Bone 110 (2018) 386–391
Contents lists available at ScienceDirect
Bone journal homepage: www.elsevier.com/locate/bone
Full Length Article
Comparison of femoral strength and fracture risk index derived from DXA-based finite element analysis for stratifying hip fracture risk: A cross-sectional study Shuman Yang a,b,c, Yunhua Luo d, Lang Yang e,f, Enrico Dall'Ara e,f, Richard Eastell e,f, Andrew L. Goertzen g, Eugene V. McCloskey h, William D. Leslie c,⁎, Lisa M. Lix b a
Department of Epidemiology and Biostatistics, School of Public Health, Jilin University, Changchun, Jilin, China Department of Community Health Sciences, University of Manitoba, Manitoba, Canada c Department of Internal Medicine, University of Manitoba, Winnipeg, Manitoba, Canada d Department of Mechanical Engineering, University of Manitoba, Winnipeg, Manitoba, Canada e Academic Unit of Bone Metabolism, Mellanby Centre for Bone Research University of Sheffield, Sheffield, UK f INSIGNEO Institute for in silico Medicine, University of Sheffield, Sheffield, UK g Department of Radiology, University of Manitoba, Manitoba, Canada h Metabolic Bone Centre, Sorby Wing, Northern General Hospital, Sheffield, UK b
a r t i c l e
i n f o
Article history: Received 11 December 2017 Revised 5 March 2018 Accepted 7 March 2018 Available online 8 March 2018 Keywords: Femoral strength Fracture risk Hip fracture Finite element analysis Osteoporosis Dual energy x-ray absorptiometry
a b s t r a c t Background: Dual-energy X-ray absorptiometry (DXA)-based finite element analysis (FEA) has been studied for assessment of hip fracture risk. Femoral strength (FS) is the maximum force that the femur can sustain before its weakest region reaches the yielding limit. Fracture risk index (FRI), which also considers subject-specific impact force, is defined as the ratio of von Mises stress induced by a sideways fall to the bone yield stress over the proximal femur. We compared risk stratification for prior hip fracture using FS and FRI derived from DXA-based FEA. Methods: The study cohort included women aged ≥65 years undergoing baseline hip DXA, with femoral neck Tscores b−1 and no osteoporosis treatment; 324 cases had prior hip fracture and 655 controls had no prior fracture. Using anonymized DXA hip scans, we measured FS and FRI. Separate multivariable logistic regression models were used to estimate odds ratios (ORs), c-statistics and their 95% confidence intervals (95% CIs) for the association of hip fracture with FS and FRI. Results: Increased hip fracture risk was associated with lower FS (OR per SD 1.36, 95% CI: 1.15, 1.62) and higher FRI (OR per SD 1.99, 95% CI: 1.63, 2.43) after adjusting for Fracture Risk Assessment Tool (FRAX) hip fracture probability computed with bone mineral density (BMD). The c-statistic for the model containing FS (0.69; 95% CI: 0.65, 0.72) was lower than the c-statistic for the model with FRI (0.77; 95% CI: 0.74, 0.80) or femoral neck BMD (0.74; 95% CI: 0.71, 0.77; all P b 0.05). Conclusions: FS and FRI were independently associated with hip fracture, but there were differences in performance characteristics. © 2018 Elsevier Inc. All rights reserved.
1. Introduction The strength of a material is its resistance to an external loading condition and depends on geometry, microstructural properties and material properties [1]. Material failure occurs when the externally applied force exceeds the strength of the material under that loading condition. Finite element analysis (FEA) is a computational method for predicting material strength and failure. Prediction of bone strength and failure by FEA is of interest for assessing fracture risk in vivo because a direct ⁎ Corresponding author at: C5121-409 Tache Ave, Department of Medicine, St. Boniface Hospital, Winnipeg, Manitoba R2H 2A6, Canada. E-mail address:
[email protected] (W.D. Leslie).
https://doi.org/10.1016/j.bone.2018.03.005 8756-3282/© 2018 Elsevier Inc. All rights reserved.
measurement of bone strength by biomechanical testing is not possible. Bone strength estimated from FEA has been shown to be useful for assessing fracture risk in human studies [2]. Most existing finite element models for assessing bone strength are constructed from threedimensional images such as computed tomography (CT) [2–4]. However, three-dimensional imaging modalities are not widely used in clinical practice due to their high cost and radiation exposure [3–6]. FEA derived from dual energy x-ray absorptiometry (DXA)-based images is attractive considering the wide availability of DXA in clinical practice [7]. Fall-induced impact force is a determinant of hip fracture independent of bone strength [8,9]. Femoral strength (FS) is the maximum force that the femur can sustain before its weakest region reaches the
S. Yang et al. / Bone 110 (2018) 386–391 Table 1 Finite element models for measuring femoral strength (FS) and fracture risk index (FRI).
Finite element mesh Region of interest (ROI) Loading condition
FS
FRI
Pixel-based mesh generated from femur DXA (see [12] for details) ROI with a specified area located in the most critical region on the femur (see [12] for details) A loading profile simulating sideways fall was applied and scaled to femur strength using the ratio of applied stress to yield stress (see [12] for details)
Geometry-based mesh generated from femur contour (see [18] for details) A ROI for the proximal femur bone (see [18] for details)
Constraint Greater trochanteric and conditions subtrochanteric regions (see [12] for details)
Subject-specific impact force in sideways fall was estimated from body weight and body height, and applied to the greater trochanteric region (see [18] for details) Femoral head and subtrochanteric regions (see [18] for details)
yielding limit. The fracture risk index (FRI) is defined as the ratio of von Mises stress induced by a sideways fall to the bone yield stress over the proximal femur. Von Mises stress is a measure of stress level in a material body under applied forces, and is widely used in engineering to assess material damage. A fundamental difference between FS and FRI is that FRI considers both bone strength and fall-induced impact force. The latter is subject-specific and determined by anthropometric and kinematic parameters such as the subject's body weight, height, body mass index (BMI) and vertical velocity in fall [10]. We compared the performance of FS and FRI derived from DXA-based FEA [11,12]. 2. Materials and methods 2.1. Study cohort We extracted our study cohort from the population-based Manitoba Bone Mineral Density Database (MBMDD), Canada, a large registrybased database of clinical DXA scans with anthropometric data (i.e., height and weight) and linkage to hospitalizations, physician visits and drug prescription records. The completeness and accuracy of the MBMDD have been validated and shown to be very high [13,14]. The Health Research Ethics Board of the University of Manitoba provided ethical approval for the research and permission for data access was provided by the Manitoba Health Information Privacy Committee. We identified all women who had baseline hip DXA (Prodigy, GE Healthcare) during 2000–2013 and were ≥65 years old, had femoral neck T-scores b−1, and no prior osteoporosis treatment. We identified women with a hip fracture before the DXA scan and then randomly selected non-fracture controls (two controls for each case). Prior hip fractures were identified from hospitalization records dating back to 1984 using a validated fracture definition [15,16]. Hip fractures associated with high-trauma codes were excluded. Non-fracture controls had no diagnosed hip, forearm, clinical spine or humerus fractures before the DXA scan. 2.2. DXA scanning and bone mineral density (BMD) data DXA scans of the hip were performed and analyzed in accordance with manufacturer recommendations using one of the Program's cross-calibrated fan-beam DXA instruments (Prodigy, GE Healthcare). Scanners and personnel are subject to province-wide quality assurance programs under the direction of a medical physicist, and include daily evaluation of densitometer stability using anthropometric spine phantoms. Long-term scanner stability was confirmed for all instruments (all coefficients of variation b 0.5%). Hip T-scores were calculated using white female reference data from the National Health and Nutrition Examination Survey (NHANES) III database [7]. A femoral neck Tscore ≤−2.5 was used to define osteoporosis [17]. In vivo inter-
387
scanner differences in femoral neck BMD and total hip BMD were negligible (b0.1 T-score units). 2.3. FS and FRI measurement We measured FS and FRI from anonymized hip DXA scans using inhouse Matlab-based programs (The MathWorks, Inc., Natick, Massachusetts). These measurements were performed by individuals blinded to fracture status and other risk factors. Details of these procedures are summarized in Table 1. The FS procedure is described in detail by Dall'Ara et al. [12]. In brief, each pixel in the DXA scan was converted into 2D 4-node plane stress elements. FS was estimated in a sideways fall configuration with the femoral shaft in 30 degree adduction and the loading force was distributed to the medial side of femoral head. Loading and constraint conditions were applied at the greater trochanteric and subtrochanteric regions. Mechanical properties (i.e., modulus of elasticity, compressive yield stress and tensile yield stress) of each element were computed as functions of areal BMD using previously published empirical relationships [19,20]. The FS of each femur was defined as the load at which a region of interest (ROI) with area equal to at least 9 mm2 and located in the anatomical region between the subcapital line and transverse line passing through the distal end of lesser trochanter reached the yield strain (7300 microstrain in tension or 10,400 microstrain in compression) [12]. The FRI procedure was developed by Luo et al. [21,22] and applied by Yang et al. [11]. In the calculation of FRI, it is assumed: 1) A sideways fall is the most critical situation leading to hip fracture; 2) The projected femur can be represented by the engineering plane stress model; and 3) Bone failure is determined by the ratio between von Mises stress and bone yield stress [18]. The finite element model is constructed from the subject's hip DXA. The DXA scan is used to generate a proximal femur bone map and the femur contour. The femur contour is then used to create a two-dimensional finite element mesh, assign material properties (Young's modulus and yield stress), apply loading/constraint conditions and calculate FRI for the proximal femur. The impact force, predicted by a whole-body dynamics simulation of sideways fall [8,10], is applied to the greater trochanter with constraint conditions applied at the femoral head and the distal femur. FRI is calculated for the proximal femur (including femoral head), which differs from previously reported site-specific FRIs for the femoral neck, intertrochanteric and subtrochanteric regions [11].
Table 2 Characteristics of the study cohort, femoral strength (FS) and fracture risk index (FRI) stratified by prior hip fracture status. Variable
Hip fracture Non-fracture cases controls
P
N Age (years) Body mass index (kg/m2) Femoral neck T-score Total hip T-score Parental hip fracture Chronic obstructive pulmonary disease diagnosis Alcohol/substance abuse diagnosis Prolonged glucocorticoid use Rheumatoid arthritis diagnosis Prior non-hip fracture Hip fracture probability (%)
324 78.3 (7.1) 24.7 (3.7) −2.7 (0.6) −2.8 (0.9) 8.6 12.4
655 74.6 (6.6) 25.6 (3.5) −2.2 (0.6) −1.9 (0.7) 5.7 8.4
NA b0.001 b0.001 b0.001 b0.001 0.077 0.049
3.1 3.1 4.9 26.9 9.3 (6.6, 15.4) 2370 (799) 0.29 (0.24, 0.37)
1.1 2.8 2.0 0 3.6 (2.1, 5.9)
0.023 0.765 0.010 NA b0.001
2897 (794) 0.22 (0.19, 0.26)
b0.001 b0.001
FS (Newtons) FRI (unitless)
NA: Not applicable. Hip fracture probability and FRI are shown as median (inter-quartile range). Unless otherwise specified, other variables are presented as mean (SD) for continuous variables and % for categorical variables.
388
S. Yang et al. / Bone 110 (2018) 386–391
2.4. Other covariates Other covariates included age, body mass index (BMI), parental hip fracture history, chronic obstructive pulmonary disease diagnosis (COPD; a proxy for smoking), alcohol/substance abuse diagnosis (a proxy for high alcohol intake), prolonged glucocorticoid use, rheumatoid arthritis diagnosis, and prior non-hip fracture. Disease diagnoses were ascertained from at least one relevant hospital record or at least two relevant physician billing claims within three years prior to the BMD test. Glucocorticoid medication use (over 90 days in the year prior to the BMD test) was determined from a province-wide retail pharmacy database. BMI was calculated from body weight and height, which were measured at the time of the DXA test. Using these covariates as inputs, ten-year hip fracture probability (with femoral neck BMD) was calculated using the Canadian FRAX tool (FRAX Desktop Multi-Patient Entry, version 3.7). For the cases, prior hip fracture was not used in computing hip fracture probability to avoid including this as both a risk factor and outcome.
2.5. Statistical analysis Fig. 1. Scatter plots demonstrating the relationships between femoral strength (FS) and fracture risk index (FRI) stratified by hip fracture cases (red cross) and non-fracture controls (gray circle). Reference lines for FS and FRI are based on the median values.
We compared cohort characteristics, FS and FRI stratified by hip fracture status. Pearson correlation coefficients for FS, FRI and BMD were estimated for the overall sample. We used multivariable logistic regression models to test the associations of FS and FRI with hip fracture. In the models, FS and FRI were treated as continuous variables
Fig. 2. Scatter plots demonstrating the relationships between femoral strength (FS) and fracture risk index (FRI) with femoral neck and total hip T-scores stratified by hip fracture cases (red cross) and non-fracture controls (gray circle). Reference lines for FS and FRI are based on the median values, and for T-scores are for the osteoporotic cutoff of −2.5.
S. Yang et al. / Bone 110 (2018) 386–391
389
Table 3 Odds ratios (ORs), c-statistics and their 95% confidence intervals (95% CIs) for logistic regression models of prior hip fracture associated with femoral strength (FS), fracture risk index (FRI) and bone mineral density (BMD). Variable
OR (95% CI) Adjusted for age and BMI
OR (95% CI) Adjusted for age, BMI and femoral neck BMD
OR (95% CI) Adjusted for age, BMI and total hip BMD
OR (95% CI) Adjusted for hip fracture probability
c-Statistic (95% CI)
FS FRI Femoral neck BMD Total hip BMD
1.86 (1.59, 2.17) 2.61 (2.18, 3.12) 2.28 (1.93, 2.70) 2.78 (2.32, 3.32)
1.29 (1.07, 1.55) 2.09 (1.66, 2.62) NA NA
1.00 (0.82, 1.23) 1.41 (1.01, 1.98) NA NA
1.36 (1.15, 1.62) 1.99 (1.63, 2.43) NA NA
0.69 (0.65, 0.72)a,b,c 0.77 (0.74, 0.80)a 0.74 (0.71, 0.77)b 0.78 (0.75, 0.81)
NA: Not applicable; BMI: Body mass index. ORs were estimated per SD. Values in boldface font are statistically significant at α = 0.05. a Significantly different at α = 0.05 as compared to femoral neck BMD. b Significantly different at α = 0.05 when compared to total hip BMD. c Significantly different at α = 0.05 for FS vs FRI.
and effects expressed as odds ratios (ORs) per SD with 95% confidence intervals (95% CIs). FRI values were converted to the logarithmic scale because of a skewed distribution. Models were adjusted for 1) age and BMI; 2) age, BMI and femoral neck BMD; 3) age, BMI and total hip BMD; and 4) hip fracture probability. We performed subgroup analyses in which we stratified the study cohort by age (b75 and ≥75 years), BMI (b25 and ≥25 kg/m2), femoral neck T-score (osteoporotic and nonosteoporotic) and hip fracture probability (b3% and ≥3%). c-Statistics and their 95% CIs for FS, FRI and BMD were also tested and compared in the logistic regression analysis. P-values for differences in the cstatistic for models containing FS, FRI and BMD were tested using the method reported by Delong et al. [23]. Net reclassification index (NRI) is a statistical method for measuring the improvement in risk assessment improvement when a new risk factor is used with established risk factors [24]. We computed category-free NRIs and 95% CIs to test the improvement in risk classification when FS or FRI, and BMD were used to estimate hip fracture risk as compared to BMD alone [25]. Before performing the NRI analysis, we used logistic regression models to calculating the probabilities for hip fracture associated with 1) FS and femoral neck BMD; 2) FRI and femoral neck BMD; 3) FS and total hip BMD; 4) FRI and total hip BMD; 5) femoral neck BMD alone; and 6) total hip BMD alone, respectively. In the category-free NRI analysis, we used the continuous scale of probabilities calculated from the above models. Then, we estimated the event, non-event and overall NRI for models 1 and 2 vs 5, and 3 and 4 vs 6 using a previously established method [25]. The Brier score measures the accuracy of probabilistic prediction models for assessing risk of categorical outcomes [26]. The range of Brier score is 0–1; lower score indicates better prediction accuracy. In our study, the Brier score was used to assess the model calibration for FS and BMD, and FRI and BMD vs BMD alone. The Brier score provides information about model calibration [27]. The probabilities for hip
fracture used for calculating Brier score were estimated from the logistic regression models. All analyses were performed using SAS (Version 9.3, SAS Institute Inc., Cary, NC). 3. Results We excluded 50 scans due to an invalid bone map contour, invalid bone map contour (i.e., missing femoral head contour) or failure of the FEA to converge that lead to either invalid FS or FRI; this affected 18 hip fracture cases (5.3% of 342 before exclusions) and 32 nonfracture controls (4.7% of 687 before exclusions). The excluded scans had similar BMI (26.3 vs 25.3 kg/m2; P = 0.063), slightly older ages (77.9 vs 75.9 years; P = 0.041) and slightly lower femoral neck Tscores (−2.6 vs −2.4; P = 0.010) when compared with included scans. The final cohort included 324 hip fracture cases and 655 nonfracture controls. Baseline characteristics of the hip fracture cases and non-fracture controls were consistent with the higher risk profile among the former (significantly older age, lower femoral neck T-score and BMI, greater prevalence of COPD diagnosis, alcohol/substance abuse diagnosis, and rheumatoid arthritis diagnosis) (Table 2). Hip fracture cases had significantly higher hip fracture probability, higher FRI, and lower FS than non-fracture controls (Table 2). The Pearson correlation coefficient (r) between FS and FRI was − 0.69 (P b 0.001; Fig. 1). Hip T-scores were positively correlated with FS (r = 0.59 with femoral neck T-score; r = 0.65 with total hip Tscore) and negatively correlated with FRI (r = −0.72 with femoral neck T-score; r = −0.88 with total hip T-score) (all P b 0.001; Fig. 2). FS and FRI had slightly stronger correlations with total hip T-score than with femoral neck T-score; FRI had stronger correlations than FS with both total hip and femoral T-score (all P for correlation difference b 0.05; Fig. 2).
Table 4 Adjusted odds ratios (ORs) and 95% confidence intervals (95% CIs) for logistic regression models of prior hip fracture associated with femoral strength (FS) and fracture risk index (FRI) in subgroup analyses. Variable
Subgroup
No. of hip fracture cases
No. of non-fracture controls
OR (95% CI) Adjusted for hip fracture probability
P-value for interaction
FS
Age b 75 years Age ≥ 75 years Body mass index b 25 kg/m2 Body mass index ≥ 25 kg/m2 Osteoporotic Non-osteoporotic Hip fracture probability b 3% Hip fracture probability ≥ 3% Age b 75 years Age ≥ 75 years Body mass index b 25 kg/m2 Body mass index ≥ 25 kg/m2 Osteoporotic Non-osteoporotic Hip fracture probability b 3% Hip fracture probability ≥ 3%
104 220 104 220 203 121 40 284 104 220 104 220 203 121 40 284
351 304 146 509 170 485 277 378 351 304 146 509 170 485 277 378
1.35 (1.00, 1.82) 1.37 (1.10, 1.70) 1.35 (1.01, 1.81) 1.34 (1.09, 1.66) 1.13 (0.88, 1.47) 1.51 (1.17, 193) 1.33 (0.90, 1.98) 1.37 (1.13, 1.66) 2.44 (1.70, 3.50) 1.85 (1.44, 2.36) 1.47 (1.07, 2.04) 2.30 (1.79, 2.97) 1.52 (1.17, 1.96) 3.08 (2.17, 4.37) 3.02 (1.75, 5.19) 2.10 (1.50, 2.94)
0.426
FRI
Values in boldface font are statistically significant at α = 0.05.
0.503 0.162 0.304 0.090 0.008 0.001 0.300
390
S. Yang et al. / Bone 110 (2018) 386–391
Table 5 Category-free net reclassification indices (NRIs) and 95% confidence intervals (95% CIs) for logistic regression models of prior hip fracture associated with femoral strength (FS) and fracture risk index (FRI) vs bone mineral density (BMD) alone. Model variables
NRI (95% CI) for hip fracture cases
NRI (95% CI) for non-fracture controls
Overall NRI (95% CI)
FS + femoral neck BMD vs femoral neck BMD alone FRI + femoral neck BMD vs femoral neck BMD alone FS + total hip BMD vs total hip BMD alone FRI + total hip BMD vs total hip BMD alone
0.14 (0.04, 0.24) 0.27 (0.17, 0.37) 0.02 (−0.03, 0.06) 0.07 (−0.03, 0.17)
0.04 (−0.03, 0.11) 0.24 (0.17, 0.31) −0.01 (−0.04, 0.02) 0.05 (−0.02, 0.12)
0.18 (0.05, 0.30) 0.51 (0.38, 0.63) 0.00 (−0.05, 0.06) 0.12 (0.00, 0.24)
Estimates in boldface font are statistically significant at α = 0.05.
In the logistic regression model adjusted for age and BMI, lower FS (OR 1.86 per SD decrease, 95% CI 1.59–2.17) and higher FRI (OR 2.61 per SD increase, 95% CI: 2.18, 3.12) were associated with significantly increased risk of hip fracture (Table 3). The association between FS and hip fracture remained statistically significant after adjusting for age, BMI, femoral neck BMD, or hip fracture probability, but was no longer significant after adjusting for age, BMI and total hip BMD. The association between FRI and hip fracture remained statistically significant in all models. The association between FS and hip fracture was not influenced by age, BMI, femoral neck BMD or hip fracture probability (Table 4). The risk for hip fracture associated with FRI was stronger in women who were younger, had higher BMI or non-osteoporotic BMD (all P for interaction b 0.1; Table 4). The associations of prior hip fracture with FS and FRI were not influenced by the time between fracture and DXA scan (all P for interaction N 0.9). The c-statistics for hip fracture associated with FS, FRI, femoral neck BMD and total hip BMD were 0.69 (95% CI: 0.65, 0.72), 0.77 (95% CI: 0.74, 0.80), 0.74 (95% CI: 0.71, 0.77) and 0.78 (95% CI: 0.75, 0.81), respectively (Table 3). The c-statistics for hip fracture associated with FS was significantly lower than the c-statistics for FRI and femoral neck BMD (all P b 0.05). FRI had higher c-statistic than femoral neck BMD for assessing hip fracture risk (P b 0.05). FS and femoral neck BMD had significantly lower c-statistics than total hip BMD (all P b 0.05); there was no significant difference between FRI and total hip BMD. Compared with femoral neck BMD alone, there was a significant increase in the overall NRI when combined with FS (NRI: 0.18, 95% CI: 0.05, 0.30) or FRI (NRI: 0.51, 95% CI: 0.38, 0.63) (Table 5). The NRI improvement for FS added to femoral neck BMD alone was mainly found for improved classification of hip fractures, but not for non-fracture controls. The NRI improvement for FRI added to femoral neck BMD alone was significant among both hip fracture cases and non-fracture controls. Compared with total hip BMD alone, only the addition of FRI showed a significant improvement in risk reclassification (overall NRI: 0.12, 95% CI: 0.00, 0.24). Models containing FS and femoral neck BMD, and FRI and femoral neck BMD had better Brier scores than femoral neck BMD alone (medians: 0.096 and 0.073 vs 0.102, respectively; all P for difference b 0.05). FRI and total hip BMD showed a significant improvement in Brier score for assessing hip fracture risk as compared with total hip BMD alone (medians: 0.023 vs 0.078; P for difference = 0.037); the score improvement was not seen for FS and total hip BMD vs total hip BMD alone (medians: 0.078 vs 0.078; P for difference = 0.521).
correlation between these two measures (r = −0.69). Consideration of subject-specific fall-induced impact force by the latter is one likely reason for the improved assessment of hip fracture risk, among other technical differences between the FS and FRI procedures (Table 1). If our results are confirmed in prospective studies, other investigators developing DXA-based FEA may need to include fall-induced impact force in the procedure. FS measured by the current method gave a lower cstatistic than femoral neck BMD (0.69 vs 0.74) for hip fracture stratification, which differs from previous human studies [28,29]. This is probably due to changes in the FEA algorithm used for calculating FS [28,29]. Our DXA-based FEA was performed based on two-dimensional images, and differs from most FEA studies which are performed on three-dimensional CT images [2–4]. Whether our results have implications for CT-based FEA is unclear. There are many other non-BMD measures (i.e., hip axis length, hip structural analysis) that can be derived from hip DXA scans [30]. Some studies have shown that longer hip axis length is associated with increased hip fracture risk in women [31,32], with limited but concordant data from men [33]. It is currently unknown whether these simple non-BMD measures of bone are independent of FS and FRI. There are several limitations to the present study. First, only a single DXA scanner configuration was studied and the FS procedure may not have been optimized for this scanner configuration (Prodigy, GE Healthcare). Second, ethnicity and sex interactions were not tested as 99% of our study population was Caucasian and only women were included. Third, prior hip fracture could lead to an accelerated bone loss [34,35]. However, we adjusted for BMD directly and for fracture probability (computed with BMD) which partially mitigates this concern. Nevertheless, this highlights the need for prospective studies. Lastly, c-statistic, NRI and Brier score results were estimated from a cross-sectional study, in which prior hip fracture was the outcome of interest. These results may not concordant with that from prospective studies, in which incident hip fracture is the outcome of interest. In conclusion, although FS and FRI were independently associated with hip fracture, there were performance differences for FS vs FRI in stratifying hip fracture risk, possibly reflecting the beneficial effect of considering fall-induced impact force. These results may help to inform future DXA-based finite element models and their associated outputs for assessing hip fracture risk. Conflict of interest
4. Discussion None of the authors have disclosures related to this work. We found significant relationships between FS and FRI with hip fracture risk in this cross-sectional study, and these were femoral neck BMD- and FRAX-independent. However, FS had a lower c-statistic than FRI for stratifying hip fracture risk and only FRI showed a significant improvement in hip fracture risk reclassification when combined with total hip BMD. This is the first study directly comparing FS and FRI for assessing hip fracture risk, and identified differences in performance characteristics. FS is a measure of the bone strength in a specific sideway fall configuration and independent of the magnitude of the impact force, whereas FRI takes the impact force into account of the impact force magnitude. The differences between FS and FRI are also partially reflected by the modest
Acknowledgements This study was funded through a Manitoba Partnership Program Grant from Research Manitoba and the Canadian Institutes of Health Research (CIHR# 326175). The authors acknowledge the Manitoba Centre for Health Policy for use of data contained in the Population Health Research Data Repository (HIPC# 2008/2009–33). The results and conclusions are those of the authors and no official endorsement by the Manitoba Centre for Health Policy, Manitoba Health, Seniors and Active Living, or other data providers is intended or should be inferred.
S. Yang et al. / Bone 110 (2018) 386–391
References [1] R.C. Hibbeler, Mechanics of Materials, 10th edition Pearson, Cambridge, UK, 2016. [2] P. Zysset, L. Qin, T. Lang, S. Khosla, W.D. Leslie, J.A. Shepherd, J.T. Schousboe, K. Engelke, Clinical use of quantitative computed tomography-based finite element analysis of the hip and spine in the management of osteoporosis in adults: the 2015 ISCD official positions-part II, J. Clin. Densitom. 18 (3) (2015) 359–392. [3] E.S. Orwoll, L.M. Marshall, C.M. Nielson, S.R. Cummings, J. Lapidus, J.A. Cauley, K. Ensrud, N. Lane, P.R. Hoffmann, D.L. Kopperdahl, T.M. Keaveny, G. Osteoporotic Fractures in Men Study, Finite element analysis of the proximal femur and hip fracture risk in older men, J. Bone Miner. Res. 24 (3) (2009) 475–483. [4] T.M. Keaveny, L.M. Marshall, C.M. Nielson, et al., Finite element analysis of proximal femur QCT scans for the assessment of hip fracture risk in older men, J. Clin. Densitom. 11 (3) (2008) 467–468. [5] D. Dragomir-Daescu, C. Salas, S. Uthamaraj, T. Rossman, Quantitative computed tomography-based finite element analysis predictions of femoral strength and stiffness depend on computed tomography settings, J. Biomech. 48 (1) (2015) 153–161. [6] M. Qasim, G. Farinella, J. Zhang, X. Li, L. Yang, R. Eastell, M. Viceconti, Patient-specific finite element estimated femur strength as a predictor of the risk of hip fracture: the effect of methodological determinants, Osteoporos. Int. 27 (9) (2016) 1–8. [7] A. El Maghraoui, C. Roux, DXA scanning in clinical practice, QJM 101 (8) (2008) 605–617. [8] Y. Luo, Image-based Multilevel Biomechanical Modeling for Fall-induced Hip Fracture, Springer International Publishing, Cham, Switzerland, 2017. [9] M.N. Sarvi, Y. Luo, Sideways fall-induced impact force and its effect on hip fracture risk: a review, Osteoporos. Int. 28 (10) (2017) 2759–2780. [10] Y. Luo, M.N. Sarvi, P. Sun, W.D. Leslie, J. Ouyang, Prediction of impact force in sideways fall by image-based subject-specific dynamics model, Int. Biomech. 1 (1) (2014) 1–14. [11] S. Yang, W.D. Leslie, Y. Luo, A.L. Goertzen, S. Ahmed, L.M. Ward, L.M. Lix, Automated DXA-based finite element analysis for hip fracture risk stratification: a crosssectional study, Osteoporos. Int. 29 (1) (2018) 191–200. [12] E. Dall'Ara, R. Eastell, M. Viceconti, D. Pahr, L. Yang, Experimental validation of DXAbased finite element models for prediction of femoral strength, J. Mech. Behav. Biomed. Mater. 63 (2016) 17–25. [13] W.D. Leslie, P.A. Caetano, L.R. MacWilliam, G.S. Finlayson, Construction and validation of a population-based bone densitometry database, J. Clin. Densitom. 8 (1) (2005) 25–30. [14] W.D. Leslie, C. Metge, Establishing a regional bone density program: lessons from the Manitoba experience, J. Clin. Densitom. 6 (3) (2003) 275–282. [15] L.M. Lix, M. Azimaee, B.A. Osman, P. Caetano, S. Morin, C. Metge, D. Goltzman, N. Kreiger, J. Prior, W.D. Leslie, Osteoporosis-related fracture case definitions for population-based administrative data, BMC Public Health 12 (2012) 301. [16] S. O'Donnell, Use of administrative data for national surveillance of osteoporosis and related fractures in Canada: results from a feasibility study, Arch. Osteoporos. 8 (1–2) (2013) 143. [17] B. Fan, Y. Lu, H. Genant, T. Fuerst, J. Shepherd, Does standardized BMD still remove differences between Hologic and GE-Lunar state-of-the-art DXA systems? Osteoporos. Int. 21 (7) (2010) 1227–1236. [18] Y. Luo, S. Ahmed, W.D. Leslie, Automation of a DXA-based finite element tool for clinical assessment of hip fracture risk, Comput. Methods Prog. Biomed. 155 (2018) 75–83.
391
[19] E.F. Morgan, H.H. Bayraktar, T.M. Keaveny, Trabecular bone modulus–density relationships depend on anatomic site, J. Biomech. 36 (7) (2003) 897–904. [20] E. Morgan, T. Keaveny, Dependence of yield strain of human trabecular bone on anatomic site, J. Biomech. 34 (5) (2001) 569. [21] Y.H. Luo, Z. Ferdous, W.D. Leslie, Precision study of DXA-based patient-specific finite element modeling for assessing hip fracture risk, Int. J. Numer. Method Biomed. Eng. 29 (5) (2013) 615–629. [22] Y. Luo, Z. Ferdous, W.D. Leslie, A preliminary dual-energy X-ray absorptiometrybased finite element model for assessing osteoporotic hip fracture risk, Proc. Inst. Mech. Eng. H 225 (H12) (2011) 1188–1195. [23] E.R. Delong, D.M. Delong, D.L. Clarke-Pearson, Comparing areas under two or more correlated receiver operating characteristics curves: a nonparametric approach, Biometrics 44 (3) (1988) 837–845. [24] K.F. Kerr, Z. Wang, H. Janes, R.L. McClelland, B.M. Psaty, M.S. Pepe, Net reclassification indices for evaluating risk prediction instruments: a critical review, Epidemiology 25 (1) (2014) 114–121. [25] M.J. Pencina, R.B. D'Agostino Sr., E.W. Steyerberg, Extensions of net reclassification improvement calculations to measure usefulness of new biomarkers, Stat. Med. 30 (1) (2011) 11–21. [26] G.W. Brier, Verification of forecasts expressed in terms of probability, Mon. Weather Rev. 78 (1) (1950) 1–3. [27] M.J. Leening, E.W. Steyerberg, B.V. Calster, R.B. D'Agostino Sr., M.J. Pencina, Net reclassification improvement and integrated discrimination improvement require calibrated models: relevance from a marker and model perspective, Stat. Med. 33 (19) (2014) 3415–3418. [28] L. Yang, L. Palermo, D.M. Black, R. Eastell, Prediction of incident hip fracture with the estimated femoral strength by finite element analysis of DXA scans in the study of osteoporotic fractures, J. Bone Miner. Res. 29 (12) (2014) 2594–2600. [29] K.E. Naylor, E.V. McCloskey, R. Eastell, L. Yang, Use of DXA-based finite element analysis of the proximal femur in a longitudinal study of hip fracture, J. Bone Miner. Res. 28 (5) (2013) 1014–1021. [30] C.P. Edmondson, E.N. Schwartz, Non-BMD DXA measurements of the hip, Bone 104 (2017) 73–83. [31] N.J. Crabtree, H. Kroger, A. Martin, H.A. Pols, R. Lorenc, J. Nijs, J.J. Stepan, J.A. Falch, T. Miazgowski, S. Grazio, Improving risk assessment: hip geometry, bone mineral distribution and bone strength in hip fracture cases and controls. The EPOS study. European Prospective Osteoporosis Study, Osteoporos. Int. 13 (1) (2002) 48–54. [32] W.D. Leslie, L.M. Lix, S.N. Morin, H. Johansson, A. Odén, E.V. Mccloskey, J.A. Kanis, Hip axis length is a FRAX- and bone density-independent risk factor for hip fracture in women, J. Clin. Endocrinol. Metab. 100 (5) (2015) 2063–2070. [33] W.D. Leslie, L.M. Lix, S.N. Morin, H. Johansson, A. Odén, E.V. Mccloskey, J.A. Kanis, Adjusting hip fracture probability in men and women using hip axis length: the manitoba bone density database, J. Clin. Densitom. 19 (3) (2015) 326–331. [34] L. Reider, T.J. Beck, M.C. Hochberg, W.G. Hawkes, D. Orwig, J.A. Yuyahiro, J.R. Hebel, J. Magaziner, Women with hip fracture experience greater loss of geometric strength in the contralateral hip during the year following fracture than age-matched controls, Osteoporos. Int. 21 (5) (2010) 741. [35] K.M. Fox, J. Magaziner, W.G. Hawkes, J. Yuyahiro, J.R. Hebel, S.I. Zimmerman, L. Holder, R. Michael, Loss of bone density and lean body mass after hip fracture, Osteoporos. Int. 11 (1) (2000) 31–35.
