality as a competing event, yielded consistent results, further supporting the robustness of the BMD–fracture association in this high-mortality population. Time-dependent ROC analysis further confirmed that BMD significantly outperformed BMMs for 5-year fracture prediction (DeLong tests, p ≤ 0.034 in three of four comparisons). Our findings align with the evolution of international guidelines. The 2009 KDIGO CKD-MBD guideline recommended against routine BMD assessment in patients with CKD stages G3a–G5D, citing limited evidence from early cross-sectional studies and the inability of DXA to distinguish histological subtypes of bone disease [6]. However, the 2017 guideline update marked a paradigm shift, informed by four prospective cohort studies demonstrating that DXA BMD predicts fracture risk across CKD stages, from G3a to G5D [7]. Among these, Iimori et al. uniquely focused on dialysis patients, reporting that BMD at the hip and femoral neck predicted fracture risk, especially in women with lower PTH levels, suggestive of low-turnover bone disease [8]. Their study also identified bone ALP as an independent predictor of fractures, regardless of PTH levels. However, an important methodological distinction exists between our study and that of Iimori et al. While they assessed BMM levels at the time of fracture occurrence, inherently enhancing discriminative performance (“incident values”), our study employed baseline measurements obtained prior to any fracture, with longitudinal prediction over a 5-year period. This reflects clinical practice more accurately, where decisions are based on pre-event risk stratification rather than biomarkers at the time of fracture. Further examination of the studies underpinning the 2017 KDIGO guideline underscores the strengths of our analysis. Naylor et al. demonstrated the utility of FRAX in predicting osteoporotic fractures in CKD patients but did not differentiate dialysis status [9]. Yenchek et al. reported that BMD predicted fracture risk in individuals with eGFR < 60 mL/min/1.73 m²; however, patients with eGFR < 15 mL/min/1.73 m² comprised only 1% of their cohort, limiting generalizability to end-stage kidney disease (ESKD) populations [10]. In contrast, our study comprehensively evaluated both dialysis and non-dialysis cohorts, providing contemporary evidence across the full spectrum of CKD severity, including patients with advanced renal dysfunction. In our study, BMMs including PTH and bone ALP were not significantly associated with incident fractures in either dialysis or non-dialysis patients. This is consistent with prior studies that failed to replicate BMM–fracture associations after adjusting for confounding variables and competing mortality [1, 4, 11]. The discordance with studies reporting significant associations may be attributed to several factors [12–14]. Prior studies have reported conflicting directions of association suggesting a potentially non-linear or U-shaped relationship between PTH and fracture risk that may not be captured by tertile-based analyses [13, 14]. The DOPPS study used total alkaline phosphatase rather than bone-specific ALP, potentially incorporating contributions from non-osseous sources [12]; our use of bone-specific ALP, while more specific, similarly failed to predict fracture risk. Significant inverse correlations between BMMs and BMD were observed in the RRT group (PTH: R = − 0.18; bone ALP: R = − 0.28) but were weaker and non-significant in the non-RRT group, suggesting that bone turnover dysregulation may play a more prominent biological role in dialysis-dependent populations. This dissociation between biological plausibility and predictive utility underscores the complexity of bone disease in CKD and highlights the limitations of single-timepoint BMM measurements as fracture risk tools. However, our study has several limitations. First, the absence of data on CKD–MBD-specific pharmacotherapy (phosphate binders, active vitamin D analogs, and calcimimetics) is an important limitation, as these agents directly modulate PTH, bone turnover, and potentially BMD; treatment-induced suppression of PTH and bone ALP could particularly attenuate expected BMM–fracture associations, although the impact on BMD–fracture associations is likely smaller. Second, the retrospective design may have introduced unmeasured confounding, including serum vitamin D status and prior anti-osteoporotic treatment, and BMD and PTH data were not available in all patients, potentially introducing selection bias. Third, the number of fracture events was relatively small (n = 83), with no fractures observed in the normal BMD category, resulting in wide confidence intervals despite Firth penalization; the hierarchical modeling framework with a parsimonious primary model yielded consistent results, though the magnitude of effect should be interpreted cautiously. Fourth, creatinine-based eGFR is not physiologically interpretable in dialysis patients and was therefore not reported for the RRT group; we also lacked data on residual kidney function, which may carry prognostic information in this population. Reliance on single-timepoint BMM measurements may also have limited the ability to capture dynamic changes in bone turnover. Lastly, as a single-center study conducted in Korea, generalizability to other populations should be approached with caution. In conclusion, our study suggests the pivotal role of BMD in fracture risk stratification across CKD stages, including dialysis-dependent patients. In contrast to BMMs, which did not predict fracture risk in either group, BMD was significantly and robustly associated with future fractures across multiple model specifications and was the strongest discriminator for 5-year fracture prediction. These findings support broader integration of BMD assessment into routine clinical practice for dialysis patients and underscore the importance of adhering to current guidelines recommending BMD assessment for fracture risk stratification. Future research is needed to identify novel biomarkers or imaging-based metrics that better reflect fracture risk in CKD populations. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1: Additional file 1: Area under the curve (AUC) for fracture prediction by BMMs and bone mineral density, stratified by dialysis status.Description of data: AUCs were derived from time-dependent ROC curves for 5-year fracture prediction. P-values from DeLong tests represent comparisons with BMD as the reference. Values in bold indicate statistical significance at p < 0.05. AUC = area under the receiver operating characteristic curve, CI = confidence interval, RRT = renal replacement therapy, BMD = bone mineral density, bone ALP = bone-specific alkaline phosphatase, PTH = parathyroid hormone. Supplementary Material 1: Additional file 1: Area under the curve (AUC) for fracture prediction by BMMs and bone mineral density, stratified by dialysis status.Description of data: AUCs were derived from time-dependent ROC curves for 5-year fracture prediction. P-values from DeLong tests represent comparisons with BMD as the reference. Values in bold indicate statistical significance at p < 0.05. AUC = area under the receiver operating characteristic curve, CI = confidence interval, RRT = renal replacement therapy, BMD = bone mineral density, bone ALP = bone-specific alkaline phosphatase, PTH = parathyroid hormone. Supplementary Material 2: Additional file 2: Sensitivity analysis comparing the main RRT analyses with the hemodialysis-only subset.Description of data: Cox proportional hazards models with Firth correction (categorical BMD), Fine–Gray subdistribution hazard models (continuous BMD), and time-dependent ROC analyses (5-year AUC) are presented for the full RRT cohort (n = 514) and for the hemodialysis-only subset (n = 452). Model 2 is the primary parsimonious model (adjusted for age, sex, BMI, and diabetes mellitus). Results were consistent between the two analyses, indicating that the BMD–fracture associations and the predictive performance of BMD were not driven by inclusion of peritoneal dialysis patients. RRT = renal replacement therapy, HD = hemodialysis, BMD = bone mineral density, HR = hazard ratio (Cox), SHR = subdistribution hazard ratio (Fine–Gray), CI = confidence interval, AUC = area under the receiver operating characteristic curve, bone ALP = bone-specific alkaline phosphatase, PTH = parathyroid hormone. Supplementary Material 2: Additional file 2: Sensitivity analysis comparing the main RRT analyses with the hemodialysis-only subset.Description of data: Cox proportional hazards models with Firth correction (categorical BMD), Fine–Gray subdistribution hazard models (continuous BMD), and time-dependent ROC analyses (5-year AUC) are presented for the full RRT cohort (n = 514) and for the hemodialysis-only subset (n = 452). Model 2 is the primary parsimonious model (adjusted for age, sex, BMI, and diabetes mellitus). Results were consistent between the two analyses, indicating that the BMD–fracture associations and the predictive performance of BMD were not driven by inclusion of peritoneal dialysis patients. RRT = renal replacement therapy, HD = hemodialysis, BMD = bone mineral density, HR = hazard ratio (Cox), SHR = subdistribution hazard ratio (Fine–Gray), CI = confidence interval, AUC = area under the receiver operating characteristic curve, bone ALP = bone-specific alkaline phosphatase, PTH = parathyroid hormone.