Prostate cancer is the second-leading cause of cancer-related death in men and is the most common visceral malignancy.1 Androgen deprivation therapy (ADT), a common treatment for metastatic and nonmetastatic prostate cancer, has increased fourfold over the past decade.2, 3 However, androgen deprivation therapy also increases fracture risk two- to fourfold.4–8 The resulting hip fractures carry an increased mortality up to 30% in men, and vertebral fractures are also associated with increased mortality and decreased quality of life.9, 10 Therefore, it is important to identify men who are at greatest risk for fractures.
The gold standard to determine fracture risk is bone mineral density assessed by dual-energy X-ray absorptiometry (DXA). However, bone mineral density only explains about 60% of fracture risk in untreated patients.11–13 We have previously shown that conventional DXA criteria for osteoporosis results in a misdiagnosis of more than 75% of men with prostate cancer on androgen deprivation therapy.14 However, other factors such as trabecular microstructure contribute significantly to bone strength and determine fracture risk.
Until relatively recently, bone microarchitecture could not be assessed except by iliac crest bone biopsy. Several recent techniques by computed tomography (CT) or magnetic resonance imaging (MRI) can examine trabecular bone microstructure.15–18 A technique known as high-resolution MRI (HR-MRI) of the radius provides a three-dimensional image of bone trabecular microarchitecture.16–18 HR-MRI examines topographical microarchitecture of the radius and examines ratios of platelike to rodlike trabecular structure. We have previously reported that in postmenopausal women receiving the bisphoshonate risedronate over 12 months, trabecular microarchitecture with HR-MRI of the wrist demonstrated improvements in trabecular microstructure, which were not appreciated by conventional bone mineral density by DXA.19 The goals of our study were to examine the cross-sectional association of HR-MRI topographical parameters with fragility fractures, conventional bone density, and biochemical markers of bone turnover in this cohort of men with prostate cancer on androgen deprivation therapy.
Materials and Methods
Design and subjects
This cross-sectional study included 137 men aged 60 years and older with nonmetastatic prostate cancer on ADT for ≥ 6 months. Participants were enrolled from urology and geriatric practices and advertisements. ADT included orchiectomy or gonadatropin releasing hormone agonists with or without antiandrogens. Men were excluded if they had metastatic prostate cancer, nonmetstatic prostate cancer with a prostate-specific antigen (PSA) level > 4 (unless therapeutic adjustments were being made), if they were currently on medications known to alter bone mineral metabolism within the past year (including bisphosphonates, corticosteroids, antiseizure medicines, or other antiresorptive therapy), or if they had diseases known to affect osteoporosis (for example, hyperparathyroidism, thyrotoxicosis). The Institutional Review Board at the University of Pittsburgh approved the study, and participants provided written informed consent.
Participants completed a food-frequency questionnaire to evaluate their total calcium and vitamin D intake from diet and vitamin supplements. Participants also completed a questionnaire regarding family history, medical, surgical, and fracture histories in addition to information on medication, activity level, tobacco, and alcohol use. Height was obtained with a Harpenden stadiometer (Holtain Ltd.-Crymych, Dyfed, UK). Weight was measured with a Health-O-Meter balance-beam scale (Sunbeam, Inc., Boca Raton, FL, USA). Body mass index (BMI) was calculated as weight/(height)2 in kg/m2. Patients were asked to report their tallest height from memory, and height loss was calculated from the tallest height and current height.
Bone mineral density
Bone mineral density of the PA spine (L1–L4), hip (total hip, femoral neck), right radius (1/3 distal radius, ultra distal radius, and total radius) was performed by DXA using a Hologic Discovery A (Hologic Inc., Bedford, MA, USA). We analyzed absolute bone mineral density in g/cm2 and calculated T-scores (number of standard deviation units from adult peak bone mass) using the Hologic male database. We used World Health Organization criteria to define osteoporosis (T-score ≤ –2.5 SD), low bone mass (T-score–2.5 to 1.0 SD), and normal (T-score ≥ –1.0 SD).20 The coefficient of variation for our DXA machine is 1.3% for the spine and 1.4% for the total hip.21
We assessed vertebral fractures utilizing Vertebral Fracture Assessment (VFA) performed with lateral spine imaging (T5–L4) on the Hologic Discovery A using the manufacturer's standard protocols. We classified vertebral fractures based on reduction in vertebral height according to the Genant semiquantitative method analysis.22 Fractures were classified as grade 1 (mild) with a 20% to 25% loss of vertebral height, grade 2 (moderate) with 25% to 40% loss of vertebral height, or grade 3 (severe) with > 40% loss of vertebral height. In addition, all fractures were confirmed by conventional lateral, thoracic, and lumbar X-rays. These were graded by a single radiologist. We have previously reported that sensitivity and specificity of VFA versus semiquantitative radiography were 100% and 95%, respectively. The Kappa statistic for agreement was 0.92.14 Participants were categorized as having a moderate-severe, mild, or no vertebral fracture based on their most severe reading in T5–L4 locations.
The fracture risk assessment algorithm was used to calculate the 10-year risk of an osteoporotic fracture and risk of hip fracture23 using the femoral neck bone mineral density (BMD; g/cm2) for each participant.
High-resolution magnetic resonance imaging
HR-MRI uses an algorithm that transforms data into a detailed three-dimensional model of bone microstructure as previously described.19, 24 Subjects had a coil placed on their right distal forearm. They were then scanned with a GE Signa 1.5 Tesla scanner (General Electric Medical Systems, Waukesha, WI, USA). The technique quantifies the degree to which trabecular plates (surfaces) have deteriorated to become rods (curves), a change that is characteristic of osteoporosis.17 Using the patented unique algorithm, each image voxel can be classified as belonging to a surface or curve junction between two topographical types. An erosion index (EI) is determined that represents the ratio of voxels expected to decrease during resorptive bone loss derived by those that are expected to increase. The topographical parameters (plate-to-rod ratio, an erosion index) are computed after resolution has been enhanced by the software, resulting in a voxel size of approximately 67 µm isotropic. The software places the region of interest automatically. The outer boundary of the region of interest is typically within 1 mm of the cortical endosteal boundary. The user may manually modify the region of interest, if needed. The indices include: the bone volume to total volume ratio (BV/TV), topographical surface density (Surface Density), topographical curve density (Curve Density), surface to curve ratio (Surf/Curv where higher values indicate a more intact trabecular network and lower values indicate a network that has deteriorated) and an erosion index (EI, a ratio of parameters expected to increase when bone trabeculae deteriorate with higher values indicating greater deterioration). The coefficient of variation for HR-MRI parameters is 4% to 7% and the reliability expressed is the interclass correlation coefficient 0.95 to 0.97 at the radius.25
Bone mineral metabolism and turnover markers
Serum 25-hydroxyvitamin D was assessed by liquid chromatography/mass spectrometry. Parathyroid hormone was measured by ELISA (ALPCO Diagnostics, Salem, NH, USA). Collagen type 1 cross-linked c-telopeptide (CTx), a marker of bone resorption, was assayed by ELISA (Crosslaps, Immunodiagnostics Systems Inc., Scottsdale, AZ, USA), and the marker of bone formation, procollagen type 1 amino-terminal propeptide (P1NP), was assessed by radioimmune assay (Immunodiagnostics Systems).
SAS® Version 9.2 (SAS Institute Inc., Cary, NC, USA) was used for all statistical analyses. We used appropriate descriptive statistics and graphical methods to summarize the participant characteristics and outcomes of bone health across the three categories of participants defined by their vertebral fracture status (none/mild/moderate-severe). All DXA- and MRI-derived measures were used as continuous variables in all analyses to ensure maximum amount of information from the measures are used in eliciting results. We used analysis of variance (ANOVA) with pairwise comparisons using Fisher's least significant difference to make comparisons of outcomes and clinical characteristics across the three vertebral fracture status categories. We used multiple coefficient of determination (R2) from ANOVA models as a measure of overall strength of association between vertebral fractures and each outcome/clinical characteristic. We used Pearson product-moment correlation coefficients (r) to examine the association between traditional BMD measures and HR-MRI measures as well as duration of ADT. Finally, we used multivariable logistic regression models to evaluate whether HR-MRI measures add substantially to the accuracy of predicting moderate-severe VFs above and beyond the combination of commonly used DXA-derived T-scores for spine, hip, and femoral neck using the unadjusted area under receiver operator characteristic curve (c-statistic or AUC) where an increase of 0.025 or greater is considered substantial.26 Specifically, using only those participants with both DXA- and MRI-derived measures, we first computed AUC for DXA-derived measures alone and then for both DXA- and MRI-derived measures together. The increase in AUC was interpreted as being attributable to additional information contained in MRI-derived measures above and beyond that contained in DXA-derived measures.
Two hundred seventeen men were telephone screened and 160 were eligible. One hundred thirty-seven were enrolled. HR-MRI was performed in 108 men. Of the 137 men enrolled, 86 had no vertebral fracture, 41 had a mild vertebral fracture, and 10 had a moderate to severe vertebral fracture. Of the 51 men with vertebral fractures, 48 (94.1%) men had silent fractures that they were unaware of and 23 (45.1%) had two or more vertebral fractures.
The average daily calcium intake, including diet plus supplementation, ranged from 1228 ± 65 to 1769 ± 163 mg/day (mean ± standard error) across the three vertebral fracture categories and was higher in those with moderate-severe VF (Table 1). There was no significant difference in age, body mass index, vitamin D intake, duration of androgen deprivation therapy, or PSA between men with and without vertebral fractures. Height loss was greater in men with moderate-severe VF compared with men with no or mild VF (both p < 0.05, Table 1).
Table 1. Clinical Characteristics, Bone Turnover Markers, and Bone Mineral Density T-Scores in Men Across Grades of Fracture
Serum calcium and 25-hydroxyvitamin D were similar across the three groups (Table 1). However, serum PTH trended higher in men with moderate-severe VF compared with those with mild VF (p < 0.05). In addition, CTx tended to be higher in men with moderate-severe VF compared with those with no or mild VF (both p < 0.05). The multiple coefficient of determination for the ANOVA models for MRI indices was also weak and ranged from R2 = 0.05 to 0.09.
Seven percent or 10 of 137 of the participants were classified as having osteoporosis by conventional bone mineral density of the spine, total hip, or femoral neck. Thirty-seven percent or 47 of the remaining 127 men without osteoporosis by bone mineral density (men who had normal to low bone mass) had a vertebral fracture identified by VFA and conventional X-rays. This suggests that 89.5% or 51 of the 57 patients with clinically defined osteoporosis would have been misclassified by bone mineral density alone. However, if we classified osteoporosis by bone mineral density of the spine, total hip, femoral neck, or 1/3 distal radius, 27% of men would have been characterized as osteoporotic.
By ANOVA comparison across VF grades, bone mineral density was lower at the PA spine, total hip, femoral neck, and 1/3 distal radius in men with moderate-severe VF compared with lesser grades (all p < 0.05, Fig. 1). Similar trends were noted for T-scores (Table 1). The multiple coefficient of determination for the ANOVA models for BMD was weak and ranged from R2 = 0.04 to 0.06. The FRAX score for the 10-year risk of a major osteoporotic fracture or a hip fracture was higher in men with moderate-severe VF compared with lesser grades (p < 0.05, Table 1).
The BV/TV was lower, surface density was lower, and erosion index was higher in men with moderate to severe VF compared with lesser grades (all p < 0.05, Fig. 2) by ANOVA comparison across VF grades. In addition, BV/TV ratio was higher, surface density was higher, the surface/curve ratio was higher, and erosion index was lower in men with moderate-severe VF compared with none (all p < 0.05) or those with mild VF (all p < 0.05, Fig. 2). The multiple coefficient of determination for the ANOVA models for MRI indices was also weak and ranged from R2 = 0.05 to 0.09.
We examined the unadjusted area under the receiver operator curve AUC for predicting a moderate-severe VF or any VF. Information contained in all MRI measures added substantially and significantly to that contained in the hip and spine T-scores. AUC for hip and spine T-scores was 0.804, which increased to 0.890 when hip and spine T-scores were considered together with MRI measures, showing a substantial increase of 0.086 (p = 0.0401) attibutable to the information contained in MRI measures above and beyond that contained in hip and spine T-scores. Similarly, the total hip, femoral neck, and spine T-scores had an AUC of 0.831, which increased to 0.902 when considered together with the MRI measures for the prediction of moderate-severe VF, resulting in a substantial increase of 0.071 (p = 0.0471). The addition of MRI measures to the total hip, femoral neck, spine, and 1/3 distal radius added substantially (AUC increased 0.829 to 0.902 by 0.073) to the prediction of moderate-severe VF, but this was not statistically significant (p = 0.1002). The addition of the MRI measures added substantially to the prediction for all VF (includes mild and moderate-severe VF) as measured by the increase in AUC magnitude ( ≥ 0.025) but was not statistically significant.
BMD at the ultradistal radius was moderately positively associated with BV/TV ratio (r = 0.65), surface density (r = 0.62), surface/curve ratio (r = 0.58), and negatively associated with erosion index (r = −0.46, all p < 0.0001, Table 2). The positive associations at other sites were weaker and ranged from r = 0.29 to 0.39 at the spine, r = 0.46 to 0.50 at the total hip, and r = 0.45 to 0.49 at the femoral neck (all p < 0.05). The indices of HR-MRI were not associated with vitamin D, PTH, or bone turnover markers. Duration of ADT was weakly and negatively associated with HR-MRI (r ranged from −0.23 to −0.27, p < 0.05) and 1/3 distal radius (r = −0.21, p < 0.05) but not at the spine or hip sites.
Table 2. Correlations of Bone Mineral Density and Indices of HR-MRI
HR-MRI scans of two representative patients are shown in Fig. 3. Although the spine and hip T-scores were similar in these two patients, patient A had a vertebral fracture and indices were suggestive of fewer trabecular plates and higher erosion index compared with patient B.
We found that 37% of men with prostate cancer on androgen deprivation therapy had vertebral fractures and the majority were silent. Furthermore, DXA-assessed bone mineral density measurements alone using the standard of care assessment of spine and hip measurements only would lead to misclassification of osteoporosis of approximately 90% of patients. We observed that men with vertebral fractures had a lower surface density suggestive of fewer trabecular plates and a higher erosion index compared with men without fracture. Previous studies have examined the importance of examining trabecular microstructure in the assessment of skeletal integrity. Wehrli and colleagues examined 79 women aged 28 to 78 years (mean age 54) with a HR-MRI of the forearm in addition to MRI assessment for vertebral fractures.17 They observed that there was no significant difference in standard DXA of the spine, femoral neck, or trochanter between women with a vertebral fracture and those without. However, they reported significant difference in the radial HR-MRI indices between those with a vertebral fracture and those without. They reported that those with vertebral deformities had a lower ratio of platelike to rodlike trabeculae to those without vertebral deformity. The number of intact trabecular rods were also lower in the vertebral deformity group. Our current study supports this information but in a cohort of men at risk for osteoporosis who are on androgen deprivation therapy. Furthermore, by ROC analysis, the addition of the HR-MRI indices to conventional DXA T-scores of the hip and spine added substantially to the prediction of moderate-severe VF.
We observed weak to moderate correlations between the HR-MRI indices and measures of DXA at spine and hip sites, in addition to the forearm. The site with the strongest correlation, the ultra distal radius, was the site where the coil was placed. Furthermore, longer duration of ADT was negatively associated with the HR-MRI indices but not the conventional measures of DXA-assessed BMD at the spine or hip. The spine may be falsely elevated because of osteoporotic calcifications, and even the hip site may be falsely elevated because of osteoarthritis. The site with the strongest association with duration of ADT was the 1/3 distal radius. We have previously demonstrated that this site may be more sensitive to loss in men on androgen deprivation therapy.21, 27 Because HR-MRI provides information on micro-structure and connectivity not available by conventional DXA, studies are needed to determine if HR-MRI in conjunction with DXA are useful in predicting future fractures.
Earlier investigations in men with prostate cancer on ADT have reported a range from 24% to 53% prevalence for DXA classification of osteoporosis.28–30 Most investigations classify osteoporosis utilizing the World Health Classification of a bone mineral density ≤−2.5 SD below peak bone mass.31 However, the classification may vary depending on the number of skeletal sites used to meet the classification (eg, spine, total hip, femoral neck, 1/3 distal radius with combinations of two to four sites), the reference database (NHANES,32 local comparison, sex, race), the study design (chart review, cross-sectional, clinical trial), patient age, duration of ADT, and inclusion criteria. Our study found a prevalence of 7% for osteoporosis with the conventional DXA of the spine, total hip, and femoral neck only, but this increased to 27% when we included the 1/3 distal radius. Adler and colleagues reported a prevalence of 33% in men with prostate cancer on ADT when he included the spine, hip, and forearm as well.29 The prevalence of osteoporosis by DXA of four distinct sections of the hip using the US NHANES male database was 3% to 6%.32 Therefore, the prevalence differs depending on the number and location of skeletal sites included.
Moreover, there is variation in the radiologic definition of a vertebral fracture. Studies may classify vertebral fractures by DXA-derived vertebral fracture assessment or conventional lateral, thoracic lumbar X-rays. Vertebral fractures may be classified based on semiquantitative, quantitative morphometric,22 algorithm-based qualitative analysis,33 clinical reading by a site radiologist, or unspecified criteria. In community dwelling men participating in the Osteoporotic Fractures in Men (MrOS) study, the prevalence of vertebral fractures ranged from 10% to 13% based on three conventional techniques, with moderate agreement (kappa = 0.42 to 0.62), but up to 74% of men had at least one vertebrae with short vertebral height that is often confused with a vertebral fracture.34 Other investigators report a prevalence range of 12% to 20% in men.35 For the current study, we used vertebral fracture assessment and conventional thoracic/lumbar X-rays using the Genant method of semiquantitative analysis, read by a single radiologist with expertise in skeletal radiology, and reported a prevalence of 37%. The prevalence values for both DXA-classified osteoporosis and vertebral fractures need to be viewed in the context of multiple ascertainment techniques and differences in analysis.
As expected, the 10-year risk of a major osteoporotic fracture or hip fracture using FRAX were higher in men with moderate-severe VF compared with lesser grades. However, there are limitations with FRAX. Although we calculated FRAX for all men, FRAX was designed to include men with low femoral neck bone mass or osteopenia rather than osteoporosis. Men with clinical osteoporosis by VF would be treated regardless of FRAX score. Furthermore, FRAX wouldn't include those with osteoporosis in the spine or forearm DXA, who would be treated.
We found that on average the total daily calcium intakes were above the current Institute of Medicine (IOM) and National Osteoporosis Foundation recommendations of 1200 mg/day and calcium intake was the highest in men with moderate-severe VF. In addition, the mean intake of vitamin D was near the IOM recommendations of 800 IU/day in those with no fracture and above 800 IU/day for those with fractures. Moreover, mean vitamin D levels were above the 20 ng/dL levels in all groups. Men with prostate cancer on ADT have been encouraged to meet these standards in clinical guidelines.36 Finally, we found that although CTx, the marker of bone resorption, was higher in men with most severe VF, P1NP, a marker of bone formation, was similar among the three groups.
We included two patients with relatively similar assessments for conventional DXA measures at the hip and spine and with T-scores that would have classified both with a “normal” BMD. However, one patient had a vertebral fracture not assessed by conventional DXA, and the HR-MRI highlighted the loss of trabecular structure not seen with standard DXA or X-rays. These types of assessments may lead to future studies to determine if therapeutic alternatives should be based on structure in addition to overall bone mass.
Our study had several limitations. First, we only examined men aged 60 years and older with prostate cancer on ADT, and results may not necessarily apply to men in general or younger men with prostate cancer on ADT. Second, our cohort was predominantly Caucasian, and the results may not necessarily apply to minorities. Third, we examined height loss assessed by memory of a participant's tallest height; however, there may be recall bias in using this method. Finally, the cross-sectional associations found in our analyses do not carry the same level of strength in evidence as a longitudinal prediction of future fractures with an independent validation cohort.
This study also has several strengths. We were able to assess conventional bone mineral density, conventional lateral spine X-rays, forearm bone mineral density, biochemical markers of bone turnover, and HR-MRI simultaneously in men with prostate cancer on ADT. Second, we were able to examine the impact of duration of therapy on these variables by including men who had been on ADT for a variable length of time. Third, we confirmed all X-rays assessed by vertebral fracture assessment with the conventional lateral X-rays using a single radiologist.
We conclude that vertebral fractures are found in 37% of men with prostate cancer on ADT. DXA-based criteria alone will under-diagnose and misclassify patients. HR-MRI provides a novel technique to assess the deterioration of micro-structural integrity in men with prostate cancer on ADT with vertebral fractures and adds additional information to conventional bone mineral density and vertebral fracture assessment.
SG serves on advisory committees for Merck and Amgen and has received research grants from Tarsa, Warner Chilcott, and Eli Lilly. SP has received past research funding to do nondrug observational research from Merck, Ortho Biotech, and Eli Lilly. All other authors state that they have no conflicts of interest.
Funding: PC060710 (DOD IDEA), 2K24DK062895-06, University of Pittsburgh Clinical Translational Research Center RFA-RM-06-002, University of Pittsburgh Department of Urology, Claude D. Pepper Center, Division of Geriatric Medicine 2 P30 AG024827-06.
Authors' roles: study design: SLG, SP, JBN, and NMR; study conduct: SLG and JW; data collection: SLG, JW, and CB; data analysis: SP; data interpretation: SLG, SP, JBN, NMR, and CB; drafting manuscript: SLG, SP, and NMR; revising manuscript: SLG and SP; approval of final version of manuscript: SLG, SP, JW, JBN, NMR, and CB. The authors take responsibility for the integrity and accuracy of the data analysis.