Evaluation of Radius Microstructure and Areal Bone Mineral Density Improves Fracture Prediction in Postmenopausal Women

Transcription

1 ORIGINAL ARTICLE JBMR Evaluation of Radius Microstructure and Areal Bone Mineral Density Improves Fracture Prediction in Postmenopausal Women Emmanuel Biver, 1 Claire Durosier-Izart, 1 Thierry Chevalley, 1 Bert van Rietbergen, 2 Rene Rizzoli, 1 and Serge Ferrari 1 1 Division of Bone Diseases, Geneva University Hospitals and Faculty of Medicine, University of Geneva, Geneva, Switzerland 2 Department of Biomedical Engineering, Eindhoven University of Technology, Eindhoven, Netherlands ABSTRACT A majority of low-trauma fractures occur in subjects with only moderate decrease of areal bone mineral density (abmd), ie, osteopenia, assessed by dual-energy X-ray absorptiometry (DXA) or low fracture probability assessed by FRAX. We investigated whether peripheral bone microstructure and estimated strength improve the prediction of incident fractures beyond central DXA and FRAX. In this population-based study of 740 postmenopausal women (aged years) from the Geneva Retirees Cohort (ISRCTN registry ), we assessed at baseline cortical (Ct) and trabecular (Tb) volumetric bone mineral density (vbmd) and microstructure by peripheral quantitative computed tomography (HR-pQCT); bone strength by micro-finite element analysis; abmd and trabecular bone score (TBS) by DXA; and FRAX fracture probability. Eighty-five low-trauma fractures occurred in 68 women over a follow-up of years. Tb and Ct vbmd and microstructure predicted incident fractures, independently of each other and of femoral neck (FN) abmd and FRAX (with BMD TBS). However, the associations were markedly attenuated after adjustment for ultra-distal radius abmd (same bone site). The best discrimination between women with and without fracture was obtained at the radius with total vbmd, the combination of a Tb with a Ct parameter, or with failure load, which improved the area under the curve (AUC) for major osteoporotic fracture when added to FN abmd (0.760 versus 0.695, p ¼ 0.022) or to FRAX-BMD (0.759 versus 0.714, p ¼ 0.015). The replacement of failure load by ultra-distal abmd did not significantly decrease the AUC (0.753, p ¼ and 0.750, p ¼ 0.509, respectively). In conclusion, peripheral bone microstructure and strength improve the prediction of fractures beyond central DXA and FRAX but are partially captured in abmd measured by DXA at the radius. Because HR-pQCT is not widely available for clinical purposes, assessment of ultra-distal radius abmd by DXA may meanwhile improve fracture risk estimation American Society for Bone and Mineral Research. KEY WORDS: BONE QCT/mCT; OSTEOPOROSIS; GENERAL POPULATION STUDIES; FRACTURE RISK ASSESSMENT Introduction Low-trauma fractures due to osteoporosis represent a major public health concern, more common than stroke, myocardial infarction, and breast cancer combined in postmenopausal women, and leading to disability and increased mortality risk. (1 3) It is estimated that a 50-year-old white woman has a 15% to 20% lifetime risk of hip fracture and a 50% risk of any osteoporotic fracture. (4,5) Although osteoporosis is defined as a decrease in bone mineral density (BMD) and an alteration of bone microstructure leading to higher bone fragility and risk of fracture, the operational definition of osteoporosis proposed by the World Health Organization is based only on the bone mass criteria (BMD T-score 2.5), with the femoral neck (FN) as the reference site. (6) The limits of this definition have rapidly been highlighted, as more than half of low-trauma fractures occur in subjects without osteoporosis on central (hip or spine) dualenergy X-ray absorptiometry (DXA). (7 9) The use of peripheral DXA (distal radius) in clinical practice has been poorly defined and is currently not recommended, except in case of impossibility to examine central bones. (10) In this context, identification of subjects at high risk of fracture remains challenging. Multiple tools have been developed, based essentially on clinical risk factors of fracture BMD. (11) The fracture risk assessment tool (FRAX) estimates the absolute fracture risk by incorporating several clinical risk factors and FN areal BMD (abmd). (12,13) More recently, the trabecular bone score (TBS), a gray-level textural index derived from lumbar spine DXA images, has been added in the FRAX algorithm to adjust FRAX probability. (14,15) However, a large proportion of Received in original form July 11, 2017; revised form September 26, 2017; accepted September 27, Accepted manuscript online September 28, Address correspondence to: Emmanuel Biver, MD, PhD, Division of Bone Diseases, Geneva University Hospitals and Faculty of Medicine, University of Geneva, 4 Rue Gabrielle Perret-Gentil, 1205 Geneva, Switzerland. Emmanuel.Biver@hcuge.ch Additional Supporting Information may be found in the online version of this article. Journal of Bone and Mineral Research, Vol. 33, No. 2, February 2018, pp DOI: /jbmr American Society for Bone and Mineral Research 328

2 low-trauma fractures occurs in patients with osteopenia or relatively few clinical risk factors, hence modest fracture probability score by FRAX. (8,16) The addition of TBS marginally improves hip fracture prediction and does not improve the performance of the tests to predict major osteoporotic fractures (MOF). (15) One of the reasons of this observation is that alterations of bone quality, especially bone microstructure (as part of the definition of osteoporosis), which contributes to bone fragility in addition to abmd, is not captured by these tools. Peripheral volumetric BMD (vbmd) and cortical (Ct) and trabecular (Tb) bone microstructure can be assessed non-invasively in vivo at the distal radius and tibia, using a high-resolution (82 mm) peripheral quantitative computed tomography (HR-pQCT) suitable for in vivo studies in humans. (17,18) Bone strength parameters (stiffness, failure load, and apparent modulus) can be estimated by micro-finite element analyses (FEA) created directly from the segmented HR-pQCT images. (19) Whereas bone microstructure and estimated strength have been associated with prevalent fractures independently of abmd in postmenopausal women and men, it remains unclear whether they may improve the prediction of incident fractures. (18,20,21) The aim of this prospective study was to investigate the independent contribution of Ct and Tb vbmd, microstructure, and estimated strength at the peripheral skeleton to incident clinical fracture risk in a cohort of community-dwelling postmenopausal women. We notably evaluated to what extent bone microstructure and strength assessment improves fracture prediction beyond central abmd, TBS, and FRAX, the reference tools used in clinical practice to assess fracture risk. Eventually, we investigated whether the more readily available measurement of radius abmd by DXA could be a surrogate for microstructural evaluation in fracture prediction. Materials and Methods Subjects GERICO (Geneva Retirees Cohort ISRCTN ) is a prospective ongoing cohort study designed to identify predictive factors of fracture risk in retired workers from the Geneva area. Healthy community-dwelling postmenopausal women (n ¼ 759) were recruited at the ages of 63 to 67 years between 2008 and 2011 by advertisement in the local newspapers, among the Geneva University Hospitals staff, or in large local companies, as previously reported. (22) The present analysis was conducted in 740 women prospectively evaluated over years after the baseline examination for the occurrence of low-trauma fracture (Supplemental Fig. S1). At baseline, body weight and standing height were measured and body mass index (BMI) calculated. Dietary calcium and protein intake, as well as physical activity, were assessed by face-to-face frequency questionnaires. (23) Tobacco and alcohol consumption were assessed, as well as menopause age and current or past drug use. The 10-year probabilities of MOF were calculated with the FRAX tool, using country-specific data, with inclusion of FN abmd in addition to the clinical risk factors (FRAX-BMD). The probability of MOF obtained was secondarily adjusted for TBS in the FRAX algorithm (FRAX-BMD þ TBS). Serum b-carboxyterminal cross-linked telopeptide of type I collagen (CTX), aminoterminal propeptide of type-i procollagen (P1NP), and 25-hydroxyvitamin D were measured at baseline on a Cobas instrument using Elecsys reagents (Roche Diagnostics, Mannheim, Germany). (24) The study protocol received the approval from the Geneva University Hospitals Ethics Committee, and all participants provided written informed consent. Fractures assessment Prior fracture history (after the age of 20 years) at baseline and incident fractures over the follow-up were recorded during faceto-face scheduled and structured interviews, in which details on fracture site, date, type and intensity of trauma, and modalities of treatment were recorded. Information on incident fracture was obtained for 98% of women who entered GERICO. Written confirmation (eg, discharge summary, radiologist report) was requested for incident fractures and all cases were adjudicated by a study medical doctor. Low-trauma (LT) fractures were defined as any fracture resulting from a fall from standing height or less, with exclusion of fractures of fingers, toes, skull, and face. Morphometric vertebral fractures (VF) were identified using vertebral fracture assessment (VFA) according to the semiquantitative method of Genant on lateral scans of the spine (T 6 to L 4 ) acquired with DXA at baseline and at time of follow-up. (25) Only grades 2 and 3 VF (25% loss of vertebral height) at baseline were recorded as morphometric VF and included in prior fracture records. Six women with vertebral deformity without clinical symptoms or radiographic confirmation during the follow-up were excluded. The primary outcome was incident clinical low-trauma fracture and the secondary outcome clinical low-trauma major osteoporotic fracture (MOF), including humerus, forearm, proximal femur, and VF. Controls were women without incident fracture except fractures of fingers, toes, skull, and face (women with other traumatic fractures were excluded from controls). BMD, bone microstructure, and bone strength assessment Lumbar spine, proximal femur, and distal radius abmd were determined at baseline by DXA using a Hologic QDR Discovery instrument (Hologic Inc., Waltham, MA, USA). Using the WHO classification, women were classified as osteoporotic if they had at least one T-score 2.5 SD at the lumbar spine, total hip, or FN. TBS was assessed using TBS insight v2.2.0 (Medimaps Group SA, Plan-les-Ouates, Switzerland) based on the antero-posterior spine DXA scan acquired at baseline. Volumetric BMD and microstructure variables, for total (Tt) bone and the Ct and Tb compartments separately, were determined at baseline at the distal radius and tibia by HRpQCT (XtremeCT, Scanco Medical, Br uttisellen, Switzerland), as previously described. (26 28) The determinations were performed on the non-dominant limb, unless a fracture was reported in the region of interest. Each scan was assessed for motion artifact according to the manufacturer recommendation and lowquality scans were excluded (39 radius scans and 16 tibia scans). Recorded variables were: Tt, Ct, and Tb vbmd (Tt.BMD, Ct.BMD, and Tb.BMD, mg/cm 3 ) and areas (Tt.Ar, Ct.Ar, and Tb.Ar, mm 2 ); Tb number (Tb.N, mm 1 ), thickness (Tb.Th, mm), and separation (Tb.Sp, mm); Tb separation SD (Tb.Sp.SD, mm), an estimate of the heterogeneity of the Tb structure; Ct thickness (mm); and Ct porosity, quantified with two different methods, an automatic segmentation of the Ct compartment using a morphological assessment (Ct.Po), and a density assessment method using StrAx10 software within the total cortex (Tt Ct.Po) or the cortex segmented into compact-appearing cortex (Comp Ct.Po), outer transitional zone (Out Ct.Po), and the inner transitional zone (Inn Journal of Bone and Mineral Research RADIUS MICROSTRUCTURE AND AREAL BMD EVALUATION IMPROVE FRACTURE PREDICTION 329

3 Ct.Po). (29) Bone strength was estimated by micro-finite element analysis (FEA) using finite element models of the radius and the tibia created directly from the segmented HR-pQCT images, as previously reported. (23,30) Recorded variables were bone predicted failure load (FL, N), stiffness (N/mm), and apparent modulus (N/mm 2 ). All micro-fea were performed using the FE solver integrated in the IPL software version 1.15 (Scanco Medical AG). Statistical analysis All data were reported as means and standard deviations (SD) and percentages. The differences between groups at baseline were assessed by a Mann-Whitney test. The Shapiro- Francia W test and Skewness/Kurtosis tests were used to test the normality of the distributions, and non-gaussian variables, including covariates, were normalized using simple mathematical transformations before testing the associations with incident fractures in linear models. We used Cox s proportional hazard models to estimate the hazard ratios (HR) for time to first incident low-trauma fracture or MOF per 1 SD impairment of each vbmd, microstructure, or estimated strength parameter. Additional models, adjusted for FN abmd, FRAX-BMD, FRAX-BMD þ TBS, or ultra-distal radius abmd were performed. These adjustments were planned a priori to test if the contribution of vbmd, microstructure, or strength parameters to fracture risk was independent of the information captured by the current tools used in clinical practice to assess fracture risk (FN abmd and FRAX) and of abmd measured at the same radius bone site (ultra-distal radius). Using forward stepwise multivariate Cox s proportional hazard regressions, we identified the strongest predictors of fracture within all the radius and tibia vbmd, microstructure, and estimated strength parameters with a p value < 0.2 in the univariate analyses. In addition, Kaplan-Meier survival analyses were applied to illustrate the predictive ability of various parameters for fracture, and log-rank test was used to compare survival curves. To assess the ability of vbmd, microstructure, or strength parameters to discriminate between women with or without incident fractures, Harrell s C statistics obtained from the Cox regression models and receiver operating characteristic (ROC) curve analyses obtained from logistic regressions adjusted for the duration of follow-up were performed. Harrell s C is a measure of predictive discrimination designed for Cox regression. (31) The C-index values were compared to determine whether any single or combination of vbmd, microstructure, or strength parameters provided better Table 1. Characteristics of Women at Baseline With and Without Incident Low-Trauma Clinical Fractures and Major Osteoporotic Fractures No incident fracture n ¼ 641 Incident clinical low-trauma fracture n ¼ 68 p Value Incident clinical major osteoporotic fracture n ¼ 31 p Value Age (years) Body mass index (kg/m 2 ) Prior low-trauma fracture 16% 47% < % <0.001 Age at menopause (years) Tobacco consumption, current 8% 10% % Alcohol consumption 3 units daily 6% 4% % Dietary calcium intake (mg/d) Dietary total protein intake (g/kg/d) Physical activity (kcal/d) Anti-osteoporotic drug during follow-up 10% 22% % Anti-osteoporotic drug or MHT during 24% 31% % follow-up Lumbar spine T-score (SD) < FN T-score (SD) Total hip T-score (SD) Osteoporotic status on DXA a Osteoporosis (%) 18% 34% % Osteopenia (%) 58% 54% 48% Normal BMD (%) 24% 12% 13% TBS < FRAX MOF with BMD (%) < <0.001 FRAX MOF with BMD þ TBS (%) < < hydroxy-vitamin D (nmol/l) CTX (ng/l) P1NP (mg/l) MHT ¼ menopausal hormone therapy; FN ¼ femoral neck; TBS ¼ trabecular bone score; FRAX ¼ fracture risk assessment tool; MOF ¼ major osteoporotic fracture; BMD ¼ bone mineral density; CTX ¼ serum b-carboxyterminal cross-linked telopeptide of type I collagen; P1NP ¼ amino-terminal propeptide of type-1 procollagen. Values are means standard deviation (SD) or number (%). a Osteoporosis defined as at least one T-score 2.5 SD and osteopenia as at least one T-score between 1 and 2.5 SD with none 2.5 SD at the lumbar spine, total hip, or femoral neck. 330 BIVER ET AL. Journal of Bone and Mineral Research

4 Table 2. Associations Between Radius vbmd, Microstructure, or Strength and the Risk of Incident Low-Trauma Clinical Fractures and Major Osteoporotic Fractures Adjusted for UD radius abmd Adjusted for FRAX-BMD þ TBS Crude associations Adjusted for FN abmd Adjusted for FRAX-BMD HR (95% CI) p Value HR (95% CI) p Value HR (95% CI) p Value HR (95% CI) p Value HR (95% CI) p Value For low-trauma fracture prediction Tb.BMD 1 SD 1.67 (1.27, 2.20) < (1.21, 2.18) (1.15, 2.03) (1.09, 1.96) (0.89, 1.97) Ct.Ar 1 SD 1.69 (1.28, 2.22) < (1.22, 2.24) (1.14, 2.02) (1.04, 1.86) (0.87, 2.04) Inn Ct.Po % 1 SD 1.76 (1.33, 2.34) < (1.27, 2.33) < (1.20, 2.16) (1.14, 2.08) (0.96, 2.22) Tt.BMD 1 SD 1.72 (1.33, 2.22) < (1.27, 2.22) < (1.20, 2.05) (1.11, 1.92) (0.95, 2.25) Failure load 1 SD 1.72 (1.33, 2.23) < (1.29, 2.39) < (1.19, 2.05) (1.14, 1.98) (0.91, 2.52) For MOF prediction Tb.BMD 1 SD 2.04 (1.31, 3.16) (1.27, 3.26) (1.18, 2.92) (1.13, 2.94) (0.76, 2.71) Ct.Ar 1 SD 2.07 (1.35, 3.19) (1.32, 3.42) (1.19, 2.90) (1.03, 2.56) (0.78, 2.92) Inn Ct.Po % 1 SD 2.35 (1.49, 3.70) < (1.47, 3.89) < (1.36, 3.45) (1.30, 3.52) (0.99, 3.77) Tt.BMD 1 SD 2.18 (1.47, 3.21) < (1.45, 3.39) < (1.32, 2.99) (1.19, 2.76) (0.96, 3.66) Failure load 1 SD 1.95 (1.31, 2.91) (1.31, 3.37) (1.16, 2.68) (1.10, 2.62) (0.60, 2.93) HR ¼ hazard ratio; CI ¼ confidence interval; SD ¼ standard deviation; Tt ¼ total bone; Ct ¼ cortical; Tb ¼ trabecular; BMD ¼ bone mineral density; Ar ¼ area; Inn Ct.Po ¼ porosity of the inner transitional zone; FN ¼ femoral neck; FRAX ¼ fracture risk assessment tool; TBS ¼ trabecular bone score; UD ¼ ultra-distal; abmd ¼ areal BMD; MOF ¼ major osteoporotic fracture. Data are hazard ratios associated with 1 SD impairment of each parameter, obtained from Cox s proportional hazard models. The five best predictors of fracture are reported in this table. All other parameters are reported in Supplemental Tables S3 and S4. fracture discrimination compared with FN abmd. Areas under the curve were compared using the non-parametric DeLong test. A p value 0.05 was considered to be significant for all analyses. With the assumptions of an expected incidence of lowtrauma fractures of 10% over 5 years and a 10% dropout rate over the follow-up, the study had 90% power to detect a 42% increase in fracture risk, and 80% power to detect a 35% increase in fracture risk, associated with 1 SD decrease of radius Tt.BMD, with a one-sided a ¼ 0.05 significance level. The data were analyzed using STATA software, version (StataCorp LP, College Station, TX, USA). Results Among 740 women prospectively evaluated over years after baseline, 141 incident fractures were recorded in 115 women. Among them, 85 low-trauma fractures occurred in 68 women (9%), including 31 women with clinical MOF (n ¼ 36) and 13 women with multiple low-trauma fractures (Supplemental Table S1). Two-thirds of these women (45/68) had no osteoporosis on DXA at baseline. Women with incident lowtrauma fractures had a higher prevalence of prior low-trauma fractures and lower abmd, TBS, and FRAX at baseline, and received more anti-osteoporotic drugs during the follow-up (Table 1). Vitamin D status and nutrition were optimal in both groups. Women with incident fracture had lower peripheral vbmd and estimated strength and significant impairment of most of the Tb and Ct microstructure parameters (Supplemental Table S2). Fracture risk significantly increased 37% to 76% by 1 SD difference in radius Tt, Ct, or Tb vbmd and microstructure parameters (except Ct.Po) and 63% to 72% by 1 SD difference for all radius strength parameters (Supplemental TableS3).Associationsofhighermagnitudewerefoundfor MOF(riskincreaseof62%to135%by1SDdifferencein radius parameters significantly associated with MOF) (Supplemental Table S4). Similar trends were observed at the tibia. The highest HR were observed at the radius with Tt.BMD, Tb. BMD, Ct area, and porosity of the inner cortex transitional zone for microstructural parameters, and with failure load for strength parameters (HR ¼ 1.67 to 1.76 for low-trauma fracture, 1.95 to 2.35 for MOF, p for all) (Table 2). These parameters remained significantly associated with fracture and MOF after adjustment for FN abmd, FRAX-BMD, or FRAX-BMD þ TBS (Table 2 and Supplemental Tables S3 and S4). In the stepwise Cox regressions, radius Tt.BMD entered the model as the unique and strongest HR-pQCT-derived parameter for both low-trauma clinical fracture and MOF prediction, even when abmd or FRAX-BMD were included in thevariablelist(table3).inmodelsincludingfrax-bmdþ TBS, radius failure load (for low-trauma clinical fractures) and heterogeneity of the trabecular structure (for MOF) were the best predictors of fracture, independently of FRAX-BMD þ TBS. Kaplan-Meier curves for incident low-trauma fractures are shown in Fig. 1 according to the median of femoral neck abmd, radius ultra-distal abmd, radius Tt.BMD, and tibia Tt.BMD. In a multivariate model combining Tb.BMD and Ct.Ar, both parameters remained independent predictors of low-trauma fracture and MOF (Supplemental Table S5). The ability of bone microstructure measurements to predict fractures was markedly attenuated when ultra-distal radius abmd measured by DXA (same bone site as HR-pQCT at the radius) was added to the models (Table 2 and Journal of Bone and Mineral Research RADIUS MICROSTRUCTURE AND AREAL BMD EVALUATION IMPROVE FRACTURE PREDICTION 331

5 Table 3. Independent Predictors of Fractures Among Radius and Tibia vbmd, Microstructure, or Strength Variables Selected From Forward Stepwise Multivariate Cox Regression Analyses Low-trauma clinical fracture prediction MOF prediction Variables selected in final model HR (95% CI) p Value Variables selected in final model HR (95% CI) p Value Variables included in forward stepwise multivariate Cox regression HR-pQCT variables a Radius Tt.BMD 1 SD 1.88 (1.44, 2.46) <0.001 Radius Tt.BMD 1 SD 2.29 (1.52, 3.46) <0.001 Spine abmd 1 SD 1.46 (1.09, 1.96) Radius Tt.BMD 1 SD 2.29 (1.52, 3.46) <0.001 HR-pQCT variables a þ FN abmd þ spine abmd þ UD radius abmd Radius Tt.BMD 1 SD 1.59 (1.20, 2.12) FRAX-BMD % 1 SD 1.38 (1.03, 1.85) Radius Tt.BMD 1 SD 2.29 (1.52, 3.46) <0.001 HR-pQCT variables a þ FRAX-BMD Radius Tt.BMD 1 SD 1.71 (1.29, 2.26) <0.001 % 1 SD 2.04 (1.34, 3.11) SD 1.63 (1.22, 2.17) Radius Tb.Sp. SD HR-pQCT variables a þ FRAX-BMDþTBS Radius failure load % 1 SD 1.77 (1.10, 2.83) % 1 SD 1.57 (1.15, 2.14) FRAX-BMD þtbs FRAX-BMD þtbs HR ¼ hazard ratio; CI ¼ confidence interval; SD ¼ standard deviation; Tt ¼ total bone; Tb ¼ trabecular; BMD ¼ bone mineral density; FN ¼ femoral neck; FRAX ¼ fracture risk assessment tool; TBS ¼ trabecular bone score; UD ¼ ultra-distal; abmd ¼ areal BMD; MOF ¼ major osteoporotic fracture. a At distal radius and tibia, with a p value < 0.2 in the univariate Cox regression analyses. Data are hazard ratios associated with 1 SD impairment of each parameter, obtained from Cox s proportional hazard models. Supplemental Tables S3 and S4). This is partly because of higher correlations between ultra-distal radius abmd and peripheral vbmd or microstructure compared with the correlations between central abmd (femoral neck) and peripheral vbmd or microstructure. For example, ultra-distal radius abmd was highly correlated with radius Tt.BMD (r ¼ 0.75, p < 0.001) and tibia Tt.BMD (r ¼ 0.63, p < 0.001), whereas femoral neck abmd was correlated, but to a lesser extent, with radius Tt.BMD (r ¼ 0.37, p < 0.001) and tibia Tt. BMD (r ¼ 0.43, p < 0.001). However, hazard ratios for radius Tt.BMD and porosity of the inner cortex transitional zone remained at the limit of the significance after adjusting for ultra-distal radius abmd in the Cox models, suggesting that these two parameters tended to be independent predictors of incident fracture risk (adjusted HR [95% CI] for low-trauma clinical fracture 1.47 [0.95, 2.25], p ¼ and 1.46 [0.96, 2.22], p ¼ 0.073, respectively; for MOF 1.87 [0.96, 3.66], p ¼ and 1.93 [0.99, 3.77], p ¼ 0.054, respectively). In a sensitivity analysis adjusting for the use of anti-osteoporotic drugs MHT during the follow-up, the results were similar to the above findings (data not shown). In subgroup analyses, we tested the associations of Tb.BMD, Ct.Ar, FL, and ultra-distal radius abmd according to osteoporotic status (at least one T-score 2.5SD at the lumbar spine, total hip or FN) and FRAX-BMD intervention threshold for age (10- year fracture probability 20%). In women without osteoporosis, radius Ct.Ar and FL were still associated with low-trauma fracture (HR [95% CI] 1.62 [1.14, 2.30], p ¼ and 1.59 [1.12, 2.26], p ¼ 0.009, respectively). In women below the FRAX intervention threshold, the four parameters were still associated with low-trauma fracture (p ¼ 0.011, 0.004, 0.002, and 0.007, respectively) (Fig. 2A). We also tested the magnitude of the associations between bone traits and the first incident lowtrauma fracture in women with multiple fractures versus with a unique fracture during the follow-up, and according to the delay of fracture onset (Fig. 2B). The associations were higher in women with multiple fractures compared with those with a unique fracture, especially for FL (HR [95% CI] 2.68 [1.51, 4.74] versus 1.57 [1.17, 2.09], p ¼ 0.023) and cortical area (2.41 [1.29, 4.49] versus 1.55 [1.14, 2.10], p ¼ 0.068). The delay of fracture onset also influenced the levels of the associations. The HRs were of higher magnitude for imminent fractures (those that occurred below the median of the delay of first fracture occurrence after baseline, ie, 3.6 years) than for delayed fractures (those that occurred after the median). The best predictors of imminent low-trauma fracture were radius Ct.Ar and FL (HR [95% CI] 2.08 [1.41, 3.08] and 2.17 [1.51, 3.12], p < for both, respectively). The performances for fracture prediction, adjusted for age, were tested for the most significant parameters alone and in combination with FN abmd (Table 4). The single parameters with the highest C indices were FL for low-trauma fracture and Ct. Area for MOF (C indices and 0.698, p ¼ and compared with FN abmd, respectively). All parameters had higher C indices than those obtained for FN abmd. Adding Tb.BMD to Ct. Ar resulted in higher C indices for low-trauma fracture than Tb. BMD or Ct.Ar alone (C indices versus and 0.648) and Tt. BMD (0.643) or ultra-distal abmd (0.649), but quite similar to those obtained with FL (0.664). These data indicate that the best prediction of fracture is obtained with the combination of a Tb with a Ct parameter, which is well reflected by FL and to a lower extent captured by Tt.BMD or by ultra-distal abmd. Adding FL to FN abmd improved fracture prediction compared with FN abmd alone (p ¼ 0.048) but not compared with FL alone. Similar results were obtained in additional ROC curve analyses obtained from logistic regressions adjusted for age and the duration of follow-up, 332 BIVER ET AL. Journal of Bone and Mineral Research

6 Fig. 1. Kaplan-Meier survival curves of incident low-trauma clinical fractures according to the median of femoral neck abmd (A), radius ultra-distal abmd (B), tibia total vbmd (C), and radius total vbmd (D). The p value of the log-rank test is indicated. BMD ¼ bone mineral density; abmd ¼ areal BMD; vbmd ¼ volumetric BMD. illustrated in Fig. 3A, B. The discrimination of women with and without incident MOF was improved when FL was added to age þ FN abmd (AUC versus 0.695, p ¼ 0.022) or FRAX-BMD (AUC versus 0.714, p ¼ 0.015). Very similar performances were obtained when ultra-distal radius abmd was added to FN abmd or FRAX-BMD instead of FL (AUC versus 0.760, p ¼ 0.747, and versus 0.759, p ¼ 0.509, respectively). For any given value of FN abmd or FRAX-BMD MOF probability, FL was lower in women who will sustain a MOF (Fig. 3C, D). Discussion In this homogenous cohort of 740 postmenopausal women followed over 5 years, we found that peripheral Tb and Ct vbmd and microstructure predict incident low-trauma fractures independently of each other and of the tools currently used in clinical practice to assess fracture risk and guide interventions with anti-osteoporotic drugs (mainly FN abmd, FRAX, and TBS). As such, the best predictive value of fracture was found for the combination of a Tb with a Ct parameter, which was well reflected in bone strength parameters (FL) and to a lower extent in ultra-distal radius abmd. The associations with fractures were of lower magnitude at the tibia than at the radius, possibly because tibia has broader cortical areas, perhaps less negatively remodeled than radius because of its weight-bearing role. The ability of FN abmd to predict fracture was very similar to those observed in the Study of Osteoporotic Fractures (6252 women 65 years or older) in which the AUC for age þ FN abmd were 0.63 for clinical fracture and 0.69 for MOF (0.71 and 0.70 for low-trauma clinical fracture and MOF, respectively, in our study). (32) To our knowledge, this is the largest prospective study using a multimodal investigation of the bone phenotype, demonstrating the independent contribution of Ct and Tb bone traits to fracture risk. Prior studies reported some abmd-independent Journal of Bone and Mineral Research RADIUS MICROSTRUCTURE AND AREAL BMD EVALUATION IMPROVE FRACTURE PREDICTION 333

7 Fig. 2. (A) Association between radius Tb.BMD, Ct.Ar, failure load, and ultra-distal abmd and the risk of incident low-trauma clinical fractures by subgroups of women according to osteoporotic status on central DXA (at least one T-score 2.5 SD at the lumbar spine, total hip, or femoral neck) and FRAX-BMD intervention threshold for age (10-year fracture probability >20%). The p value of the interaction between groups is indicated. (B) Association between radius Tb.BMD, Ct.Ar, failure load, and ultra-distal abmd and various fracture outcomes: first low-trauma clinical fractures in women with multiple incident fractures and low-trauma clinical fracture in women with a unique incident fracture; imminent and delayed low-trauma clinical fractures (the cut-off was 3.6 years, corresponding to the median of the delay for the occurrence of the first incident fracture after baseline). The p value of the difference between the outcomes is indicated. Tb ¼ trabecular; BMD ¼ bone mineral density; Ct ¼ cortical; UD ¼ ultra-distal; abmd ¼ areal BM; FRAX ¼ fracture risk assessment tool. Table 4. Harrell s C Indices Showing Ability of Various Cox Regression Models Compared With FN abmd to Predict Incident Low-Trauma Clinical Fractures and Major Osteoporotic Fractures Low-trauma clinical fracture MOF Variables C indices (95% CI) p Value a C indices (95% CI) p Value a Age þ FN abmd (0.495, 0.663) Ref (0.427, 0.689) Ref Age þ Rad Tb.BMD (0.547, 0.703) (0.524, 0.738) Age þ Rad Ct.Ar (0.575, 0.720) (0.603, 0.793) Age þ Rad Inn Ct.Po (0.567, 0.719) (0.590, 0.773) Age þ Rad Tt.BMD (0.567, 0.720) (0.584, 0.789) Age þ Rad Tb.BMD þ Rad Ct.Ar (0.585, 0.733) (0.602, 0.787) Age þ Rad Failure load (0.588, 0.741) (0.594, 0.794) Age þ Rad UD abmd (0.575, 0.723) (0.565, 0.781) Age þ FN abmd þ Rad Tb.BMD (0.550, 0.706) (0.524, 0.738) Age þ FN abmd þ Rad Ct.Ar (0.575, 0.721) (0.603, 0.793) Age þ FN abmd þ Rad Inn Ct.Po (0.569, 0.722) (0.588, 0.773) Age þ FN abmd þ Rad Tt.BMD (0.566, 0.721) (0.583, 0.787) Age þ FN abmd þ Rad Tb.BMD þ Rad Ct.Ar (0.588, 0.734) (0.602, 0.788) Age þ FN abmd þ Rad Failure load (0.588, 0.741) (0.593, 0.799) Age þ FN abmd þ Rad UD abmd (0.568, 0.718) (0.566, 0.781) CI ¼ confidence interval; FN ¼ femoral neck; BMD ¼ bone mineral density; abmd ¼ areal BMD; Tt ¼ total bone; Ct ¼ cortical; Tb ¼ trabecular; Ar ¼ area; Inn Ct.Po ¼ porosity of the inner transitional zone; UD ¼ ultra-distal; MOF ¼ major osteoporotic fracture. a Compared with FN abmd alone. 334 BIVER ET AL. Journal of Bone and Mineral Research

8 Fig. 3. (A, B) Receiver operating characteristic curves of age þ FN abmd radius failure load or ultra-distal abmd (A) or of FRAX-BMD radius failure load or ultra-distal abmd (B), adjusted for the duration of follow-up for incident MOF discrimination. The AUC for age þ radius failure load was 0.756, p ¼ versus age þ FN abmd, p ¼ versus FRAX-BMD alone. The AUC for age þ radius ultra-distal abmd was 0.749, p ¼ versus age þ FN abmd, p ¼ versus FRAX-BMD alone. (C, D) Prediction of radius failure load (FL) along with 95% confidence interval (short dash lines) from linear regressions of radius failure load on FN T-score (C) or 10-year MOF probability with FRAX-BMD (D), in women with incident MOF and women without fracture. For a given value of FN abmd or of 10-year MOF probability, radius FL at baseline was always lower in women with incident MOF compared with those without fracture. The differences were particularly found in women who did not reach the intervention thresholds usually used in clinical practice (FN abmd T-score 2.5; 10-year fracture probability 20% at age 65 years) and in whom the majority of fracture occurred over the follow-up (81% of incident MOF in women with FN abmd T-score > 2.5 at baseline and 77% of incident MOF in women with FRAX 10-year fracture probability <20% at baseline). FN ¼ femoral neck; BMD ¼ bone mineral density; abmd ¼ areal BMD; rad ¼ radius; UD ¼ ultra-distal; FRAX ¼ fracture risk assessment tool; AUC ¼ area under the curve; MOF ¼ major osteoporotic fracture. associations between bone microstructure (at various compartments and sites) and incident fractures. They highlighted either Ct.Ar at the radius in women and at the tibia in men, or Tb.BMD and microstructure in women. (33 35) In addition, the combination of FN abmd with a radius strength parameter derived from pqct was previously reported to give better performance than FN abmd alone for the prediction of nonvertebral fractures in men. (36) We found that the magnitude of the associations between peripheral bone microstructure and incident fracture was higher in women with multiple fractures and imminent fractures. The concept of imminent fracture risk has recently emerged, with two main predictors, an index recent low-trauma fracture or a recent history of fall. (37,38) Our data suggest that alteration of bone microstructure and strength may further identify subjects at imminent and/or multiple fracture risk. Bone microstructure also predicted incident fractures in subgroups of women that would not have been qualified at high risk of fracture with the classical tools (non-osteoporotic women on DXA; women with FRAX score below the intervention threshold for age). As previously reported, the majority of low-trauma fractures occurred in these subgroups in our cohort. (8) The performances of cortical parameters were particularly good for MOF prediction. At the age of 65 years, the majority of bone loss occurs at the Ct compartment. (39) The characteristics of Ct porosity differed according to the method of assessment (range at the radius 1% to 13% for Scanco Ct.Po, 38% to 77% for Strax Tt Ct.Po). Ct porosity assessed with the density-based method (Tt Ct.Po) was highly associated with incident fractures, but the magnitude of the associations was very similar to those observed with Ct.BMD (HR 1.54 and 1.53, respectively, at the radius; 1.59 and 1.57, respectively, at the tibia). HRs for radius and tibia Tt Ct.Po were no more significant after adjustment for Ct.BMD (Supplemental Table S5). These observations suggest Journal of Bone and Mineral Research RADIUS MICROSTRUCTURE AND AREAL BMD EVALUATION IMPROVE FRACTURE PREDICTION 335

9 that Ct porosity of the total cortex assessed with the densitybased method mainly captures Ct.BMD. Among all the recorded bone parameters, the highest HRs were obtained for the Inn Ct. Po of the radius, which might reflect both the Ct and the Tb compartments, as illustrated in multivariate models including Inn Ct.Po in Supplemental Table S5. Its performance for fracture prediction was, however, lower than with the combination of a Tb with a Ct parameter or with strength estimates (FL). In a clinical perspective, although combining abmd or FRAX with Ct and/or Tb bone traits resulted in better fracture prediction than abmd or FRAX alone, the combination did not provide better performances than the respective microstructure or strength parameters alone (Table 4 and Fig. 3), suggesting that these parameters capture all the abmd or FRAX variability associated with fracture risk. Conversely, abmd assessed by DXA at the ultra-distal radius showed very good performance for fracture prediction, approaching those obtained with the combination Tb.BMD þ Ct.Ar or FL (Fig. 3). The UD radius site represents not only the same anatomic bone site as where bone microstructure and strength were assessed by HR-pQCT but also provided better fracture discrimination than distal 1/3 radius or total radius (Supplemental Table S2). Its good performance to predict fracture might be explained by the fact that it combines both trabecular and cortical bone, contrary to proximal forearm, which is primarily cortical bone. These data suggest that measuring UD radius abmd in addition to FN abmd could be useful to refine fracture prediction in this age range of postmenopausal women, especially in patients in whom bone microstructure cannot be specifically assessed. HR-pQCT devices indeed are not widely available and are currently dedicated to research purpose. In this context, there is an ongoing interest to develop surrogate tools reflecting bone microstructure and easily applicable to clinical practice. Ct area, whose performance was very good for MOF, might be more easily and largely assessed by other devices of lower resolution (pqct, ultrasounds). Although TBS has been proposed as a surrogate of Tb microstructure, the association of Tb and Ct bone traits with fracture remained significant after adjustment for FRAX-BMD þ TBS. These data indicate that TBS does not substitute for the evaluation of Tb microstructure to fracture risk, contrary to ultradistal radius abmd. We acknowledge a number of limitations to our findings. First, the majority of fractures included in this study were nonvertebral fractures (only 3 vertebral fractures), with very few hip fractures, and a short duration of follow-up. Another limitation is the multiple associations because of the high number of variables obtained from the HR-pQCT analysis, most of them being intercorrelated. In addition, the number of incident fractures was relatively low because of the quite short length of follow-up (5 years) and rather small sample size of the study, so that it was not possible to examine the association by fracture type. We also cannot exclude potential selection biases in this healthy community-dwelling population of postmenopausal women. The significant findings of this exploratory study need, therefore, to be replicated in other cohorts and confirmed by studies in more heterogeneous populations, including men, before generalization. In conclusion, the prediction of low-trauma fractures in postmenopausal women can be improved beyond DXA and FRAX with the assessment of peripheral Ct and Tb vbmd and microstructure. Ct.Ar and Tb.BMD predict low-trauma fractures independently of each other, and the best prediction is obtained with the combination of a Tb with a Ct parameter at the radius. The associations are higher for the risk of multiple and imminent low-trauma fractures. Estimated bone strength has similar performance than the combination of Ct.Ar with Tb.BMD, but abmd measured by DXA at the same bone site (ultra-distal radius) captures a substantial proportion of the information provided by this combination or FL. Before tools assessing peripheral bone microstructure and strength are made more largely available for clinical practice, ultra-distal radius abmd might help to improve fracture prediction, in particular in women without osteoporosis on central DXA (spine þ hip) and in which the majority of fractures occur. Disclosures BvR reports personal fees from Scanco Medical AG, outside the submitted work. All other authors state that they have no conflicts of interest. Acknowledgments We thank Ms F Merminod, M Hars, C Genet, M-A Schaad, and A Sigaud, and Dr J Pepe and A de Sire for the management of GERICO cohort participants, Mr G Conicella for HR-pQCT measurements, and Dr R Zebaze and Prof E Seeman for cortical porosity quantification with StrAx1.0 software. We thank the Geneva University Hospitals and Faculty of Medicine Clinical Research Centre, the HUG Private Foundation, and the BNP- Paribas Foundation for their support. Authors roles: Study design: EB, RR, and SF. Study conduct: CDI and EB. Data collection: CDI and EB. HR-pQCT data and analysis: EB and BvR. Data analysis: EB and TC. Data interpretation: EB, TC, RR, and SF. Drafting manuscript: EB and SF. Revising manuscript content: CDI, TC, BvR, and RR. Approving final version of manuscript: EB, CDI, TC, BvR, RR, and SF. EB takes responsibility for the integrity of the data analysis. References 1. Watts NB, Manson JE. Osteoporosis and fracture risk evaluation and management: shared decision making in clinical practice. JAMA. 2017;317(3): Bliuc D, Alarkawi D, Nguyen TV, Eisman JA, Center JR. Risk of subsequent fractures and mortality in elderly women and men with fragility fractures with and without osteoporotic bone density: the Dubbo Osteoporosis Epidemiology Study. J Bone Miner Res. 2015;30(4): Bliuc D, Nguyen ND, Milch VE, Nguyen TV, Eisman JA, Center JR. Mortality risk associated with low-trauma osteoporotic fracture and subsequent fracture in men and women. JAMA. 2009;301(5): Cummings SR, Black DM, Rubin SM. Lifetime risks of hip, Colles, or vertebral fracture and coronary heart disease among white postmenopausal women. Arch Intern Med. 1989;149(11): Cummings SR, Melton LJ. Epidemiology and outcomes of osteoporotic fractures. Lancet. 2002;359(9319): Kanis JA, McCloskey EV, Johansson H, Oden A, Melton LJ 3rd, Khaltaev N. A reference standard for the description of osteoporosis. Bone. 2008;42(3): Siris ES, Miller PD, Barrett-Connor E, et al. Identification and fracture outcomes of undiagnosed low bone mineral density in postmenopausal women: results from the National Osteoporosis Risk Assessment. JAMA. 2001;286(22): BIVER ET AL. Journal of Bone and Mineral Research

10 8. Schuit SC, van der Klift M, Weel AE, et al. Fracture incidence and association with bone mineral density in elderly men and women: the Rotterdam Study. Bone. 2004;34(1): Sornay-Rendu E, Munoz F, Garnero P, Duboeuf F, Delmas PD. Identification of osteopenic women at high risk of fracture: the OFELY study. J Bone Miner Res. 2005;20(10): Hans DB, Shepherd JA, Schwartz EN, et al. Peripheral dual-energy X-ray absorptiometry in the management of osteoporosis: the 2007 ISCD Official Positions. J Clin Densitom. 2008;11(1): Rubin KH, Friis-Holmberg T, Hermann AP, Abrahamsen B, Brixen K. Risk assessment tools to identify women with increased risk of osteoporotic fracture: complexity or simplicity? A systematic review. J Bone Miner Res. 2013;28(8): Kanis JA, Johnell O, Oden A, Johansson H, McCloskey E. FRAX and the assessment of fracture probability in men and women from the UK. Osteoporos Int. 2008;19(4): Siris ES, Adler R, Bilezikian J, et al. The clinical diagnosis of osteoporosis: a position statement from the National Bone Health Alliance Working Group. Osteoporos Int. 2014;25(5): Silva BC, Leslie WD, Resch H, et al. Trabecular bone score: a noninvasive analytical method based upon the DXA image. J Bone Miner Res. 2014;29(3): McCloskey EV, Oden A, Harvey NC, et al. A meta-analysis of trabecular bone score in fracture risk prediction and its relationship to FRAX. J Bone Miner Res. 2016;31(5): Jiang X, Gruner M, Tremollieres F, et al. Diagnostic accuracy of FRAX in predicting the 10-year risk of osteoporotic fractures using the USA treatment thresholds: a systematic review and meta-analysis. Bone. 2017;99: Geusens P, Chapurlat R, Schett G, et al. High-resolution in vivo imaging of bone and joints: a window to microarchitecture. Nat Rev Rheumatol. 2014;10(5): Boutroy S, Khosla S, Sornay-Rendu E, et al. Microarchitecture and peripheral BMD are impaired in postmenopausal white women with fracture independently of total hip T-score: an international multicenter study. J Bone Miner Res. 2016;31(6): Vilayphiou N, Boutroy S, Sornay-Rendu E, et al. Finite element analysis performed on radius and tibia HR-pQCT images and fragility fractures at all sites in postmenopausal women. Bone. 2010;46(4): Szulc P, Boutroy S, Vilayphiou N, Chaitou A, Delmas PD, Chapurlat R. Cross-sectional analysis of the association between fragility fractures and bone microarchitecture in older men: the STRAMBO study. J Bone Miner Res. 2011;26(6): Edwards MH, Robinson DE, Ward KA, et al. Cluster analysis of bone microarchitecture from high resolution peripheral quantitative computed tomography demonstrates two separate phenotypes associated with high fracture risk in men and women. Bone. 2016;88: Hars M, Biver E, Chevalley T, et al. Low lean mass predicts incident fractures independently from FRAX: a prospective cohort study of recent retirees. J Bone Miner Res. 2016;31(11): Durosier-Izart C, Biver E, Merminod F, et al. Peripheral skeleton bone strength is positively correlated with total and dairy protein intakes in healthy postmenopausal women. Am J Clin Nutr. 2017;105(2): Durosier C, van Lierop A, Ferrari S, Chevalley T, Papapoulos S, Rizzoli R. Association of circulating sclerostin with bone mineral mass, microstructure, and turnover biochemical markers in healthy elderly men and women. J Clin Endocrinol Metab. 2013;98(9): Genant HK, Wu CY, van Kuijk C, Nevitt MC. Vertebral fracture assessment using a semiquantitative technique. J Bone Miner Res. 1993;8(9): Biver E, Durosier C, Chevalley T, Herrmann FR, Ferrari S, Rizzoli R. Prior ankle fractures in postmenopausal women are associated with low areal bone mineral density and bone microstructure alterations. Osteoporos Int. 2015;26(8): Pepe J, Biver E, Bonnet N, et al. Within- and across-sex inheritance of bone microarchitecture. J Clin Endocrinol Metab. 2017;102(1): Chevalley T, Bonjour JP, van Rietbergen B, Ferrari S, Rizzoli R. Fracture history of healthy premenopausal women is associated with a reduction of cortical microstructural components at the distal radius. Bone. 2013;55(2): Zebaze R, Ghasem-Zadeh A, Mbala A, Seeman E. A new method of segmentation of compact-appearing, transitional and trabecular compartments and quantification of cortical porosity from high resolution peripheral quantitative computed tomographic images. Bone. 2013;54(1): Pistoia W, van Rietbergen B, Lochmuller EM, Lill CA, Eckstein F, Ruegsegger P. Estimation of distal radius failure load with microfinite element analysis models based on three-dimensional peripheral quantitative computed tomography images. Bone. 2002;30(6): Pencina MJ, D Agostino RB. Overall C as a measure of discrimination in survival analysis: model specific population value and confidence interval estimation. Stat Med. 2004;23(13): Ensrud KE, Lui LY, Taylor BC, et al. A comparison of prediction models for fractures in older women: is more better? Arch Intern Med. 2009;169(22): Dennison EM, Jameson KA, Edwards MH, Denison HJ, Aihie Sayer A, Cooper C. Peripheral quantitative computed tomography measures are associated with adult fracture risk: the Hertfordshire Cohort Study. Bone. 2014;64: Ohlsson C, Sundh D, Wallerek A, et al. Cortical bone area predicts incident fractures independently of areal bone mineral density in older men. J Clin Endocrinol Metab. 2017;102(2): Sornay-Rendu E, Boutroy S, Duboeuf F, Chapurlat RD. Bone microarchitecture assessed by HR-pQCT as predictor of fracture risk in postmenopausal women: the OFELY study. J Bone Miner Res. 2017;32(6): Sheu Y, Zmuda JM, Boudreau RM, et al. Bone strength measured by peripheral quantitative computed tomography and the risk of nonvertebral fractures: the Osteoporotic Fractures in Men (MrOS) study. J Bone Miner Res. 2011;26(1): Bonafede M, Shi N, Barron R, Li X, Crittenden DB, Chandler D. Predicting imminent risk for fracture in patients aged 50 or older with osteoporosis using US claims data. Arch Osteoporos. 2016; 11(1): Roux C, Briot K. Imminent fracture risk. Osteoporos Int. 2017;28(6): Zebaze RM, Ghasem-Zadeh A, Bohte A, et al. Intracortical remodelling and porosity in the distal radius and post-mortem femurs of women: a cross-sectional study. Lancet. 2010;375(9727): Journal of Bone and Mineral Research RADIUS MICROSTRUCTURE AND AREAL BMD EVALUATION IMPROVE FRACTURE PREDICTION 337