Discriminants of Prevalent Fractures in Chronic Kidney Disease

Abstract

Patients with chronic kidney disease (CKD) have higher rates of fracture than the general population. Increased bone remodeling, leading to microarchitectural deterioration and increased fragility, may accompany declining kidney function, but there are no reliable methods to identify patients at increased risk for fracture. In this cross-sectional study of 82 patients with predialysis CKD, high-resolution imaging revealed that the 23 patients with current fractures had significantly lower areal density at the femoral neck; total, cortical, and trabecular volumetric bone density; cortical area and thickness; and trabecular thickness. Compared with levels in the lowest tertile, higher levels of osteocalcin, procollagen type-1 N-terminal propeptide, and tartrate-resistant acid phosphatase 5b were associated with higher odds of fracture, even after adjustment for femoral neck T-score. Discrimination of fracture prevalence was best with a femoral neck T-score of −2.0 or less and a value in the upper two tertiles for osteocalcin, procollagen type-1 N-terminal propeptide, or tartrate-resistant acid phosphatase 5b; these values corresponded to the upper half of the normal premenopausal reference range. In summary, these cross-sectional data suggest that measurement of bone turnover markers may increase the diagnostic accuracy of densitometry to identify patients with CKD at high risk for fracture.

Fracture rates are elevated in patients with predialysis chronic kidney disease (CKD).^1–3 Cross-sectional data suggest fracture risk increases as kidney function declines and by CKD stage 4 approximates that for ESRD.² Furthermore, fracture-related mortality for patients with predialysis CKD is approximately twofold higher than the general population.⁴ These data are alarming because CKD and osteoporosis are highly co-prevalent,⁵ and incidence and prevalence of predialysis CKD, osteoporosis, and fragility fracture are expected to increase exponentially as the population ages. Reliable methods to identify patients with CKD at risk for fracture are lacking.⁶ Thus, there is a critical need to develop diagnostic tests and assess their utility to detect increased fracture risk in the CKD population.

Bone strength is determined by the amount of bone present and the quality of that bone. In the clinic, bone mass is usually assessed by dual energy x-ray absorptiometry (DXA), which provides an estimate of areal bone mineral density (aBMD, g/cm²). Bone quality is related to other material properties of the skeleton, including bone microarchitecture and remodeling activity. Microarchitecture and remodeling are linked; low and high remodeling lead to loss of bone structural integrity and increased fracture risk.^7–11

High-resolution peripheral computed tomography (HR-pQCT; XtremeCT; Scanco Medical AG, Brüttisellen, Switzerland; voxel size approximately 82 μm) is a new imaging technique that measures true volumetric bone mineral density (vBMD, g/cm³) and microarchitecture of the distal skeleton (radius and tibia). Bone remodeling can also be assessed with reasonable accuracy by measurement of bone turnover markers (BTMs) in patients with normal kidney function.^12,13 Certain BTMs, including osteocalcin and the C-terminal telopeptides of type I collagen (CTX), are renally cleared. Others, such as bone-specific alkaline phosphatase (BSAP), procollagen type-1 N-terminal propeptide (P1NP), and tartrate-resistant acid phosphatase-5b (Trap-5b), are metabolized by nonrenal mechanisms. Regardless of whether they are renally excreted, BTMs are higher in patients with more severe CKD and are associated with lower aBMD by DXA.⁷ It is possible that nonrenally cleared BTMs provide more accurate assessments of fracture risk in CKD.

We previously reported that patients with predialysis CKD and a history of fragility fracture have lower aBMD by DXA and lower vBMD and abnormal cortical and trabecular microarchitecture by HR-pQCT compared with CKD patients without fracture.¹⁴ However, the diagnostic accuracy of DXA and HR-pQCT was not sufficiently high to identify reliably patients with prevalent fractures, except in those with longstanding (>7 years) CKD. We hypothesized that the microarchitectural abnormalities that we detected in CKD patients with fracture were related to abnormal bone remodeling. Furthermore, we hypothesized that calciotropic hormones and BTMs, particularly those that are not renally cleared, would discriminate fracture status independent of bone mass.

RESULTS

Subject Characteristics According to Fracture Status

We recruited participants from the nephrology clinics of Columbia University Medical Center (CUMC). Of 82 patients, 23 had prevalent fracture: 14 vertebral (18 mild, 3 moderate) and 14 nonvertebral (2 rib, 1 humerus, 5 forearm, 1 femur, 4 ankle, 1 metatarsal). Five patients had multiple nonvertebral fractures. Median time (interquartile range) from fracture was 7.9 (2.1 to 9.8) years. Groups with and without fracture (Table 1) were similar with respect to body mass index, kidney function, CKD duration, race, ethnicity, use of calcium and noncalcium phosphate binders, and parent and active forms of vitamin D. No patients took cinacalcet. Patients with fracture were more likely to be female, although this difference was NS, and were significantly older and less likely to have diabetes.

Table 1.

Baseline characteristics of patients with fracture ≥1 year in comparison to those without fracture

	Fracture	Nonfracture	Unadjusted P	Age-, Gender-, and Diabetes-Adjusted P
n	23	59
eGFR, ml/min, median (IQR)	25 (14 to 35)	28 (19 to 41)	NS
Age, years, median (IQR)	78 (66 to 84)	69 (62 to 76)	0.006
BMI, median (IQR)	27.1 (24.2 to 30.9)	29.3 (26.1 to 33.8)	NS
Years since nonspine fracture, median (IQR)	7.9 (2.1 to 9.8)	NA
Gender, % female	57	39	NS
Race, %
white	48	59	NS
black	42	40	NS
ethnicity, %
Hispanic	52	58	NS
non-Hispanic	48	52	NS
Diabetes, %	39	68	0.02
Duration of CKD, years, median (IQR)	5.0 (3.4 to 10.0)	5.5 (3.4 to 9.3)	NS
Laboratory parameters, median (IQR)
serum calcium, mg/dl	9.4 (8.9 to 9.8)	9.3 (9.0 to 9.7)	NS
serum phosphorus, mg/dl	3.7 (3.3 to 4.7)	3.6 (3.0 to 4.2)	NS
serum bicarbonate, mM/L	22 (21 to 26)	23 (22 to 25)	NS
Medications, %
calcium-containing agents	26	21	NS
noncalcium-containing phosphate binding agents	17	18	NS
vitamin D—parent	70	63	NS
vitamin D—active	35	32	NS
Behaviors, %
current alcohol use	13	10	NS
current tobacco use	9	7	NS
DXA measurements, median (IQR)a
LS aBMD, g/cm2	0.938 (0.853 to 1.057)	1.060 (0.957 to 1.216)	0.006	NS
LS T-score	−1.4 (−2.1; −0.2)	−0.1 (−1.1 to 1.2)	0.01	0.04
TH aBMD, g/cm2	0.783 (0.728 to 0.878)	0.943 (0.805 to 1.069)	0.004	NS
TH T-score	−1.5 (−2.1; −0.6)	−0.6 (−1.6 to 0.3)	0.007	NS
FN aBMD, g/cm2	0.621 (0.569 to 0.666)	0.747 (0.667 to 0.848)	0.0009	NS
FN T-score	−2.2 (−2.7; −1.4)	−1.4 (−2.1; −0.8)	0.002	0.03
1/3R aBMD, g/cm2	0.630 (0.549 to 0.714)	0.773 (0.634 to 0.805)	0.001	NS
1/3R T-score	−1.8 (−3.1; −0.6)	−0.9 (−2.1 to 0.06)	0.006	0.02
UDR aBMD, g/cm2	0.340 (0.302 to 0.415)	0.442 (0.369 to 0.503)	<0.0001	0.005
UDR T-score	−2.2 (−3.4; −1.2)	−1.2 (−1.8; −0.5)	0.009	0.007
HR-pQCT measurements distal radius, median (IQR)
cross-sectional area, mm2	278 (235 to 314)	300 (236 to 357)	NS	NS
cortical area, mm2	40 (35 to 48)	55 (46 to 72)	0.0001	0.01
cortical thickness, μm	610 (480 to 710)	750 (600 to 930)	<0.0001	0.02
total density, mg HA/cm3	248 (209 to 299)	308 (268 to 347)	<0.0001	0.006
cortical density, mg HA/cm3	795 (724 to 850)	842 (781 to 910)	0.006	NS
trabecular density, mg HA/cm3	114 (86 to 161)	154 (130 to 189)	0.002	0.02
trabecular number, 1/mm	1.7 (1.4 to 2.1)	2.0 (1.8 to 2.3)	0.04	NS
trabecular thickness, μm	59 (48 to 69)	65 (60 to 68)	0.02	0.03
trabecular heterogeneity (μm)	237 (165 to 359)	178 (151 to 201)	0.03	NS
HR-pQCT measurements distal tibia, median (IQR)
cross-sectional area, mm2	783 (624 to 866)	782 (642 to 863)	NS	NS
cortical area, mm2	80 (56 to 89)	113 (84 to 138)	<0.0001	0.01
cortical thickness, μm	690 (500 to 810)	1005 (820 to 1210)	<0.0001	0.01
total density, mg HA/cm3	222 (186 to 244)	257 (224 to 304)	0.0002	0.03
cortical density, mg HA/cm3	718 (662 to 802)	797 (752 to 846)	0.002	0.04
trabecular density, mg HA/cm3	150 (119 to 158)	161 (131 to 186)	NS	NS
trabecular number, 1/mm	1.7 (1.5 to 1.9)	1.9 (1.7 to 2.2)	NS	NS
trabecular thickness, μm	69 (60 to 73)	69 (61 to 77)	NS	NS
trabecular heterogeneity (μm)	245 (206 to 295)	193 (161 to 264)	NS	NS

BMI, body mass index; IQR, interquartile range.

^aT-scores are only adjusted for age and diabetes status.

Consistent with our previous report,¹⁴ measures of aBMD by DXA and of vBMD, size, and microarchitecture by HR-pQCT were associated with fracture (Table 1). Fracture patients had significantly lower aBMD at the lumbar spine (LS), total hip (TH), femoral neck (FN), and 1/3 and ultradistal radius (1/3R and UDR, respectively). After adjustment for group differences, aBMD remained significantly lower only at the UDR site. Unadjusted T-scores at all sites were associated with fracture history. After adjustment for age and diabetes, only FN, 1/3R, and UDR T-scores were associated with fracture history.

At the radius, several HR-pQCT measures of vBMD, size, and trabecular microarchitecture were associated with fracture (Table 1). After adjustment for group differences, total and trabecular vBMD, cortical area, and trabecular thickness remained significantly lower in the fracture group. At the tibia, cortical parameters (vBMD, thickness, and area) were significantly lower in the fracture group and remained so after adjustment; trabecular parameters did not differ, possibly because of the mitigating effects of weight bearing.

Calciotropic Hormones, Fibroblast Growth Factor-23, and Biochemical Markers of Bone Turnover by Fracture Status

Median levels for intact parathyroid hormone (iPTH), fibroblast growth factor-23 (FGF-23), and both resorption markers were above the normal range for healthy premenopausal women (Table 2). Although large, the differences between fracture and nonfracture groups in median levels of iPTH and FGF-23 were NS, likely because of heterogeneity in measurement distribution or the relatively small number of subjects with fracture. BSAP levels did not differ. In contrast, osteocalcin and P1NP were significantly higher in patients with fracture, by 52% and 42% respectively; each SD increase in their levels corresponded with 2.6- and 3.2-fold increased odds of fracture, respectively. Trap-5b was 29% higher in the fracture group; each SD increase was associated with a 2.3-fold increased odds of fracture. CTX was 19% higher in the fracture group, but by univariate logistic regression the association was NS. Adjustment for age, gender, and diabetes did not materially change the results. However, the between-group difference in CTX achieved significance.

Table 2.

Calciotropic hormones and biochemical markers of bone turnover among patients with and without fracture

	Fracture	Nonfracture	Percent Difference between Means	Univariate Logistic Regression OR (95% CI) for each SD increase in BTM	Unadjusted P	Age-, Gender-, and Diabetes-Adjusted P	Normal Range for Premenopausal Women
iPTH, pg/ml	105 (40 to 249)	68 (39 to 138)	42	1.24 (0.77 to 2.00)	NS	NS	14 to 66
25OHD, ng/ml	29 (20 to 37)	32 (22 to 38)	−4	0.96 (0.56 to 1.63)	NS	NS	>30
FGF-23, RU/ml	175 (115 to 578)	162 (107 to 303)	44	1.67 (0.74 to 3.84)	NS	NS	<100
BSAP, U/L	30.3 (24.9 to 45.4)	29.6 (20.8 to 40.7)	8	1.21 (0.72 to 2.05)	NS	0.059	11.6 to 29.6
Osteocalcin, ng/ml	41.8 (24.1 to 66.0)	24.2 (14.2 to 41.6)	52	2.66 (1.27 to 5.54)	0.006	0.003	8.4 to 33.9
P1NP, μl/L	79.5 (58.0 to 92.0)	54.0 (39.0 to 79.0)	42	3.25 (1.50 to 7.05)	0.0008	0.0007	19 to 83
CTX, ng/ml	1.24 (0.77 to 1.47)	0.87 (0.52 to 1.35)	19	1.78 (0.94 to 3.38)	0.047	0.03	0.11 to 0.74
Trap-5b, U/L	5.33 (4.01 to 7.47)	4.41 (3.38 to 5.36)	29	2.31 (1.21 to 4.42)	0.01	0.006	1.03 to 4.15

Values presented as median (IQR).

Prevalence of Hyperparathyroidism and Relationships among Parathyroid Hormone and Calciotropic Hormones, FGF-23, and BTMs

On the basis of the Kidney Disease Outcomes Quality Initiative (KDOQI) and Kidney Disease Improving Global Outcomes (KDIGO) guidelines, 40% and 52%, respectively, of the subjects had hyperparathyroidism. Unadjusted serum iPTH levels correlated directly with 25-hydroxyvitamin D (25OHD), FGF-23, and BTMs. After estimated GFR (eGFR) adjustment, iPTH correlated inversely with 25OHD (r = −0.28, P = 0.01) and directly with BSAP (r = 0.30, P = 0.007), osteocalcin (r = 0.50, P < 0.0001), P1NP (r = 0.26, P = 0.02), CTX (r = 0.43, P < 0.0001), and Trap-5b (r = 0.31, P = 0.005). FGF-23 and iPTH did not correlate after eGFR adjustment.

Relationships between Calciotropic Hormones, FGF-23, and BTMs and Measures of Bone Size, Density, and Microarchitecture

To assess mechanisms associated with fracture, we investigated relationships between eGFR-adjusted biochemical parameters and aBMD, vBMD, bone size, and microstructure (Table 3). Higher iPTH levels were associated with lower aBMD by DXA at the TH, FN, and UDR and with lower trabecular vBMD by HR-pQCT at radius and tibia. Higher iPTH levels were also associated with trabecular microarchitectural deterioration—thinner trabeculae at the radius and a more widely separated, heterogeneous trabecular network at the tibia. There were no relationships between iPTH and cortical parameters.

Table 3.

Pearson correlations between BTMs and measures of bone mass, geometry, and microarchitecture adjusted for eGFR

	PTH	25OHD	FGF-23	BSAP	Osteocalcin	P1NP	CTX	Trap-5b
Age	−0.1	0.17	0.18	−0.33b	−0.004	−0.02	−0.08	0.03
BMI	0.1	−0.18	−0.12	0.12	0.004	0.03	−0.03	−0.04
DXA
LS	−0.21	0.0004	0.05	0.05	−0.14	−0.12	−0.06	−0.1
TH	−0.26a	−0.01	−0.22	−0.14	−0.36b	−0.21	−0.26a	−0.14
FN	−0.29b	0.06	−0.18	−0.11	−0.37c	−0.27a	−0.25a	−0.16
1/3R	−0.19	0.01	−0.03	−0.17	−0.25a	−0.23a	−0.07	−0.30b
UDR	−0.27a	−0.08	−0.16	−0.12	−0.28a	−0.29a	−0.2	−0.30b
Distal radius HR-pQCT
cross-sectional area, mm2	−0.001	−0.01	0.03	−0.05	−0.06	−0.03	−0.06	−0.04
cortical area, mm2	−0.11	−0.12	0.02	−0.03	−0.08	−0.19	−0.01	−0.15
cortical thickness, μm	−0.15	−0.13	−0.01	−0.02	−0.1	−0.2	0.002	−0.18
total density, mg HA/cm3	−0.27a	−0.03	−0.06	−0.12	−0.27a	−0.33b	−0.16	−0.28a
cortical density, mg HA/cm3	−0.14	−0.11	−0.04	0.05	−0.06	−0.13	0.09	−0.17
trabecular density, mg HA/cm3	−0.27a	0.05	−0.07	−0.17	−0.36b	−0.33b	−0.30b	−0.25a
trabecular number, n/mm	−0.16	0.05	−0.04	−0.15	−0.24a	−0.17	−0.22	−0.23a
trabecular thickness, μm	−0.29b	0.04	−0.08	−0.15	−0.34b	−0.36b	−0.27a	0.2
trabecular separation, μm	0.17	−0.07	−0.08	0.12	0.27a	0.19	0.21	0.18
trabecular heterogeneity, μm	0.16	−0.06	−0.06	0.09	0.25a	0.19	0.22	0.27a
Distal tibia HR-pQCT
cross-sectional area, mm2	−0.04	0.03	0.11	−0.07	−0.004	0.08	0.03	0.006
cortical area, mm2	−0.03	−0.11	−0.07	0.01	−0.14	−0.25a	−0.07	−0.14
cortical thickness, μm	−0.01	−0.11	−0.1	0.02	−0.16	−0.28a	−0.09	−0.15
total density, mg HA/cm3	−0.21	−0.01	−0.11	−0.1	−0.26a	−0.32b	−0.24a	−0.25a
cortical density, mg HA/cm3	−0.2	−0.06	−0.09	0.03	−0.17	−0.26a	−0.05	−0.17
trabecular density, mg HA/cm3	−0.34b	0.08	−0.05	−0.16	−0.27a	−0.23a	−0.28a	−0.23a
trabecular number, n/mm	−0.31b	0.12	0.1	−0.2	−0.21	−0.17	−0.21	−0.22
trabecular thickness, μm	−0.18	−0.02	−0.26a	−0.01	−0.15	−0.14	−0.19	−0.12
trabecular separation, μm	0.34b	−0.17	−0.06	0.21	0.23	0.2	0.26a	0.27a
trabecular heterogeneity, μm	0.34c	−0.24a	−0.01	0.17	0.18	0.17	0.19	0.26a

OR, odds ratio; CI, confidence interval.

P value ^a<0.05,

^b<0.01,

^c<0.001;

^d<0.0001.

Higher formation markers were associated with lower aBMD by DXA and vBMD and microarchitectural deterioration by HR-pQCT. Osteocalcin and P1NP correlated inversely and significantly with FN, 1/3R, and UDR aBMD; osteocalcin also correlated with TH aBMD. Osteocalcin and P1NP correlated inversely with total and trabecular vBMD at radius and tibia. In terms of microarchitecture, higher osteocalcin was associated with lower trabecular number and a more widely separated, heterogeneous trabecular network at the radius, but not at the tibia. P1NP was only associated with thinner trabeculae at the radius. Higher P1NP levels were also only associated with lower cortical area, density, and thickness at the tibia.

Higher resorption markers were also associated with lower aBMD, vBMD, and microarchitectural deterioration. Higher CTX was associated with lower TH and FN aBMD, whereas higher Trap-5b was associated with lower 1/3R and UDR aBMD. CTX and Trap-5b correlated inversely with trabecular vBMD at the radius and tibia. Although relationships with microarchitectural parameters were not consistently significant, in general, CTX and Trap-5b were associated with trabecular but not cortical microarchitectural deterioration.

Discrimination of Fracture Status by Biochemical Markers of Bone Turnover, Bone Mass, and Microarchitecture

Although HR-pQCT measures were strongly associated with fracture, the XtremeCT is available at few academic centers. In contrast, DXA is widely available and FN aBMD and T-score are used in the World Health Organization (WHO) Fracture Risk Assessment Tool (FRAX) to predict absolute 10-year risk of hip and major osteoporotic fracture.¹⁵ We therefore investigated whether any DXA, HR-pQCT, or biochemical parameter provided better discrimination of fracture status than FN T-score using receiver-operator characteristic (ROC) curve analysis of the most robust predictors of fracture by univariate analyses. Specifically assessed were (1) osteocalcin, P1NP, and Trap-5b; (2) FN and UDR T-score; and (3) radius total vBMD and tibial cortical thickness by HR-pQCT. Areas under the curve (AUCs) ranged from 0.68 (Trap 5b) to 0.78 (tibia cortical thickness) (Figure 1), but differences were NS.

Figure 1.

In comparison to FN T-Score, there were no differences in AUCs for BTMs, aBMD by DXA, and vBMD and bone geometry by HR-pQCT.

Relationships among Fracture, Biochemical Markers of Bone Turnover, and Bone Mineral Density

To determine whether BTMs measured in the CKD subjects discriminated fracture status independent of bone mineral density (BMD), we performed multiple logistic regression analyses adjusting for age, diabetes, and FN T-score (Table 4). BTMs associated with fracture status by univariate analysis (osteocalcin, P1NP, and Trap-5b) were divided into tertiles. In unadjusted and adjusted analyses, the highest tertiles of all three BTMs were associated with fracture status. In addition, combining BTMs with FN T-score increased the R² value, a measure of how much the data contribute to the variance of the outcome (fracture status), and the model-specific AUC, a measure of discrimination.

Table 4.

Association between tertiled levels of BTMs and fracture after adjustment for FN T-score

	FN T-Score Model	BTM Models	FN T-Score and BTM Models
	R2 0.28
	AUC 0.78
FN T-score	1.88 (1.02 to 3.49)
Osteocalcin
		R2 0.33	R2 0.38
		AUC 0.81	AUC 0.83
FN T-score			1.84 (0.94 to 3.60)
OC tertiles
7.4 to 20.6		Reference	Reference
20.8 to 37.1		3.68 (0.78 to 17.34)	3.51 (0.71 to 17.31)
38.7 to 138.1		6.69 (1.41 to 31.84)	5.47 (1.11 to 27.05)
P1NP
		R2 0.35	R2 0.39
		AUC 0.81	AUC 0.82
FN T-score			1.85 (0.94 to 3.66)
P1NP tertiles
22 to 46		Reference	Reference
48 to 78		3.54 (0.71 to 17.61)	3.66 (0.70 to 19.10)
79 to 176		8.17 (1.67 to 39.90)	7.02 (1.36 to 36.28)
Trap-5b
		R2 0.31	R2 0.37
		AUC 0.79	AUC 0.83
FN T-score			1.98 (1.04 to 3.76)
Trap-5b tertiles
1.8 to 3.7		Reference	Reference
3.8 to 5.2		1.76 (0.40 to 7.79)	1.96 (0.42 to 9.17)
5.3 to 10.7		4.93 (1.15 to 21.23)	5.16 (1.10 to 24.17)

All models adjusted for age and diabetes status.

We next evaluated the co-prevalence of low FN T-scores and high BTMs among subjects with a history of prior fracture (Figure 2A through 2c). FN T-score was stratified as (1) −2.0 or less, (2) more than −2.0 to −1.0 or less, and (3) greater than −1.0 because these approximated the tertile cutoffs. We dichotomized osteocalcin, P1NP, and Trap-5b at the upper limit of the lowest tertile, which approximates the middle of the normal range for a healthy premenopausal reference population. We selected this cutoff because the clinical target for monitoring antiresorptive therapy is to reduce BTMs to the lower half to one third of the premenopausal reference range and no reference population has been established for men.^16,17 Most patients (55% to 60%) with fractures had a FN T-score of −2.0 or less and serum BTMs (Figure 2a through 2c) in the upper two tertiles of the CKD subject range.

Figure 2.

Percent prevalence of patients with fracture (n = 23) according to BTM levels and FN T-scores. FN T-score is divided at −2.0 or less, −2 to −1.0, and more than −1. BTM levels are dichotomized at </> the first tertile. (A) osteocalcin, (B) P1NP, and (C) Trap-5b.

Finally, we determined the proportion of CKD patients, with and without fracture, according to whether FN T-score was less than or equal to or more than −2.0 and whether BTMs were in the lowest tertile or the upper two tertiles of the range of the CKD subject values, approximating the middle of the premenopausal normal range. For each BTM, the proportion of patients with fractures was markedly higher in those with FN T-scores of −2.0 or less and with serum osteocalcin, P1NP, or Trap-5b level in the upper two tertiles (Figure 3a to 3c, respectively). Although diagnostic test characteristics demonstrated poor sensitivity, there were moderate specificity, negative predictive values, and positive and negative likelihood ratios (Table 5). When we performed this analysis with tibia cortical thickness rather than FN T-score, the results were similar.

Figure 3.

Prevalence of patients with CKD with and without fracture according to whether the FN T-score is −2 or less or more than −2.0 and BTMs are in the upper two tertiles versus the lowest tertile. (A) osteocalcin tertile 1, 7.4 to 20.6; tertile 2, 20.8 to 37.1; and tertile 3, 38.7 to 138.1. (B) P1NP tertile 1, 22 to 46; tertile 2, 48 to 78; and tertile 3, 79 to 176. (C) Trap-5b tertile 1, 1.8 to 3.7; tertile 2, 3.8 to 5.2; and tertile 3, 5.3 to 10.7.

Table 5.

Diagnostic test characteristics for fracture: BTM levels in the upper two tertiles and a FN T-score of −2 or less versus other BTM and FN T-score groups

	Sensitivity (95% CI)	Specificity (95% CI)	PPV (95% CI)	NPV (95% CI)	Positive Likelihood Ratio (95% CI)	Negative Likelihood Ratio (95% CI)
Osteocalcin ≥ 21 and FN T-score of −2 or less	56 (37 to 79)	81 (69 to 90)	54 (33 to 74)	84 (72 to 92)	3.17 (1.68 to 5.99)	0.50 (0.69 to 0.84)
P1NP ≥ 46 and FN T-score of −2 or less	63 (41 to 82)	86 (74 to 93)	64 (41 to 82)	86 (74 to 93)	4.61 (2.25 to 9.45)	0.42 (0.24 to 0.74)
Trap-5b ≥ 3.7 and FN T-score −2 or less	59 (37 to 79)	81 (68 to 90)	54 (33 to 74)	84 (71 to 92)	3.12 (1.65 to 5.88)	0.50 (0.30 to 0.84)

PPV, positive predictive value; NPV, negative predictive value.

DISCUSSION

These results confirm our previous findings that CKD patients with prevalent fragility fractures have lower aBMD by DXA and lower vBMD and abnormal microarchitecture by HR-pQCT.¹⁴ They extend those findings by demonstrating that certain BTMs distinguish between patients with and without fracture. The highest tertile of formation (osteocalcin, P1NP) and resorption (Trap-5b) markers were independently associated with increased odds of prior fracture. Combining the highest tertile level of osteocalcin, P1NP, or Trap-5b with FN T-score improved discrimination of fracture history over measurement of FN T-score or BTMs alone. Fracture prevalence was highest in CKD patients with a FN T-score of −2.0 or less and either osteocalcin, P1NP, or Trap-5b in the highest two tertiles. Indeed, for any combination of elevated BTM and low FN BMD, specificity and negative predictive value for fracture were >80%. These diagnostic test characteristics exceed those of prostate-specific antigen testing for prostate cancer screening.¹⁸ These data support our hypothesis that predialysis CKD patients can be risk-stratified for fragility by combining BTMs and aBMD. Interestingly, whether BTMs were renally excreted did not affect fracture discrimination. In terms of pathogenesis, high iPTH and BTM levels were associated with low bone mass, smaller bone size, and a disrupted trabecular network, abnormalities we^14,19 and others^20–23 have found to be associated with fracture.

Contrary to our hypotheses, BSAP, a nonrenally cleared BTM, was not associated with bone mass, size, or microarchitecture or fracture. These findings may be explained by the cross-sectional design or small sample size of the study. They are in agreement with a cross-sectional study by Tsuchida et al.,²⁴ who reported that BSAP did not correlate with cortical or trabecular vBMD measured by peripheral computed tomography in predialysis CKD. In prospective studies, high BSAP levels predicted abnormalities associated with fracture, such as bone loss,^7,25–27 turnover, and mineralization defects,^28–31 but have not been shown to predict fracture per se.

To our knowledge, these are the first data demonstrating utility for BTMs in discriminating fracture among predialysis CKD patients over age 50. Whether BTMs, measured a median of 7.9 years after fractures occurred, reflect remodeling activity at the time of fracture is unknown. Remodeling rates may change over time, particularly given likely deterioration of kidney function between time of fracture and BTM measurement. However, multiple prospective studies have shown that BTMs predict fracture in elderly women.^32–35 In elderly women with a history of fragility fracture, elevated urinary CTX was superior to FN T-score for fracture prediction.³⁴ BTMs also predicted fractures that occurred up to 6.5 years in the future.¹² Moreover, a past history of fragility fracture is a marker of overall poor bone quality and an important risk factor for future fracture, as evidenced by inclusion of this historical feature in the WHO FRAX.¹⁵ Although our data suggest that a combination of BTMs, FN BMD, and a history of fracture may be useful to identify CKD patients who might benefit from fracture prevention strategies, prospective studies are needed to assess the utility of these parameters for fracture prediction in the CKD population.

We know of no data regarding FGF-23 and fracture in CKD. In children with ESRD, FGF-23 has recently been correlated with abnormal mineralization,³⁶ which could lead to increased fragility. Vitamin D deficiency is a risk factor for fracture.^37,38 However, our results did not detect any relationship between 25OHD and fracture risk, possibly because of equivalent vitamin D supplementation in each group. Furthermore, most subjects had an eGFR < 45 ml/min, a level at which serum 25OHD weakly correlates with hyperparathyroidism.³⁹

In patients with ESRD, data regarding iPTH and fracture are conflicting. Some studies found no association between iPTH and fracture^40,41 whereas others suggest that low^42–44 or high^44,45 iPTH levels are associated with fracture. Studies in patients with predialysis CKD demonstrated that higher iPTH is inversely associated with aBMD by DXA^7,25,27,46 and total and cortical vBMD by HR-pQCT.⁴⁷ We also found inverse relationships between iPTH and aBMD by DXA and trabecular vBMD and microarchitecture by HR-pQCT. Bone biopsy studies have demonstrated that very low or very high iPTH levels are moderately predictive of turnover abnormalities^29–31; in turn, abnormal turnover may cause low BMD, microarchitectural deterioration, and decreased bone strength that could lead to fracture.^21,48–53 However, in our study, the relationship between iPTH and fracture history was NS. There are several possible reasons for this: The number of patients with fracture was relatively small, iPTH levels were outside of the range shown to predict low or high turnover,^29–31 and the iPTH assay used detects the intact (1-84) parathyroid hormone (PTH) molecule and PTH fragments that accumulate in kidney failure and may not accurately reflect bone turnover.⁵⁴ PTH assays that measure only the intact (1-84) molecule may provide superior discrimination for fracture.

Although hyperparathyroidism is classically associated with cortical thinning and porosity, we noted more consistent, robust relationships between iPTH and BTMs and trabecular vBMD and microarchitecture than with cortical vBMD or thickness. This may be because hyperparathyroidism affects more metabolically active trabecular bone and causes cancellization of the cortical endosteal surface. In hyperparathyroid states, the Laplace–Hamming filter and threshold,⁵⁵ used by HR-pQCT to delineate the trabecular-endosteal surface boundary, may not be reliable. Furthermore, the resolution of HR-pQCT is not sufficient to detect small cortical pores.

Measurement of FN BMD by DXA is widely available, strongly predicts fracture in postmenopausal women,^56,57 and is the main driver of the WHO FRAX tool. FRAX has not been validated in CKD. It is not known if FRAX can be used to determine fracture risk or decide upon initiating fracture prevention treatments in CKD. Although more research is needed to develop a “CKD-specific” FRAX tool, our data are the first to suggest that fracture prediction is enhanced by combining certain BTMs with FN BMD in CKD patients.

Limitations of this study include its cross-sectional design, which prevents assessment of relationships between BTMs and incident fractures. In addition, BMD and BTMs were measured years after fractures occurred and may not reflect levels at the time of fracture. The small sample size may have limited the detection of relationships between fracture and BSAP and iPTH. Levels of PTH and FGF-23 are altered with declining kidney function because of accumulation of degradation products.⁵⁴ The assays used for these biomarkers do not distinguish between various forms of these proteins. The fracture and nonfracture groups were not balanced for predictors of fragility; however, after adjustment for these differences, relationships remained significant between fracture and osteocalcin, P1NP, and Trap-5b. Bone biopsies, the gold standard to assess turnover, were not performed. We used the HR-pQCT manufacturer's algorithm to separate cortical from trabecular regions, which may affect the accuracy of cortical measures.

In conclusion, commercially available assays for osteocalcin, P1NP, and Trap-5b discriminated fracture status in predialysis CKD patients independent of BMD. Combining a BTM with FN BMD provided greater utility to identify CKD patients with fractures than either test alone. If confirmed by longitudinal studies with fracture outcomes, these data suggest that combining a BTM with measurement of FN BMD may prove useful for fracture risk screening in CKD patients. Longitudinal studies are needed to determine whether BTM assessment predicts incident fracture, whether measuring bone turnover and BMD improves fracture prediction, and mechanisms by which elevated bone turnover increases propensity for fracture in predialysis CKD.

CONCISE METHODS

Subjects

Predialysis CKD patients with eGFR < 90 ml/min, with and without a history of fracture, and enrolled in an ongoing longitudinal study of relationships between kidney function and bone structure and strength were included (n = 82). Patients with nonspine fracture were ≥1 year from fracture occurrence. Participants were recruited from the general nephrology clinics of CUMC between August 2006 and September 2010. All patients referred to the nephrology clinics and meeting study inclusion criteria were eligible. The CUMC nephrology clinics serve as a referral center for patients with CKD from the northern Manhattan, Bronx, Queens, southern New York State and Connecticut, and northern and central New Jersey areas. We did not exclude patients on the basis of the etiology of CKD. eGFR was determined by the Modification of Diet in Renal Disease short formula.⁵⁸ Patients with a history of kidney transplantation; malignancy; bilateral lower extremity amputations; residing in a nursing home; requiring a wheelchair; and those taking bisphosphonates, gonadal steroids, aromatase inhibitors, and anticonvulsants that induce hepatic cytochrome P450 enzymes were excluded. Two patients had a remote history of glucocorticoid use (>1 year before the study visit) for treatment of IgA nephropathy and idiopathic focal segmental glomerulosclerosis. Therefore, the cohort included in this investigation represents the wide spectrum of ambulatory patients with CKD and is generalizable to those patients typically treated in a general nephrology clinic. The CUMC Institutional Review Board approved the study and all subjects provided written informed consent.

Laboratory Measurements

Routine laboratory parameters were measured by Quest diagnostics. Serum creatinine was determined by the Jaffe reaction and serum calcium, phosphorus, and bicarbonate were measured by spectrophotometry. Calciotropic hormones and BTMs were measured at CUMC in a specialized research laboratory. Serum 25OHD was measured by RIA (Diasorin, Stillwater, MN); intra- and interassay precision are 10.5% and 11.0%, respectively, and the normal range is ≥30 ng/ml. iPTH was measured by RIA (Scantibodies, Santee, CA); intra- and interassay precision are 4.8% and 6.8%, respectively, and the normal range is 14 to 66 pg/ml. C-terminal FGF-23 was measured by ELISA (Immutopics, San Clemente, CA); intra- and interassay precision are 2.4% and 4.7%, respectively, and normal values are <100 RU/ml. BSAP activity was measured by immunoassay (Metra BAP, Quidel Corporation, San Diego, CA); inter- and intra-assay variabilities are 7.6% and 3.9%, respectively, and the normal range is 11.6 to 29.6 U/L in premenopausal women. Osteocalcin was measured by ELISA (N-mid Osteocalcin, IDS, Ltd., Scottsdale, AZ); the inter- and intra-assay variability is 2.7% and 1.8%, respectively, and the normal range is 8.4 to 33.9 ng/ml in premenopausal women. P1NP was measured by RIA (IDS, Ltd., Fountain Hills, AZ); inter- and intra-assay variabilities are 8.3% and 6.5%, respectively, and the normal range is 19 to 83 pg/ml for premenopausal women. Serum CTX was measured by ELISA (IDS, Ltd., Scottsdale, AZ); the inter- and intra-assay variabilities are 9.75 and 1.7%, respectively, and the normal range is 0.112 to 0.738 ng/ml for premenopausal women. TRAP-5b was measured by ELISA (IDS, Ltd., Scottsdale AZ); intra- and interassay precisions are 4.7% and 9.0%, respectively, and the normal range is 1.03 to 4.15 U/L in premenopausal women.

Assessment of CKD, Fracture Status, Date of Fracture Occurrence, Factors Associated with BMD, and Definitions of Hyperparathyroidism

CKD duration was estimated for all patients by review of medical records and prior measurements of serum creatinine. The online medical records system at CUMC contains clinical and demographic data dating to 1990, and this system was utilized to determine CKD duration for patients seen within the CUMC system (approximately 95% of this cohort). The onset of CKD was determined from the earliest serum creatinine that consistently corresponded to an eGFR < 90 ml/min, in compliance with KDOQI guidelines.⁵⁹ For patients lacking historical information (n = 7; 8%), duration of kidney function was based on the date of first presentation to a nephrologist. In patients with inadequate historical clinical data, the serum creatinine at nephrology presentation was uniformly abnormal (range 1.5 to 6.0 mg/dl). In patients when the earliest serum creatinine was abnormal, and no normal range of serum creatinines was available, duration of CKD was based on the date of the earliest abnormal serum creatinine. Etiology of kidney disease was grouped into three categories: (1) diabetic and hypertensive kidney disease, (2) glomerular causes of CKD (nephritis or nephrosis), and (3) other causes of CKD (polycystic kidney disease, tubularinterstitial, or unknown).

Fragility fracture was defined as a fracture associated with trauma equivalent to or less than a fall from a standing height; fractures associated with major trauma (e.g., motor vehicle accidents) and skull and digit fractures were excluded. Vertebral and nonvertebral fragility fractures were included. Nonvertebral fractures were ascertained by self-reported history during our standardized interview and confirmed by review of radiographs or radiology reports whenever possible. Nine patients had 14 nonvertebral fractures: 6 fractures were confirmed by radiographs and 8 were included based on patient report, which included being casted or having undergone radiographic procedures and being told they had a broken bone. All study subjects underwent lateral spine x-rays. Vertebral fractures were identified by a skeletal radiologist (R.B.S.) and graded by the semiquantitative method on spine radiographs performed according to the protocol of the Study of Osteoporotic Fractures.⁶⁰ Because all vertebral fractures were identified at the study visit, it was not possible to determine their occurrence in relationship to the duration of CKD or the study visit. There were a total of 21 vertebral fractures in 14 patients: 8 patients with a single vertebral fracture, 5 patients with two vertebral fractures, and 1 patient with three vertebral fractures. Of these, 18 fractures were mild and 3 were moderate.

Current alcohol consumption was defined as one or more drinks per day. Current tobacco use was reported as having smoked tobacco in the 5 years before the study visit. Vitamin D supplementation was defined as any use of ergocalciferol (n = 25), cholecalciferol (n = 30), paricalcitol (n = 14), doxercalciferol (n = 8), or calcitriol (n = 4). Phosphate binders included calcium acetate (n = 2), sevelamer (n = 13), or lanthanum carbonate (n = 1). No patient was taking cinacalcet or aluminum-containing phosphate binding agents. Eight patients were taking calcium carbonate.

Hyperparathyroidism was defined by KDOQI and KDIGO guidelines. KDOQI defines hyperparathyroidism on the basis of CKD stage⁶¹: (1) patients with CKD stages 3 and 4 who have plasma levels of iPTH > 70 pg/ml (7.7 pmol/L; stage 3) or >110 pg/ml (12.1 pmol/L; stage 4) on more than two consecutive measurements, and (2) patients with CKD stage 5 who have elevated plasma levels of iPTH > 300 pg/ml (33.0 pmol/L). Because of the cross-sectional design of this investigation, a single iPTH measurement was used. KDIGO defines hyperparathyroidism as being above the upper limit of normal for the assay.⁶² The iPTH assay utilized in this investigation had an upper limit of normal of 66 pg/ml.

Measurement of aBMD by DXA

aBMD by DXA was measured at the total LS (L1 through L4), TH, FN, and nondominant 1/3R and UDR using a Hologic QDR 4500 densitometer (Hologic, Inc., Waltham, MA) in the array (fan beam) mode. In our laboratory, short-term, in vivo precision is 0.68% for the spine, 1.36% for the FN, and 0.70% for the radius. T-scores compared subjects to data from young-normal populations of the same race and sex provided by the manufacturer (spine and forearm) and by the National Health and Nutrition Examination Survey III (TH and FN).

HR-pQCT Imaging of the Radius and Tibia

HR-pQCT (XtremeCT; Scanco Medical AG, Brüttisellen, Switzerland) of the nondominant forearm and leg was performed and analyzed as described previously.^{14,55,63–66} In brief, all imaging was performed in our laboratory by a dedicated research densitometrist. HR-pQCT of the dominant limb was performed only if there was previous fracture or an arteriovenous fistula or graft in the nondominant limb. The arm or leg was positioned in the scanner and the region of interest was defined on a scout film by manual placement of a reference line at the endplate of the radius or tibia, with the first slice 9.5 and 22.5 mm proximal to the reference line at the radius and tibia, respectively. A stack of 110 parallel computed tomography slices was acquired at the distal end of both sites with a slice thickness of 82 μm, an image matrix size of 1024 × 1024, and a nominal voxel size of 82 μm. Attenuation data were converted to equivalent hydroxyapatite (HA) densities. A phantom was scanned daily for quality control. Image processing and calculation of numerical values were performed using Scanco software. Briefly, the volume of interest is automatically separated into cortical and trabecular regions using a Laplace–Hamming filter and threshold. Mean cortical thickness is defined as the mean cortical volume divided by the outer bone surface. Trabecular bone density is defined as the average bone density within the trabecular volume (TV) of interest; bone volume (BV)/TV (%) is derived from trabecular density (Dtrab) assuming that the density of fully mineralized bone was 1.2 g HA/cm³ (BV/TV = 100 × Dtrab/1200 mg HA/cm³). Because measurements of trabecular microstructure are limited by the resolution of the XtremeCT, which approximates the width of individual trabeculae, trabecular structure is assessed using a semiderived algorithm.^63,64 Trabeculae are identified by a mid-axis transformation method and the distance between them is assessed by the distance-transform method.⁶⁷ Trabecular number is defined as the inverse of the mean spacing of the mid-axes. Trabecular thickness and trabecular separation are derived from BV/TV and trabecular number using formulas from traditional quantitative histomorphometry: trabecular thickness = (BV/TV)/trabecular number and trabecular separation = (1 − BV/TV)/trabecular number. The intraindividual distribution of separation (μm), a parameter that reflects the heterogeneity of the trabecular network, is also measured.

Statistical Analysis

Statistical analyses were conducted using SAS (version 9.2, SAS Institute, Cary, NC). Categorical data were compared using the χ² test. Continuous data were evaluated for normality before statistical testing and log-transformed when appropriate. Group differences were determined by t test for unequal variances. Generalized linear models were used to adjust for imbalances in covariate structure between fracture and nonfracture groups (Table 1). Preliminary models including gender and race did not demonstrate effects on associations between BTMs and fracture; therefore, these covariates were not included in the final adjusted models. Pearson correlation coefficients were determined after adjustment for kidney function defined by eGFR. Univariate logistic regression was performed to determine univariate relationships between fracture and measures of BTMs after mean standardization. Multiple logistic regression models were used to evaluate independent relationships between BTMs and fracture after adjustment for FN T-score by DXA and unbalanced covariates in this cohort, including age and diabetes status. Because of moderate and significant correlations between BTMs with FN BMD by DXA, BTMs were tertiled before inclusion in multiple logistic models to avoid issues of multicolinearity. Standard ROC curve analysis was performed to determine the ability of DXA, HR-pQCT, and BTMs to discriminate fracture status. In multiple logistic and ROC models, FN T-score was chosen as the referent group because of its common use in clinical practice to evaluate for osteoporosis and its inclusion in the FRAX tool as the skeletal site at which to assess absolute 10-year hip and major osteoporotic fracture risk.¹⁵ Diagnostic test characteristics were also determined for a combination of low FN T-score, defined as being −2.0 or less and a BTM level in the upper two tertiles. Data are presented as sensitivity, specificity, and positive and negative predictive values and positive and negative likelihood ratios with their respective 95% confidence intervals.

DISCLOSURES

T.L.N. has a consulting agreement with Roche Diagnostics.