Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Magnetic resonance imaging as an instrument to assess the association between femoral neck bone geometry… Manske, Sarah Lynn 2005

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
831-ubc_2005-0547.pdf [ 12.19MB ]
Metadata
JSON: 831-1.0092401.json
JSON-LD: 831-1.0092401-ld.json
RDF/XML (Pretty): 831-1.0092401-rdf.xml
RDF/JSON: 831-1.0092401-rdf.json
Turtle: 831-1.0092401-turtle.txt
N-Triples: 831-1.0092401-rdf-ntriples.txt
Original Record: 831-1.0092401-source.json
Full Text
831-1.0092401-fulltext.txt
Citation
831-1.0092401.ris

Full Text

MAGNETIC RESONANCE IMAGING AS AN INSTRUMENT TO ASSESS THE ASSOCIATION BETWEEN FEMORAL NECK BONE GEOMETRY AND STRENGTH OF THE PROXIMAL FEMUR by Sarah Lynn Manske BSc. Honours (Kinesiology), Simon Fraser University, 2003 A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE in THE FACULTY OF GRADUATE STUDIES (Experimental Medicine) THE UNIVERSITY OF BRITISH C O L U M B I A OCTOBER^2005 1 © Sarah Lynn Manske, 2005 ABSTRACT Introduction: Hip fractures are an increasing health and economic burden. Dual energy x-ray absorptiometry (DXA) is the instrument currently used to diagnose osteoporosis, however, there are limitations associated with using D X A to predict fracture risk and to measure response to therapeutic interventions. Magnetic resonance imaging (MRI) is an emerging instrument to assess bone, however the ability of MRI measurements of femoral neck geometry to predict bone strength has not been previously assessed. Objectives: To evaluate the association of femoral neck cross-sectional geometry measured with MRI with failure load in cadaveric femora, and to compare this association with D X A and Hip Structural Analysis (HSA). The secondary objective was to compare reliability of femoral neck geometry measured with 3 Tesla (T) MRI and 1.5 T MRI systems. Methods: Thirty-six human cadaveric proximal femora underwent D X A and MRI imaging. D X A images were also analyzed with HSA. Areal B M D (aBMD) was evaluated with D X A and HSA. Cross-sectional geometry (area, second area moment of inertia - Ix, and section modulus) were evaluated with MRI (femoral neck region) and HSA (narrow neck and intertrochanteric regions). Inter-analysis and inter-acquisition reliability were compared between measurements with 1.5 T and 3 T MRI systems. The femora were loaded to failure in a fall configuration. Results: Femoral neck cortical cross-sectional area and Ix, measured with MRI, were strongly associated with failure load (R 2 = 0.47 for both measures, p < 0.001). The predictive ability of Ix was lower than trochanteric aBMD (R = 0.70), p = 0.10. Ix significantly contributed to the variance explained in failure load after accounting for femoral neck aBMD (R2-change = 0.14, p = 0.01), but not after accounting for trochanteric aBMD (R2-change = 0.03, p = 0.23). Cross-sectional geometry, e.g. Ix, measured with MRI explained similar variance in failure load (R 2 = 0.47) as cross-sectional geometry estimated with HSA (R = 0.31). Inter-acquisition and inter-analysis reliability were similar for 3 T and 1.5 T MRI systems. Summary and Conclusion: Femoral neck cross-sectional geometry assessed with MRI and HSA, and aBMD by D X A were similarly associated with failure load ex vivo. MRI holds promise for the in vivo assessment of cortical bone geometry at the proximal femur, as neither D X A nor HSA are capable of measuring these parameters without major assumptions. However, a targeted program of research that aims to improve and standardize MRI image acquisition and analysis is warranted. n T A B L E O F C O N T E N T S A B S T R A C T ii T A B L E O F C O N T E N T S ii i L I S T OF T A B L E S : vii L I S T O F F I G U R E S ix G L O S S A R Y & A B B R E V I A T I O N S xi A C K N O W L E D G M E N T S xiv 1. INTRODUCTION 1 1.1 Need to Identify Individuals at Risk for Hip Fracture 1 1.2 Limitations of Current Fracture Prediction Methodologies 1 1.3 MRI as an Instrument to Improve Measurement of Fracture Risk 3 2. L I T E R A T U R E R E V I E W 5 2.1 Bone Biology and Mechanics 5 2.1.1 Bone Biology 5 2.1.1.1 Cortical Bone Biology 6 2.1.1.2 Trabecular Bone Biology 7 2.1.1.3 Bone Growth 7 2.1.2 Bone Mechanical Properties 8 2.1.2.1 Bone Structural Properties 9 2.1.2.2 Bone Material Properties 10 2.1.2.3 Bone Cross-Sectional Geometry 12 2.1.3 Contributions of Cortical Bone to Strength in the Proximal Femur 13 2.1.4 Bone Biology and Mechanics Summary 15 2.2 Proximal Femur Fracture 15 2.2.1 Epidemiology 15 2.2.2 Clinical Fractures of the Proximal Femur 17 2.2.3 Immediate Cause of Proximal Femur Fractures 18 2.2.4 Laboratory Simulation of Proximal Femur Fractures 19 2.2.5 Proximal Femur Fracture Summary 22 2.3 Bone Imaging to Predict Hip Fracture 22 2.3.1 Dual Energy x-Ray Absorptiometry (DXA) 22 2.3.1.1 Description 22 2.3.1.2 Advantages of DXA 23 2.3.1.3 Limitations of DXA 24 2.3.1.4 Hip Structural A nalysis (HSA) 29 2.3.2 Quantitative Ultrasound (QUS) 31 2.3.3 Radiography 31 2.3.4 Quantitative Computed Tomography (QCT) 33 2.3.4.1 Description 33 2.3.4.2 A dvantages of QCT 33 2.3.4.3 Limitations of QCT 35 2.3.4.4 Advantages of pQCT 36 2.3.4.5 Limitations of pQCT 37 2.3.5 Magnetic Resonance Imaging (MRI) 39 2.3.5.1 Description 39 2.3.5.2 Signal to Noise Ratio in MRI 40 iii 2.3.5.3 Advantages of MRI 41 2.3.5.4 Limitations of MRI 46 2.3.6 Summary of Limitations in Existing Bone Imaging Technologies 47 2.4 Age-Related Differences in Bone 48 2.4.1 Age-Related Differences in Bone Density 49 2.4.2 Age-Related Differences in Proximal Femur Cross-Sectional Geometry 49 2.4.3 Age-Related Differences in Bone Material Properties 51 2.5 Summary of Literature Review and Limitations in Existing Knowledge 51 3. R E S E A R C H QUESTIONS 53 3.1 Rationale, Objectives and Hypotheses for Primary Aim 53 3.2 Rationale, Objective and Hypothesis for Secondary Aim 55 4 M E T H O D S 56 4.1 Specimens 56 4.2 Dual energy x-Ray Absorptiometry (DXA) 56 4.2.1 DXA Acquisition 56 4.2.2 DXA Analysis 57 4.2.3 DXA Independent Variables 57 4.3 Hip Structural Analysis (HSA) 57 4.3.1 HSA Independent Variables 58 4.4 Magnetic Resonance Imaging (MRI) 59 4.4.1 1.5 TMRIAcquisition 59 4.4.2 3 T MRI Acquisition 60 4.4.3 MRI Analysis 61 4.4.4 MRI Reliability 64 4.4.5 MRI Independent Variables 64 4.5 Mechanical Testing - Sideways Impact Configuration ....64 4.5.1 Testing Protocol 64 4.5.2 Displacement Rate 66 4.5.3 Dependent Variables 66 4.5.4 Fracture Classification 68 4.6 Sample Size 68 4.7 Statistical Analyses 69 4.7.1 Primary Aim 69 4.7.1.1 Descriptives 69 4.7.1.2 Association Between Independent Variables and Failure Load 69 4.7.2 Secondary Aim 71 4.7.2.1 Differences Between 3 T MRI and 1.5 T MRI 71 4.7.2.2 Reliability Analyses 71 5. R E S U L T S 72 5.1. Descriptives 72 5.1.1. Specimens and Fracture Types 72 5.1.2. Fracture Types 72 5.1.3. Independent Variables 72 5.1.3.1. Comparison Between Independent Variables Measured with MRI 74 5.1.3.2. Comparison Between Independent Variables Measured with Different Imaging Systems 76 5.1.4 Dependent Variables 79 iv 5.1.5 Side-to-Side Comparisons 80 5.2 Relationship of Independent Variables with Failure Load 80 5.2.1 Correlations 80 5.2.2 Comparison of Correlations Between MRI, DXA and HSA 87 5.2.3 Hierarchical Linear Regression Models 87 5.2.3.1 MRI + DXA Trochanteric Region- Hierarchical Model 87 5.2.3.2 MRI + DXA Femoral Neck Region- Hierarchical Model 88 5.2.3.3 HSA Intertrochanteric Region- Hierarchical Model 88 5.2.3.4 HSA Narrow Neck Region - Hierarchical Model 89 5.2.3.5 Comparison Between Hierarchical Models 89 5.3 Comparison of 3 T with 1.5 T MRI 90 5.3.1 Relationships Between Independent Variables Measured by 1.5 T and 3 T 90 5.3.2 Is Inter-Acquisition Reliability Greater with 3 T than 1.5 T? 93 5.3.2.1 Inter-Acquisition Reliability with 3 T 93 5.3.2.2 Inter-Acquisition Reliability with 1.5 T 93 5.3.2.3 Comparison of Inter-Acquisition Reliability Between 3 T and 1.5 T 93 5.3.3 Is Inter-Analysis Reliability Greater with 3 T than 1.5 T? 95 5.3.3.1 Inter-Analysis Reliability with 3 T 95 5.3.3.2 Inter-Analysis Reliability with 1.5T 95 5.3.3.3 Comparison of Inter-Analysis Reliability Between 3 T and 1.5 T 96 6. D ISCUSSION 98 6.1. The Use of Mechanical Testing to Induce Fracture 98 6.2. The Role of MRI Assessment of the Femoral Neck to Predict Failure Load 100 6.2.1. MRI Assessment of the Femoral Neck is Associated with Failure Load 100 6.2.2. Comparison Between MRI, DXA and HSA Imaging Systems 101 6.2.2.1. Comparison of Associations Between Independent Variables Measured with MRI DXA and HSA, and Failure Load 101 6.2.2.2. Addition of Femoral Neck Geometry to DXA-Derived Regression Models. 102 6.2.3. Secondary Findings of Interest 107 6.2.3.1. Regional-Specificity of Failure Load Prediction 107 6.2.3.2. High Rate of Intertrochanteric Fractures 109 6.2.3.3. Left-Right Differences in Fracture Types 112 6.2.3.4. Bending Strength Indices and Cortical Area are Similarly Associated with Failure Load 112 6.3. Comparisons Between 1.5 and 3 T MRI 114 6.3.1. Systematic Differences Between 1.5 T and 3 T MRI 114 6.3.2. Reliability of MRI Measures of Cross-Sectional Geometry 114 6.3.2.1. Inter-Acquisition Reliability for 3 T MRI 115 6.3.2.2. Inter-Analysis Reliability for 3 T MRI 115 6.3.3. Reliability was Similar for a 3 T MRI System and a 1.5 T MRI System. 116 6.3.4. Comparison of Reliability Results with Previous Research 116 6.4. Limitations Related to the Study Methodology 118 6.4.1. Cadaveric Sample 118 6.4.1.1. Limited Knowledge of Medical Histories 118 6.4.1.2. Specimen Storage 119 6.4.1.3. Missing Anatomical Structures 119 6.4.2. The Trochanteric Region was not Assessed with MRI 120 v 6.4.3. Limitations of MRI Measurements 121 6.4.3.1. Accuracy Unknown, Moderate Reliability 121 6.4.3.2. Presence of Gas in the Marrow Cavity 121 6.4.4. Collinearity 121 6.4.5. Paired Samples 122 6.5. Limitations in Applying Results to Future In Vivo Studies 122 6.5.1. Limitations when Applying the MRI Protocol In Vivo 122 6.5.2. Mechanical Testing 122 6.6. Future Directions 123 6.6.1. Recommendations for Improving Techniques to Assess Bone with MRI 124 6.6.1.1. Recommendations for Assessing the Trochanteric Region with MRI 124 6.6.1.2. Improving Reliability 125 6.6.1.3. Assessing Trabecular Architecture with MRI 125 7. S U M M A R Y A N D C O N C L U S I O N S 126 7.1. Summary (Primary Objectives) 126 7.2. Summary (Secondary Objective) 126 7.3. Conclusions 126 R E F E R E N C E S 127 A P P E N D I X 1: C A L C U L A T I O N O F D I S P L A C E M E N T R A T E 136 A P P E N D I X 2: C O R R E L A T I O N S B E T W E E N INDEPENDENT V A R I A B L E S 138 A P P E N D I X 3: L I N E A R R E G R E S S I O N A N A L Y S E S W I T H M U L T I - L E V E L M O D E L L I N G 140 vi L I S T O F T A B L E S Table 1. Hierarchical anatomical structure of bone. Reproduced from Hollister, 2005 [29] 6 Table 2. Fracture types in simulated sideways falls. The references provided are examples only, and are not a complete review. In all studies, impact was applied to simulate a sideways fall on the greater trochanter (GT). Orientation of the femur varied slightly in each study. FN = femoral neck, IT = intertrochanteric 21 Table 3. The effect of changing MRI properties on the signal-to-noise ratio. Based on Pooley et al, 2001 and Hashemi et al, 2004 [102, 103] 41 Table 4. Mean values for independent variables of interest measured with MRI, D X A and HSA for each region of interest (N = 27) 73 Table 5. Mean and standard error of the mean (SEM) for the dependent variables obtained from the load-displacement curves in mechanical testing (N = 27) 79 Table 6. Correlations (R) and R 2 between independent variables measured with each imaging system and failure load (N = 27). *p < 0.05, **p < 0.01 82 Table 7. Hierarchical linear regression model summary for trochanteric (TR) a B M D D X A and IXMRI (N = 27). B is the unstandardized coefficient, (3 is the standardized coefficient 88 Table 8. Hierarchical linear regression model summary for femoral neck (FN) a B M D D x A and IXMRI (N = 27). B is the unstandardized coefficient, (3 is the standardized coefficient 88 Table 9. Hierarchical linear regression model summary for the intertrochanteric (IT) region measured by HSA (N = 27) 89 Table 10. Hierarchical linear regression model summary for the narrow neck (NN) region measured by HSA (N = 27) 89 Table 11. Independent variables measured with 3 T MRI and 1.5 T MRI. Differences are presented as (1.5 T - 3 T), therefore a positive difference indicates a larger value measured by 1.5 T 90 Table 12. Inter-acquisition reliability for a) 3 T MRI (n = 12) and b) 1.5 T MRI (n = 12). Differences are presented as (2 n d scan - 1 s t scan). Percent differences are presented as (2 n d scan - 1s t scan/mean of two scans) 94 Table 13. 95% limits of agreement for inter-acquisition reliability 95 Table 14. Inter-analysis reliability for a) 3 T MRI (n = 12) and b) 1.5 T MRI (n = 12). Differences are presented as (2 n d scan - 1s t scan). Percent differences are presented as (2 n d scan - 1s t scan/mean of two scans) 96 Table 15. 95% limits of agreement for inter-analysis reliability 97 Table 16. Mean failure loads and correlations with D X A aBMD measured in different proximal femur subregions, as well as percentages of femoral neck (FN) and intertrochanteric (IT) fractures as reported by studies that loaded human cadaveric femora to failure in a sideways fall simulation 100 Table 17. Comparison of reliability for ToCSA and CoCSA (coefficient of variation - C V , and 95%o Limits of Agreement - LOA) reported in the current study (3 T MRI) and reported by McKay et al. (1.5 T MRI) [25] 117 Table 18. Correlation coefficients (R) between independent variables (N = 27). TR = trochanteric region, N N = narrow neck region, IT = intertrochanteric region. *p < 0.05, **p <0.01 138 Table 19. Regression coefficients and standard errors for the relationship between trochanteric aBMD and failure load. Two analysis models are presented, 1) treating the specimens as vn independent samples, and 2) treating the data as pairs, where appropriate, in a random effects model (N = 27) 142 Table 20. Regression coefficients and standard errors for the relationship between femoral neck aBMD and failure load. Two analysis models are presented, 1) treating the specimens as independent samples, and 2) treating the data as pairs, where appropriate, in a random effects model (N = 27) 142 Table 21. Regression coefficients and standard errors for the relationship between CoCSA measured by MRI and failure load. Two analysis models are presented, 1) treating the specimens as independent samples, and 2) treating the data as pairs, where appropriate, in a random effects model (N = 27) 143 viii L I S T O F F I G U R E S Figure 1. Typical load-deformation curve to describe the structural behaviour of a specimen, identifying the elastic, plastic and yield regions. The yield point and stiffness calculation are included. Reproduced from Bouxsein et al, 1996 [34] 9 Figure 2. Typical stress-strain curve describing the mechanical behaviour of a material 11 Figure 3. Cross-sectional moment of inertia (mm4) for cylindrical beams. Reproduced from Hayes, 1991 [28]. 10a) describes a solid cylinder, 10b) describes a cylinder with a thin shell, similar to a long bone which has a dense cortical bone located at a distance from the neutral axis 13 Figure 4. Classifications of proximal femur fracture by region. Reproduced from Zuckerman, 1996 [53] 17 Figure 5. A proximal femur loaded in a one-legged stance configuration. Reproduced from Lochmuller et al, 2002 [11] 20 Figure 6. A proximal femur loaded in a sideways fall configuration. The load is applied to the greater trochanter. Reproduced from Lochmuller et al, 2002 [11] 20 Figure 7. Schematic of D X A image to represent bone 26 Figure 8. A model of the cross-section of the femoral neck derived using HSA software from the bone mass profile obtained in a D X A scan of the proximal femur 30 Figure 9. The D X A total proximal femur ROI and its subregions. Reproduced from Hologic User's Guide [121] 57 Figure 10. Proximal femur image from a Hologic D X A scanner showing positions of HSA analysis and bone mass profiles of the two ROIs investigated in this study- narrow neck and intertrochanteric 58 Figure 11. A scout scan of the proximal femur ex vivo. The solid yellow line indicates the longitudinal femoral neck axis. The solid red line indicates the location of the cross-section shown along the FN shown in Figure 12 below 59 Figure 12. A representative MRI cross-section of the femoral neck 60 Figure 13. MRI coronal view of the proximal femur labelled with the medial (a) and lateral (b) borders of the F N region of interest 62 Figure 14. MRI cross-section of the femoral neck segmented using Analyze software, a) the green line defines the outer, periosteal border, b) The red line defines the inner, endocortical border 63 Figure 15. A schematic of the mechanical testing apparatus. Bearings between the surfaces allowed the plates to translate freely. The distal end of the femur was free to pivot 66 Figure 16. Failure load determined from the load displacement curve, as defined as the first local maximum 67 Figure 17. Failure load determined by the load-displacement curve, as defined by the maximum load. In this case, the failure load dropped less than 10% of the first local maximum after reaching the first local maximum 68 Figure 18. CoCSA measured with MRI vs. ToCSA measured with MRI 74 Figure 19. Ix measured with MRI vs. CoCSA measured with MRI 75 Figure 20. Sx measured with MRI vs. CoCSA measured with MRI 75 Figure 21. Narrow neck CSA measured by HSA vs. CoCSA measured by MRI. Narrow neck CSAHSA includes cortical and trabecular bone areas 76 Figure 22. Narrow neck I measured by HSA vs. Ix measured by MRI 77 Figure 23. Narrow neck S measured by HSA vs. Sx measured by MRI 77 Figure 24. Femoral neck aBMD measured with D X A vs. CoCSA measured with MRI 78 ix Figure 25. Femoral neck aBMD measured with D X A vs. Ix measured with MRI 78 Figure 26. A load-displacement curve for a 73 year old female cadaveric femur. The specimen was loaded to failure by simulating a sideways fall on the greater trochanter 79 Figure 27. Mean failure load by type of fracture (Mean + SEM). * indicates significant difference between failure load for stable intertrochanteric (IT) and unstable fractures (p < 0.05) 80 Figure 28. Failure load vs. CoCSA measured with MRI 83 Figure 29. Failure load vs. Ix measured with MRI 83 Figure 30. Failure load vs. Sx measured with MRI 84 Figure 31. Failure load vs. trochanteric aBMD measured with D X A 84 Figure 32. Failure load vs. femoral neck aBMD measured with D X A 85 Figure 33. Failure load vs. intertrochanteric aBMD measured with HSA 85 Figure 34. Failure load vs. intertrochanteric I measured with HSA 86 Figure 35. Failure load vs. narrow neck aBMD measured with HSA 86 Figure 36. Failure load vs. narrow neck I measured by HSA 87 Figure 37. Bland-Altman (difference vs. mean) plot for ToCSA measured in the same specimens with 1.5 T MRI and 3 T MRI (N = 20). The difference is calculated as 1.5 T - 3 T. The solid red line indicates the mean difference. The dashed red lines indicate the mean difference + 1.96 x standard deviation of the difference 91 Figure 38. Bland-Altman plot (difference vs. mean) for CoCSA measured in the same specimens with 1.5 T MRI and 3 T MRI (N = 20). The difference is calculated as 1.5 T - 3 T. The solid red line indicates the mean difference. The dashed red lines indicate the mean difference + 1.96 x standard deviation of the difference 91 Figure 39. Bland-Altman plot (difference vs. mean) for Ix measured in the same specimens with 1.5 T MRI and 3 T MRI (N = 20). The difference is calculated as 1.5 T - 3 T. The solid red line indicates the mean difference. The dashed red lines indicate the mean difference + 1.96 x standard deviation of the difference 92 Figure 40. Bland-Altman plot (difference vs. mean) for Sx measured in the same specimens with 1.5 T MRI and 3 T MRI (N = 20). The difference is calculated as 1.5 T - 3 T. The solid red line indicates the mean difference. The dashed red lines indicate the mean difference + 1.96 x standard deviation of the difference 92 Figure 41. Venn diagram illustrating the variance in failure load explained by IXMRI and femoral neckaBMDDxA. 103 Figure 42. Venn diagram illustrating that Ix shares a large portion of the variance with trochanteric aBMD 104 Figure 43. Schematic showing the hierarchical structure of the data 140 Figure 44. Failure load vs. CoCSA, measured with MRI. Each specimen is treated as an independent sample. The donor pairs are highlighted by the colour indicated in the legend. The regression equation treats each specimen as an independent sample 141 Figure 45. Failure load vs. CoCSA for paired samples only. Regression lines are plotted for each donor 142 x G L O S S A R Y & A B B R E V I A T I O N S Biomechanics Bone Strength: In this thesis, bone strength will be defined as the ultimate (maximum) failure load of a whole bone structure. Strength is often defined as the ultimate stress of bone as a material. Bone Structural Properties: These properties reflect the shape of the bone, the distribution of material within the bone envelope and the bone material itself. Bone Material Properties: Reflects the behaviour of the tissue independent of its shape and geometry. Bone Geometry (cross-sectional geometry): Reflects the distribution of bone material within a bone cross-section. Stiffness: The resistance of a structure to deformation when loaded. Measured in Newtons per metre (N/m), and determined from the slope of a load-deformation curve. Stress: The intensity of internal forces at a particular point in a material. Measured in Newtons per metre squared (N/m ) or Pascals (Pa), as the ratio of the applied load to the cross-sectional area over which the load is applied, in compression, tension and shear. Strain: The relative deformation of a material that occurs as a load is applied. Expressed as a percentage or as a ratio of the amount of deformation to the original length. Young's Modulus: The resistance of a material to deformation when loaded. Measured in Pascals (Pa), and determined from the slope of a stress-strain curve. Toughness: The amount of work per unit volume of material that is required to reach a particular point (often the failure point). Measured in Pascals (Pa) and defined as the area under a stress-strain curve. Bone Imaging DXA: Dual energy x-ray absorptiometry QCT: Quantitative computed tomography pQCT: Peripheral quantitative computed tomography MRI: Magnetic resonance imaging aBMD: areal bone mineral density as measured by D X A in grams per centimetre squared (g/cm2). This value of a pixel in a D X A image represents the attenuation of a mass of xi hydroxyapatite (e.g. 0.5 g) spread over the area of the pixel (e.g. 1 cm x 1 cm), producing a value in g/cm (e.g. 0.5 g/cm ). BMC: Bone mineral content, measured in grams (g). When measured by DXA, BMC is computed as aBMD x bone projectional area. vBMD: volumetric bone mineral density. The ratio of BMC to the volume of bone. Measured in grams per centimetre cubed (g/cm ). ROI: Region of interest ToCSA: Total bone cross-sectional area (ToCSA was chosen rather than ToA to emphasize that this represents cross-sectional rather than projectional area), measured in mm . ToCSA, in this thesis, represents the area within the bone envelope, including the cortical area, trabecular area, and area of the marrow cavity. CoCSA: Cortical bone cross-sectional area, measured in mm . I: Cross-sectional moment of inertia, or second area moment of inertia, measured in mm4. This represents the distribution of bone in a cross-section with respect to the neutral axis of bending. Computed about the appropriate axis: Ix = \/dA I = [x2dA y JA where dA is an element of Area A and x and y are the coordinates of each element of area. S: Section modulus, measured in mm3. In a bone cross-section, the section modulus is the ratio of the I to the distance of the bone from the neutral axis of bending. Computed about the appropriate axis: r y r X where Ix is the moment of inertia about the x-axis, and ry is the maximum distance from the centroid to outer edge of the bone in the y direction. S and I determine the stress due to an applied bending moment. xii Epidemiology Relative Risk (RR): This is defined as the ratio of the incidence of a disease in exposed individuals to the incidence of disease in unexposed individuals. In terms of fracture risk, RR is often presented in the following manner, for a particular variable (e.g. aBMD): _ Incidence of fracture in people 1 SD below the mean Incidence of fracture in people less than 1 SD below the mean RR can only be evaluated prospectively, when the incidence in the entire population being examined is known. Odds Ratio (OR): OR is used as an estimate of relative risk when the incidence in a population is unknown, and is calculated from case-control studies. In terms of fracture risk, it is defined as: _ Odds of fracture in people 1 SD below the mean Odds of fracture in people less than 1 SD below the mean Sensitivity: The ability of a screening test to correctly identify individuals who will fracture. It is the proportion of true positives among all individuals who did fracture. Specificity: The ability of a screening test to correctly identify individuals who will not fracture. It is the proportion of true negatives among all individuals who did not fracture. Positive Predictive Value: Proportion of true positives (fracture cases) among all individuals with positive screening tests. Population Attributable Risk Percent: The proportion of risk of fracture which can be attributed to a positive screening result (e.g. low aBMD) in the total population. The equation is presented below: Incidence in Total Population - Incidence in Population without low aBMD = xlOO Incidence in Total Population Terms I will use throughout the thesis: 1) Femoral neck fracture rather than cervical fracture. 2) Intertrochanteric fracture rather than trochanteric fracture. 3) Cortical bone rather than compact bone 4) Trabecular bone rather than cancellous bone. xiii A C K N O W L E D G M E N T S My ability to produce this thesis has been a truly inter-disciplinary, team effort. I would like to thank my supervisor, Dr. Heather McKay, for all her valuable instruction, guidance and patience. Her motivation and encouragement has been inspiring. In addition to her personal support, she has brought and coordinated an incredible team of researchers to her lab, which has allowed me to learn more than I ever imagined. I would also like to thank the members of my committee, who have all been extremely supportive and generous with their time and thoughts: Dr. Tom Oxland, for always making our problems and solutions seem so clear; Dr. Pierre Guy for his excitement, and interest in the project; and finally, Dr. Teresa Liu-Ambrose for her daily support and discussion which has been invaluable. Many people have been extremely help for each step of the process. I would like to thank Dr. Bruce Forster for his review of x-rays and discussion regarding imaging methodology, as well as the MRI technologists, Sylvia Renneberg and Trudy Harris. As well, I would like to thank Dr. Karim Khan for his valuable expertise and constant encouragement. The co-operative atmosphere in the Bone Health Research Group and the Division of Orthopaedic Engineering has been inspiring, and the assistance plentiful. I greatly appreciate the mentoring I have received along the way, particularly from Dr. Saija Kontulainen and Dr. Danmei Liu. As well, I would like to thank Peter de Bakker for building the loading apparatus and spending countless hours assisting with testing and answering my questions. Also thanks to Cecelia Tang for assistance with mechanical testing, particularly for showing us the ropes and helping develop the testing protocol. Finally, Derek Wilson and Emily Mc Walter provided much needed assistance with development of imaging analysis protocols. While the lab has become my family over the last two years, I also need to thank the people who helped me get here. My dad is the one who has truly inspired me to become involved in health research. Although he has encouraged and allowed me take my own path, it is one that constantly leads back in his direction. My mom has taught me perseverance and what should be important in life. Finally, thanks to Asher Lipson all of his support throughout the last couple of years, including the technical assistance with M A T L A B that saved me countless hours of time. xiv Introduction M A G N E T I C R E S O N A N C E I M A G I N G AS A N I N S T R U M E N T TO ASSESS T H E A S S O C I A T I O N B E T W E E N F E M O R A L N E C K B O N E G E O M E T R Y A N D S T R E N G T H O F T H E P R O X I M A L F E M U R 1. INTRODUCTION 1.1 Need to Identify Individuals at Risk for Hip Fracture In Canada, over 24 000 hip fractures occur each year [1], resulting in health care costs of $650 million per year [2]. By 2040, the incidence of hip fracture is expected to increase to nearly 90 000 fractures per year, at a cost of $2.4 billion. Current strategies to prevent low-trauma hip fractures include pharmacologic intervention (e.g. bisphosphonates), diet (e.g. vitamin D and calcium supplements), exercise to promote bone health and fall prevention strategies [3]. A recent editorial identified the need for a universal risk prediction instrument that moves beyond assessment of areal bone mineral density (aBMD) by dual energy x-ray absorptiometry (DXA) to identify fracture risk so that treatment can be cost-effective [4]. Such an instrument would allow us to identify individuals with the highest fracture risk and appropriately target prevention efforts. Curiously, while the incidence of hip fractures rose by 100% in Malmo, Sweden between 1970 and 2001 in women over 50, the prevalence of osteoporosis did not increase [5]. There is, therefore, a strong need to identify high hip fracture risk over and above the risk associated with osteoporosis alone. The discrepancy between hip fracture rate and prevalence of osteoporosis may be associated with bone strength and/or external factors such as fall risk. 1.2 Limitations of Current Fracture Prediction Methodologies Clinicians and researchers rely heavily on medical imaging technologies to characterize bone properties in vivo. A number of parameters derived from bone imaging techniques have been used to a) predict fracture risk [6], b) describe changes in bone with aging [7], and c) evaluate treatments designed to strengthen bone [8]. Areal B M D measured by D X A remains the 'gold' standard imaging technology for osteoporosis diagnosis at the hip [9, 10]. Areal B M D , as measured by D X A , is able to explain a high percentage of the variance in failure load when proximal femur specimens are mechanically tested to failure ex vivo [11, 12]. But this technology is an imperfect predictor of fracture. Although a meta-analysis of prospective cohort studies and case-control studies concluded that 1 Introduction individuals with low aBMD had a high incidence of fracture [13], a large proportion of fractures also occurred in individuals who did not have low aBMD. Furthermore, women with aBMD more than 2.5 SD below the young normal mean, the World Health Organization's current operational definition of osteoporosis, contribute to only 25-50% of all fractures in the population [14]. Thus, detection of fracture risk based on low aBMD has high specificity, but low sensitivity, low positive predictive value, and low population attributable risk [13]. For a screening program to be considered cost-effective, the screening test must have a high positive predictive value and population attributable risk. There are also a considerable number of limitations associated with D X A as an instrument to measure bone properties. Areal B M D represents a composite measure of many bone properties. To illustrate, D X A images are two dimensional coronal, or planar, views of a bone. Because the images are planar, it is not possible to assess bone cross-sectional geometry nor can one differentiate cortical from trabecular bone. Also, there are substantial inaccuracies in aBMD measurement due to variations in soft tissue anthropometry within an image [15]. Finally, many interventions that have shown no changes in aBMD measurement actually promoted increased v B M D in the vertebrae [16], increased cortical cross-sectional area and B M C in the radius [17] and increased failure load in rat femora [18]. Thus, aBMD measurements cannot capture all changes in bone. Because bone strength is partially determined by its cross-sectional geometry, representing the distribution of bone throughout a cross-section should improve the ability to assess hip fracture risk [19]. Recent editorials have indicated that "apportioning the causation of bone fragility to specific material and structural abnormalities is likely to help us identify the persons who are at risk for fracture and thus to target therapy to those who are most likely to benefit" (p. 323) [20], and that ".. .methods must be sought to allow accurate measures of biomechanically effective femoral neck size in vivo" (p. 952) [21]. Thus, there is a need to identify novel measurement techniques that assess various components of proximal femur bone strength, including bone geometry. 2 Introduction Experiments, ex vivo, have shown that cross-sectional bone geometry as well as v B M D at the femoral neck assessed by quantitative computed tomography (QCT), are highly related to proximal femur failure load [22-24]. However, QCT exposes patients to considerably more ionizing radiation than a D X A scan. It is also necessary to consider the anatomical location at which bone measurements are made. Measures of aBMD at the proximal femur are more highly associated with proximal femur failure load than a) peripheral measurements of volumetric bone mineral density (vBMD) at the radius, tibia and spine by peripheral quantitative computed tomography (pQCT) or, b) broadband ultrasound attenuation at the calcaneus by quantitative ultrasound (QUS) [12]. Thus with technologies available, to date, direct assessment of the proximal femur best predicts fracture at that site. 1.3 MRI as an Instrument to Improve Measurement of Fracture Risk Magnetic resonance imaging (MRI) can be used to measure the cross-sectional geometry of whole bones, including the proximal femur [25-27]. It generates no ionizing radiation. To my knowledge, no previous study has determined the relation of MRI measures of cross-sectional geometry at the proximal femur to its mechanical strength. Thus, the primary aim of this thesis is to evaluate MRI as an instrument to assess proximal femur geometry as a means to predict the mechanical strength in human cadaveric femora. In order to utilize MRI measurements of bone geometry to predict fracture risk, these measurements must be reliable. A previous study found that reliability of MRI measures of femoral neck total cross-sectional area was high, but reliability of measures of cortical cross-sectional area were low [25]. Thus the secondary aim is to determine whether a 3 T MRI system can be used to measure cortical geometry with greater reliability than a 1.5 T MRI system. The broad purpose of this thesis is to evaluate the utility of MRI to contribute to the determination of bone strength. This thesis includes a comprehensive review of relevant literature in the fields of: 1) bone biology, mechanics and determinants of bone strength, 2) epidemiology and biomechanics of 3 Introduction p r o x i m a l f e m u r f r a c t u r e s , a n d f i n a l l y 3) b o n e i m a g i n g a n d i ts r e l a t i o n to p r o x i m a l f r a c t u r e p r e d i c t i o n ( C h a p t e r 2 ) . C h a p t e r 3 p r e s e n t s t h e r a t i o n a l e f o r t h e s t u d y , t h e s p e c i f i c r e s e a r c h q u e s t i o n s a n d c o r r e s p o n d i n g h y p o t h e s e s . C h a p t e r 4 d e s c r i b e s t h e m e t h o d o l o g y u s e d to test t h e s e h y p o t h e s e s . C h a p t e r 5 p r e s e n t s t h e r e s u l t s . C h a p t e r 6 p r o v i d e s a d i s c u s s i o n o f t h e s e r e s u l t s , c o m p a r i s o n to p r e v i o u s s t u d i e s , a n d p r o p o s e s f u t u r e d i r e c t i o n s f o r t h i s a r e a o f r e s e a r c h . C h a p t e r 7 s u m m a r i z e s t h e r e s u l t s a n d p r e s e n t s c o n c l u s i o n s . 4 Literature Review 2. L I T E R A T U R E R E V I E W 2.1 Bone Biology and Mechanics Bone, the primary structural element of the human body, performs mechanical functions including bearing load by supporting posture, permitting movement, and protecting internal organs. Bone also performs metabolic functions by regulating calcium homeostasis and acting as the primary site of haematopoiesis. Unlike other engineering structural materials, bone is capable of repairing itself and can alter its material properties and geometry in response to mechanical loading [28]. The following sections provide an overview of bone biology and mechanics, particularly with respect to the determinants of bone strength in the proximal femur. 2.1.1 Bone Biology The anatomical structure of bone and its development are important to an understanding of how bones develop and maintain strength. Whole bones contain two bone compartments: cortical and trabecular bone. Cortical bone, also called dense or compact bone, comprises the diaphysis (shaft) of long bones, and is the dense shell that surrounds the metaphyses. In contrast, trabecular bone, also called spongy or cancellous bone, is continuous with the inner surface of the metaphyses as well as the proximal femur and vertebrae. Cortical bone resists bending and torsional loads in addition to compressive loads. In the metaphyses, trabecular bone is designed to primarily resist compressive loads. The anatomy of cortical and trabecular bone can be described as a hierarchical structure (Table 1). 5 Literature Review Table 1. Hierarchical anatomical structure of bone. Reproduced from Hollister, 2005 [29]. Level Cortical Structure Size Range Trabecular Structure Size Range 0 Solid material >3000 urn Solid material >3000 um 1 Secondary osteons Primary osteons Interstitial bone 100 to 300 pm Secondary trabeculae Primary trabeculae Trabecular packets 75 to 200 um 2 Lamellae Lacunae Cement lines Canaliculi 3 to 20 urn Lamellae Lacunae Cement lines Canaliculi 1 to 20 um 3 Collagen - mineral composite 0.06 to 0.6 um Collagen - mineral composite 0.06 to 0.4 um 2.1.1.1 Cortical Bone Biology At the solid material level in cortical bone, level 0 (Table 1), the inner surface of the cortical bone adjacent to the bone marrow is termed the endosteal or endocortical surface. The outer surface of cortical bone adjacent to surrounding soft tissues is termed the periosteal surface. The periosteum surrounds the outer surface of most bones, and acts as a transitional region between cortical bone and surrounding soft tissue [30]. It is a thin layer containing osteogenic and fibroblastic cells, and is highly vascularized. Although presence of a periosteal layer on the femoral neck has historically been questioned, more recent histological studies indicate it is present [30]. The endosteum lines the inner surface of the bone and plays a role similar to the periosteum. However, the cells in the endosteum have a different physical environment, as they are embedded in haematopoietic marrow. The periosteal and endosteal cells work together to regulate the thickness of the cortex. At the first level of cortical bone, the major structural unit is the osteon. An osteon is a cylindrical column of bone (lamellae) surrounding a vascular channel, oriented in the long axis of the bone. Primary osteons are formed during growth, where bone was not previously present. In contrast, secondary osteons are formed by replacing existing bone. Also at the first level is interstitial bone which is composed of lamellae between Haversian systems. A Haversian system is comprised of an osteon and its vascular channel. 6 Literature Review At the second level of cortical bone are the lamellae, canaliculi and cement lines. Lamellae are the concentric bone columns surrounding the central Haversian canal and form the osteons. Lacunae are holes within the bone matrix that contain bone cells and extracellular fluid. Canaliculi connect the lacunae to allow cell travel and communication. Cement lines are formed in secondary osteons as a result of the remodelling process, at the point where resorption ends and formation begins. Cement lines are believed to slow and/or stop crack growth in cortical bone. At the third level of cortical bone are Type I collagen fibres packed with mineral and ground substance (glycosaminoglycans and glycoproteins). The mineral phase of bone, composed of calcium hydroxyapatite crystals and calcium phosphate, makes up approximately 50% of bone by volume and 75% by weight [28]. 2.1.1.2 Trabecular Bone Biology For the most part, trabecular bone is composed of the same basic structures as cortical bone [29]. The major difference between trabecular and cortical bone is found at the first and second levels. To my knowledge, cortical and trabecular bone are identical at the third level. At the first level, trabecular bone is much more porous than cortical bone, and forms a three-dimensional lattice oriented along lines of stress. Unlike cortical bone, trabecular bone does not have central canals with blood vessels. Distinctions between primary and secondary trabeculae are similar to those in cortical bone, in that secondary trabeculae are formed during remodelling. Trabecular packets are the product of bone remodelling in trabecular bone, and are thus found only in secondary trabeculae. In the 2 n d level of trabecular bone, lamellae are arranged longitudinally along the trabeculae rather than concentrically. The third level of trabecular bone is the same as the third level of cortical bone. 2.1.1.3 Bone Growth As previously mentioned, bone cells are embedded in the lacunae. These cells, including osteoblasts, osteoclasts and osteocytes, are primarily responsible for bone modelling and remodelling. Osteoblasts are the bone-forming cells which synthesize osteoid, the protein 7 L i t e r a t u r e Review component of bone tissue. The osteoid synthesized by osteoblasts becomes calcified after about ten days. Osteoblasts eventually become osteocytes or flat-lining cells [31]. Osteocytes are mature cells which, in communication with osteoblasts, are believed to be responsible for mechanosensitivity [32]. Osteoclasts are the cells responsible for bone resorption, which involves the removal of old bone. Bone modelling occurs during skeletal growth and in response to external stimuli or the internal hormonal milieu and may reconfigure bone structure to enhance bone strength. Bone mass is added at the periosteal surface. In modelling, both the periosteal and endosteal diameters are increased. In contrast, the function of bone remodelling is to replace damaged bone with new bone, through the organized activity of bone resorption by osteoclasts and bone formation by osteoblasts. Remodelling occurs in an equal balance in the young, healthy skeleton; however resorption outweighs formation in the aging and osteoporotic skeleton. 2.1.2 Bone Mechanical Properties The anatomical structure of bones results in distinctive mechanical properties. Because of their material composition and structural design, long bones have the combined properties of stiffness, flexibility, strength and lightness [33]. Bones are normally subjected to a combination of axial compressive forces, muscular forces, and bending forces applied by the weight of the body. When bone fracture occurs, it is as a result of failure of the bone tissue at the material level and of failure of the whole bone at the structural level [28]. For a given load, the structural behaviour of a whole bone is influenced by both its geometric and material properties [34]. For example, the stiffness of a whole bone subjected to bending reflects both the material stiffness of bone and the geometric distribution of bone at a distance from neutral bending axis. The mechanical properties of a bone can be described at the structural, geometric and material level. 8 Literature Review 2.1.2.1 Bone Structural Properties In the bone literature, the term 'structure' is often used synonymously with 'geometry'. However in engineering terms, geometric properties are independent of the material, whereas true structural properties are not. Structural properties are typically examined in an intact whole bone (i.e. at level 0 in the hierarchical anatomical structure of bone). The mechanical behaviour of an intact whole bone as a structure is typically described by a load-deformation curve (Figure 1). The deformation that occurs in response to an applied load is linearly related to that load (termed the elastic region) until the yield point is reached, at which point the slope of the load-deformation curve decreases. In the elastic region, the structure will return to its original shape if unloaded. Beyond the yield point is the plastic region, where addition of greater load will cause permanent deformation. Eventually, further increase in applied load will reach the failure load (N), the point at which the structure fails catastrophically, and bone fracture is said to occur. to failure to yield Deformation (m) Figure 1. Typical load-deformation curve to describe the structural behaviour of a specimen, identifying the elastic, plastic and yield regions. The yield point and stiffness calculation are included. Reproduced from Bouxsein etal., 1996 [34]. 9 Literature Review The stiffness (N/m) of a structure represents the amount of force required to cause deformation, and is measured as the slope of the load-deformation curve in the linear elastic region. The energy to yield (J) and energy to failure (J) are defined as the area under the load-displacement curve to the yield point and failure loads, respectively. Bone is anisotropic, meaning that the elastic and strength properties of the bone depend on the direction of the applied loading [28]. The anisotropy is evident in that cortical bone is stiffer and stronger in the longitudinal direction, which is the direction of osteon orientation, than in the transverse direction. 2.1.2.2 Bone Material Properties Material properties of a tissue are independent of its geometry and size. Thus, mechanical properties of bone as a material are determined by testing small uniform bone specimens in well-defined loading conditions. Examination of material properties typically begins at level 1 in the hierarchical structure of bone. Material properties cannot currently be assessed in vivo; however the material content can be described. The mechanical behaviour of a material is typically represented by a stress-strain curve (Figure 2). The strain is the relative deformation that occurs as a load is applied (the deformation divided by the original length), and is expressed as either a percentage of the original length or as a ratio of the amount of deformation to the original length. The stress, defined as the ratio of the applied load to the bone cross-sectional area (N/m 2 or Pa), indicates the intensity of internal forces at a particular point. The stress-strain curve is analogous to the load-deformation curve, and similar properties can be derived from it. Young's modulus is a material property which describes the resistance of a material to loading, measured as the slope of a stress-strain curve in the elastic region (Pa). Toughness (Pa), defined as the area under the stress-strain curve is the amount of work per unit volume of material that is required to reach the yield point or failure load. Brittleness is the inverse of toughness, and reflects the tendency of a bone to fracture with a small plastic region. The stress and strain at the yield point (where the stress-strain curve is no longer linear and permanent damage begins) are called the yield stress and yield strain, respectively 10 Literature Review [35]. The stress and strain at the maximum load are called the ultimate stress and ultimate strain, respectively. Strain Figure 2. Typical stress-strain curve describing the mechanical behaviour of a material. Because bone is a viscoelastic, or time-dependent, material, its stress-strain characteristics and strength properties depend on the applied strain rate [28]. There are many determinants of the material properties of cortical and trabecular bone. In cortical bone, the degree of mineralization and porosity are highly related to its material properties [33]. In addition, the histologic structure (primary or osteonal bone), collagen content and orientation, the number and composition of cement lines and the occurrence of microdamage due to fatigue may alter the material properties of cortical bone [34]. In trabecular bone the apparent density and microarchitecture of the trabeculae are the primary influences on the material properties [34]. The apparent density is determined by the number and thickness of the trabeculae, as well as the degree of mineralization. The compressive strength (measured as ultimate stress) of trabecular bone is proportional to the square of the apparent density [28]. The Young's modulus of trabecular bone is related to apparent density with a 11 Literature Review relationship between squared and cubic (i.e. Young's modulus a apparent density 2 1 0 3) [28]. Therefore, small changes in trabecular bone apparent density result in large changes in stress and Young's modulus. It is from this data that the notion that 'bone density' has a large impact on bone strength has been widely adopted. 2.1.2.3 Bone Cross-Sectional Geometry In a whole bone, the strength of the bone increases with the amount of material in that bone. However, increased material also results in increased mass which requires increased energy consumption for body movements. To optimize quantity of material for strength, within a bone cross-section, the highly mineralized cortical bone is distributed at a distance from the long axis of the bone. The properties associated with this distribution are typically referred to as the cross-sectional geometry of a bone, and are described at level 0 in cortical bone. This distribution results in a geometry of the bone that is both light, as the cross-section is not filled entirely with bone, and resistant to bending, as the bending stiffness is proportional to the fourth power of the distance from the neutral axis of the bone. The cross-sectional moment of inertia (I), also referred to as the second area moment of inertia, is the geometric parameter that represents the distribution of the cross-sectional area with respect to the neutral axis of bending (the unstressed axis of a beam) (Figure 3). An area at a greater distance from the neutral axis is more resistant to bending in that axis. 12 Literature Review a) Figure 3. Cross-sectional moment of inertia (mm4) for cylindrical beams. Reproduced from Hayes, 1991 [28]. 10a) describes a solid cylinder, 10b) describes a cylinder with a thin shell, similar to a long bone which has a dense cortical bone located at a distance from the neutral axis. The design shown in Figure 3b) is a lighter structure than Figure 3a), with minimal consequences on bending stiffness. Section modulus (S) is the parameter that is directly associated with stress in bending, and is often used as a surrogate for bending strength. Section modulus is the ratio of area moment of inertia to the maximum distance of the bone from the neutral axis (R). The equation is presented below: R4 -r4 I S = n = —, measured in mm 3 4R R In order to account for the contribution of geometric properties to bone strength for use in estimating fracture risk and evaluating treatments, images of bone cross-sections must be obtained. 2.1.3 Contributions of Cortical Bone to Strength in the Proximal Femur Cortical and trabecular bone compartments significantly contribute to bone strength, particularly in the proximal femur. Mechanically, cortical and trabecular bone can be considered the same material with variable apparent density [36]. This was determined by isolating cortical and 13 Literature Review trabecular bone samples in a study of bone material properties [36]. After adjustment for apparent density (defined in this study as the ratio of wet weight to bone tissue volume), it was determined that the ultimate stress of both cortical and trabecular bone have the same relationship with strain rate. This relationship is the apparent density proportional to the strain rate raised to the power of 0.06. This indicates that when comparing cortical and trabecular bone, the difference in strength of these two compartments results from the difference in apparent density. However, the varied distribution of the two compartments within a bone structure and between bones results in different contributions of cortical and trabecular bone to whole bone strength. Removal of the central portion of the trabecular bone in the proximal femur has been shown to reduce whole bone strength [37]. In two femur pairs tested, the trabecular bone at the center of the femoral head, neck and intertrochanteric regions was removed with a hollow drill on one side of each pair. The cortical bone remained intact on the paired side. Both sides of the pair were loaded to failure. The failure load of the intact femur was approximately 50% greater than the failure load of the contralateral drilled femur. Thus both trabecular and cortical bone are important for strength. However, the relative importance of the two compartments depends on the location within the proximal femur. A finite element analysis of the proximal femur suggested that cortical bone supports 30% of the load in the subcapital region, 50% of the load at the mid-neck, 96% of the load at the base of the neck, and 80% of the load in the intertrochanteric region during both gait and sideways falls [38]. While both cortical and trabecular bone compartments contribute to strength of the proximal femur, few researchers have investigated their relative importance in predicting proximal femur fracture. One study attempting to discern their relative importance found that biopsies of the femoral neck demonstrated no significant difference in the amount of trabecular bone in the femoral neck fracture cases compared with female post-mortem controls [39]. However, the proportion of cortical bone in the inferior, inferioranterior and anterior regions of the neck was significantly lower in female fracture cases compared with controls [39]. In addition, the mean 14 Literature Review and minimum cortical widths were lower in female fracture cases than controls in the inferoanterior, anterior and superoposterior sections of the femoral neck. Finally, it is believed that proximal femur fractures are initiated in the cortex rather than in the trabecular compartment [40]. The evidence presented here indicates that cortical bone is an important determinant of proximal femur strength and fracture. Thus, research that targets the role of the cortex in bone strength and ultimately failure is needed. Therefore, it is essential to develop imaging techniques that safely, precisely, and accurately describe the cortex in the proximal femur. 2.1.4 Bone Biology and Mechanics Summary Cortical and trabecular bone compartments are important determinants of bone strength. These compartments can be described anatomically as a hierarchical structure. At the smallest level, cortical and trabecular bone do not differ. The strength of a whole bone is influenced by its material properties and cross-sectional geometry. 2.2 Proximal Femur Fracture In this section, I will provide an overview of the epidemiology and the burdens associated with proximal femur fractures. I will then describe the role of falls in fracture causation. I finish by presenting previous work which has simulated proximal femur fractures in the laboratory. 2.2.1 Epidemiology The importance of the rapidly increasing increase in the incidence of proximal femur fractures is high. The lifetime risk of proximal femur fracture has been assessed at 17% for women and 6% for men [41]. The incidence of proximal femur fracture increases exponentially with age [1], while the incidence of distal forearm fractures plateaus after 65 years of age [42]. The incidence of fractures is increasing, it appears, without a corresponding increase in the prevalence of osteoporosis [5]. The age-adjusted rate of proximal femur fracture in Canada in 1993-1994 for women and men over the age of 65 was 479 and 187 per 100 000, respectively [1]. Based on demographic 15 Literature Review projections by Statistics Canada, there will be a substantial increase in the percentage of the population aged 65 and older, and aged 85 and older, resulting in a projected increase in proximal femur fractures between 79 000 and 104 000 by 2041 [1]. Currently the annual cost of proximal femur fracture in Canada is $650 million, and the projected cost in 2041 is $2.4 billion [2]. In other countries, similar projections are equally staggering. In California in 1998, nearly half (27 000) of all osteoporosis-related hospital discharges (55 000) were proximal femur fractures [43]. Worldwide, the total number of proximal femur fractures in 1990 was estimated at 460 000 and 1 200 000 for men and women, respectively [41]. Projected to 2050, there will be an approximately 4-fold increase in proximal femur fractures in men and women [41]. A recent Canadian study indicated that proximal femur fracture rates in Ontario have actually been decreasing slightly since 1997 [44]. The authors attributed the slight decrease to increased diagnosis and treatment of osteoporosis. This would suggest that even if the original projections were pessimistic, fracture risk identification and subsequent application of therapy can have an effect and could be improved to further decrease proximal femur fracture rates. The consequences of proximal femur fractures are more severe than the consequences of other low trauma fractures (e.g. vertebral and radial fractures). Though the incidence of vertebral fractures is estimated to be at least twice that of proximal femur fractures [45], the clinical implications including health care burden and mortality rate as a result of proximal femur fracture are more severe [46]. The mortality rate following a proximal femur fracture is the highest of any fracture, as 10% to 20% more women die in the first year after a proximal femur fracture than is expected for their age [47]. In addition to direct health care costs, health-related quality of life and independence is much lower in individuals who have experienced proximal femur fracture than those without fracture [48]. Thus, proximal femur fractures are frequent, costly, debilitating and increasingly prevalent. Development of an effective screening instrument will aid in identifying individuals at risk for fracture. 16 Literature Review 2.2.2 Clinical Fractures of the Proximal Femur The clinical consequences of hip fractures stem from the nature of the fracture. Proximal femur fractures are classified broadly into three types by anatomical region: femoral neck, intertrochanteric, and subtrochanteric (Figure 4) [49]. Femoral neck fractures are often termed subcapital fractures. These fractures lie completely within the capsule of the hip joint. The blood supply to the femoral head can be compromised placing the patient at risk of avascular necrosis [49]. Intertrochanteric fractures occur at the level of the greater and lesser trochanters, in the intertrochanteric region. Frequently these fractures are comminuted (where the bone is broken into many pieces) and displacement of the greater and lesser trochanters often occurs. Intertrochanteric fractures have a high healing potential [49]. Subtrochanteric fractures occur distal to the lesser trochanter, frequently with considerable comminution [49]. True basicervical fractures are uncommon [50], but are defined as proximal femur fractures through the base of the femoral neck at its junction with the intertrochanteric region [51]. They are typically treated as intertrochanteric fractures [50]. Isolated GT fractures also occur occasionally [52]. I 1 Figure 4. Classifications of proximal femur fracture by region. Reproduced from Zuckerman, 1996 [53]. Clinically, approximately 49% of hip fractures are intertrochanteric, 37% femoral neck, and 14% are subtrochanteric [54]. Whether different aetiology or mechanisms underlie these different fracture types is unknown [55], however a prospective study demonstrated that women with Subtrochanter ic region Intertrochanteric region \ , Grea te r trochanter 17 Literature Review intertrochanteric fractures were older than those with femoral neck fractures, and after adjustment for age, were more likely to be dependent for activities of daily living [56]. After 65 years of age, the ratio of intertrochanteric to femoral neck fractures is approximately 2.8% greater for each year of age in Caucasian women [57]. In contrast, Caucasian men and African-American men and women have shown no trend in fracture type with age. 2.2.3 Immediate Cause of Proximal Femur Fractures For fracture to occur, the applied load must be greater than the load which the bone can withstand. Approximately 90% of proximal femur fractures are associated with trauma due to a fall [53]. There is substantial evidence that falls are the immediate cause of proximal femur fracture, as falls produce applied loads that are much greater than those typically experienced in everyday activities [38]. A nested case-control study (a sample of fallers from the prospective Study of Osteoporotic Fractures) of 891 community-dwelling women over age 65 compared fallers with a hip fracture and fallers with a wrist fracture separately to fallers with no fracture. In this group of fallers, those who fell to the side or straight down were more likely to sustain fractures (Odds Ratio, OR = 3.3) [58]. Also, women who fell on or near the hip were more likely to sustain fractures (OR = 32.5). A case-control study of ambulatory community dwelling elderly fallers (N = 149), demonstrated that a fall to the side was the only fall characteristic that could significantly discriminate (OR = 4.9) between fallers with hip fracture and fallers without hip fracture [59]. Similar results were observed in a case-control study of ambulatory nursing home residents (N = 132) with similar methodology, where a fall to the side was the only fall characteristic significantly associated with hip fracture (OR = 4.6) [60]. These studies both demonstrated that sideways falls contributed to the risk of hip fracture in fallers independent of aBMD. Finally, a large multicentre prospective study of ambulatory women over 75 years of age (N = 7575) found that fall-related risk factors (small calf circumference, gait speed, visual acuity, and dynamic balance) contributed to fracture incidence (range of relative risk, RR = 1.2 to 2.0) 18 Literature Review independent of femoral neck aBMD [61]. Thus, sideways falls on the proximal femur appear to be the most common route to proximal femur fracture. A finite element model of the proximal femur suggested that the peak compressive stresses in a sideways fall occur in the superior-posterior neck and posterior trochanteric region, whereas in gait the peak compressive stresses occurred at the base of the femoral neck and medial intertrochanteric region [38]. In addition, the peak stresses that occur within the femoral neck are four times greater in a sideways fall than the stresses that occur in normal gait [38]. Thus the stresses which result from these two loading mechanisms (gait and falling) differ considerably. Fracture patterns and the association of bone cross-sectional geometry with bone strength would likely differ if a bone is loaded to simulate a fall rather than to simulate gait. 2.2.4 Laboratory Simulation of Proximal Femur Fractures As the immediate cause of proximal femur fracture and typical clinical fracture types are known, researchers have developed several methods to induce clinically relevant proximal femur fractures in the laboratory. In general, two different methodologies have been used to reproduce fractures in the proximal femur in human cadaveric specimens. The first method loads the femur in a stance configuration, with the load applied vertically to the superior aspect of the head of the femur (Figure 5) [11]. This configuration mimics the normal loading conditions in a one-legged stance, which is considered to produce the greatest stresses on the femur in normal daily activities [62, 63]. Often the fractures produced with this technique differ from those observed clinically [64]. This loading configuration would also differ greatly from what is observed clinically. 19 Literature Review Figure 5. A proximal femur loaded in a one-legged stance configuration. Reproduced from Lochmuller et at., 2002 [11]. Based on the literature presented in subsection 2.2.3, an alternative laboratory method to induce proximal femur fracture mimics a sideways fall impacting on the greater trochanter (Figure 6). The load is typically applied to either the head [65] or to the greater trochanter [11]. In a fall on the greater trochanter there is an impact to the head from the pelvis [66], however, the initial impact is on the greater trochanter. While both impact directions (load applied to the head or greater trochanter in a sideways fall configuration) have produced clinically relevant fracture types, applying the load to the greater trochanter should theoretically produce failure loads that most closely represent a clinical scenario. Figure 6. A proximal femur loaded in a sideways fall configuration. The load is applied to the greater trochanter. Reproduced from Lochmuller et at., 2002 [11]. 20 Literature Review Attempts have been made to load femora in mechanical testing in the same orientation as would occur in a sideways fall. Kinematic studies of individuals falling to the side voluntarily have shown that impact occurs when the trunk is nearly vertical (21.7° +13.3°) [67]. However, to date, no researchers have determined the orientation of the femur in a fall in vivo. Ex vivo testing of specimens in a simulated fall with specimens at internal rotations of 0°, 15° and 30° has determined that impact direction has a significant effect on failure load [68]. Specimens tested at 30° failed at a 24% lower load than specimens tested at 0°. There were no differences between loading directions and fracture patterns. Most studies that have induced fracture by simulating a sideways fall internally rotated the femur 15°, and adducted the shaft 10°, as this positioning is believed to approximate the position of the femur in a typical fall [11, 65, 69]. Testing protocols that simulate sideways falls to induce low-trauma fractures in vivo have been used extensively for the mechanical testing of cadaveric specimens. They produced fracture types similar to that observed in population based studies, although with much wider variation (Table 2). Loading rates of 2 mm/s and 100 mm/s produced similar fracture patterns and levels of comminution [65]. Thus the combination of a novel imaging instrument and mechanical testing of cadaveric specimens can together be used to develop an ex vivo fracture risk model, prior to undergoing a lengthy and costly population-based study. Table 2. Fracture types in simulated sideways falls. The references provided are examples only, and are not a complete review. In all studies, impact was applied to simulate a sideways fall on the greater trochanter (GT). Orientation of the femur varied slightly in each study. FN = femoral neck, IT = intertrochanteric. Author N Load Applied To FN IT Unknown or Discarded Courtney et al. [65] 40 Head 58% 37% 5% Bouxsein et al. [70] 16 Head 69% 31% -Eckstein et al. [69] 108 GT 67% 19% 14% Lochmuller et al. [12] 241 GT 50% 24% 26% Beck etal. [71] 22 GT 36% 64% -Keyak et al. [72] 36 Head 26% 74% 11% Includes all fractures classified as cervical, basicervical, midcervical, and subcapital, includes all fractures classified as unknown, subtrochanteric, shaft, head, or crush. 21 Literature Review 2.2.5 Proximal Femur Fracture Summary Clinically, most proximal femur fractures occur in the femoral neck and intertrochanteric regions. The immediate cause of proximal femur fractures appears to be sideways falls on the proximal femur. Methods have been devised to mimic clinically relevant proximal femur in the laboratory. 2.3 Bone Imaging to Predict Hip Fracture In the following sections I will review medical imaging techniques (DXA, QCT and MRI) that are currently used to assess bone. I will discuss quantitative ultrasound and radiography only briefly based on their limited use. For each imaging technique, I will present both relevant population-based studies and laboratory studies of bone strength that relate imaging techniques to hip fracture prediction. 2.3.1 Dual Energy x-Ray Absorptiometry (DXA) 2.3.1.1 Description Currently D X A is the imaging system used clinically to assess bone mass and is considered the 'gold' standard for osteoporosis diagnosis [10]. D X A operates on the principle that radiation of two different energies is attenuated by different tissues to varying degrees, and can thus separate mineralized and soft tissues. The value of a pixel in a D X A image represents the attenuation of a mass of hydroxyapatite along the path of the x-ray beam, expressed in g/cm2. That value is termed the areal bone mineral density (aBMD, g/cm ). An aBMD value of 0.5 g/cm , for example, is equivalent to the attenuation of 0.5 g of hydroxyapatite spread over an area of 1 cm by 1 cm. As such, aBMD is a 2-dimensional representation of a 3-dimensional structure, rather than a representation of volumetric bone density (g/cm ). D X A scans of the proximal femur are 2-dimensional coronal images. Therefore in D X A (aBMD), one dimension of geometry, the cross-sectional diameter, remains. Areal B M D also measures of bone density with D X A reflects the density of integral bone, which includes the degree of mineralization, porosity as well as the apparent density and trabecular microarchitecture. D X A can not discriminate the relative contribution of these component parts, or the independent contributions of cortical and trabecular bone. 22 Literature Review The aBMD for a region of interest is the mean value of all the pixels in that region. D X A scanners also report the projectional area of the bone in a region of interest, the area of bone in a coronal view (BA, cm ). Bone mineral content (BMC, g) is calculated from the product of aBMD and bone area. 2.3.1.2 Advantages of DXA Clinical and research use of D X A has many advantages. These include the ease of use, the ability to predict failure load in laboratory studies and the ability to predict fracture in population-based studies. D X A scanners are readily available, and relatively inexpensive for clinical use, and scans cost approximately $80 to $100 (in Ontario) [73]. The Medical Services Plan (MSP) in BC covers the cost of D X A scans when "medically necessary", but not for screening purposes only [74]. Because of the prevalence of this technology, a great deal of normative data exists. While patients are exposed to radiation during a scan, the effective dose is extremely low (1-3 uSv for common procedures of the proximal femur) [75]. In side impact tests loading femur cadaveric specimens to failure, D X A measures at the proximal femur have explained a range from 55% to 81% of the variance in failure load [12, 65, 70]. A major advantage of using D X A to predict proximal femur fractures is that the proximal femur can be assessed directly. The importance of direct proximal femur assessment is apparent when the association with failure load of proximal femur aBMD measurements are compared with assessment of peripheral sites (e.g. radius, tibia). Studies that have made these comparisons are discussed in subsection 2.3.4.5. Numerous prospective cohort studies have demonstrated that low aBMD is strongly associated with fracture risk. One classic epidemiological study demonstrated that for each SD in femoral neck aBMD below the age-adjusted mean, the relative risk of hip fracture was 2.6 [76]. A meta-analysis of prospective cohort studies measuring proximal femur aBMD showed that the relative risk for individuals with aBMD one SD below the age-adjusted mean ranged from 2.0 to 3.5 [13]. Thus, at a population level, aBMD measured by D X A is an excellent predictor for fracture risk. 23 Literature Review In prospective cohort studies, proximal femur aBMD (and/or T-score) measurements by D X A have consistently shown to be better predictors of proximal femur fractures than aBMD measurements at other sites [13, 76, 77]. Thus, direct measurements of the proximal femur are important, and can be performed with D X A . 2.3.1.3 Limitations of DXA I will highlight several limitations associated with D X A aBMD measures. These include the inability to predict fracture risk in individuals, measurement of aBMD rather than vBMD, the inability to distinguish between cortical and trabecular bone compartments, the accuracy errors and inability to identify changes in bone. A meta-analysis of prospective studies was performed to determine the sensitivity, specificity, positive predictive value, and population attributable risk percent for D X A aBMD to predict proximal femur fracture [13]. If one accepts that the relative risk of hip fracture increases 2.6 times i f aBMD is one SD below the age-adjusted mean, and that the lifetime incidence of hip fracture of 15%, the calculated sensitivity of D X A aBMD measurements to predict hip fracture (percentage of individuals correctly predicted to fracture among the individuals who actually fracture) is 37%. The specificity (percentage of individuals correctly identified as not experiencing fracture among individuals who do not fracture) is 88%. The positive predictive value (the percentage of individuals correctly predicted to fracture among the individuals identified with aBMD one SD below the mean) is 36%). The population attributable risk percent (the percent of risk of hip fracture attributable to aBMD one SD below the mean among the total population, PAR%) is 26% [13]. The high specificity of bone density assessment indicates that most individuals who are identified as not having low aBMD are highly unlikely to fracture (low rate of false negatives). The low sensitivity and PAR% of aBMD measures indicate that aBMD has less ability to identify individuals who will fracture. The Study of Osteoporotic Fractures (SOF) evaluated the PAR% for their sample population of women aged 65 years and older [77]. When osteoporosis was defined as having a Total Hip T-score < -2.5, only 28% of hip fractures were attributable to osteoporosis (PAR% = 28%, 95% CI 22-33%). When defining osteoporosis 24 Literature Review more conservatively (T-score < 1.5), the PAR% increased to 51%. In consideration of the low PAR% for low aBMD, D X A aBMD is not recommended as a screening tool [13]. De Laet et al. demonstrated that hip fracture risk increased 13-fold from age 60 to 80 years, but the decline in bone density with age only accounted for a doubling of fracture risk [78]. Thus, there are factors other than aBMD that explain the incidence of hip fracture. For women with the same aBMD level, the risk of fracture is eight to ten times higher in women over 80 years of age compared to women less than 45 years of age [79]. Again, this evidence suggests that risk factors other than aBMD that contribute significantly to fracture. Some of these risk factors may relate to bone strength, while others may relate to falls and applied loads on the bone. D X A measures aBMD rather than vBMD because D X A obtains a coronal "projectional", or "planar" image of the proximal femur. In other terms, this projectional image is a 2D image that does not provide an indication of the depth, of the bone in the sagittal plane (Figure 7). Thus, D X A measurements scale for size with projectional area rather than volume, thus aBMD lacks the third dimension, cross-sectional diameter (volume = area x cross-sectional diameter). 25 Literature Review Cylindrical representation of a long bone Height 1 Depth Length Representation of the long bone in a DXA planar image i Height Length Figure 7. Schematic of DXA image to represent bone. Consider two individuals who have the same vBMD, but different cross-sectional diameters, or depth, in the anterior-posterior direction (or cross-sectional areas). In this situation, the individual with the greater diameter will appear to have a larger aBMD as measured by DXA. Conversely, the individual with the smaller diameter will have a lower aBMD. In an extreme case, this could result in an individual with a normal vBMD being diagnosed with osteoporosis, or alternatively an individual with low vBMD but a large cross-sectional area not being diagnosed with osteoporosis. As previously mentioned, both cortical and trabecular bone compartments make independent contributions to bone strength. By imaging the bone in the coronal plane (projectional image) one cannot determine their relative difference between individuals and within individuals over time. 26 Literature Review Another issue associated with the D X A image plane are the necessary assumptions relating to soft tissue in the scan region. D X A uses the absorption of two different energies (dual-energy) to distinguish between different tissue types in an image. However, there are more than two "absorptiometrically distinguishable" tissue types in each scan - bone, lean tissue (muscle, connective tissue, etc.), extraosseous fat, and marrow (an unknown mix of red and yellow marrow) [15]. Because D X A images are coronal projections, the attenuation related to soft tissue must be subtracted from the attenuation related to bone. It is assumed throughout the region of interest (ROI) that the percentage of fat and lean mass is homogenous. In addition, the scanner cannot account for the attenuation related to the bone marrow as it lies within the bone tissue. A simulation study which varied the ratio of fat to lean tissue, actual aBMD, and the ratio of red to yellow marrow found that "measured" aBMD varied up to 30% from the true aBMD value [15]. Importantly, the inaccuracies due to soft tissue inhomogeneities were largest for the smallest aBMD values, suggesting that aBMD may be underestimated more for osteoporotic and osteopenic individuals, as well as children. A further study that performed D X A scans with phantoms representing varying ranges of soft tissue and bone sizes came to similar conclusions -that for individuals with low actual aBMD, higher yellow marrow content (typical in postmenopausal and elderly women) or low body fat percentages, the inaccuracy of D X A aBMD can be as high as 30% - 50% [80]. This study emphasized that accuracy errors may arise when the relative distribution of fat and lean tissue are not homogenous [81]. Finally, at bone sites with contributions from both cortical and trabecular bone compartments, such as the femoral neck, the thinner the cortical bone, the greater the percent error in aBMD [82]. Because of the inherent inaccuracies in D X A measurements, there are consequential limitations when evaluating differences between individuals and within individuals over time. As previously described, the marrow composition and fat to lean mass ratio can affect the aBMD values [80]. The marrow composition can vary considerably with age, drug therapy, physical activity levels, for example. In addition, these factors and dietary changes can also affect the ratio of fat to lean mass. Thus apparent changes in aBMD values in response to intervention therapies may actually be an artefact that can be explained by changes in soft tissue. In addition, differing changes in 27 Literature Review cortical and trabecular bone cannot be distinguished and may negate each other (e.g. an increase in trabecular density with a concomitant decrease in cortical density may appear as no change in aBMD). Several intervention studies demonstrated changes in other bone parameters that were not evaluated by D X A measurements. These studies are discussed below. Adami et al. [17] performed a prospective six-month controlled exercise intervention in postmenopausal women (N = 250) that targeted the distal radius. Investigators found no significant between group differences in aBMD at trial completion. However, increased cortical area (+3%), cortical B M C (+3%) and decreased trabecular B M D (-4%) were observed in the exercise group compared with the control group at the distal radius, measured by pQCT. Thus, D X A assessment was unable to represent the specific effects of exercise on bone geometry and material density in the cortical and trabecular compartments. In fact, the opposing changes observed in cortical and trabecular bone showed no effect on aBMD for the intervention. Jarvinen et al. [18] examined the effects of sudden impact loading on rat femora. They demonstrated that there were no significant differences in D X A B M C measurements between groups (sedentary, exercise, and exercise plus impact). However, when mechanical tests were performed, the failure load and energy to failure were greater in the exercise plus impact group than the sedentary group. Genant et al. [16] recently performed the first study to use quantitative computed tomography (QCT) to evaluate vertebral bone responses to a pharmaceutical intervention (raloxifene) in a subset (n = 58) of a larger sample (N = 7705). In this subset of the study sample there was no significant increase in vertebral aBMD after two years of raloxifene treatment, compared to the placebo group. However, there were significant increases in trabecular v B M D in all regions of interest evaluated by QCT. A meta-analysis of 16 published clinical trials evaluated the effects of anti-resorptive drugs on vertebral and non-vertebral fractures. While these drugs reduced the risk of vertebral and non-vertebral fractures, there was insufficient evidence to suggest that changes in spine or hip aBMD were responsible for the reduced fracture risk [83]. The authors suggested that aBMD cannot be 28 Literature Review used as a surrogate measure for fracture risk reduction. Similarly, an earlier meta-analysis of 13 trials showed that decreased risk for vertebral fracture observed in clinical trials of anti-resorptive drugs was twice as large as would have been expected from the change observed in vertebral aBMD [84]. These studies clearly demonstrate that bone strength can increase without corresponding increases in DXA measurements. 2,3.1.4 Hip Structural Analysis (HSA) Dr. Thomas J. Beck (Johns Hopkins Outpatient Centre, Baltimore, MD) developed Hip Structural Analysis (HSA) software to address the inability to evaluate cross-sectional geometry of the proximal femur with DXA images. HSA is a computer software algorithm designed to estimate cross-sectional geometry of the proximal femur [64]. The software uses bone mass profiles obtained from digital images (x- or gamma-ray, most commonly from DXA) of the hip acquired in the frontal plane (Figure 8) [64]. To derive cross-sectional properties, the bone cross-sectional diameter is computed at each bone pixel. The diameter is calculated based on the bone mass determined by beam attenuation and the assumption that vBMD is a fixed value (the average vBMD for a fully mineralized adult bone, 1.05 g/cm3). The value of each pixel (in g/cm2) is divided by the fixed vBMD (in g/cm3) to obtain the bone diameter in the sagittal plane (in cm) at the location of that pixel. The cross-sectional area is the integral of the cross-sectional diameter values. Geometric parameters assessed by HSA include cross-sectional area (combined cortical and trabecular bone), area moment of 29 Literature Review inertia, section modulus and cortical thickness. Distance (cm) Figure 8. A model of the cross-section of the femoral neck derived using HSA software from the bone mass profile obtained in a D X A scan of the proximal femur. To compute cortical thickness, HSA models the shape of the regions. The neck is modelled as a circle, with 60% of bone mass contained in the cortex. The intertrochanteric region is modelled as an ellipse with 70% of the mass in the cortex. Like DXA, use of HSA has several advantages and limitations. An early version of the HSA software predicted failure load ex vivo, evaluated with mechanical testing of 22 cadaveric femora in a stance configuration [64]. A combination of parameters assessed with HSA was used to estimate the ultimate tensile stress and the ultimate compressive stress. HSA-derived tensile stress had the highest correlation with failure load (R2 = 0.79). The correlation between aBMD at the femoral neck by DXA and failure load was lower (R2 = 0.62), although the statistical significance of this difference was not reported. A cross-sectional study of 213 postmenopausal women and 200 men (greater than 50 years of age) compared previous fracture incidence with DXA scans analyzed with HSA software [85]. 30 Literature Review Results were reported for both the standard D X A analysis and HSA. A multivariate model demonstrated that the standard femoral neck aBMD was the best predictor of osteoporotic fracture (moderate or minimal trauma fracture of the hip, spine or forearm) in women (OR = 3.3, 95% CI = 2.0 to 5.2). However there was little decline in predictive power if femoral neck buckling ratio was substituted in the model, suggesting that HSA-estimated structural variables were at least as good at predicting fracture risk. This study was limited as it included all osteoporotic fracture types (radial, vertebral and femoral). It is possible that i f a study examined hip fractures only, different relationships with HSA variables would become evident. The main limitation of HSA is that direct measurements of bone cross-sectional geometry cannot be performed. While HSA has provided a good estimate of bone geometry, the necessary assumptions limit the applicability of HSA. Because geometry is derived from D X A images, HSA is hampered by the same inaccuracies as D X A . In addition, a recent study found the HSA-derived variables had poorer short-term reliability compared to traditional D X A manufacturer-derived variables [86]. Coefficients of variation (CV) for Hologic scanners varied from 2.8% for narrow neck aBMD to 5.1% for narrow neck section modulus. In comparison, the CV for D X A manufacturer-derived femoral neck aBMD for all Hologic scanners was 2.3%. Traditional D X A measures are less sensitive to errors in positioning than HSA measures. 2.3.2 Quantitative Ultrasound (QUS) Ultrasound is used clinically to measure the velocity of ultrasonic pulse and the attenuation of the pulse through bone, typically the calcaneus. In vitro studies have demonstrated that ultrasound is not able to provide comparable or additional information to predict femur fracture compared to D X A measurement of the proximal femur [12, 70, 87]. As QUS was not utilized in this thesis, I will not discuss it further. 2.3.3 Radiography Bone geometric properties have traditionally been derived from plain radiographs [88, 89]. A nested case-control study used data from participants in the Study of Osteoporotic Fractures (46 31 Literature Review proximal femur fracture cases, 262 randomly selected age-matched controls) to estimate femoral neck-shaft angle and the length, outer diameter and thickness of the cortex at the femoral neck from pelvic radiographs acquired at baseline. Although fracture subjects had lower mean values for cortical thickness, only aBMD and height were significantly associated with hip fracture status in an age-adjusted logistic regression model [88]. In a more recent case-control study by Pulkkinen et al. [89], bone geometry was assessed in anteroposterior radiographs of the pelvis in 74 proximal femur fracture cases and 40 age-matched controls. They found that the medial calcar femoral cortex width and neck/shaft angle in combination with aBMD in the trochanteric region and Ward's triangle resulted in a greater sensitivity in predicting hip fracture (92.5%) than by trochanteric aBMD alone (52.5%). There are several possible reasons for the discrepancy between the results of Pulkinnen et al. [89] and those of Robinovitch et al. [88]. Pulkkinen et al. included aBMD in Ward's triangle in their model. Areal B M D measured in Ward's triangle is not typically reported because it cannot be assessed reliably. The radiographs used by Robinovitch et al. were obtained during baseline measurements, whereas Pulkkinen et al. obtained the radiographs after fracture. Thus the results of Robinovitch et al. provide stronger evidence that cortical geometry measures cannot predict fracture. In addition, case-control studies typically sample data from more controls than cases. Pulkkinen et al. used data from just more than half the number of controls than cases. As data from cases is easier to obtain, it is a relatively simple method to increase the confidence in the results. Finally, Pulkkinen et al. did not indicate whether their controls were randomly selected from the population, thus it is possible that there was bias in this selection. Although more research is warranted, planar radiographic measurements of bone geometry are unable to represent the distribution of bone within a cross-section and thus share several limitations with D X A measurements. In addition, in comparison with a D X A scan, there is considerable exposure to ionizing radiation with radiographs. 32 Literature Review 2.3.4 Quantitative Computed Tomography (QCT) 2.3.4.1 Description Quantitative computed tomography (QCT) measures the apparent volumetric bone density (vBMD, g/cm3) of cortical or trabecular bone at any skeletal site. These density measures are considered apparent density rather than true density because they include the density pf tissues other than bone, e.g. the pores in cortical bone. Calibration phantoms are required to convert the attenuation measured in Houndsfield units into vBMD (g/cm ). With QCT images, one can differentiate cortical and trabecular bone compartments based on density thresholds. In addition based on these thresholds, the cross-sectional geometry of cortical and trabecular bone can be determined. Geometric parameters assessed by QCT include total bone area, cortical area, cross-sectional moment of inertia and section modulus. Currently, QCT is primarily used clinically to measure trabecular v B M D in the spine [90]. Peripheral QCT (pQCT) is an instrument with a smaller gantry being used in research studies to assess v B M D and cross-sectional geometry in peripheral limbs such as the radius and tibia. Micro-CT (pCT) uses the same technology as a traditional CT scanner with a small field of view to acquire very high resolution images of small bones (e.g. small animal bones). Micro-CT will not be discussed further in this thesis as it is not currently used to measure the proximal femur or to relate its measures with proximal femur fracture risk. The advantages and limitations of QCT and pQCT will be discussed below. 2.3.4.2 Advantages of QCT Development of multislice and spiral CT systems allows the acquisition of images of the proximal femur within a few seconds and reconstruction of these images in any plane [91]. High resolution CT imaging also allows the evaluation of trabecular architecture. Helical CT systems and increasingly powerful computers may make these approaches more clinically applicable in the future. Further advantages include the magnitude of the association between measurements in the proximal femur and bone strength, as well as the precision of measurements in vivo. Cadaveric studies have been used to demonstrate the association of direct measurements of the proximal femur with failure load of the proximal femur. Lotz and Hayes [24] examined the 33 Literature Review contribution of bone vBMD and geometric properties measured by QCT in predicting failure load in 24 human cadaveric femora loaded in a fall configuration. Based on earlier finite element modelling of a fall with impact on the lateral aspect of the greater trochanter, they determined that local stresses were highest within the trabecular bone of the subcapital and intertrochanteric regions, and within the cortical bone at the base of the femoral neck [92, 93]. Thus, at each of three locations, they examined a single slice: 1) the subcapital region, perpendicular to the neck axis, at the junction of the femoral head and neck, 2) the region of Ward's triangle, perpendicular to the neck axis, at the base of the femoral neck, and 3) the intertrochanteric region in the plane that bisected the cervical axis and the plane perpendicular to the shaft axis at the superior aspect of the lesser trochanter. The highest correlation between bone strength indices and failure load were variables measured in the intertrochanteric region (R 2 = 0.48 to 0.93), and the lowest were variables measured in the subcapital region (R = 0.37 to 0.54). The product of average trabecular v B M D and total area in the intertrochanteric region explained the greatest variance in failure load (R 2 = 0.93). In 64 proximal femora, Cheng et al. [94] measured total, cortical and trabecular vBMD (in Hounsfield units) and geometry with QCT, as well as aBMD with D X A . In this study the trochanteric region in QCT was defined as one 10 mm slice perpendicular to the axis of the shaft, and the femoral neck region as the narrowest slice perpendicular to the femoral neck axis. The QCT images were acquired and analyzed in the plane of acquisition, without reconstruction. They used mechanical testing methodology similar to that described by Lotz and Hayes [24]. Trochanteric CoCSA (measured with QCT) and trochanteric aBMD (measured with D X A ) had similar associations with failure load (R 2 = 0.83 and R 2 = 0.88, respectively). To perform the measurements described by Lotz and Hayes, and Cheng et al. in vivo, it would be necessary to acquire the images in the axial plane and reconstruct the images in the plane of interest. This is more feasible with new CT scanners and more powerful computers than were previously available. Lang et al. [22] compared QCT measures of cortical and trabecular vBMD, B M C , volume and cross-sectional geometry of proximal femora with mechanical testing of human cadaveric femora in a fall loading configuration (N = 26). Unlike the two previous studies described, the images 34 Literature Review were acquired in the axial plane, and reconstructed along the femoral neck axis, as would be done in vivo. Volumes, rather than single slices, were analyzed. Three volumes of interest (VOIs) were defined based on cross-sectional area and neck axis position: 1) femoral neck - the volume within 25% and 75% of the distance between the maximum intertrochanteric cross-sectional area and maximum femoral head cross-sectional area, 2) trochanteric - the volume between the lateral femoral neck border and the lateral border of the bone, 3) total proximal femur - femoral neck VOI + trochanteric VOL Trochanteric trabecular v B M D explained the most variation in failure load (R = 0.87) [22], In a regression model of the femoral neck region, addition of cross-sectional area at the narrowest location of the neck after accounting for trabecular v B M D significantly improved the correlation between density and failure load (from R 2 = 0.59 to R 2 = 0.87). In summary, both bone geometry and density as assessed by QCT can explain a substantial proportion of variance in failure load. Based on these findings, a combination of cross-sectional geometry as assessed by MRI in combination with aBMD assessed by D X A may provide enhanced proximal femur fracture prediction; however this has not yet been evaluated. Lang et al. also evaluated precision of their volumetric QCT protocol in vivo [22]. Precision errors in vivo were quite low, ranging from 0.8% for the total proximal femur volume of interest (VOI) to 3.0%) for the femoral neck VOI. Proximal femur assessment with QCT appears to be a promising instrument to predict fracture, however, a number of limitations to its use have been identified. 2.3.4.3 Limitations of QCT There is currently no clinically accepted technique to assess the proximal femur with QCT [95]. To date, no studies have prospectively evaluated the ability of QCT to predict femur fracture, and no normative QCT data exists for the proximal femur. In individuals with low cortical thickness in some regions, the partial-volume effect (the relative abundance of pixels not filled completely with bone), may result in artificially low vBMD values [96]. 35 Literature Review The main limitation that will prevent QCT from widespread use is the radiation exposure associated with a scan of the proximal femur. "The benefits for the patient from the correct diagnosis must far outweigh any radiation risk" [97]. The Effective Dose Equivalent (H e , Sv or Rem) is the scientific term used to describe the radiation dose as it affects humans. H e is a weighted average that accounts for the relative risk associated with radiation exposure to specific tissues. H e is corrected for body attenuation, the type and volume of tissue being exposed, and the reproductive capacity of the subject. Thus it is difficult to calculate the exact H e for each type of scan, as it differs for each patient. The H e from the average annual background radiation in the United States is 300 uSv. The effective dose of a D X A whole body scan is approximately 4 pSv and about one pSv for a proximal femur scan, which is equivalent to 0.001 months of natural background exposure in Vancouver [97]. Peripheral QCT is associated with approximately 1 u.Sv per scan. For a QCT scan of the proximal femur with a slice thickness of 0.625 mm, acquired from the superior surface of the proximal femur to the inferior of the lesser trochanter, the effective dose is approximately 4 to 4.6 mSv, as calculated by a radiologist (personal communication with Dr. John Aldrich, [98]). In comparison, a one-way trip from Vancouver to Halifax is associated with 0.02 mSv. Radiation exposure is an issue in research studies where benefits to the participants may not exceed risk of exposure. Prospective studies serve as the gold standard to determine whether bone assessment techniques can predict fracture. To my knowledge, no studies have prospectively or retrospectively evaluated the ability of proximal femur assessment with QCT to predict hip fracture in a large population trial. 2.3.4.4 Advantages of pQCT Peripheral QCT is becoming an increasingly common research instrument, however the advantages and disadvantages of this instrument which are different from a traditional QCT assessment. Peripheral QCT scanners measure the peripheral skeleton, such as the radius and tibia, as the gantry size of in vivo models is considerably smaller than a traditional CT scanner. Peripheral QCT evaluates the same density and geometric parameters as QCT. The effective 36 Literature Review radiation dose associated with a typical pQCT scan is considerably lower than a traditional CT scan of a central region, at approximately one pSv per scan. 2.3.4.5 Limitations of pQCT The major limitation in predicting proximal femur fracture with pQCT is the inability to assess the proximal femur. With the exception of a sample of Japanese women [99], the small gantry sizes and orientation of pQCT scanners have prevented its use for measurement of the proximal femur in vivo. There are several limitations associated with using peripheral measurements to predict proximal femur fractures. Recent studies have examined the ability of assessment of bone structure and density as assessed in peripheral limbs of human cadaveric samples to predict femur fracture. Augat et al. [23] compared the ability of cortical vBMD, cortical cross-sectional geometry and trabecular v B M D assessed at the radius, lumbar vertebrae and femoral neck by pQCT to predict failure load at the femoral neck in 20 embalmed human cadaveric specimens. The femoral neck region of interest was composed of five consecutive 3.0 mm slices. The best predictor of proximal femur failure load, tested in a stance configuration was femoral neck cross-sectional moment of inertia about the minimum axis ( I m j n ) (R 2 = 0.71). I m i n measured at the 30% site at the radius explained 76% of the variance in proximal femur failure load - higher than direct measurement of the femoral neck, although this difference may not have been significant. This result may be partially explained by the methodological limitation of assessing the femoral neck with pQCT. Peripheral QCT is not designed to evaluate the proximal femur, as it is difficult to position cadaveric specimens in the gantry. Lochmuller et al. [11] measured total, cortical and trabecular v B M D , B M C and geometry, as well as trabecular v B M D and B M D with pQCT at peripheral sites (radius, tibia) and in the proximal femur (femoral neck) in 105 embalmed human cadavers. Hip axis length was also measured from radiographs. The femora were loaded to failure in both stance and fall configurations. I will present results for femora tested in the fall configuration only. The greatest amount of variance in failure load was explained by a combination of hip axis length, femoral neck trabecular density and femoral neck polar moment of inertia (R 2 - 0.60). The greatest amount of variance explained 37 Literature Review in failure load from peripheral measurements was cortical thickness at the distal radius (R = 0.42), and the combination of trabecular cross-sectional area and total B M D at the distal tibia (R = 0.42). They did not state whether the predictive ability of direct femur measurements were statistically greater than the predictive ability of peripheral measurements. However, it appeared that direct measurement of the femoral neck using pQCT better predicted proximal femur failure load than peripheral measurement by pQCT. Finally, Lochmuller et al. [12] related total, cortical and trabecular vBMD, B M C and geometry at the distal radius measured with pQCT and calcaneal speed of sound, broadband ultrasound attenuation and stiffness index measured with QUS, with aBMD measurements at the proximal femur and radius (the spine was also included, but I will not discuss it here). One hundred and twenty human cadaveric femora were tested in a fall configuration. Cortical thickness at the radius and broadband ultrasound attenuation at the calcaneus explained considerably less of the 2 2 variance in femur failure load (R = 0.41 and 0.28, respectively) than femoral neck aBMD (R = 0.53). Direct measurement of aBMD at the proximal femur better predicted proximal femur failure load than peripheral measurements of any kind. The reason for the discrepancy in predictive ability of direct measurements of the femoral neck (vBMD and geometry) with pQCT reported by Augat et al. [23] and Lochmuller et al. [11, 12] is unclear. However the loading configuration (stance in Augat et al. vs. fall in Lochmuller et al.) may have contributed to the differences reported. In addition, given the sample size used by Augat et al., the differences in peripheral vs. femoral neck prediction (R = 0.76 vs. 0.71) reported by Augat et al. were unlikely to be statistically significant. Overall, measurements of properties of the proximal femur appear to be more highly associated with proximal femur fracture than measurements of properties in peripheral bones. Augat et al. [100] later performed a cross-sectional study to compare hip fracture cases vs. controls. They measured total, cortical and trabecular vBMD, B M C and geometry at the distal radius with pQCT, aBMD and B M C at the proximal femur with D X A , and speed of sound at the tibia with QUS. They showed that odds ratios for hip fractures cases based on hip aBMD in various subregions ranged from 3.0 to 3.8, whereas vBMD, B M C and geometry measurements at 38 Literature Review the distal radius and speed of sound at the tibia were associated with odds ratios ranging from 1.0 to 1.8. Therefore, direct measurement of the hip with D X A could better discriminate hip fracture cases from controls than peripheral measurements of separate bone compartments with pQCT or speed of sound with QUS [100]. In conclusion, with the exception of one study [23], direct assessment of the proximal femur with pQCT or D X A provided the best predictors of proximal femur fracture. The importance of site-specific measurement is explained in part by Wolffs law - bone structure adapts through local responses to the local stress environment [101]. As discussed earlier, one of the main advantages of MRI is its capacity to assess the proximal femur directly. 2.3.5 Magnetic Resonance Imaging (MRI) 2.3.5.1 Description Magnetic resonance imaging (MRI) has traditionally been utilized for the evaluation of soft tissue. MRI is increasingly becoming available clinically, with seven MRI systems currently being operated for clinical use in Vancouver. More recently it has been used to assess bone cross-sectional geometry. The key characteristic of MRI is the application of a strong magnetic field that causes the nuclei with an odd number of protons and/or neutrons to align themselves with the magnetic field and to rotate at a particular frequency known as the Larmor frequency. Typically hydrogen is used as the proton of interest because of the abundance of water in the human body. When a radiofrequency is transmitted at the Larmor frequency, the nuclei will absorb the energy, become excited and then relax by emitting radio waves. The energy emitted as the nuclei relax is the signal that can be detected by a radio receiver. In the resulting image, the brightness of a particular pixel depends on the way mobile protons' response to the static and fluctuating magnetic fields. These responses are a result of the chemical and biophysical environment of the protons, and thus vary for each tissue. The T l (spin-lattice) relaxation time reflects the rate of return of nuclei to alignment with the static magnetic field. In other words, the T l of a tissue depends on how easily the protons can give off 39 Literature Review their energy to the surrounding lattice, or to absorb energy from the lattice [102]. The T2 (spin-spin) relaxation time reflects the relationship of the proton to the surrounding nuclei, orthogonal to the magnetic field. The pixel intensities in an image depend on the tissue scanned, as well as the scan parameters used. In a Tl-weighted scan, the pixel intensity depends on the T l characteristic of the tissue in that pixel. MRI pulse sequences are the 'recipe' for application of radiofrequency (RF) pulses, the sequencing of gradient pulses in all 3 directions, and for the acquisition of the signal [103]. Various pulse sequences (i.e. manipulations of the echo time and repetition time) can be applied to vary signal intensities from tissues. The separation of the 90° RF pulses which tip the nuclei into the transverse plane that are applied to the sample is called the repetition time (TR), and the separation of the 90° and 180° pulses which cause formation of successive signals for detection is called the echo time (TE). A Tl-weighted pulse sequence uses short TR and TE to elicit differences based on T l times, causing fat to have the highest intensity signal, solid tissue to have an intermediate intensity signal and water to have the lowest intensity signal [102]. This is the sequence which my lab has used to examine bone in vivo [25, 104]. Cortical bone has a long TI and short T2, which both produce low signal intensity [102]. In contrast, muscle has intermediate T l and T2, while fat has short T l and short T2, resulting in relatively higher signal intensity. Therefore cortical bone can be distinguished from surrounding fat and muscle with a Tl-weighted sequence because of bone's relatively longer T l time. 2.3.5.2 Signal to Noise Ratio in MRI The signal-to-noise ratio (SNR) in MRI depends on properties such as the strength of the magnetic field, the field of view and coil sizes, and the pulse sequence used (Table 3). Use of a quadrature coil, such as has been used in previous in vivo studies of the tibia, increases the SNR by a factor of the square root of 2 by placing two receivers present 90° to one another [102]. An MRI system with a stronger magnetic field (e.g. 3 Tesla rather than 1.5 Tesla) should produce images with a higher SNR. Because the cortical shell is so thin in the femoral neck (a pQCT 40 Literature Review study reported 1.27 mm on average in females [11]), improvements in SNR may increase the ability to separate cortical bone from surrounding tissue. Table 3. The effect of changing MRI properties on the signal-to-noise ratio. Based on Pooley et al, 2001 and Hashemi et al, 2004 [102,103|. Property | SNR Magnetic Field T Field of view 1 Coil size 1 TR T TE 1 # of Averages T Phase Matrix i Frequency Matrix 2.3.5.3 Advantages of MRI One major advantage of MRI is the ability to obtain images in any plane, as the RF can be applied and detected in any direction. Cross-sectional images can therefore be produced perpendicular to the shaft of any long bone, or along any desired axis. For example, images perpendicular to the long axis of the femoral neck can be obtained, producing cross-sections of the femoral neck. The operator is able to adjust the alignment of the axis with a localizer view prior to scanning. In comparison, direct measurement along any axis is not possible with pQCT or QCT, although image reconstruction allows the images to be realigned in any plane after scanning, provided the scanning parameters are nearly cubic voxels. A second advantage is that currently, MRI provides the only assessment of bone without exposure to ionizing radiation. A magnetic field strength of 4 Tesla is currently accepted as the limit below which no biological effects will occur [103]. While time-varying magnetic fields have the potential to induce electric currents in biological tissue, particularly in the heart and nervous system, the rate of magnetic field change currently used is lower than the threshold for biologic effects (less than 400 T/s) [105]. The absence of ionizing radiation when using this 41 Literature Review measure provides the opportunity to safely assess bone geometry. Finally, cortical bone geometry and trabecular architecture can be assessed with MRI. MRI has been used to evaluate cortical bone geometry primarily in paediatric populations. Most of these studies have measured cortical bone geometry at sites with primarily cortical bone. A series of studies by a group of researchers in Sydney, Australia have evaluated total, cortical and medullary cross-sectional areas and volumes in the middle third of the femoral shaft to compare prepubertal children and young adults [106] and to compare bone cross-sectional properties between young athletes in various sports (swim, bike, run, triathlon, aged 15-18) and controls [107]. Heinonen et al. [104] performed comparisons of the relationships between bone cross-sectional geometry and muscle in the tibia at a site 60% of the tibial endplate in prepubertal and early pubertal girls (N = 17). This was the first study to examine both bone and muscle cross-sections with MRI. Bass et al. [108] examined the maturity effects (pre-, peri- and post-puberty) of loading in females (N = 47) through comparisons of the loaded and unloaded humeri in competitive tennis players. To my knowledge, Arokoski et al. [109] were the first to assess femoral neck cross-sectional geometry with MRI. They measured total cross-sectional area only, and did not attempt to segment cortical bone. They acquired M R images in vivo in the coronal plane with a 1.5 T scanner, using a T l F L A S H 2D sequence, TR = 770 ms, TE = 11 ms, slice thickness 3.0 mm, and in-plane resolution = 1.09 x 0.82 mm. These images were acquired to match the D X A femoral neck region of interest. Images were reconstructed from the coronal plane into axes parallel and perpendicular to the femoral neck. The quality of the images provided in the manuscript was surprisingly high given the image resolution and need for image reconstruction. They calculated the total volume of the femoral neck by measuring the area and taking the sum of the volume of five slices in the femoral neck. Our lab [25] examined femoral neck total and cortical cross-sectional areas with MRI, and compared D X A , HSA and MRI measures in early and pre-pubertal girls (N = 18). Participants 42 Literature Review were scanned in a 1.5 T MRI with a torso-phased array coil, TR = 600 ms, TE = 14 ms, slice thickness = 3.0 mm, and in-plane resolution 0.49 mm 2, with images acquired perpendicular to the femoral neck axis. This protocol did not require reconstruction. The study showed positive changes in bone structure with both HSA and MRI over seven months. The precision portion of this study will be described below. This protocol was different from that performed by Arokoski et al. [109] in that the images were acquired directly in the plane of interest, and with much higher in-plane resolution, which should have produced higher quality femoral neck images. Thus, very few MRI studies of bone geometry have been performed to date, and there is no standard protocol, particularly for evaluation of the proximal femur. The studies that have been performed have been able to demonstrate changes in bone geometry with growth and exercise that could not otherwise have been assessed. However the relationship of these geometrical parameters with bone strength and fracture risk is unknown. The accuracy of cortical bone measures with MRI has been evaluated in two studies in phantoms and at sites with primarily cortical bone. Woodhead et al. [26] compared measurements from M R images of phantoms and the middle third of the shaft of venison femora (acquired in a 1.5 T MRI with a body coil with a proton-density weighted 2D turbo spin echo sequence, TR = 1600 ms, TE = 15 ms, slice thickness = 6 mm, in-plane resolution = 0.488 mm ) with gold standards for width, cross-sectional areas and volumes. In custom-made cylindrical phantoms, the mean percent difference between MRI measurement and the gold standard (measurement by vernier caliper), was low for all measurements, ranging from - 0.4% (SD 2.6%) for cortical width to - 2.3%) (SD 1.5%) for cortical cross-sectional area. In the venison femora, differences were computed between MRI-derived values and the gold standard (volumes by water displacement, total width and cortical width by vernier caliper, total cross-sectional area and cortical cross-sectional area by digital images). Accuracy was high for total (1.8% difference, SD 2.6%) and cortical (1.6% difference, SD 4.2%) cross-sectional areas. For cortical width, accuracy was much lower and significantly different from the gold standard (- 16.4% difference SD 14.9%). While accuracy appears to be high for measures of cross-sectional areas, these accuracy measurements were performed at sites with only cortical bone. 43 Literature Review A recent study by Kontulainen et al. (unpublished data) evaluated the accuracy of cross-sectional geometry measures in human cadaveric distal tibia at a site 25% of the distance from the distal tibial endplate [110]. The M R images were acquired with a quadrature head coil in a 1.5 T MRI with a spin-echo scan sequence, TR = 600 ms, TE =14 ms, slice thickness = 3.0 mm, in-plane resolution = 0.5 mm 2. MRI measurements of ToCSA and CoCSA were compared to direct histomorphometric measurement of the same site. MRI measured both ToCSA and CoCSA accurately, slightly overestimating ToCSA by 0.1% and slightly underestimating CoCSA by 0.7%. While these results demonstrate the promise of MRI as an instrument to assess cortical bone, the femoral neck contains a significantly greater portion of trabecular bone and a thinner cortex (1.27 mm, SD 0.74 mm for women [11]) than the 25% site of the tibia (3.71 mm, SD 0.92 mm, unpublished data) and the midshaft of the femur (2.3 mm, SD 0.67 mm for women [11]). Therefore, it is inappropriate to apply the results from this distal tibia study to the proximal femur. Accuracy and validity of MRI measurement of bone geometry at the proximal femur have not been evaluated to date. Reliability of MRI evaluation of bone geometry was evaluated in the middle third of the femoral shaft in 13 participants with a wide age range (9 to 82 years), three of whom were diagnosed with osteoporosis according to the WHO criteria [26]. The scan protocol used was the same as described previously for Woodhead et al. [26] for the femoral shaft. They evaluated short-term reliability by performing scans on two occasions, intra-rater repeatability by analyzing each scan twice, and inter-rater repeatability as a comparison between observers. Intra-rater coefficient of variation values for all subjects ranged from 0.5% (SD 0.5%) for total volume to 3.1% (SD 3.1%) for cortical width in the femoral shaft [26]. Inter-rater coefficient of variation values for all subjects ranged from 0.6% (SD 0.5%) for total volume to 3.6% (SD 3.1%) for cortical width in the femoral shaft [26]. Again, these measurements were performed at a site with only cortical bone. Our lab evaluated in vivo precision of cortical bone measurements with MRI at the femoral neck in ten participants (age range 13 to 46 years) [25]. Participants were scanned three times: 1) 44 Literature Review initial scan, 2) second scan without repositioning, and 3) third scan after repositioning. Precision was also compared for analysis of 1) a single slice (the slice with the smallest area), and 2) multiple slices (the mean of the smallest slice and the two adjacent slices). Without repositioning, the coefficient of variation (CV) for total area was 1.5% for analysis of a single slice and 1.3% for analysis of multiple slices. With repositioning, the CV was slightly higher, 4.2%) for a single slice and 2.5% for multiple slices. For cortical area the C V were high without repositioning (7.7% for a single slice and 6.1% for multiple slices) and with repositioning (13.9%o for a single slice and 9.7% for multiple slices). They concluded that reporting the average of three adjacent slices was much more precise than reporting area of a single slice. The high C V for cortical area measurements suggests that precision was too low to assess cortical area. In conclusion, MRI has been shown to be an accurate and reliable instrument to measure bone geometry in sites containing primarily cortical bone. However, in the femoral neck, cortical bone assessment is not reliable with the protocol used previously. In addition, accuracy of MRI measures of bone cross-sectional geometry in the femoral neck has not been determined. To my knowledge, MRI assessment of femoral neck geometry has not previously been compared to mechanical outcomes such as fracture incidence in vivo or failure load ex vivo. MRI has been used by several research groups to assess trabecular bone architecture. MRI has been to used to quantify trabecular bone structures such as trabecular bone volume fraction, trabecular thickness, trabecular spacing, trabecular number, connectivity, fractal dimension, tubularity and maximal entropy [91]. To obtain these parameters accurate segmentation of the bone and bone marrow compartments were required. Majumdar et al. [ I l l ] introduced the concept of "apparent" trabecular bone network which acknowledges that the trabecular structure reconstructed from high resolution M R does not reflect, but is highly correlated with, the true histologic structure. The inability to assess true histologic structure occurs because the spatial resolution of MRI images is comparable to the thickness of trabecular bone. This results in a substantial partial volume effect, where a particular voxel could represent both bone and marrow [HI]-45 Literature Review Link et al. were the first to examine trabecular architecture with MRI. They compared D X A aBMD measurements and M R structural imaging with mechanical testing in a stance configuration in 31 human cadaveric specimens [91, 112]. High-resolution images were obtained in the femoral head, neck and trochanter with a field of view 75 x 100 mm, in-plane resolution 0.195 x 0.195 mm and slice thicknesses of 0.3 mm and 0.9 mm. Trabecular bone volume fraction in the femoral head, neck and trochanter were highly associated with failure load (R = 0.69, 0.47 and 0.62, respectively). Multiple linear regression models combining D X A femoral neck aBMD and MRI femoral head trabecular bone volume fraction, trabecular thickness and trabecular model explained more of the variance in failure load (R 2 = 0.86) than femoral neck aBMD (R 2 = •y 0.61) or MRI parameters (highest R = 0.45 for femoral head trabecular bone volume fraction) alone. The high predictive ability of the femoral head in this study is surprising as the femoral head is not typically a location of clinical fracture. This study indicated that trabecular architecture is highly correlated with bone strength. While assessment of trabecular architecture with MRI appears to predict failure load in cadaveric specimens, technical difficulties such as the need for a small field of view currently limit the utility of these measurements in vivo at the proximal femur. Trabecular architecture has been . studied extensively in vivo at the distal radius and calcaneus, as these are sites with considerable trabecular bone and which are accessible for the specialized coils required, and subjects are easily able to tolerate the immobilization required [113]. However, this requires peripheral measurements to predict proximal femur fracture. Finally, analysis of trabecular structure alone ignores contributions of cortical bone geometry to whole bone strength. 2.3.5.4 Limitations of MRI The limitations of MRI as an instrument to evaluate bone geometry include the lack of signal from cortical bone, the inability to assess bone density, and practical issues such as cost. The signal emitted by cortical bone is very weak and nearly undetectable because of the low proportion of free electrons, and thus it appears black on an image. As a result, the relaxation times are not measurable or are extremely short, and the detection of edges of cortical bone in vivo are obtained from the contrast provided by the surrounding soft tissue (grey) and marrow 46 Literature Review (white), for the periosteal and endocortical surfaces, respectively. Ex vivo, with the soft tissue removed, the bone specimen must be placed in a water bath, as air also appears black. The water produces a grey colour in the image that provides contrast with the black periosteal border of the cortex. Because MR images are a grey scale representing hydrogen nuclei rather than attenuation of radiation, the density of the structures cannot be evaluated. Thus, any assessment of fracture risk using MRI may be enhanced by DXA assessment of aBMD. Practically, the cost of MRI is high (approximately $600 CDN per scan), and access to MRI time is limited due to clinical demands. It is estimated that approximately 10% of patients will suffer from claustrophobia during MRI scanning. Individuals would be excluded from scanning if they had any contraindications for MRI, such as implanted medical devices, or foreign bodies such as metal shavings in the eyes. In addition, the time required to acquire a high resolution MRI scan can be considerable (e.g. several minutes) in comparison to scan acquisition with the same resolution with QCT (e.g. thirty seconds on a multi-slice scanner). Thus, a patient is required to lie without moving for a considerable amount of time. Similar to CT, there are no normative data for MRI-derived variables. Prospective studies serve as the gold standard to determine whether bone assessment techniques can predict fracture. To my knowledge, no studies have prospectively or retrospectively evaluated the ability of proximal femur assessment with MRI to predict hip fracture in a large population trial. Fracture risk screening with MRI would not be associated with negative long-term health risks to the patient. Thus changes in bone geometry with age and exercise or other interventions could be safely evaluated longitudinally. However, an understanding of how MRI-derived bone geometry relates to fracture is required to ensure this instrument is used to its full capacity. 2.3.6 Summary of Limitations in Existing Bone Imaging Technologies Each medical imaging technique has limitations in its ability to assess bone and/or its practicality for use in population-based fracture prediction and/or research studies. 47 Literature Review DXA • Measures density based on projectional bone area • Cannot distinguish between cortical and trabecular bone • Cannot measure cross-sectional geometry • Accuracy affected by soft tissue thickness HSA • Estimates bone cross-sectional geometry from DXA images • Makes many assumptions to derive structural properties pQCT • Limited to measuring peripheral limbs in vivo QCT • Exposure to ionizing radiation (to a greater extent than DXA) MRI • Cannot measure density • Accuracy at the proximal femur and ability to predict fracture is unknown Determination of the association of bone geometry measurements measured with MRI with failure load of the proximal femur is a novel question. 2.4 Age-Related Differences in Bone In the following subsections, I will describe age-related differences in the mechanical properties of bone that occur as a result of differences in material and geometry. Riggs et al. summarized the three major processes that contribute to bone loss: 1) decrease in trabecular vBMD through trabecular thinning, disruption of trabecular microstructure and loss of trabeculae; 2) resorption at the endocortical surface, resulting in expansion of the marrow cavity or trabecular area; and 3) decrease in cortical vBMD, most likely through increased porosity from an increase in resorption cavities, and increased number of incompletely closed osteons [7]. As technologies to measure bone improve, the understanding of age-related changes in bone will continue to improve. While most studies to date provide cross-sectional data regarding age-related differences in bone geometry and density, no longitudinal studies across the entire lifespan have been performed. Descriptions of age-related differences in bone material properties require destructive testing and 48 Literature Review therefore cannot be studied longitudinally. Thus, one is unable to assess age-related changes in bone properties although many cross-sectional studies frequently use the term "change" to describe cross-sectional differences between age groups. I will discuss these age-related differences in more detail and are described quantitatively below, with a focus on the differences in the proximal femur. 2.4.1 Age-Related Differences in Bone Density Differences in aBMD between young and old individuals are greater in sites with trabecular bone (such as the femoral neck) than in long bone shaft sites that are comprised primarily of cortical bone [114]. To date, Riggs et al. [7] performed the only study that assessed age-related differences in v B M D with QCT. This was a cross-sectional study of 696 individuals which performed QCT scans at the lumbar spine and femoral neck, as well as pQCT scans of the distal radius and distal tibia. I present results for the femoral neck only. Researchers found that femoral neck cortical vBMD in women over age 80 was 24% lower than in women aged 20 to 29 years, and 13% lower in same aged men. Femoral neck trabecular v B M D was 56% lower in women over 80 years, compared with the 20 to 29 year age group, and 45% lower in same-aged men. This explains that differences in aBMD, where cortical and trabecular bone compartments are combined, are greater at bone sites with trabecular bone, as the difference between older and younger individuals is much greater for trabecular vBMD than cortical vBMD. 2.4.2 Age-Related Differences in Proximal Femur Cross-Sectional Geometry Changes also occur with age in the distribution of bone about its cross-section. HSA was used by Beck et al. [114] to estimate differences in the cross-sectional geometry of the femoral neck and shaft across the lifespan using D X A hip scans acquired from a cross-sectional study of 5603 non-Hispanic whites in the Third National Health and Nutrition Examination Survey (NHANES III). Data were presented as means by decade beginning at 20 years of age to 80+ years of age. For men, weight-adjusted section modulus in the femur shaft was 9% greater in men over 80 years compared with men aged 20 to 29 years. However weight-adjusted section modulus in the femoral neck was 4.5% lower in men over 80 years compared with men aged 20 to 29 years. This difference was small compared to a 22% lower narrow neck aBMD in the same study reported for the older age group of men. 49 Literature Review For women, weight-adjusted section modulus in the femur shaft was 3% lower in women over age 80 years compared to women 20 to 29 years, and 13% lower in the femoral neck. However there was little difference between the younger female age groups until after age 49, implying that the decline may not begin until 50 years of age. Similar to men, the difference in section modulus between young and older women is small compared to the 32% lower narrow neck aBMD observed in women over 80 years compared with women 20 to 29 years in the same study. The lesser difference between ages for section modulus compared to aBMD suggests that periosteal expansion may serve to partially compensate for the loss in bone strength due to loss in aBMD. Beck et al. did not examine changes in bone cross-sectional area, however they did report changes in bone widths [115]. Subperiosteal width and endocortical width were greater in older than younger individuals in both sexes at the femoral shaft and neck. Cortical thickness was narrower in the older compared with younger subjects in both sexes at the femoral shaft and neck. This suggests that although periosteal expansion appears to continue with aging, it is of lesser magnitude than medullary expansion, which results in thinning of the cortex. The cross-sectional QCT study by Riggs et al. discussed previously also examined age-related differences in cross-sectional geometry [7]. Results for the femoral neck only are presented below. Total and trabecular cross-sectional areas were greater for 90 year old women (13% and 30%), respectively) and men (7%> and 25%), respectively) compared with same sex participants aged 20 to 29, suggesting that both periosteal and endosteal expansion occur at the femoral neck with age. As differences in trabecular cross-sectional area were greater than differences in total cross-sectional area, cortical cross-sectional areas were 23% and 14% lower for 90 year old men and women, respectively, compared with same sex participants aged 20 to 29. This again supports the notion that with aging, medullary expansion occurs at a greater magnitude than periosteal expansion, resulting in decreased cortical area with age. Periosteal expansion with age may represent the whole bone's attempt to accommodate for decreased mass by enhancing geometry. Periosteal apposition places bone further from the 50 Literature Review neutral bending axis, thus improving the bending strength of the whole bone. However, recent histological examination of femoral neck sections showed that, on average, only 20% of the femoral neck surface was covered with cellular periosteum [116]. In contrast, the amount of cellular periosteum on the femoral shaft was at least two fold higher in each sample compared to its femoral neck. This may indicate that periosteal expansion does not occur to the same extent at the femoral neck as compared with other anatomic regions [116]. 2.4.3 Age-Related Differences in Bone Material Properties In addition to changes in bone density and geometry, the material properties of cortical and trabecular bone change with age. Bone becomes more brittle, as it is able to absorb less energy before failure. The ultimate stress of human cortical bone in compression is 2.5% lower every decade after age 20, while the ultimate strain is 5% lower each decade and the energy to failure is 7% lower each decade [117]. Tensile strength and modulus are approximately 2% lower each decade from age 20 to age 90 [117]. With age, cracks in cortical bone are more likely to be initiated and propagated. This is evidenced by the lower stress required to initiate a macrocrack in older individuals, as the critical stress intensity level (K c) is approximately 3.7% lower per decade [118]. Once a macrocrack is initiated, the energy required to propagate that crack also decreases with age, by 8.7% per decade [118]. The material properties of trabecular bone also decline with age [34]. The thickness of horizontal trabeculae in the lumbar vertebrae are significantly lower in older than younger individuals, while there is no difference in the thickness of vertical trabeculae [119]. The number of both horizontal and vertical trabeculae are lower in older than younger individuals, while separation between trabeculae is greater in older than younger individuals [119]. As a result of the discrepancy in age-related differences between vertical and horizontal trabeculae, the degree of anisotropy is also greater in older than younger individuals. It is unknown whether these properties are similar in the proximal femur. 2.5 Summary of Literature Review and Limitations in Existing Knowledge Proximal femur fractures are a serious health and economic concern. Cortical cross-sectional geometry contributes to bone strength. D X A , the current gold standard for bone assessment, is 51 Literature Review unable to assess these properties. MRI is currently the only instrument that has the potential to measure cortical cross-sectional geometry in the femoral neck without exposure to ionizing radiation. However, the relationship between MRI measures of cross-sectional geometry and proximal femur fracture is unknown. 52 Research Questions 3. RESEARCH QUESTIONS 3.1 Rationale, Objectives and Hypotheses for Primary Aim The incidence of hip fracture in Canada is increasing, and the health and economic consequences are severe [1, 47]. An instrument that improves fracture risk prediction might provide a partial solution for offsetting this burden. Currently D X A assessment of aBMD is the gold standard used to identify individuals at risk for fracture [10]. However, D X A is incapable of characterizing bone geometry or of separating cortical from trabecular bone [19], and is susceptible to many inaccuracies [82]. Thus it is not possible to characterize the specific bone characteristics that underpin D X A ' s ability to predict fracture or to characterize the contributions of bone geometry to bone strength. QCT has been shown to evaluate geometric properties that are independently associated with failure of the proximal femur [22, 92], however the radiation exposure associated with these CT scans limits its use in longitudinal studies. Peripheral QCT is capable of measuring bone geometry with less radiation exposure, however, its use is limited to peripheral limbs such as the radius and tibia and it is unable to evaluate the more clinically relevant proximal femur. Direct measurements of the proximal femur explain a greater variance in femur failure load [11, 23] and better predict hip fracture cases than peripheral measurements of the radius or tibia [77]. Therefore, an instrument that might be used to screen for hip fracture and monitor changes must measure the proximal femur directly. This instrument would likely be enhanced by characterizing bone density and cross-sectional geometry rather than either alone. MRI is a instrument that can accurately assess bone geometry in the femoral shaft without the patient being exposed to ionizing radiation [26]. MRI also reliably assesses total area in vivo at the femoral neck [25]. Therefore, the primary aim was to determine the feasibility of using MRI to assess femoral neck geometry and its role in predicting strength of the proximal femur. 53 Research Questions Primary Objectives: 1) To determine whether MRI assessment of femoral neck geometry is associated with the failure load of cadaveric femora tested in a fall configuration. 2) To compare associations of MRI measures of femoral neck geometry with failure load with the associations of D X A or HSA associations with failure load. 3) To determine whether the combination of MRI assessment of femoral neck geometry and D X A assessment of aBMD or B M C explains a greater proportion of variance in failure load than either assessment alone. 4) To determine whether the combination of MRI assessment of femoral neck geometry and D X A assessment of aBMD or B M C explains a greater proportion of variance in failure load than HSA assessment of proximal femur geometry and aBMD. Hypotheses: 1) MRI measures of femoral neck bone geometry (total cross-sectional area, cortical cross-sectional area, area moment of inertia and section modulus) will show significant, positive correlations with failure load in mechanical testing. 2) MRI measures of femoral neck geometry will be comparable in predictive ability of D X A and HSA measures of aBMD, and superior to the predictive ability of HSA measures of cross-sectional geometry. 3) Hierarchical linear regression models that include geometrical parameters from MRI and aBMD from D X A will explain greater variance in failure load (R 2) linear regression models comprised only of variables measured with each of the following instruments alone: a) D X A b) HSA c) MRI The results of this study will provide knowledge regarding the utility of MRI assessment of femoral neck bone geometry in an older population. 54 Research Questions 3.2 Rationale, Objective and Hypothesis for Secondary Aim MRI is a novel instrument only recently adapted to evaluate bone geometry. The only study that evaluated reliability of femoral neck cortical geometry assessment with a 1.5 T MRI system reported that cortical area could not be measured reliably [25]. A 3 T MRI offers improved signal to noise ratio and enhanced image quality. Improved image quality should in turn, enhance the accuracy and reliability of cortical bone measurement compared with 1.5 T MRI. Because cortical bone makes an important contribution to bone strength [11, 92] it is important to include a reliable assessment of cortical bone geometry in a fracture prediction model. Thus, I will also compare bone geometry outcomes in the femoral neck between a 1.5 T MRI and a 3 T MRI. Secondary Objective: 1) To compare bone cross-sectional geometry assessed with a 3 T MRI system with the same parameters assessed with a 1.5 T MRI system. 2) To compare reliability of cross-sectional geometry assessed with a 3 T MRI system with the reliability of the same parameters assessed with a 1.5 T MRI system Hypotheses: 1) There will be a significant correlation between independent variables measured by 1.5 T and 3 T MRI. 2) Reliability of image acquisition and image analysis will be greater for variables measured with a 3 T MRI system than with a 1.5 T MRI system. 55 Methods 4 METHODS 4.1 Specimens Thirty-eight unembalmed, previously frozen, human proximal femur specimens were obtained from the Faculty of Medicine, University of British Columbia (n = 20) and LifeLegacy Foundation (Tuscon, Arizona) (n = 18). The specimens were stored in a freezer at - 20° C until their use. Conventional anteroposterior (AP) radiographs were obtained using standard femur protocols. A radiologist (Dr. Bruce Forster) screened the films to identify any metabolic bone disease or previous proximal femur fracture. One specimen had a femoral neck fracture, thus it was excluded from the study. Therefore, 37 specimens were included in the study. My sample included 12 specimen pairs (left and right limbs from the same donor), as well as 12 specimens for which only one limb was available. One additional specimen was scanned with all imaging techniques, but not mechanically tested. The mean age of the 36 specimens was 68.9 years (SD 16.4 years, range 22 to 92). Twenty-one of these specimens were female. The two youngest specimens were excluded due to age (22 years). The mean age of the 27 specimens used in the final analysis was 71.1 years (SD 12.4 years, range 36 to 9 2 ) O f these 27 specimens, 12 were paired (six pairs) and 15 were unpaired. There were nine male and 18 female specimens. Ethics approval for this study was obtained from the Clinical Ethics Review Board, University of British Columbia. 4.2 Dual energy x-Ray Absorptiometry (DXA) 4.2.1 DXA A cquisition A l l specimens were scanned in array mode using standard protocol for the proximal femur region with a Hologic QDR 4500W bone densitometer (Hologic Inc., Waltham, MA) . The specimens were aligned with the femoral neck axis parallel to the table (internally rotated 0°) and the shaft axis parallel to the scan line on the screen (0° adduction). Rice bags of constant thickness were placed above and below the bone to simulate constant soft tissue thickness [120]. 1 Age was unknown in one pair of specimens, thus this number is based on N = 25. 56 Methods 4.2.2 DXA Analysis The total proximal femur region of interest (ROI) and its femoral neck, intertrochanteric and trochanteric subregions were analyzed using standard Hologic analysis protocol (Figure 9) [121]. The total proximal femur region is the sum of the femoral neck, trochanteric and intertrochanteric subregions. The bottom line of the femoral neck sub region of interest (subROI) is centered 0.75 cm below the narrowest point of the femoral neck and is perpendicular to the axis of the neck. The femoral neck subROI is 1.5 cm high and 5 cm wide. The trochanter subROI is defined as a region that includes the borders of the inferolateral edge of the neck box and a line connecting the midpoint of the neck axis to the base of the greater trochanter (where the edge of the femur changes curvature). The intertrochanteric subROI lies inferior to the trochanteric ROI. Global Region of Interest .SvmmetryAxis Ward 's Triangle Search Region [M]' Femoral Neck \~~\ Ward 's Triangle HI Trochanteric Region 111 Inter-Trochanteric Region Figure 9. The DXA total proximal femur ROI and its subregions. Reproduced from Hologic User's Guide [121 ] . 4.2.3 DXA Independent Variables For the total proximal femur and its subregions, I report bone mineral content (BMC, g) and areal bone mineral density (aBMD, g/cm ). 4.3 Hip Structural Analysis (HSA) I utilized Hip Structural Analysis (HSA) (Johns Hopkins University, Baltimore MD) to estimate bone structure at the proximal femur from the bone mass profiles obtained from D X A scans. The 57 Methods method has been described elsewhere [71, 114, 122]. I assessed two ROIs in the proximal femur (Figure 10): 1) the narrow neck - the narrowest segment of the femoral neck, and 2) the intertrochanteric region - along the bisector of the neck shaft angle. Each ROI was 5 mm wide. The ROIs defined with HSA are different from the ROIs defined using the traditional DXA analysis. Figure 10. Proximal femur image from a Hologic DXA scanner showing positions of HSA analysis and bone mass profiles of the two ROIs investigated in this study- narrow neck and intertrochanteric. 4.3.1 HSA Independent Variables For each of the two ROIs, the cross-sectional area (CSA, mm2), mean cortical thickness (CoTh, mm) and cross-sectional moment of inertia (IHSA, mm4) for bending in the image plane were derived from the bone mass profiles obtained from the DXA scans. Section modulus (SHSA, mm3) was calculated (S = I/r) where r = Vi subperiosteal width for the neck and r = the distance from the centroid to the lateral cortical margin for the intertrochanteric region. Areal BMD (g/cm2) was also measured for each ROI. Differences between standard D X A analysis of aBMD and HSA aBMD are expected due to the slightly different positions and size of the ROIs. 58 Methods 4.4 Magnetic Resonance Imaging (MRI) 4.4.1 1.5 T MRI Acquisition MRI examinations were performed with a 1.5 T system (GE Signa, GE Medical Systems, Milwaukee, Wisconsin) at the University of British Columbia Hospital with a subset of 20 specimens. Al l specimens were thawed prior to performing scans. The specimens were submerged in water to simulate soft tissue and to provide a contrasting shade (grey) surrounding cortical bone (black) and soft tissue (shades of grey and white). The specimens were placed in a quadrature head coil. Two trained technologists first performed localizer scans in the axial and coronal planes to landmark for later scans. They acquired high resolution oblique axial images perpendicular to the axis of the femoral neck, as cross-sectional images of the femoral neck. We defined this femoral neck axis as the line connecting the top of the fovea of the femoral head to the base of the greater trochanter in the localizer scan (Figure 11, yellow line). A representative cross-section of the femoral neck is shown in Figure 12, and its location in the coronal view in Figure 11. We obtained 25 consecutive 3.0 mm slices from the center of the femoral head to the base of the greater trochanter. Figure 11. A scout scan of the proximal femur ex vivo. The solid yellow line indicates the longitudinal femoral neck axis. The solid red line indicates the location of the cross-section shown along the FN shown in Figure 12 below. 59 Methods Figure 12. A representative MRI cross-section of the femoral neck. The high resolution cross-sectional images were acquired with a Tl-weighted spin-echo (SE) sequence with a TR = 600 ms and TE = 14 ms. The field of view was 15 cm x 15 cm, with a matrix size of 512 x 512, resulting in an in-plane pixel size of 0.29 mm2. Including preparation and positioning of the specimen, as well as acquisition of 20 additional high resolution images in another plane, total scan time was approximately 30 minutes per specimen. 4.4.2 3 T MRI Acquisition All 37 specimens were scanned with a 3.0 T MRI system (Philips Gyroscan Intera, Netherlands) at the High Field Magnetic Imaging Centre at the University of British Columbia Hospital. The femur set-up and protocol were similar to the protocol used for the 1.5 T system. Small differences between measurement protocols, including improved image resolution, are described below. Specimens were submerged in ultrasound gel to provide a contrasting shade (grey) surrounding cortical bone (black) and soft tissue (shades of grey and white) to simplify preparation. The 60 Methods specimens were placed in a SENSE-head coil. High resolution scans were performed perpendicular to the femoral neck axis as per the 1.5 T protocol. Cross-sectional images were acquired using a Tl-weighted spin-echo sequence with a TR = 700 ms and TE = 15 ms. The field of view was 12 cm x 12 cm, with a matrix size of 512 x 512, resulting in an in-plane pixel size of •y 0.23 mm . Forty-four consecutive images were obtained at a slice thickness of 2 mm. The slice thickness used for the 3 T images was smaller (2 mm) than the slice thickness used for the 1.5 T images (3 mm). Due to increased inhomogeneities in the 3 T scanner, an artefact was present which affected the appearance of fat tissue surrounding the bone. A fat shift was applied in the superior direction to minimize this artefact. 4.4.3 MRI Analysis A l l scans from the 3 T and 1.5 T MRI systems were analyzed using the same protocol. The femoral neck ROI was defined from the coronal view and the cross-sectional images were acquired perpendicular to the femoral neck axis. Slices within the following boundaries were included (Figure 13): a. Medial border - lateral to the subcapital region, where the superior and inferior surfaces were near parallel to the femoral neck axis. b. Lateral border - the last slice medial to the lesser trochanter. On average, five slices (range four to ten) were included in this ROI for each specimen scanned by the 3.0 T MRI. For the 1.5 T MRI, on average, four slices (range three to six) were included in the femoral neck ROI for each specimen. 61 Methods Figure 13. MRI coronal view of the proximal femur labelled with the medial (a) and lateral (b) borders of the FN region of interest. Scans obtained from both systems were segmented to differentiate between soft tissue/water, cortical bone, and trabecular bone/marrow by defining the periosteal (Figure 14a) and endocortical (Figure 14b) borders on each cross-sectional image in the ROI. Segmentation was performed using Analyze software (Analyze 6.0, Mayo Clinic, Lenexa, KS). The endocortical border and thus the trabecular/marrow cavity (white and grey) was identified using a threshold-driven region-growing algorithm in the "Image Edit" module. The periosteal border was identified manually using a spline from the endocortical border and moving it outwards towards the periosteal border (where soft tissue and water appeared white and grey) in locations where cortical bone was present (black). Using the "Region of Interest" module, Total Area (ToCSA), and Trabecular Area (TrCSA) were defined for each slice from the area within the periosteal border, and the area within the endocortical border, respectively. Cortical Area (CoCSA) was derived as the area between the periosteal and endocortical borders (ToCSA - TrCSA). 62 Methods a) b) Figure 14. MRI cross-section of the femoral neck segmented using Analyze software, a) the green line defines the outer, periosteal border, b) The red line defines the inner, endocortical border. The coordinates of the total area and trabecular area were exported to compute the second area moment of inertia (I) for bending about the x-and y-axes (Ix and Iy, respectively), passing through the geometric centroid. The x-axis was defined as the horizontal axis in the bone cross-section (anterior-posterior), while the y-axis was defined as the vertical axis in the bone cross-section (superior-inferior). The following equations were used to compute I for the cortical bone area [CoCSA) in the xy plane: /, = \fdA I, = [x>dA where dA is an element of Area A and x and y are the coordinates of each element of area. The calculation of Ix and Iy assumed that the bone was hollow, with the cortical bone of constant density as the shell. The section modulus was calculated about the x- and y-axes (Sx and Sy, respectively), according to the following equations: 63 Methods where and rx and ry are the maximum distances from the centroid to outer edge of the bone in the x and y directions, respectively. 4.4.4 MRI Reliability Twelve specimens were scanned twice, on different days, to evaluate reliability of M R image acquisition in the 3 T MRI and in the 1.5 T MRI (inter-acquisition reliability). Two different subgroups of 12 specimens were used for each imaging system. I also analyzed the first scans of these specimens twice to evaluate reliability of image analysis (inter-analysis reliability). 4.4.5 MRI Independent Variables Dependent variables for MRI were total bone cross-sectional area (ToCSA, mm2), cortical cross-sectional area (CoCSA, mm2), area moment of inertia about the x-axis (Ix, mm4), area moment of inertia about the y-axis (Iy, mm4), section modulus about the x-axis (S x, mm3), and section modulus about the y-axis (Sy, mm ). I previously defined these variables in subsection 4.4.3. Mean values for each of these independent variables in the femoral neck ROI are reported. 4.5 Mechanical Testing - Sideways Impact Configuration 4.5.1 Testing Protocol Thirty-six of the 37 specimens were loaded to failure using the Instron materials testing system (Instron 8841, Norwood, MA) in a simulated sideways fall configuration . I induced proximal femur fractures by simulating a sideways fall based on the strong evidence that hip fracture frequently occurs as the result of a sideways fall and the resulting impact on the greater trochanter. The set-up was modified from several studies [11, 65, 71] that produced clinically relevant fracture types (e.g. cervical and intertrochanteric fractures) (Figure 15). The loading apparatus was constructed by an Engineering MASc student, Peter de Bakker. The testing protocol was refined through pilot testing of animal specimens and composite ceramic bones (Sawbones, Pacific Research Laboratories, Vashon, Washington). 2 One specimen was imaged, but was not loaded to failure because of its use in a Finite Element Model Study. 64 Methods The femur was cut at the midshaft (50% of the total femur length, measured from the superior aspect of the greater trochanter to the inferior aspect of the lateral condyle). The distal portion ( l /6 t h of the total femur length) of the remaining shaft was potted in P M M A to prevent rotation along the shaft axis. The femur shaft was adducted 10° to the horizontal and internally rotated 15°, as described by Pinilla et al. [68]. The load was applied to the greater trochanter through a P M M A cup moulded to the greater trochanter to prevent local crush fractures. The femoral head was placed inside half a tennis ball with lubricant to prevent crushing of the head, as per Lochmuller et al. [11] and Eckstein et al. [123]. A foam pad was placed below the tennis ball to further prevent crushing of the head3. The distal and proximal ends of the femur were allowed to translate independently of each other through the use of ball bearings below the two independent plates (one small plate below the head, which rested on a larger plate) (Figure 15). The femur was also allowed to translate in the vertical plane through a pivot at the distal end. The distance from the pivot to the superior aspect of the greater trochanter was 2/3 of the total femur length. The mechanical testing was performed with the assistance of a research engineer (Cecelia Tang) and an Engineering MASc student (Peter de Bakker). Pilot testing using composite ceramic bones indicated that cracks frequently propagated through the head. The pad was used to ensure the human specimens were not exposed to a stress concentration at the head. 65 Methods Applied Load mm Bearings Base of Instron Figure 15. A schematic of the mechanical testing apparatus. Bearings between the surfaces allowed the plates to translate freely. The distal end of the femur was free to pivot. 4.5.2 Displacement Rate Robinovitch et al. [124-126] measured the stiffness of soft tissues above the greater trochanter, the total effective stiffness of the body during impact and the rate of the load applied to the bone in a fall from standing height. Based on these measures, the approximate displacement rate in an average individual that would occur in a fall on the greater trochanter from standing height was derived. Calculations used to derive this displacement rate are presented (Appendix 1). Based on these calculations and the ability to control the Instron system, I applied a displacement rate of 100 mm/s. The chosen displacement rate (100 mm/s) was within the order of magnitude expected of a displacement during a fall from standing height (340 mm/s). Carter and Hayes [36], reported that the strength of bone is proportional to the displacement rate, to the power of 0.06, which indicates that a small change in displacement rate will have a minimal effect on the strength of the bone. Thus a relatively small discrepancy between the calculated displacement rate and the chosen displacement rate (~3x) was not anticipated to significantly affect failure load. 4.5.3 Dependent Variables Failure load (N) and energy to failure (J) were evaluated. Failure load was defined as the first local maximum load (Figure 16), except in cases where the drop in load after the first local 66 Methods maximum was less than 10% of the load at the first local maximum. In these cases, the failure load was defined as the maximum load. Failure load was used as the primary dependent variable. 5000 i 14 Displacement (mm) Figure 16. Failure load determined from the load displacement curve, as defined as the first local maximum. 67 Methods 4000 -i Max Load 3205 N First peak 2437 N o o 2 4 6 8 10 12 14 Displacement (mm) Figure 17. Failure load determined by the load-displacement curve, as defined by the maximum load. In this case, the failure load dropped less than 10% of the first local maximum after reaching the first local maximum. 4.5.4 Fracture Classification All specimens were visually inspected by an orthopaedic surgeon (Dr. Pierre Guy) after mechanical testing to classify the fracture type. Dr. Guy was blinded to the failure load and all other outcomes prior to classification. Fractures were classified anatomically based on regions defined by Zuckerman et al. [53]: femoral neck, basicervical, intertrochanteric, subtrochanteric or greater trochanter crush. Intertrochanteric fractures were classified as stable or unstable based on a classification scheme described by Evans [127]. 4.6 Sample Size The primary aim in this study was to identify whether a linear regression model incorporating MRI-derived ToCSA (shown to be a reliable measure [25]) and DXA-derived aBMD or BMC has a greater predictive ability than DXA aBMD alone. The variance in failure load explained by aBMD measured by DXA in the trochanteric region using the loading apparatus after which the current apparatus was modeled was 61% (R = 0.78) [12]. The required sample size for this study was then estimated based on this 5% increase in the total variance in failure load to be explained 68 Methods by ToCSA measured by MRI in the femoral neck. A 5% greater predictive ability would result in R = 0.64 for the correlation between the MRI + D X A model and failure load. To detect a difference of 0.03 (0.64 - 0.61) in R 2 between a MRI -DXA model and a D X A model alone, for a power of 0.88 at an alpha level < 0.05, the required sample size was 30. Assuming technical difficulties with 20% of the specimen, the sample size required was 36. 4.7 Statistical Analyses A l l statistical analyses were performed using SPSS 12.0 for Windows (SPSS Inc.). Where applicable, the level of significance was set at p < 0.05. 4.7.1 Primary A im Measurements obtained from the 3.0 T MRI, D X A and HSA were used as the independent variables to answer questions posed in the primary aim. As stated previously, all specimens were scanned in the 3.0 T MRI. 4.7.1.1 Descriptives For independent and dependent variables of interest, I report the mean + standard error of the mean (SEM). A two-tailed independent samples t-test was used to determine the difference in failure load between males and females. The difference between right and left limbs from the same specimen is reported as the mean of the left-right difference and as the mean of the absolute value of the left-right difference. 4.7'.1.2 Association Between Independent Variables and Failure Load The relationships between key independent variables, between and within imaging systems are presented in scatterplots. R 2 values are also presented. The level of the association between independent variables and the dependent variable, failure load, was determined using Pearson product moment correlations (R) and R . A test for correlated correlations [12, 128] was used to determine whether the variance explained by MRI variables differed significantly from the variance explained by the D X A and HSA variables. This statistical test, endorsed by Steiger [128], allows comparison of correlations that are based on the same sample and share a variable. 69 Methods Hierarchical linear regression models were developed to determine the relationship between mean failure load and independent variables from combinations of density and geometry variables from each imaging system4. Hierarchical regression modelling was chosen as it allowed input of variables based on a theoretical order, rather than solely based on statistical criteria. As variables are entered in blocks, this modelling technique provides information regarding the independent contribution of the variable in each block to the model. I created models including femoral neck aBMD and trochanteric aBMD for each group of variables (corresponding aBMD values were chosen for HSA - narrow neck rather than femoral neck and intertrochanteric rather than trochanteric). In each model, aBMD was entered into the model first. A variable representing geometry was then added in the second block. The independent variable representing geometry was chosen for input into the model based on several criteria: 1) a high association with failure load and 2) a moderate or lower association with the density variable. In the case where several independent variables were similarly associated with failure load, the independent variable with the lowest association with aBMD was selected. Thus, I created four hierarchical linear regression models: 1) DXA + MRI a. Trochanteric aBMD measured with DXA (aBMDoxA) + femoral neck area moment of inertia about the x-axis measured with MRI (IXMRI) b. Femoral neck aBMD measured with DXA (aBMDoxA) + femoral neck area moment of inertia about the x-axis measured with MRI (IXMRI) 2) HSA a. Intertrochanteric aBMD measured with HSA (aBMDHSA) + intertrochanteric area moment of inertia estimated with HSA (IHSA) b. Narrow neck aBMD measured with HSA (aBMDHSA) + narrow neck area moment of inertia estimated with HSA (IHSA) The first variable listed was entered in the first block, and the second variable listed was entered in the second block. The extra sums of squares test was used to determine whether the addition of Linear regression analyses were also performed with multilevel modelling techniques to account for the relationships between paired specimens. These models are described in Appendix 2. 70 Methods the structural variable significantly contributed to the variance explained, at a level of p < 0.05. For each model, I report the variance in failure load explained (R ), the unstandardized B coefficient, the standard error of the estimate (SEE) and the standardized P coefficient. A test for correlated correlations [128] was used to determine whether the variance explained by the D X A + MRI models differed significantly from the variance explained by the HSA models. 4.7.2 Secondary Aim 4.7.2.1 Differences Between 3 T MRI and 1.5 T MRI For the subsample of specimens for which MRI scans were obtained in the 1.5 T and 3 T, I report the mean + S E M for each independent variable measured with each system. In addition, I report the mean difference, percent difference, and R 2 within each independent variable measured by both systems. For each independent variable, the difference between measurement systems against the mean of the two measurements were plotted (Bland-Altman plots) [129]. The 95% limits of agreement (i.e. mean difference + 1.96 multiplied by the standard deviation of the difference) were also plotted in the same figure to indicate the difference in values for most specimens. 4.7.2.2 Reliability Analyses For all independent variables measured with 1.5 T and 3 T MRI, I calculated the mean difference, 95% CI for the mean differences, percent differences, R 2 , and the within-specimen coefficient of variation (CV). I computed the intra-class correlation coefficient (ICC) using a two-way mixed model and a single measure. These analyses were performed separately for inter-acquisition and inter-analysis reliability. Finally, I calculated the 95% limits of agreement (LOA) for the differences observed in repeated measurements of a given independent variable (i.e. mean difference ± 1.96 multiplied by the standard deviation of the difference) [129]. The L O A provides information on reliability in absolute terms, and indicates how large a change at the individual level should be observed in order to be confident that a real change in that given variable has occurred [130]. 71 Results 5. RESULTS 5.1. Descriptives 5.1.1. Specimens and Fracture Types Of the 36 specimens loaded to failure, I excluded nine specimens. The two youngest specimens were excluded due to age (22 years). Seven additional specimens were excluded based on fracture types (n = 2 crush fractures at the greater trochanter, 6%; n = 5 in the subtrochanteric region, 15%)5. 5.1.2. Fracture Types Among the 27 specimens, three fractures occurred at the femoral neck (9%), three were basicervical (9%), 14 were stable intertrochanteric (41%) and seven were unstable intertrochanteric (21%)6. 5.1.3. Independent Variables7 Mean values for selected independent variables of interest are presented in Table 4. In general, the same properties measured in the same regions by different imaging systems provided similar values. For example, mean femoral neck aBMD measured by D X A was 0.66 g/cm , while mean narrow neck aBMD measured by HSA was 0.66 g/cm . 5 Percentages were calculated based on N = 36. 6 Percentages were calculated based on N = 36. 7 All data presented for MR] measurements were obtained from the 3.0 T MRI 72 Results Table 4. Mean values for independent variables of interest measured with MRI, D X A and HSA for each region of interest (N = 27). 3 T MRI Femoral Neck Region ToCSA (mm2) CoCSA (mm2) TrCSA (mm2) Ix (mm4) iy (mm4) Sx (mm3) Sy (mm3) Mean S E M 749.4 34.05 132.5 6.94 616.9 31.51 15031 1262 9759 917 635.1 46.6 599.5 42.9 D X A Femoral Neck Region aBMD (g/cm2) B M C (g) Mean S E M 0.66 0.03 3.32 0.21 HSA Narrow Neck Region aBMD (g/cm2) CSA (mm2) CoTh (mm) I (mm4) S (mm3) Mean S E M 0.66 0.04 197,5 11.5 1.3 0.1 16224 1376 947.0 66.1 D X A Trochanteric Region aBMD (g/cm2) Troc B M C (g) FN aBMD (g/cm2) F N B M C (g) Mean S E M 0.61 0.03 6.50 0.49 0.66 0.03 3.32 0.21 HSA Intertrochanteric Region aBMD (g/cm2) CSA (mm2) CoTh (mm) I (mm4) S (mm3) Mean S E M 0.69 0.03 340.5 18.9 2.7 0.2 86075 6919 2909.0 188.8 73 Results 5.1.3.1. Comparison Between Independent Variables Measured with MRI The correlation matrix for all independent variables of interest is presented in Appendix 3. Within each imaging system, most variables measured by that system were moderately or highly correlated with each other. CoCSA was moderately associated with ToCSA (R = 0.45, R 2 = 0.21, Figure 18). Ix was strongly associated with CoCSA (R = 0.82, R 2 = 0.67, Figure 19). Sx was also strongly associated with CoCSA (R = 0.79, R 2 = 0.62, Figure 20). ° R S q Lineat = 0.206 O ~~i 1 1 1 1 1 1 r~ 500 600 700 SOO 900 1000 1100 1200 T o C S A (nun2) Figure 18. CoCSA measured with MRI vs. ToCSA measured with MRI. 74 Results 35000 H 30000-25000-j 1 1 20000 H 15000 H 10000-5000-RSq Linear =0.674 — i 1 1 1 1 1 r 75 100 125 150 175 200 225 CoCSA (inm2) Figure 19. Ix measured with MRI vs. CoCSA measured with MRI. 12C 1000-800-4 02 600H 400 H 200-o O RSqLme«r- 0.619 — i f i 1 1 1 r 75 100 125 150 175 200 225 CoCSA («un2) Figure 20. Sx measured with MRI vs. CoCSA measured with MRI. 75 Results 5.1.3.2. Comparison Between Independent Variables Measured with Different Imaging Systems When comparing imaging systems, many variables were also highly correlated. The following plots compare measurements of similar variables measured with M R I and H S A . Narrow neck CSAHSA was highly associated with COCSAMRI (R = 0.59, R 2 = 0.35, Figure 21). Narrow neck IHSA was strongly associated with IXMRI (R = 0.77, R 2 = 0.60, Figure 22). Narrow neck SHSA was also highly associated with SXMRI (R = 0.62, R 2 = 0.39, Figure 23). 350 H 1 < te O z. < X 300 H 250 H 200-150-100 H 50-R Sq L i n e a r = 0349 ~i 1 1 i — i 1 r ~ 50 100 150 200 250 300 350 MRI CoCSA (mm2) Figure 2 1 . Narrow neck C S A measured by HSA vs. C o C S A measured by MRI. Narrow neck C S A H S A includes cortical and trabecular bone areas. 76 Results 1 30000-25000H M CP 2 20000H o 15000H 10000-5000- RSq Linear = 0.396 1 1 1 1 1 1 1 5000 10000 15000 20000 25000 30000 35000 M R I Ix (nun4) Figure 22. Narrow neck I measured by HSA vs. Ix measured by MRI. i 3 t/3 u OJ 3 < s 1600-1200 H 800-400 H oo w o T 400 RSq Linear = 0.39 T 800 1200 M R I Sx (iran3) 1600 Figure 23. Narrow neck S measured by HSA vs. Sx measured by MRI. Figure 24 provides an example of the high correlation between aBMD and CoCSA observed in this study (R = 0.64, R 2 = 0.41). The relationship was similar, but slightly lower, between femoral neck aBMD and Ix (R = 0.57, R 2 = 0.32, Figure 25). 77 % 1.0-a OS —j 05 1 0.6-o s 3 Q 0.4-O O / r j » cp O O Results RSq Linear =0.405 75 100 125 150 •~r~ 175 200 225 MRI CoCSA («un2) Figure 24. Femoral neck aBMD measured with D X A vs. CoCSA measured with MRI. l_| 1 ( ( ( r_ ^ 5000 10000 15000 20000 25000 30000 35000 MRI Ix (mrn4) Figure 25. Femoral neck aBMD measured with D X A vs. Ix measured with MRI. 78 Results 5.1.4 Dependent Variables A typical load displacement curve is shown in Figure 26. The mean failure load was 4353 N (SE 363 N) (Table 5). The mean energy to failure was 13.25 J (SE 1.48 J). 3500 i 3000 H 0 2 4 6 8 10 12 Displacement (mm) Figure 26. A load-displacement curve for a 73 year old female cadaveric femur. The specimen was loaded to failure by simulating a sideways fall on the greater trochanter. Table 5. Mean and standard error of the mean (SEM) for the dependent variables obtained from the load-displacement curves in mechanical testing (N = 27). Failure Load (N) Energy to Failure (Nm) Mean 4353 13.25 S E M 363 1.48 Failure load for each fracture type is described in Figure 27. The mean failure load for male cadaveric specimens was 5290 N (SE 655 N), and for female cadaveric specimens was 3884 N (SE 403). A one-tailed independent t-test indicated that this difference was significant (t.05,25 = 1.9, p = 0.03). 79 Results n= 14 Neck Basicervical IT stable IT unstable Fracture Type Figure 27. Mean failure load by type of fracture (Mean + SEM). * indicates significant difference between failure load for stable intertrochanteric (IT) and unstable fractures (p < 0.05). 5.1.5 Side-to-Side Comparisons Only two of six pairs fractured in the same location on both sides (both pairs experienced intertrochanteric fractures). In the paired specimens, the systematic mean difference for failure load between sides (left - right) was - 207.9 N (SD 1043.1 N), or - 3.4% (SD 18.9%). The random mean difference in failure load between sides was 803.8 N (SD 656.1 N), or 16.5% (SD 8.5%). As the standard deviation for failure load in the whole sample was 1885.8 N , the variability in failure load in the whole sample was much greater than the variability within paired specimens. 5.2 Relationship of Independent Variables with Failure Load 5.2.1 Correlations Failure load was strongly correlated with many independent variables from each imaging modality (Table 6). Specifically, MRI variables CoCSA (Figure 28), Ix (Figure 29) and Sx (Figure 30) were highly correlated with failure load, (range R - 0.65 to 0.69). Within variables measured with D X A , trochanteric aBMD was most highly correlated with failure load (R = 0.84) (Figure 31). 8000 7000 6000 X 5000 -2 4000 a 3000 u. 2000 1000 n = 3 80 Results Trochanteric B M C was also highly associated with failure load. D X A intertrochanteric and femoral neck aBMD and B M C were slightly less correlated with failure load than the same variables measured in the trochanteric region. Figure 32 shows a plot of femoral neck aBMD versus failure load. Within variables measured with HSA, intertrochanteric aBMD, intertrochanteric CSA, intertrochanteric I and intertrochanteric S were strongly correlated with failure load. Figures 33 to 36 show plots of selected HSA variables versus failure load. Variables from the HSA narrow neck region were slightly less correlated with failure load, e.g. narrow neck aBMD (R = 0.50). 81 Results Table 6. Correlations (R) and R 2 between independent variables measured with each imaging system and failure load ( N = 27). *p < 0.05, * *p < 0.01 Imaging System Independent Variable R R 2 MRI ToCSA 0.44* 0.20 CoCSA 0.68** 0.47 TrCSA 0.33 0.11 Ix 0.69** 0.47 iy 0.66** 0.43 Sx 0.65** 0.42 Sy 0.65** 0.43 D X A Femoral neck aBMD 0.63** 0.39 Femoral neck B M C 0.67** 0.45 Trochanteric aBMD 0.84** 0.70 Trochanteric B M C 0.73** 0.53 Intertrochanteric aBMD 0.74** 0.54 Intertrochanteric B M C 0.67** 0.45 HSA Narrow neck aBMD 0.50** 0.25 Narrow neck CSA 0.60** 0.36 Narrow neck CoTh 0.49* 0.24 Narrow neck I 0.55** 0.31 Narrow neck S 0.57** 0.33 Intertrochanteric aBMD 0.78** 0.60 Intertrochanteric CSA 0.77** 0.60 Intertrochanteric CoTh 0.72** 0.51 Intertrochanteric I 0.69** 0.47 Intertrochanteric S 0.77** 0.60 82 Results O Figure 28. Failure load vs. CoCSA measured with MRI. O Figure 29. Failure load vs. Ix measured with MRI. 83 p - , , , , n 200 400 600 800 1000 1200 MRI Sx (iom3) Figure 30. Failure load vs. Sx measured with MRI. ~I r — — | , , r—— j -~r 03 0,4 0,5 0,6 0.7 0.8 0.9 1.0 DXA Trochanteric aBMD (g/cm2) Figure 31. Failure load vs. trochanteric aBMD measured with DXA. 84 i i i i i i 1 0.4 0.3 0.6 0.7 0.8 0.9 1.0 1.1 H S A Intertrochanteric aBMD (g/cin2) Figure 33. Failure load vs. intertrochanteric aBMD measured with HSA. 85 Results O T 1 1 1 1 1 1 r 4 6 S 10 12 14 16 18 HSA Intertrochanteric I (min4) Figure 34. Failure load vs. intertrochanteric I measured with HSA. 86 Results 0 sooo o o « o ^ 6000-r | 4000 -j O o o o 2000 H o o R Sq Linear =* 0.306 O 5000 10000 15000 20000 25000 30000 35000 HSA Narrow Neck 1 (mm4) Figure 36. Failure load vs. narrow neck I measured by HSA. 5.2.2 Comparison of Correlations Between MRI, DXA and HSA The association of IXMRI with failure load did not differ significantly from the association of trochanteric aBMD with failure load (Z = - 1.66, p = 0.10) or femoral neck aBMD with failure load (Z = 0.46, p = 0.64). Similar results were observed for comparisons between MRI and HSA variables. The association of COCSAMRI with failure load did not differ significantly from the association of CSAHSA with failure load (Z- 0.66, p = 0.51). The association between IXMRI with failure load did not differ significantly from the association of IHSA with failure load (Z= 1.28, p = 0.20). The association between SXMRI with failure load did not differ significantly from the association of SHSA with failure load (Z= 0.58, p = 0.56). 5.2.3 Hierarchical Linear Regression Models 5.2.3.1 MRI + DXA Trochanteric Region- Hierarchical Model Area moment of inertia about the x-axis (IXMRI) was not a significant independent predictor of failure load after accounting for trochanteric aBMD, R-squared change = 0.03, F-change = 2.6, p = 0.12 for 1, 24 df (standardized /3 = 0.23) (Table 7). Therefore addition of IXMRI did not explain 87 Results significantly more variance in failure load. This model accounted for 73% of the total variance in failure load. Table 7. Hierarchical linear regression model summary for trochanteric (TR) a B M D D X A and Ix M R i ( N = 27). B is the unstandardized coefficient, B is the standardized coefficient. R2 R2 Change S E E B S E of B B p-value Independent Variable Model 1 070 0/70 1050.1 < 0.001 TR aBMD 10945 1427 0.84 Model 2 0.73 0.03 1018.1 TR aBMD 8951 1856 0.69 < 0.001 Ix 0.07 0.04 0.23 0.120 5.2.3.2 MRI + DXA Femoral Neck Region- Hierarchical Model Area moment of inertia about the x axis (IXMRI) was a significant independent predictor of failure load after accounting for femoral neck aBMD, R-squared change = 0.16, F-change = 8.67, p = 0.01 for 1, 24 df (Table 8). When accounting for both variables, both femoral neck aBMD and Ix were significant independent predictors of failure load, t = 2.1, p = 0.05, and t = 2.9, p = 0.01 for 24 df, respectively. The final model explained 55% of the total variance in failure load. Table 8. Hierarchical linear regression model summary for femoral neck ( F N ) a B M D D X A and Ix M R i ( N = 27). B is the unstandardized coefficient, B is the standardized coefficient. I n d e P e n c j e n t r 2 R 2 C h S E E B S E of B fi p-value Variable c r r Model 1 039 039 1501.2 < 0.001 FN aBMD 6997.52 1747.95 0.63 Model 2 0.55 0.16 1313.4 FN aBMD 3894.29 1857.33 0.35 0.05 Ix 0.14 0.05 0.49 0.01 5.2.3.3 HSA Intertrochanteric Region- Hierarchical Model Of the HSA-derived variables, intertrochanteric aBMD showed the strongest association with failure load, F = 37.70, p < 0.001 for 1, 25 df. Intertrochanteric SHSA was a significant independent predictor of failure load after accounting for intertrochanteric aBMD, R-squared change = 0.07, F-change = 4.71, p = 0.04 for 1, 24 df (Table 9). In the final model, standardized (3 coefficients were 0.58 and 0.29 for intertrochanteric aBMD, and intertrochanteric SHSA, respectively. The final model accounted for 67% of the total variance in failure load. 88 Results Table 9. Hierarchical linear regression model summary for the intertrochanteric (IT) region measured by HSA (N = 27). Independent Variable R2 R2 Change SEE B SE of B B p-value Model 1 0.60 0.60 1214.4 < 0.001 I T aBMDnsA 8380 1365 ° - 7 8 Model 2 0.67 0.07 1133.3 I T aBMDH S A 4679 2129 0.43 0.038 ITSHSA 821.41 378.57 0.43 0.040 5.2.3.4 HSA Narrow Neck Region - Hierarchical Model An additional HSA model was derived using variables from the narrow neck region only (Table 10). Narrow neck aBMD was strongly associated with failure load, F = 8.35, p = 0.01 for 1, 25 df. Area moment of inertia (IHSA) was a significant independent predictor of failure load after accounting for narrow neck aBMD, R-squared change = 0.12, F-change = 4.36, p = 0.05 for 1, 24 df. In the combined model, the standardized /? coefficients were 0.29 and 0.40 for narrow neck aBMD and narrow neck IHSA, respectively. After addition of narrow neck IHSA, narrow neck aBMD was no longer a significant independent predictor of failure load, t = 1.50, p = 0.15 for 24 df. The final model accounted for 37% of the total variance in failure load. Table 10. Hierarchical linear regression model summary for the narrow neck (NN) region measured by HSA (N = 27). Independent r 2 r 2 SEE B SE of B B p-value Variable ° r r Model 1 0\25 025 1665.1 0.008 NN aBMDnsA 4833 1672 0.50 Model 2 0.37 0.12 1563.3 NN aBMDHsA 2776 1853 0.29 0.147 N N I H S A 1056.88 506.10 0.40 0.048 5.2.3.5 Comparison Between Hierarchical Models The predictive ability of the model combining Ix measured by MRI and aBMD in the femoral neck measured by DXA was not significantly different from a model composed of variables measured in the intertrochanteric region with HSA (aBMD and IHSA) ,Z=- 0.94, p = 0.34. There was no significant difference in the predictive value between the models that combined IXMRI and femoral neck aBMD measured by DXA and the model comprised of variables measured in the narrow neck region with HSA (aBMD and IHSA), Z = 1.87, p = 0.06. 89 Results 5.3 Comparison of 3 T with 1.5 T MRI 5.3.1 Relationships Between Independent Variables Measured by 1.5 T and 3 T Total cross-sectional area (ToCSA) and trabecular cross-sectional area (TrCSA) as measured by 1.5 T MRI and by 3T MRI were highly associated (R 2 = 0.99 and 0.98, respectively) (Table 11). The association between derived variables (i.e. measures derived from ToCSA and TrCSA) were more moderate, e.g. R = 0.60 for CoCSA. Based on two-tailed paired samples t-tests, the differences between imaging systems were significant (p < 0.05) for CoCSA, Iy, and Sy only. Table 11. Independent variables measured with 3 T MRI and 1.5 T MRI. Differences are presented as (1.5 T - 3 T), therefore a positive difference indicates a larger value measured by 1.5 T. ToCSA (mm2) CoCSA (mm2) TrCSA (m Ix (m iy (m Sx (mi Sy (mm3) m2) m4) ( m4) ( m3) 1.5 T Mean + S E M 3 T Mean + S E M Mean Difference % Difference R2 794.32 + 44.14 787.29 + 43.29 7.04 1% 0.99 144.87 + 9.99 130.04 + 8.79 14.83 11% 0.60 649.45 + 36.69 657.25 + 38.84 -7.80 - 1% 0.98 18222 + 2389 15772 + 2081 2451 13% 0.63 11761 + 1410 9856 + 1070 1906 15% 0.78 754.62 + 76.58 638.90 + 63.51 115.72 16% 0.35 691.75 + 60.58 564.58 + 41.16 127.17 18% 0.62 Bland-Altman plots for ToCSA, CoCSA, Ix and Sx are shown (Figure 37 to Figure 40). These figures indicate that ToCSA, CoCSA, Ix and Sx were systematically larger when measured with the 1.5 T MRI system compared to the 3 T MRI system. The difference for one specimen falls outside the 95% limits of agreement for each variable. 90 Results 60 + 1.96 SD 40 20 -20 200 400 600 800 1000 1200 1400 .40 J- -1.96SD -60 Mean: ToCSA(1.5 T) and ToCSA(3 T) (mm2) Figure 37. Bland-Altman (difference vs. mean) plot for ToCSA measured in the same specimens with 1.5 T MRI and 3 T MRI (N = 20). The difference is calculated as 1.5 T - 3 T. The solid red line indicates the mean difference. The dashed red lines indicate the mean difference + 1.96 x standard deviation of the difference. 120 - i 100 80 60 ^ 40 20 0 -20 4 -40 4 -60 + 1.96SD . mean 50 100 J50 200 250 300 -1.96SD Mean: CoCSA(1.5 T) and CoCSA(3 T) (mm2) Figure 38. Bland-Altman plot (difference vs. mean) for CoCSA measured in the same specimens with 1.5 T MRI and 3 T MRI (N = 20). The difference is calculated as 1.5 T - 3 T. The solid red line indicates the mean difference. The dashed red lines indicate the mean difference + 1.96 x standard deviation of the difference. 91 Results 25000 20000 S 15000 a E 10000 5000 0 -5000 -10000 -15000 + 1.96 SD 1 1 1 1 10000 * 50000 30000 40000 * 50000 -1.96 SD Mean: Ix (1.5 T) and Ix (3 T) (mm4) Figure 39. Bland-Altman plot (difference vs. mean) for Ix measured in the same specimens with 1.5 T MRI and 3 T MRI (N = 20). The difference is calculated as 1.5 T - 3 T. The solid red line indicates the mean difference. The dashed red lines indicate the mean difference + 1.96 x standard deviation of the difference. 1200 1000 -800 -600 -400 200 H E e P 52, x CC I p SC 0 -200 i -400 -600 J + 1.96 SD . mean 300 500 700 900 1100 1300 1500 -1.96 SD Mean: Sx (1.5 T) and Sx (3 T) (mm4) Figure 40. Bland-Altman plot (difference vs. mean) for Sx measured in the same specimens with 1.5 T MRI and 3 T MRI (N = 20). The difference is calculated as 1.5 T - 3 T. The solid red line indicates the mean difference. The dashed red lines indicate the mean difference + 1.96 x standard deviation of the difference. 92 Results 5.3.2 Is Inter-Acquisition Reliability Greater with 3 T than 1.5 T? 5.3.2.1 Inter-Acquisition Reliability with 3 T For measurements made with the 3 T MRI, the association between images acquired at different times was high for ToCSA and TrCSA (R 2 = 0.98 for both) (Table 12). The associations for derived variables were generally lower (e.g. Sx, R = 0.55). The 95% confidence intervals crossed zero for all variables measured with both systems, suggesting there was no systematic difference between measurement times. Percent differences between scans were low for most variables, ranging from 0.2% to 6.0%. Within-specimen coefficients of variation (CV) were low for directly measured variables (2.0% for ToCSA). However, CVs were higher for derived variables, ranging from 4.6% for CoCSA to 8.5% for Ix. Intraclass correlation coefficients were high for all variables measured. 5.3.2.2 Inter-Acquisition Reliability with 1.5 T For measurements made with the 1.5 T MRI, the association between images acquired at different times was high for ToCSA and TrCSA (R 2 = 0.99 and 0.98, respectively) (Table 12). The associations for derived variables were generally lower (e.g. Sx, R = 0.75). The 95% confidence intervals crossed zero for all variables measured with both systems, suggesting there was no systematic difference between measurement times. Percent differences between scans were low for most variables, ranging from 0.8% to 5.9%. Within-specimen coefficients of variation (CV) were low for directly measured variables (e.g. 1.7% for ToCSA). However, CVs were higher for derived variables, ranging from 5.6% for CoCSA to 10.1% for Sy. Intraclass correlation coefficients were high for all variables measured. 5.3.2.3 Comparison of Inter-Acquisition Reliability Between 3 T and 1.5 T Similar percent differences, R , CV and ICCs for variables measured with both systems indicate that inter-acquisition reliability did not differ between 1.5 and 3 T MRI systems. Limits of agreement (LOA), at the 95% level, are presented in Table 13. The L O A for each independent variable was similar for each imaging system, indicating that inter-acquisition reliability did not differ between systems. 93 Results Table 12. Inter-acquisition reliability for a) 3 T MRI (n = 12) and b) 1.5 T MRI (n = 12). Differences are presented as (2 n d scan - 1st scan). Percent differences are presented as (2n d scan - Is' scan/mean of two scans). a) 3 T M R I R 2 Difference (mm2) 95% CI % Difference CV ICC ToCSA (mm2) 0.98 20.0 - 17.9 to 10.0 - 0.6% 2.0% 0.99 CoCSA (mm2) 0.77 9.0 -5.5 to 7.4 1.1% 4.6% 0.95 TrCSA (mm2) 0.98 15.4 - 15.4 to 5.6 - 1.0% 1.9% 0.99 Ix (mm4) 0.68 1957 -2098 to 1237 -1.9% 8.5% 0.89 I y A (mm4) 0.89 1024 -216 to 1346 4.2% 6.5% 0.95 Sx (mm4) 0.55 73.0 -61.1 to 58.7 0.2% 7.7% 0.91 Sy (mm4) 0.77 59.0 -36.7 to 64.6 1.0% 6.8% 0.93 b) 1.5 T M R I R 2 Difference (mm2) 95% CI % Difference CV ICC ToCSA (mm2) 0.99 18.2 - 18.6 to 4.6 - 0.9% 1.7% 0.99 CoCSA (mm2) 0.87 10.1 -8.6 to 6.8 - 1.2% 5.6% 0.93 TrCSA (mm2) 0.98 16.8 -18.7 to 6.5 - 0.8% 1.9% 0.99 Ix (mm4) 0.91 1677 - 1030 to 1609 1.4% 8.6% 0.96 iy (mm4) 0.93 1062 -943 to 773 - 3.0% 8.7% 0.96 Sx (mm4) 0.75 80.3 -49.3 to 94.1 2.6% 9.1% 0.87 Sy (mm4) 0.83 79.4 -82.0 to 32.1 - 5.9% 10.1% 0.91 94 Results Table 13. 95% limits of agreement for inter-acquisition reliability. 3 T (n=12) 1.5 T (n=12) ToCSA (mm2) 48.4 40.3 CoCSA (mm2) 22.4 26.6 TrCSA (mm2) 36.4 43.6 Ix (mm4) 5775 4569 Iy (mm4) 2705 2973 Sx (mm4) 207.6 248.4 Sy (mm4) 175.5 197.7 5.3.3 Is Inter-Analysis Reliability Greater with 3 T than 1.5 T? 5.3.3.1 Inter-Analysis Reliability with 3 T For measurements made with the 3 T MRI system, the associations between repeated analyses ToCSA and TrCSA were high (R 2 = 0.99 for both) (Table 14). The associations between derived variables were slightly lower (e.g. Sx, R = 0.73). 95% confidence intervals did not include zero for ToCSA, Ix, Iy, and Sy. This suggests that the second measurements for these variables were systematically larger than the first measurements. Percent differences between scans were low for most variables from both systems, ranging from 0.9%> to 6.0%). Within-specimen coefficients of variation (CV) were low for directly measured variables (e.g. 0.9%> for ToCSA). However, CVs were slightly higher for derived variables, ranging from 3.0% for CoCSA to 5.9% for Sx. Intraclass correlation coefficients were very high for all variables measured. 5.3.3.2 Inter-Analysis Reliability with 1.5T For measurements made with the 3 T MRI system, the associations between repeated analyses ToCSA and TrCSA were high ToCSA and TrCSA were high (R 2 = 0.99 for both) (Table 14). The associations between derived variables was also high (e.g. Sx, R 2 = 0.94). The 95% confidence intervals crossed zero for all variables measured with both systems, suggesting there was no systematic difference between analysis times. Percent differences between scans were low for most variables from both systems, ranging from 0.4%) to 7.0%). Within-specimen coefficients of variation (CV) were low for directly measured variables (e.g. 0.5%> for ToCSA). However, CVs were slightly higher for derived variables, 95 Results ranging from 3.3% for CoCSA to 6.6% for Sx. Intraclass correlation coefficients were very high for all variables measured. Table 14. Inter-analysis reliability for a) 3 T MRI (n = 12) and b) 1.5 T MRI (n = 12). Differences are presented as (2 n d scan - 1st scan). Percent differences are presented as (2 n d scan - 1st scan/mean of two scans). a) 3 T MRI R2 Difference (mm2) 95% CI % Difference CV ICC ToCSA (mm2) 0.99 7.6 3.4 to 10.3 0.9% 0.8% 1.00 CoCSA (mm2) 0.84 4.9 - 1.8 to 6.8 2.2% 3.1% 0.98 TrCSA (mm2) 0.99 5.6 -0.1 to 8.7 0.6% 0.8% 1.00 Ix (mm4) 0.84 947 82 to 1439 5.5% 5.2% 0.99 iy (mm4) 0.92 466 54 to 715 3.9% 3.9% 0.99 Sx (mm4) 0.73 44.7 -0.9 to 68.5 5.9% 5.9% 0.98 Sy (mm4) 0.86 28.1 2.1 to 43.0 3.8% 3.8% 0.99 b) 1.5 T MRI R2 Difference (mm2) 95% CI % Difference CV ICC ToCSA (mm2) 0.99 4.8 -0.5 to 5.8 0.4% 0.5% 1.00 CoCSA (mm2) 0.96 5.6 -6.3 to 1.8 - 2.0% 3.3% 0.98 TrCSA (mm2) 0.99 5.9 2.1 to 7.8 0.9% 0.7% 1.00 Ix (mm4) 0.98 1060 -470 to 981 - 7.0% 5.8% 0.99 iy (mm4) 0.98 563 -494 to 326 - 0.9% 5.1% 0.99 Sx (mm4) 0.94 53.0 -29.0 to 43.4 -6.1% 6.6% 0.97 S y . (mm4) 0.96 36.4 -33.2 to 19.5 0.4% 4.8% 0.98 5.3.3.3 Comparison of Inter-Analysis Reliability Between 3 T and 1.5 T -y Similar percent differences, R , CV and ICCs for variables measured with both systems suggest that inter-analysis reliability did not differ between 1.5 and 3 T MRI systems. 96 Results Limits of agreement (the LOA), at the 95% level, are presented in Table 15. The L O A for each independent variable was similar for each imaging system, indicating that inter-analysis reliability did not differ between systems. Table 15. 95% limits of agreement for inter-analysis reliability. 3 T (n=12) 1.5 T (n=12) ToCSA (mm2) 11.6 11.0 CoCSA (mm2) 14.3 14.0 TrCSA (mm2) 14.6 9.8 Ix (mm4) 2252 2512 Iy (mm4) 1097 1419 Sx (mm4) 115.0 125.4 Sy (mm4) 67.7 91.3 97 Discussion 6. DISCUSSION In this thesis I addressed the question of whether MRI had potential clinical utility to predict hip fracture. The results of this study indicate that MRI measures of cortical geometry in the femoral neck are highly associated with failure load in human cadaveric proximal femora. I will discuss these results and the implications of these findings in detail below. 6.1. The Use of Mechanical Testing to Induce Fracture This study mechanically tested cadaveric specimens to induce clinically relevant fractures in the proximal femur. Prior to discussing the results relating to the specific objectives and hypotheses investigated in this study, I present comparisons of the results of mechanical testing with previous studies, specifically relating to the fracture types and failure load. Four key studies (Table 16) previously induced fracture in a sideways falling configuration in human cadaveric specimens. The distribution of fracture types was different for each study. Lochmuller et al. [12] used a loading apparatus which provided a prototype for the design for the current apparatus. They reported only 24% intertrochanteric fractures, compared with 62%> in the present study. Courtney et al. [65] tested their specimens at 100 mm/s, the same displacement rate as used in this study. In ten specimens, Courtney et al. observed one subcapital fracture, four femoral neck fractures (40%>), four intertrochanteric fractures (40%>) and one fracture unidentifiable by radiograph [65]. In contrast, Cheng et al. [94] reported 59% intertrochanteric fractures. Thus the current study produced more intertrochanteric and fewer femoral neck fractures than observed in previous studies using similar methodology. Eckstein et al. [69] used the same loading apparatus as Lochmuller et al. [12], and thus the same loading apparatus which provided a prototype for the current design8. The failure load in the current study was slightly higher (4353 N , SD 1886 N) than the failure load observed by Eckstein et al. (3926 N , SD 1656 N). However, Eckstein et al. loaded the specimens at 6.6 mm/s, compared with 100 mm/s used in the current study. Courtney et al. [65] were the only other researchers to load femora to failure with a displacement rate of 100 mm/s. When comparing mean failure load between studies, it is important to consider the displacement rate as bone is a Failure load was not reported by Lochmuller et al. 2003 98 Discussion viscoelastic material. The mean failure load observed by Courtney et al. 9 was similar to that observed in this study, approximately 4250 N (SD 1500 N). Thus the distribution of fracture types differed from studies using the same loading apparatus. However, results from other studies are not consistent. The mean failure load observed in this study was consistent with earlier studies. 9 Values were not reported in the text of this paper, but have been estimated from Figure 2. 99 Discussion Table 16. Mean failure loads and correlations with D X A aBMD measured in different proximal femur subregions, as well as percentages of femoral neck (FN) and intertrochanteric (IT) fractures as reported by studies that loaded human cadaveric femora to failure in a sideways fall simulation. Author Failure load (N) R with R with (Mean ± SD) Troc aBMD F N aBMD Fracture Similarity to current Distribution study Courtney et al. [65]1 0 73 + 7 yrs (n=10) Lochmiiller etal. [12] 80+ 10 yrs (n = 126) Eckstein et al. [69] 79 + 11 yrs (n= 108) Cheng et al. [94] 69+ 15 yrs (n = 64) Manske et al. 69+ 16 yrs (n= 36) 4250+ 1500 0.45 0.72 0.48 0.53 3926+ 1656 3978+ 1414 0.88 0.71 4353 + 1886 0.71 0.39 40% FN Displacement rate 40% IT 100 mm/s 50% FN 24% IT 67% FN 19% IT 41% FN 59% IT 9% FN" 62% IT Same loading apparatus Same loading apparatus More IT than FN fractures 6.2. The Role of MRI Assessment of the Femoral Neck to Predict Failure Load 6.2.1. MRI Assessment of the Femoral Neck is Associated with Failure Load Cortical bone geometry is strongly associated with proximal femur strength [94], however the association of MRI measures of femoral neck geometry with failure load has not previously been evaluated. I found that measures of femoral neck cross-sectional geometry obtained from 3 T MRI systems are strongly associated with failure load in human cadaveric femora. CoCSA and Ix explained significant variance in failure load (R z = 0.47 for both). 1 0 Courtney et al. tested both old and young cadaveric specimens. All references to Courtney et al. discuss the older Decimens . Basicervical fractures (9%) have not beei femoral neck or intertrochanteric fractures speci e s. " een included in this classification as they are inconsistently grouped with 100 Discussion Although no published studies have examined the relationship between femoral neck cortical geometry and proximal femur failure load, these MRI results concur with previous QCT and pQCT studies that reported the importance of the contribution of cortical bone at the femoral neck to strength. Augat et al. [23] indicated that femoral neck cortical area explained 44% of the variance in failure load when measured by pQCT. Cheng et al. [94] determined that the cortical area measured in the femoral neck by QCT explained 66%> of the variance in failure load. 6.2.2. Comparison Between MRI, DXA and HSA Imaging Systems 6.2.2.1. Comparison of Associations Between Independent Variables Measured with MRI, DXA and HSA, and Failure Load From a clinical perspective, prospective epidemiologic studies have indicated that proximal femur aBMD, measured by D X A , is an important predictor of hip fracture [61, 76]. From an engineering perspective, the cross-sectional geometry (e.g. cross-sectional area and bending indices such as area moment of inertia), determines the stresses on a structure in bending and axial loading [131]. Thus, it was important to evaluate and compare the association of both aBMD and cross-sectional geometry to failure load of the proximal femur. I found that trochanteric aBMD measured with D X A had the highest association with failure load (R 2 = 0.70). This association, although greater in magnitude, was not significantly different than the association between Ix measured with MRI and failure load (R = 0.47), although there was a trend towards significance, p = 0.10. Areal B M D encompasses many bone properties and thus is capable of explaining more variance in failure load than a single bone property such as CoCSA or Ix. The primary reason for the lack of significance between D X A trochanteric aBMD and MRI Ix associations with failure load may have been the small sample size. With only 27 specimens included in the analysis, there may not have been sufficient power to demonstrate a statistically significant difference between predictive ability of independent variables. As comparison between models was not my primary a priori question, sample size was calculated based on increasing the explanatory power within individual models by 3%>. 101 Discussion Additionally, there was no significant difference between the association of femoral neck aBMD measured with D X A and failure load (R 2 = 0.39), and the association between Ix measured with MRI and failure load (R 2= 0.47), p = 0.64. Thus, in the femoral neck, IXMRI and aBMDoxA have similar predictive capabilities. Conversely, there were no significant differences between the correlations of geometry variables measured with MRI and failure load, and the correlations of geometry variables measured with HSA and failure load. Thus, the predictive abilities of geometry variables measured with MRI are similar to the predictive abilities of geometry variables estimated with HSA. Despite the similarities in predictive abilities of variables measured with MRI and HSA, it may be important to measure cross-sectional geometry (e.g. CoCSA) directly using MRI rather estimating cross-sectional geometry with approaches such as HSA (e.g. CSA). For example, MRI provides the ability to distinguish between cortical and trabecular bone compartments, whereas C S A H S A is a measure that includes both cortical and trabecular bone. Therefore, i f a change is observed in C S A H S A , one is unable to determine whether that change occurred in the cortical or trabecular bone compartment. In contrast C O C S A M R I provides a means to determine the geometric basis that explains change in bone strength. That said, given the considerable expense and limited accessibility to MRI systems, D X A and HSA would be the clinical instrument of choice at this point. 6.2.2.2. Addition of Femoral Neck Geometry to DXA-Derived Regression Models Regression models including v B M D and cross-sectional geometry measured with QCT in the proximal femur have, in some studies, explained more variance in failure load than either property alone [22, 24]. Thus, I chose to construct hierarchical regression models comprised of aBMD and one cross-sectional geometry variable. In general, aBMD was strongly associated with measures of cross-sectional geometry (e.g. Figure 24, page 78). I found that the combination of MRI assessment of femoral neck geometry (Ix) and femoral neck aBMD by D X A explained a greater proportion of variance in failure load (R = 0.55 for a model including Ix) than femoral neck aBMD alone (R 2 = 0.39). In contrast, MRI assessment of femoral neck geometry (Ix) did not significantly contribute to variance in failure load in 102 Discussion combination with trochanteric aBMD (R2-change = 0.03). Thus, the variance explained by a combination of femoral neck aBMD and Ix was greater than the variance explained by either alone. However, the variance explained by the combination of trochanteric aBMD and Ix was not greater than the predictive value of trochanteric aBMD alone. There are several explanations for the difference between models derived from femoral neck aBMD and trochanteric aBMD. First, with the femoral neck aBMD model, area moment of inertia (Ix) had a slightly higher association with failure load (R2 = 0.47) than femoral neck aBMD (R2 = 0.39), although this difference was not significant. The contribution of the independent variables to the variance in failure loads is illustrated by the Venn diagram below (Figure 41). Figure 41. Venn diagram illustrating the variance in failure load explained by IxM R I and femoral neck aBMDnxA. Second, in the trochanteric aBMD model, aBMD alone accounted for 70% of the total variance in failure load. In fact, trochanteric aBMD accounted for more variance in failure load than any other variable measured with any imaging system. As bone cross-sectional area (in the sagittal plane) is included in the measurement of aBMD, these variables are by nature not independent of each other. This is illustrated visually in Figure 42. This 'size artefact' within aBMD has been discussed at length in the literature [132, 133]. 103 Discussion ic aBMD Figure 42. Venn diagram illustrating that Ix shares a large portion of the variance with trochanteric aBMD.. Simply, aBMD is a composite measure of bone, obtained from planar projectional images of the proximal femur. Thus, aBMD is comprised of bone cross-sectional area in the sagittal plane and volumetric bone mineral density. Volumetric BMD, and thus aBMD, when measured in this way is an amalgamation of bone mineral content, cortical porosity, degree of mineralization, trabecular number and thickness. Areal BMD therefore represents bone cross-sectional area as well as bone density and its sub-properties. It is therefore not surprising that Ix explained little or no additional variance in failure load when trochanteric aBMD was included in the model. It is important to note that each model created with both MRI and D X A variables included only one independent variable from each imaging system. Within each imaging system, the independent variables that were strongly associated with failure load were also very highly correlated with each other. Within variables measured with MRI, for example, Ix was strongly associated with failure load (R = 0.69) and also with CoCSA (R = 0.82) and Sx (R = 0.91), Appendix 2. Within variables measured with DXA, femoral neck aBMD was highly associated with failure load (R = 0.63), and also with trochanteric BMC (R = 0.93). Cross-regional similarities in bone properties have been reported extensively [12, 65]. Between systems, I X M R I was also associated with femoral neck aBMD (R = 0.57). Although models could have been constructed in a variety of ways I selected a variable that represented cross-sectional geometry 104 Discussion that was highly associated with failure load, but less associated with aBMD, for input into the regression model. One specific limitation of these prediction models was that I was unable to evaluate cross-sectional geometry in the trochanteric region from the M R images. This limitation is discussed further in subsection 6.4.2.C heng et al. [94] found in 64 cadaveric specimens that trochanteric CoCSA measured with QCT and trochanteric aBMD measured with D X A were both associated with failure load (R 2 = 0.83 and 0.88, respectively). Thus, I hypothesize that the relationship between trochanteric cross-sectional geometry measured with MRI and failure load may be similar to the relationship between trochanteric aBMD measured by D X A and failure load. Recommendations for measuring the trochanteric region using MRI are discussed further in the subsection 6.6.1.1. However, several QCT studies evaluated the contribution of cross-sectional geometry to failure after accounting for vBMD. Lang et al. [22] tested 26 specimens and found that in the femoral neck adding the minimum total cross-sectional area after accounting for trabecular v B M D significantly explained more of the variance in failure load. However, in the trochanteric region trabecular v B M D explained the most variance in failure load. Adding minimum total cross-sectional area did not significantly improve the variance explained in this model. In contrast, Lotz and Hayes [24] tested 22 specimens and found that in the trochanteric region, a regression model that included total vBMD and total cross-sectional area assessed by QCT explained a greater proportion of variance in failure load than models that included only density or only geometric parameters. Thus, results that demonstrated the importance of accounting for density and geometry in the trochanteric region are somewhat inconsistent. To my knowledge, this is the first study to evaluate whether the contribution of cross-sectional geometry properties, measured by MRI, improves failure load prediction after accounting for aBMD by D X A . My findings using a novel instrument to assess cross-sectional geometry supports the findings of Lang et al. with QCT that both density and cross-sectional geometry are associated with failure load in the femoral neck, but not the trochanteric region. Measurements of the trochanteric region with MRI would confirm or disprove this result. 105 Discussion Finally, the strong relationship between trochanteric aBMD and failure load was also likely related to the high percentage of intertrochanteric fractures that occurred in this study. This relationship will be discussed further in section 6.2.3. In summary, cross-sectional geometry assessed by MRI significantly improved the prediction of failure load after accounting for aBMD in the femoral neck. In the trochanteric region, aBMD was the sole predictor of failure load, among those geometric and density variables assessed by MRI and D X A . 6.2.2.3. Comparison of DXA + MRI Models to HSA Models Hip Structural Analysis (HSA) has been widely adopted as an instrument to estimate geometric properties of the proximal femur from D X A data [114, 132, 134]. Limitations of this approach have been discussed at length, but it remains a popular approach in research studies given the accessibility and affordability of D X A instruments. However, to my knowledge, only one study has evaluated the ability of HSA outcomes to predict failure at the hip, and none have compared HSA to proximal femur geometry assessed by MRI. In bivariate models, femoral neck CoCSA measured with MRI explained a greater proportion of variance in failure load (47%) than CSA estimated by HSA at the narrow neck (36%>). However, this difference in predictive ability was not significant. There were also no significant differences in Ix and Sx in predicting failure load as measured by MRI and HSA. These results were surprising, as it appears as though an estimation of cross-sectional geometry predicts failure to a similar extent as more costly direct measurement. I addressed the question of whether a model that combined several variables (aBMD measured with D X A and Ix measured with MRI) explained more of the variance in failure load a combination of the same variables measured with HSA. A model that included femoral neck aBMDoxA and IXMRI explained a greater proportion of variance in failure load (55%>) than a model that included narrow neck aBMD H sA and IHSA (37%), but this difference in variance accounted for between the models was not significant. In contrast, a combination of parameters estimated by HSA in the trochanteric region (aBMDnsA, IHSA) explained a greater proportion of variance in failure load (64%>) than the model 106 Discussion that combined femoral neck IXMRI and aBMDoxA (55%). However, there was no significant difference between these two models. Thus, the variance in failure load explained by the models developed from variables measured by two instruments, D X A and MRI, did not differ from models developed using HSA variables. There are several possible reasons for the lack of a significant difference in predictive ability between HSA and MRI + D X A models. First, as I am measuring bone geometry and density in both models (although in slightly different ways) I would expect some consistency between models. Second, the study may have lacked sufficient sample size to demonstrate a statistically significant difference between models. As this was not my a priori question, sample size was calculated based on increasing the explanatory power of individual models by 3%, and not on a comparison between models. That said, for models of the femoral neck, the p-value of the difference between correlations was close to achieving significance (p = 0.06). To my knowledge, no studies have examined linear regression models derived from HSA variables and their association with proximal femur failure load. In addition this is the first study to compare the ability of HSA with MRI to predict proximal femur fracture. However, estimates of predicted tensile strength in the femoral neck with an earlier version of the HSA software explained 79% of variance in failure load in cadaveric femora tested to failure in a stance configuration [64]. This estimation actually derived the tensile stress based on estimated cross-sectional properties. In summary, models that incorporated direct measures of bone cross-sectional geometry assessed by MRI were not statistically better predictors of failure load than models that incorporated estimated cross-sectional geometry by HSA. However, this study may have been underpowered to demonstrate a difference between models. 6.2.3. Secondary Findings of Interest 6.2.3.1. Regional-Specificity of Failure Load Prediction It has been suggested in D X A studies that regional measures of aBMD are the best predictors of fractures in that region [13]. That is, proximal femur aBMD is a better predictor of proximal femur failure. The contributions of subregions of interest (e.g. the trochanteric region) within a global region of interest (e.g. the total proximal femur) to failure are less clear [6, 135]. 107 Discussion In this study density and geometry measured in the trochanteric region explained more variability in failure load when compared with the same measures assessed in the femoral neck region. More specifically, D X A trochanteric aBMD and HSA-derived intertrochanteric aBMD, CSA and section modulus were most highly correlated with failure load (R 2 = 0.60 to 0.70). Many population-based studies reported that measurements closest to the site of fracture (e.g. trochanteric region for intertrochanteric fractures, femoral neck region for femoral neck fractures) are the best predictors of fracture at that site [6, 76]. In the current study, 62% of fractures occurred in the intertrochanteric region, which might explain the strong relationship . with trochanteric aBMD and HSA-derived variables in the intertrochanteric region. This finding is in line with previous clinical observations that fracture prediction using outcomes from imaging techniques may be region-specific. Three key studies (Table 16, page 100) examined the association between failure load and aBMD in the proximal femur regions in human cadaveric specimens. Courtney et al. [65], tested their specimens at 100 mm/s, the same displacement rate as used in this study. They found that femoral neck aBMD was explained a greater proportion of failure load than trochanteric aBMD. They reported approximately equal numbers of femoral neck and intertrochanteric fractures. Lochmuller et al. [12] used a loading apparatus which provided a prototype for the current design. They reported only 24% intertrochanteric fractures, and a slightly higher association between femoral neck aBMD and failure load than trochanteric aBMD and failure load. In contrast, Cheng et al. [94] reported a greater percentage of intertrochanteric than femoral neck fractures, and found that trochanteric aBMD explained a greater proportion of the variance in failure load than femoral neck aBMD. In addition to these three key studies that examined the association between aBMD and failure load of the proximal femur, a QCT study conducted by Lotz and Hayes [24] is also relevant. Lotz and Hayes reported 50% intertrochanteric fractures12 and 25% femoral neck fractures, and showed that variables measured in their intertrochanteric slice were more highly associated with failure load (up to R 2 = 0.93 for average density x total cross-sectional area) than variables measured in their basicervical slice (up to R 2 = 0.54 for average density x area moment of inertia). Region-specific parameters may be associated with 1 2 In this study, no description of classification was provided, thus it is inappropriate to say whether the basicervical fractures could be grouped with femoral neck or intertrochanteric fractures. 108 Discussion the type of fracture that is sustained, although the literature was not consistent. I didn't have a sufficient sample size to draw conclusions regarding regional-specificity in this study. 6.2.3.2. High Rate of Intertrochanteric Fractures I observed a high proportion of intertrochanteric fractures (62%) in the present study. In this section, I will explore possible explanations for this finding. The first question is whether the rate of intertrochanteric fractures observed in this study was higher than might be expected from the literature, or whether the rate of femoral neck fractures was low. A high percentage of intertrochanteric fractures may be a function of: 1) specimens that were weak in the trochanteric region, or 2) the loading configuration I utilized to induce fracture. In the present study, 62%> of fractures occurred in the intertrochanteric region. In contrast, only 9% of fractures occurred in the femoral neck. This distribution of fracture types is different from typical clinical presentations [54]. Epidemiological studies of adults aged 50 years and older have reported the percentage of intertrochanteric fractures to range from 36%) [136] to 53%) [137], and for neck fractures to range from 37% [54] to 64%) [136]. Thus, my study had a considerably higher percentage of intertrochanteric fractures than would be expected in a clinical population. Two human cadaveric studies that used a similar loading apparatus to the one used in the present study observed lower percentages of intertrochanteric fracture - 19%) by Eckstein et al. [69] and 24%o by Lochmuller et al. [12]. In contrast, Keyak et al. [72] produced 53%> intertrochanteric fractures and Cheng et al. [94] 59%) intertrochanteric fractures. Both of these studies used a biomechanical configuration that loaded the femoral head rather than the greater trochanter. Although the loading configuration was similar to that of Eckstein et al. [69] and Lochmuller et al. [12], there were differences in the testing protocol that may have contributed to the higher percentage of intertrochanteric fractures. There are a number of methodological differences in specimens and testing protocol that likely contribute to the disparate findings I report between the present study, and the studies performed by Lochmuller et al. [12] and Eckstein et al. [69]. These include the age, sex and preservation of the specimens, specimen positioning, and the displacement rate. 109 Discussion Lochmiiller, Eckstein and coworkers used embalmed (formalin-fixed) cadavers in their studies, while I used unembalmed, previously frozen cadavers. Lochmuller, Eckstein and coworkers reported that embalming does not affect the measurement of B M C by D X A [9]. However, there is no evidence that demonstrates whether it may affect other bone properties that play a considerable role in determining fracture type and location. The advantage of using embalmed specimens is that they are readily available compared with unembalmed specimens. However, to my knowledge, no study has reported the effects of formalin-fixation on the mechanical properties of bone. There may also have been subtle differences in specimen positioning between studies as these methods were not reported by Lochmuller, Eckstein and coworkers. For example, they did not report the location where the femoral shaft was cut, nor did they report the length or percentage of shaft length that was potted. Thus, the length of the exposed portion of the femur may have differed between studies. Lochmuller, Eckstein and coworkers reported internally rotating the femoral neck by 15 degrees, as was done in the present study. However, in the only figure they provided to demonstrate positioning (Figure 2, [69]), the femoral neck appears externally rotated by 15 degrees. Thus, I may have observed more intertrochanteric fractures than Lochmuller, Eckstein, and coworkers due to internally rotating the femurs rather externally rotating them. It is worth noting that Pinilla et al. [68] found that varying the angle of internal rotation from 0° to 30°, testing 11 specimens per group, had no effect on fracture patterns, but did significantly affect failure load. However, they did not test any specimens in external rotation [68]. As previously mentioned, the failure load in my study was slightly higher than the failure load observed by Eckstein et al. [69] (Table 16, page 100). However, the failure load observed in the present study with a displacement rate of 100 mm/s was similar to that observed by Courtney et al. [65] at the same loading rate. Findings reported in a conference proceeding suggested that the association between femoral neck aBMD and failure load was higher at a low loading rate (R 2 = 0.89 at 0.7 mm/s) than at a higher loading rate (R 2 = 0.60 at 4000 mm/s) [138]. To my knowledge, no studies has investigated the effect of loading rate on fracture location in the proximal femur. The loading 110 Discussion rate may also be a possible explanation for the difference in the proportion of intertrochanteric fractures I observed compared to Lochmuller, Eckstein, and co-workers. Another possible reason for the high percentage of intertrochanteric fractures may have been that the specimen sample was disproportionably weak in the trochanteric region. Theoretically this would be reflected by lower trochanteric aBMD values than femoral neck aBMD values as compared with a same age and sex normative sample. However, the D X A results do not support this hypothesis. 2 2 Mean trochanteric aBMD for males in this study was 0.66 g/cm (SD 0.13 g/cm ). This is lower than mean trochanteric aBMD (0.84 g/cm , SD 0.008 g/cm ) reported in a sample of 272 men (mean age = 74 + 4.5 years) in the Framingham study [139]. Mean femoral neck aBMD for males in this study was 0.69 g/cm2 (SD 0.13 g/cm2). This is also lower than femoral neck aBMD 2 2 reported in the Framingham study (0.89 g/cm , SD 0.008 g/cm ). For females in this study, mean trochanteric aBMD was 0.57g/cm z(SD 0.13), while mean femoral neck aBMD was 0.64 g/cm2 (SD 0.17). These values are comparable to those measured in women 65 years of age and older in the Study of Osteoporotic Fractures (trochanteric aBMD 0.56, SD 0.10 g/cm2, femoral neck aBMD 0.65, SD 0.11 g/cm2). [76]. In addition, the means for males and females for both the trochanter and the femoral neck would be considered osteopenic according to the NHANES III reference database [140]. Although mean trochanteric aBMD was slightly lower than mean femoral neck aBMD for both males and females, these are consistent with the lower trochanteric aBMD than femoral neck aBMD reported in reference data [77, 139] and osteopenic cutoffs [140]. Therefore, D X A data suggest that the specimens in my study were not disproportionably weak in the trochanteric region compared with the femoral neck region. There are limitations to drawing these conclusions based solely on D X A data, however, no referent data exists for other imaging systems. In summary, it is likely that methodological, but not biological, factors contributed to the differences I observed in fracture patterns compared with the published literature. However, it is not possible to identify with certainty the specific factors that explain this difference. Ill Discussion 6.2.3.3. Left-Right Differences in Fracture Types Four of six specimen pairs in the final analysis experienced different fracture types. While the mean side-side difference in failure load was similar to that reported by Eckstein et al. [69], I expected that there would be greater similarity in fracture types between pairs. This highlights the potential variability in setting up the loading configuration, which could have influenced the fracture type. 6.2.3.4. Bending Strength Indices and Cortical Area are Similarly Associated with Failure Load Bending strength indices such as area moment of inertia (Ix, R 2 = 0.47) and section modulus (Sx, 2 2 R z = 0.42) were associated with failure load similarly to cortical cross-sectional area (CoCSA, R = 0.47). In the following section, I discuss the assumptions made in calculating each of the bending indices, Ix and Sx , which may explain these similar relationships. For a structure loaded in pure bending, Ix and Sx are the primary predictors of bending strength [131]: _My _M ^"bending j ^ where Obending is the bending stress, M i s the bending moment andjy is the distance from the neutral axis. In calculating these variables I assumed that the most important stresses in the femoral neck were bending stresses. However, a combination of axial loading and bending occurred during mechanical testing of these specimens. One of the primary determinants of stress in axial loading is cross-sectional area [131]: F °0X101 ~CSA where a axial is the axial stress, F is the applied load, and CSA is the cross-sectional area. Therefore, the high association between failure load and femoral neck C o C S A is quite reasonable when given the loading conditions. There were other key assumptions made when calculating area moment of inertia and section modulus. First, I assumed the femoral neck to be a shell with a constant cortical density and a hollow centre. This assumption is valid for regions composed of primarily cortical bone (e.g. the 1 31 am discussing Ix and Sx only for simplicity. Iy and Sy were similarly correlated with failure load. 112 Discussion femoral shaft). However, the femoral neck is comprised of approximately 60% trabecular and 40% cortical bone [141]. Trabecular bone and bone marrow likely contribute to bending strength in this region. In addition, the apparent cortical density may differ to some degree throughout the cortical compartment but it is unlikely that a small variation would significantly affect the strength indices. Second, I assumed that the neutral axis passed through the centroid of each bone cross-section. This assumption is only valid for a beam subjected to pure bending when the stresses remain in the elastic range [131]. Therefore, this assumption is violated when the proximal femur is mechanically loaded to failure. Third, I did not attempt to calculate the bending strength indices about the principal axes. I calculated Ix (where the x-axis was the anterior-posterior axis) and Iy (where the y-axis was the superior-inferior axis) rather than about the principal axes (Imax and Imin) for several reasons: 1) the computation was simpler, as the images were anatomically oriented along these axes, 2) detailed investigation of an archaeological sample found that in the femoral neck, Imax is almost directly in the superior-inferior direction [142], and 3) computing both Ix and Imax with femoral neck images obtained with QCT demonstrated that Ix and Imax differed less than 5 mm 4 (the mean value calculated in this sample was approximately 15 000 mm4) in the same specimen. In addition to these assumptions, area moment of inertia and section modulus rely on the accurate identification of cortical bone. When a pixel is inappropriately pixel defined or not defined as cortex, the magnitude of the error is exponential when calculating area moment of inertia. As indicated in the equation below, the area moment of inertia is proportional to the square of the distance from the neutral axis: '• - [yldA It is possible that by improving the accuracy of cortical area measurements, area moment of inertia and section modulus would also be measured more accurately. This would result in improving the bivariate association between these variables and failure load. In summary, it appears reasonable that CoCSA and bending strength indices explain similar proportions of variance in failure load. In vivo, proximal femur fractures typically occur as the 113 Discussion result of falls. Since these falls, and thus the directions of applied loads differ considerably between individuals, axes about which bending occurs may also differ. Therefore, CoCSA may be a better predictor of fractures in a population. 6.3. Comparisons Between 1.5 and 3 T MRI 6.3.1. Systematic Differences Between 1.5 T and 3 T MRI A previous study reported poor reliability for measurement of cortical cross-sectional area (CoCSA) with 1.5 T MRI [25]. An MRI system with a stronger magnet (e.g. 3 T) should improve image quality, and thus potentially improve reliability. The secondary aim of this study was to determine whether femoral neck geometry could be assessed more reliability with a 3 T MRI system than a 1.5 T MRI system. In addition, I determined the difference in measurements between the two systems. CoCSA, Iy, and Sy, assessed by a 3 T MRI system were significantly lower than the same variables assessed by a 1.5 T MRI system. There was no statistical difference between systems for measures of ToCSA, TrCSA, Ix or Sx. There are a number of differences between these two imaging systems that might account for these differences. Greater in-plane resolution (0.23 mm x 0.23 mm) and smaller slice thicknesses (2.0 mm) were used in the 3 T system compared with the 1.5 T system (0.29 mm x 0.29 mm, slice thickness 3.0 mm). As the partial volume effect decreases with increased resolution, I hypothesize that measurements made with the 3 T were more accurate. Qualitatively, the endocortical and periosteal borders appeared sharper in images from the 3 T system than in images from the 1.5 T system. However, as this study did not assess the true bone cross-sectional areas using histomorphometry or another criterion method, I cannot compare the accuracy between the two MRI systems. 6.3.2. Reliability of MRI Measures of Cross-Sectional Geometry There are a number of potential sources of error in any imaging technique. These include both acquisition (positioning and system error), and analysis error. To address this I assessed the reliability of each of these in a number of ways. 114 Discussion 6.3.2.1. Inter-Acquisition Reliability for 3 T MRI I demonstrated that reliability for image acquisition (inter-acquisition reliability) was highest for measurements of total cross-sectional area (ToCSA) and trabecular cross-sectional area (TrCSA). For CoCSA and all other derived variables, inter-acquisition reliability was lower. The increased error for derived variables reflects the compounded error from each of variable used to derive that variable. For example, for CoCSA measured with 3 T MRI, the CV was 4.6% with repositioning, compared to 2.0% for ToCSA. 95% confidence intervals did not cross zero for inter-analysis reliability, indicating there was no systematic difference between acquisition times. Examining the limits of agreement (LOA) can help understand the absolute error in repeated measurements. In order to be a confident that a real change has occurred in repeated measurements of an individual over time, the change in CoCSA would have to be greater than the 95% LOA for inter-acquisition reliability (26.62 mm2). In the present sample, where the mean CoCSA was 132.52 mm2, one would need to see a change greater than 20% in an individual in order to be confident that a real change has occurred. In comparison, for ToCSA 2 2 (95% LOA for inter-acquisition reliability was 40.25 mm ), the mean ToCSA was 749.41 mm , thus one would need to see a change greater than 5% in an individual in order to be confident that a real change has occurred. Inter-acquisition reliability was similar for variables measured with 1.5 T MRI. Therefore, I will not discuss specific trends in the results here. 6.3.2.2. Inter-Analysis Reliability for 3 T MRI I demonstrated that reliability for image analysis (inter-analysis reliability) was higher for ToCSA and TrCSA than for variables derived from these measures. For example, the CVs were 0.8% for ToCSA, but 3.1% for CoCSA. 95% confidence intervals did not cross zero for ToCSA, Ix, Iy, and Sy. This suggests that values for the second measurements were systematically higher than the first measurements. In four of 12 specimens the difference in ToCSA between analyses was greater than 10 mm . Thus, the systematic difference observed in Ix, Iy and Sy can be explained by the variability in defining the periosteal border. 115 Discussion Conversely, significant differences were not observed for inter-acquisition reliability, although the percent differences, CVs and 95% LOAs were higher for inter-acquisition than inter-analysis reliability. It is possible that there was bias in my segmentation over the approximately 11 week analysis time frame. However, this bias was not observed when analyzing images for inter-acquisition reliability. When the sample size is small, large differences defining borders in one or a few specimens can cause large differences in the mean value. Finally, defining the periosteal and endosteal borders in M R images is a subjective process. Together, these factors could potentially contribute to measurement reliability. Inter-analysis reliability was similar for variables measured with the 1.5 T MRI system. The exception was that the 95% confidence intervals crossed zero for all variables measured with the 1.5 T MRI system. I will not discuss specific trends further. 6.3.3. Reliability was Similar for a 3 T MRI System and a 1.5 T MRI System. The main advantage of a 3 T MRI system is the greater signal-to-noise ratio than a 1.5 T MRI system, which produces greater image quality. However, the 1.5 T MRI systems are currently more common and thus more accessible. To my knowledge, bone measurement outcomes between systems have not previously been compared. There appeared to be no difference in inter-analysis or inter-acquisition reliability between 3 T and 1.5 T MRI systems. The R 2 , mean differences, CV, ICC, and LOA were similar for each independent variable measured with each system. This was the case for images acquired at different times, after repositioning (inter-acquisition reliability) and for images analyzed twice (inter-analysis reliability). This finding does not support my hypothesis that the higher image resolution in the 3 T MRI would result in improved reliability. For measurements with both systems, specimens were positioned and images were analysed in the same manner. As reliability was no different between the two systems (and image resolutions), it appears as though sources of error, reproducibility of specimen positioning and image analysis, are common in both. Thus to improve reliability the user should focus on these parameters. 6.3.4. Comparison of Reliability Results with Previous Research Reliability, presented as coefficients of variation, of femoral neck ToCSA and CoCSA were slightly better in this study than previously reported, in vivo, by my supervisor and colleagues 116 Discussion (McKay et al. [25]). The limits of agreement with repositioning are very similar between the studies. McKay et al. also assessed short-term precision without repositioning to assess the system errors, and findings are presented below in Table 17. Table 17. Comparison of reliability for ToCSA and C o C S A (coefficient of variation - C V , and 95% Limits of Agreement L O A ) reported in the current study (3 T MRI) and reported by McKay et al. (1.5 T MRI) [25]. Mans he et al. (n = 12) McKay et al. (n = 10) Var iab le With Between With Without repositioning Analyses repositioning repositioning ToCSA C V 2.0% 0.8% 2.5% 1.3% 95 % L O A 48.4 mm 2 11.6 mm 2 46 mm 2 21 mm 2 CoCSA C V 4.6% 3.1% 9.7% 6.1% 95 % L O A 22.4 mm 2 14.3 mm 2 26 mm 2 20 mm 2 Several methodological differences between my study and that of McKay et al. may account for the differences and similarities in results. First, although the CV was lower with repositioning, limit of agreement was similar. McKay et al. used younger participants (aged 10 to 46 years), thus the absolute areas measured by McKay et al. would have been smaller (mean values were not reported). The CVs may thus appear higher for the same absolute error. Second, the present study used different software (Analyze, Mayo Clinic) than McKay et al. (custom software developed for PV Wave, Visual Numerics, Boulder, Co). The custom software used by McKay et al. required that both periosteal and endocortical borders be defined manually. Using Analyze, I was able to apply a semi-automated analysis protocol that used a region-growing algorithm to define the endocortical border. However, I defined the periosteal border manually. Third, I am comparing results obtained from cadaveric specimens ex vivo to an in vivo study. This has several implications. First, a head coil was used with both the 3 T and 1.5 T systems for the cadaveric specimens, whereas the in vivo scanning required use of a larger torso-phased array coil. Head coils have greater signal-to-noise ratio than torso coils, resulting in greater image quality with the head coil. The field of view is also smaller in a head coil than in a torso -y coil, which results in a greater in-plane resolution for the same matrix size (0.29 mm vs. 0.49 mm 2 for 1.5 T MRI in the current study and McKay et al., respectively), which also results in greater image quality. Finally, in vivo scans are susceptible to movement which may occur 117 Discussion during the scan. Therefore, the improved reliability found in my study may not be observed in an in vivo study using the same protocol. The only other study to assess femoral neck geometry with a 1.5 T MRI was performed by Arokoski et al. [109]. They measured total volume in the femoral neck in four participants (in vivo), and found a CV of 2.01% with repositioning. Arokoski et al. used a different measurement approach to what I adopted. They segmented each slice five times, and reported the mean value. Thus, the difference assessed with repositioning was actually a difference between two means. Although using the mean of five analyses of the same image likely improved the reliability of analysis, it would substantially increase the amount of time required for segmentation. In summary, femoral neck cortical geometry is highly associated with failure load. However, i f these measures are to be used to assess fracture risk they must be highly reliable. As there was no difference in reliability at two different image resolutions (3 T and 1.5 T), improvements in positioning, localization and analysis techniques may be required to improve overall reliability. In future, the development of reliable, automated analysis methods would contribute significantly to this research area. 6.4. Limitations Related to the Study Methodology This study has several key limitations. I will discuss these in the following section with reference to the limitations associated with using cadaveric specimens, as well as the limitations associated with the measurements I made using MRI. 6.4.1. Cadaveric Sample To address the objectives of this study it was appropriate to use cadaveric specimens. However, there are limitations to using cadaveric specimens which I will discuss in the sections below. 6.4.1.1. Limited Knowledge of Medical Histories I had little or no knowledge of the medical histories of the specimens. While a radiologist screened x-rays of specimens for previous fracture and metastatic diseases, I had no knowledge regarding key information pertinent to one's bone health, such as the amount of bed rest and mobility immediately prior to death, as these could affect structural properties. Thus, it is 118 Discussion difficult to ascertain which sub-population of older adults my sample of cadaveric specimens represents. 6.4.1.2. Specimen Storage While one advantage of this study was the use of unembalmed, previously frozen cadaveric specimens rather than embalmed specimens, this conversely meant that specimens were subject to degradation with each period of thawing and refreezing. While thawing was required only for MRI scanning and mechanical testing, several thaw/freeze cycles may have occurred in each specimen (the number of specimens which were thawed for each procedure is indicated in brackets): 1) scanning in the 3 T MRI (all specimens), 2) scanning in the 1.5 T MRI in addition to the 3 T MRI (20 specimens) and, 3) MRI scans for reliability (24 specimens), and 4) partial thawing on another occasion (all specimens). Therefore, the specimens were thawed and refrozen once, at minimum, and four times, at maximum. It is accepted that storing bones at - 20°C has little effect on the mechanical properties of cortical bone [143] or whole bones [144]. Additionally, eight freeze-thaw cycles at - 20° C were shown to have no effect on the ultimate stress in a sample of bovine trabecular bone [145]. In a sample of canine cortical bone, five freeze-thaw cycles were shown to have no significant effect on ultimate stress [146]. Apart from concerns regarding the effects of freezing and thawing on bone mechanical properties, there were concerns with respect to imaging. When the specimens were thawed, degeneration of the marrow produced gas bubbles in the marrow cavity. Gas bubbles were visible in the MR images, which made segmentation slightly more challenging than it would have been in vivo. The accuracy of the segmentation is unknown, but because most of the gas bubbles were found in the marrow/trabecular compartment, the effect on cortical area measurements was likely minimal. 6.4.1.3. Missing Anatomical Structures The pelvis and majority of the soft tissue were absent from my cadaveric sample. In DXA and HSA analyses, the software uses the position of the pelvis to assist in anatomical landmarking. As a result, the femoral neck axis had to be manually adjusted in several specimens for both 119 Discussion standard DXA and HSA analyses. The lack of a pelvis should not have affected MRI acquisition or analysis. Attenuation of the x-ray beam in a DXA scan is affected by the amount of soft tissue surrounding the bone. The major source of the error is due to the variation in distribution of fat and muscle throughout the scan region of interest. Rice bags were placed above and below the bone, simulating a more homogenous tissue than that found in vivo. Thus, I speculate that the error in measured aBMD values observed in this study is lower than in vivo. In an MR image, identification of the periosteal border relied on the contrast between cortical bone and the material which surrounded it. Because cortical bone and air both appeared black, I provided a surrounding medium of water or ultrasound gel for contrast. While both mediums (water and ultrasound gel) provided sufficient contrast for my MRI analysis, intact soft tissue may have provided a more distinct contrast. A more distinct contrast would have been particularly useful in analyzing regions where there was little or no visible cortical bone. In this region there was very little contrast between the trabecular compartment and the surrounding aqueous environment. Improved contrast may result in higher precision and accuracy. 6.4.2. The Trochanteric Region was not Assessed with MRI MRI assessment was limited to the cortical bone in the femoral neck as I was unable to assess the trochanteric region from the MR images due to acquisition protocol. The MR images were acquired perpendicular to the axis of the femoral neck in order to obtain cross-sectional images of the femoral neck. While these images were acquired along this axis for the entire proximal femur, the centre of the field of view was placed on the femoral neck. Due to the size and placement of the field of view, the inferior-medial border of the intertrochanteric region was not captured in approximately one half of the specimens. Reconstruction of these images in a different plane was not a suitable option as the data were acquired in voxels at a resolution of 0.23 x 0.23 x 2.0, which were not cubic. Considerable interpolation is required if in-plane resolution differs considerably from the slice thickness and this would significantly decrease the quality of the image. 120 Discussion 6.4.3. Limitations of MRI Measurements 6.4.3.1. Accuracy Unknown, Moderate Reliability Accuracy of measurements at the femoral neck with MRI is unknown. Therefore, while I have shown that cortical bone geometry measurements in the femoral neck are associated with failure load, I do not know the accuracy of these measurements. Thus I do not know the contribution of accuracy to the findings reported. Because of the complex structure of a femoral neck cross-section, segmentation to identify cortical bone in the MR images was challenging. There were locations where cortical bone, between the trabecular/marrow and soft tissue/water compartments, was not visible. Thus the use of automated segmentation techniques, such as threshold-driven, region-growing algorithms was difficult, as the region 'grew' outside of the trabecular/marrow compartment when a seed was planted in that compartment. Region limits could be applied in order to use this technique in the trabecular/marrow compartment. In the cortical bone compartment, I could not apply this technique as the cortical bone borders were not visibly continuous. Therefore, to define the periosteal border, I manually moved points from the endocortical border to represent the periosteal border. This requirement for manual segmentation of the periosteal border likely reduced the reproducibility of the measurements. Finally, manual segmentation required a significant amount of time, approximately one hour per specimen, which would limit its utility in a large-scale study. 6.4.3.2. Presence of Gas in the Marrow Cavity Several cadaveric studies [11, 64] reported 'degassing' the specimens prior to imaging to remove gas bubbles in the specimens which may have formed during breakdown of the tissues after death. One of the effects of degassing would be drainage of an unknown proportion of the marrow. While removal of gas bubbles would have improved image contrast in the MRI, one must note that the fatty marrow also provides contrast within the image. While it is currently unknown whether degassing the specimens prior to MR imaging would increase or decrease the quality of the image, this method should be considered in future studies. 6.4.4. Collinearity Most of the independent variables measured in this study were highly correlated with each other. When redundant variables are entered in a regression model, they do not improve prediction, and 121 Discussion i n f l a t e t h e s i z e o f t h e e r r o r t e r m s [ 1 4 7 ] . T h i s w a s a n i m p o r t a n t c o n s i d e r a t i o n w h e n d e v e l o p i n g r e g r e s s i o n m o d e l s w i t h c o m b i n a t i o n s o f v a r i a b l e s . T h u s a d d i t i o n o f s u p p l e m e n t a r y v a r i a b l e s c o u l d n o t r e l i a b l y e x p l a i n m o r e o f the v a r i a n c e i n f a i l u r e l o a d . V i o l a t i o n o f t h e s e r e c o m m e n d a t i o n s r e s u l t s i n a w i d e n i n g o f t h e c o n f i d e n c e i n t e r v a l s , i n d i c a t i n g tha t t h e r e i s l e s s c e r t a i n t y w h e r e t h e i n t e r c e p t a n d s l o p e o f t h e p r e d i c t i o n e q u a t i o n l i e [ 1 4 7 ] . T h u s , m o d e l s w e r e d e v e l o p e d b a s e d b o t h o n t h e k n o w n a s s o c i a t i o n o f t h e i n d e p e n d e n t v a r i a b l e s w i t h f a i l u r e l o a d a n d o n t h e s t a t i s t i c a l r e l a t i o n s h i p b e t w e e n i n d e p e n d e n t v a r i a b l e s . 6.4.5. Paired Samples I u s e d s e v e r a l s p e c i m e n p a i r s ( s i x p a i r s i n t h e f i n a l a n a l y s e s ) i n t h i s s t u d y . It w a s h y p o t h e s i z e d tha t t h e u s e o f p a i r e d s p e c i m e n s w o u l d d e c r e a s e t h e v a r i a b i l i t y i n t h e s a m p l e a n d a r t i f i c i a l l y i n f l a t e t h e a s s o c i a t i o n o f i m a g i n g p a r a m e t e r s w i t h f a i l u r e l o a d . T o a d d r e s s t h i s i s s u e , I r a n s e c o n d a r y a n a l y s e s u s i n g m u l t i - l e v e l m o d e l l i n g to s t a t i s t i c a l l y a c c o u n t f o r t h e r e l a t i o n s h i p b e t w e e n t h e p a i r s . T h e r e s u l t s o f t h e s e a n a l y s e s a r e p r e s e n t e d i n A p p e n d i x 2 . T h e r e a p p e a r e d to b e o n l y s m a l l d i f f e r e n c e s i n t h e c o e f f i c i e n t s i n t h e r e g r e s s i o n e q u a t i o n s w h e n a l l o w i n g a r a n d o m e f f e c t b e t w e e n t h e p a i r s . In f a c t , a c c o u n t i n g f o r t h e p a i r e d s a m p l e s a c t u a l l y d e c r e a s e d t h e r e s i d u a l e r r o r w h e n a m u l t i - l e v e l m o d e l w a s a p p l i e d , i n d i c a t i n g tha t m o r e o f t h e v a r i a n c e i n f a i l u r e l o a d w a s e x p l a i n e d b y t h e m u l t i - l e v e l m o d e l . 6.5. Limitations in Applying Results to Future In Vivo Studies 6.5.1. Limitations when Applying the MRI Protocol In Vivo T h e M R I p r o t o c o l I u s e d w a s d e s i g n e d to p r o v i d e o p t i m a l i m a g e q u a l i t y i n c a d a v e r i c s p e c i m e n s . T h u s , t h e s c a n t i m e ( 1 6 m i n u t e s ) is l i k e l y t o o l e n g t h y f o r i n v i v o a p p l i c a t i o n . I f a n i d e n t i c a l p r o t o c o l w a s a p p l i e d to t h e f e m o r a l n e c k r e g i o n o n l y , t h e s c a n t i m e w o u l d b e r e d u c e d b y h a l f (8 m i n u t e s ) . F i n a l l y , t h e h e a d c o i l u s e d i n t h i s s t u d y is i n a p p r o p r i a t e f o r a n i n v i v o s c a n o f t h e p r o x i m a l f e m u r . A s u r f a c e c o i l w o u l d b e r e q u i r e d i n s t e a d a n d t h i s c o u l d p o t e n t i a l l y r e d u c e t h e i m a g e q u a l i t y d u e to a r e d u c t i o n i n t h e s i g n a l - t o - n o i s e r a t i o . 6.5.2. Mechanical Testing W h i l e m e c h a n i c a l t e s t i n g p r o v i d e s a n i n d i c a t i o n o f h o w b o n e s m a y f a i l , p r e d i c t o r s o f f r a c t u r e i n v i v o m a y d i f f e r f r o m t h e c o n t r o l l e d s i t u a t i o n i n the l a b o r a t o r y . In r e a l i t y , e a c h f a l l i s u n i q u e , 1 2 2 Discussion resulting in a unique combination of applied loads in each instance. The loading conditions may poorly represent the typical fall, thus failure loads should be interpreted with caution. Finally, the specimens were loaded to failure. In many cases, the failure load would exceed the applied load in a fall. 6.6. Future Directions The results of this study demonstrate that MRI can be used to assess femoral neck cortical bone geometry ex vivo, and that these parameters are strongly related to failure load. However, MRI measures were no better than predictions of failure load using D X A or HSA. Although predictive ability was greater with D X A , assessment of changes in the proximal femur with D X A aBMD is confounded by many limitations. Using D X A aBMD as a surrogate measure, it is impossible to know the reason for change or lack of change in fracture risk. An increase in trabecular number with a concomitant increase in cortical porosity may appear as an increase, decrease or no change in aBMD depending on the relative proportions of these changes. Alternatively, a decrease in trabecular thickness with a concomitant increase in cortical cross-sectional area may appear as a decrease in aBMD. This may lead investigators of a clinical trial to stop the trial prematurely, concluding an intervention had a negative impact on the bone. It is thus important to be able to assess each of these bone properties individually to determine a) whether an intervention may be affecting fracture risk before waiting for fracture outcomes b) the mechanism by which an intervention is affecting fracture risk, and c) the mechanism by which an intervention is affecting bone growth. D X A and HSA make a critical assumption that all the attenuation through a bone cross-section is due to the presence of bone. In fact, the marrow within the bone will also affect the attenuation. When evaluating longitudinal changes in bone with D X A and HSA, one must assume that there are no changes occurring in the marrow that would affect attenuation. Evidence presented in the literature review [80] indicated that changes in marrow can occur and can significantly impact the aBMD measurements. Thus developing a technique which can measure bone changes without these limitations will be important. In addition to fracture risk prediction, MRI may be useful for monitoring intervention effects. Many interventions are being developed, particularly pharmaceutical and physical activity prescription, as an attempt to reduce the risk of fracture. Ultimately, in order to evaluate the 123 Discussion success of these interventions, a sufficient number of participants need to be followed for a sufficient duration to assess fracture as an outcome. Currently, aBMD by DXA is the most prominent measure used as a surrogate, or short-term outcome for fracture risk [148, 149]. While other technologies are being used for more peripheral measurements, such as the QCT at the vertebrae [16] and the tibia [150], direct measurement of the proximal femur would provide the best surrogate of fracture risk at the proximal femur. As stated in the introduction, QCT is a technology capable of measuring vBMD and cortical geometry in cross-sectional images. However, in a longitudinal study to assess changes in these measures, an individual would require repeated scans over time, which would expose the individual to considerable doses of ionizing radiation. There is a need, therefore, to pursue the development of less invasive techniques such as MRI that assess proximal femur bone geometry reliably and accurately. In future, MRI could be developed to expand the analysis to the trochanteric region, to improve reliability, to improve analysis techniques, and to include the assessment of trabecular architecture. I discuss my recommendations related to each of these below. 6.6.1. Recommendations for Improving Techniques to Assess Bone with MRI 6.6.1.1. Recommendations for Assessing the Trochanteric Region with MRI This study has demonstrated that the trochanteric region plays an important role in the strength of the proximal femur. It therefore seems important to acquire and analyze MR images of the trochanteric region in a plane chosen specifically for that region. The optimal plane of acquisition is unknown. Several studies have used the plane of the bisector of the femoral neck axis and shaft axis [24, 132], however no comparisons have been made to other planes which may be simpler to define. It may be possible to determine an appropriate plane with images of the trochanteric region acquired with cubic voxels, so that the images can be reconstructed in any plane. Upon acquisition of images of the trochanteric region, the images must be segmented into cortical and trabecular bone compartments. The irregular shape of this region, with the greater and lesser trochanters, presents challenges to standardizing both segmentation techniques and 124 Discussion ROI definition. A research program that addresses these acquisition and analysis issues would be valuable for advancing research in this area. 6.6.1.2. Improving Reliability Reliability of cortical geometry measurements is important to assess differences between individuals and within individuals over time. I recommend that MRI scan protocols that maximize image quality be developed to improve reliability. Most importantly, MRI techniques that allow cortical bone (if present) to be visible around the perimeter of the bone would provide increased opportunities to assess the contribution of this compartment to bone strength. Improved image quality would also enhance segmentation techniques. Automated, threshold-driven, region-growing algorithms can be applied easily when a tissue is homogenous and continuous through the region which is being segmented. As it may not be possible to visualize the cortical border in a MR image when the cortex is thin, automated techniques which allow manual intervention to 'connect' non-continuous regions would likely improve reliability of analysis. Finally, defining a region of interest of a constant length about a quantitative landmark (e.g. the slice with the smallest total cross-sectional area) may improve reliability of the measurements with repositioning. This speculation is supported by current standards of D X A analyses. The length of the femoral neck region of interest defined by D X A is the same for all scans, and is positioned automatically by the software. The precision of D X A femoral neck measurements is quite high [86]. 6.6.1.3. Assessing Trabecular Architecture with MRI Other researchers have developed techniques to assess trabecular architecture in the proximal femur [112]. A major advantage of assessing the two bone compartments separately is that changes in the two compartments would be assessed independently over time, providing a better perspective on growth or change in the bone in response to an intervention. To assess trabecular architecture, a very high resolution is required (in the order of 0.15 mm x 0.15 mm x 0.25 mm) [151]. This resolution can only be obtained with specialized coils with a small field of view. Use of this specialized equipment may also improve assessment of cortical bone. 125 Summary and Conclusions 7. SUMMARY AND CONCLUSIONS 7.1. Summary (Primary Objectives) 1) MRI assessment of femoral neck geometry, particularly cortical area and area moment of inertia about the x-axis, is strongly associated with the failure load of cadaveric femora tested in a fall configuration. 2) The strength of the association between MRI measures of cortical geometry with failure load was similar to the strength of the association between D X A measures of aBMD and failure load. 3) The combination of femoral neck area moment of inertia measured with MRI and femoral neck aBMD explained a greater proportion of variance in failure load than either measure alone. The combination of femoral neck area moment of inertia measured with MRI with trochanteric aBMD did not explain a greater proportion of variance in failure load than trochanteric aBMD alone. 4) The combination of MRI assessment of femoral neck geometry and D X A assessment of aBMD did not explain a greater proportion of variance in failure load than HSA assessment of proximal femur geometry and aBMD. 7.2. Summary (Secondary Objective) 1) Cortical area, area moment of inertia about the y-axis and section modulus about the y-axis were significantly smaller when assessed with a 3 T MRI system than the same parameters assessed with a 1.5 T system. 2) Inter-acquisition and inter-analysis reliability were similar when assessed with 3 T MRI and 1.5 T MRI systems. 7.3. Conclusions Cortical bone geometry in the femoral neck assessed by MRI is strongly associated with failure load in human cadaveric proximal femora. The relationship between cortical bone geometry measured with MRI and failure load is similar to the relationship between femoral neck aBMD and failure load. Improved techniques that assess the trochanteric region with MRI would be a valuable contribution to this research area. With current positioning and analysis techniques, 3 T and 1.5 T MRI systems can be used to assess femoral neck cortical bone geometry with similar reliability. 126 REFERENCES 1. Papadimitropoulos EA, Coyte PC, Josse RG, Greenwood CE. Current and projected rates of hip fracture in Canada. Cmaj 1997; 157(10): 1357-63. 2. Wiktorowicz M E , Goeree R, Papaioannou A, Adachi JD, Papadimitropoulos E. Economic implications of hip fracture: health service use, institutional care and cost in Canada. Osteoporos Int 2001; 12(4): 271-8. 3. Khan K, McKay HA, Kannus P, Bailey DA, Wark JD, Bennell K L , Physical activity and bone health. 2001, Champaign, IL: Human Kinetics. 4. Melton LJ, 3rd, Johnell O, Lau E, Mautalen CA, Seeman E. Osteoporosis and the global competition for health care resources. J Bone Miner Res 2004; 19(7): 1055-8. 5. Ahlborg HG, Johnell O, Jarvinen TL, Karlsson M K . Prevalence of low bone mass in women - secular trends over 30 years, in ASBMR. 2004. Seattle, WA. 6. Gluer CC, Cummings SR, Pressman A, L i J, Gluer K, Faulkner K G , Grampp S, Genant HK. Prediction of hip fractures from pelvic radiographs: the study of osteoporotic fractures. The Study of Osteoporotic Fractures Research Group. J Bone Miner Res 1994; 9(5): 671-7. 7. Riggs BL , Melton Iii LJ, 3rd, Robb RA, Camp JJ, Atkinson EJ, Peterson JM, Rouleau PA, McCollough CH, Bouxsein M L , Khosla S. Population-based study of age and sex differences in bone volumetric density, size, geometry, and structure at different skeletal sites. J Bone Miner Res 2004; 19(12): 1945-54. 8. Riggs BL , Hodgson SF, O'Fallon W M , Chao E Y , Wahner HW, Muhs JM, Cedel SL, Melton LJ, 3rd. Effect of fluoride treatment on the fracture rate in postmenopausal women with osteoporosis. N Engl J Med 1990; 322(12): 802-9. 9. Lochmuller E M , Krefting N , Burklein D, Eckstein F. Effect of fixation, soft-tissues, and scan projection on bone mineral measurements with dual energy X-ray absorptiometry (DXA). Calcif Tissue Int 2001; 68(3): 140-5. 10. Genant HK, Engelke K, Fuerst T, Gluer CC, Grampp S, Harris ST, Jergas M , Lang T, Lu Y , Majumdar S, Mathur A, Takada M . Noninvasive assessment of bone mineral and structure: state of the art. J Bone Miner Res 1996; 11(6): 707-30. 11. Lochmuller E M , Groll O, Kuhn V , Eckstein F. Mechanical strength of the proximal femur as predicted from geometric and densitometric bone properties at the lower limb versus the distal radius. Bone 2002; 30(1): 207-16. 12. Lochmuller E M , Muller R, Kuhn V, L i l l CA, Eckstein F. Can novel clinical densitometric techniques replace or improve D X A in predicting bone strength in osteoporosis at the hip and other skeletal sites? J Bone Miner Res 2003; 18(5): 906-12. 13. Marshall D, Johnell O, Wedel H. Meta-analysis of how well measures of bone mineral density predict occurrence of osteoporotic fractures. B M J 1996; 312(7041): 1254-1259. 14. Seeman E. An exercise in geometry. J Bone Miner Res 2002; 17(3): 373-80. 15. Bolotin HH. Inaccuracies inherent in dual-energy X-ray absorptiometry in vivo bone mineral densitometry may flaw osteopenic/osteoporotic interpretations and mislead assessment of antiresorptive therapy effectiveness. Bone 2001; 28(5): 548-55. 16. Genant HK, Lang T, Fuerst T, Pinette K V , Zhou C, Thiebaud D, Diez-Perez A. Treatment with raloxifene for 2 years increases vertebral bone mineral density as measured by volumetric quantitative computed tomography. Bone 2004; 35(5): 1164-8. 17. Adami S, Gatti D, Braga V, Bianchini D, Rossini M . Site-specific effects of strength training on bone structure and geometry of ultradistal radius in postmenopausal women. J Bone Miner Res 1999; 14(1): 120-4. 127 18. Jarvinen TL, Kannus P, Sievanen H, Jolma P, Heinonen A, Jarvinen M. Randomized controlled study of effects of sudden impact loading on rat femur. J Bone Miner Res 1998; 13(9): 1475-82. 19. Ferretti JL, Peripheral quantitative computed tomography for evaluating structural and mechanical properties of small bone. 2000. 20. Seeman E. Periosteal bone formation—a neglected determinant of bone strength. N Engl J Med 2003; 349(4): 320-3. 21. Orwoll ES. Toward an expanded understanding of the role of the periosteum in skeletal health. J Bone Miner Res 2003; 18(6): 949-54. 22. Lang TF, Keyak JH, Heitz MW, Augat P, Lu Y, Mathur A, Genant HK. Volumetric quantitative computed tomography of the proximal femur: Precision and relation to bone strength Bone 1997; 21(1): 101-108. 23. Augat P, Reeb H, Claes LE. Prediction of fracture load at different skeletal sites by geometric properties of the cortical shell. J Bone Miner Res 1996; 11(9): 1356-63. 24. Lotz JC and Hayes WC. The use of quantitative computed tomography to estimate risk of fracture of the hip from falls. J Bone Joint Surg Am 1990; 72(5): 689-700. 25. McKay HA, Sievanen H, Petit MA, MacKelvie KJ, Forkheim K M , Whittall KP, Forster BB, Macdonald H. Application of magnetic resonance imaging to evaluation of femoral neck structure in growing girls. J Clin Densitom 2004; 7(2): 161-8. 26. Woodhead HJ, Kemp AF, Blimkie CJR, Briody JN, Duncan CS, Thompson M, Lam A, Howman-Giles R, Cowell CT. Measurement of midfemoral shaft geometry: repeatability and accuracy using magnetic resonance imaging and dual-energy X-ray absorptiometry. J Bone Miner Res 2001; 16(12): 2251-9. 27. Kroger H, Vainio P, Nieminen J, Kotaniemi A. Comparison of different models for interpreting bone mineral density measurements using DXA and MRI technology. Bone 1995; 17(2): 157-159. 28. Hayes WC, Biomechanics of cortical and trabecular bone: implications for assessment of fracture risk, in Basic Orthopaedic Biomechanics, V.C. Mow and W.C. Hayes, Editors. 1991, Raven Press, Ltd.: New York. 29. Hollister SJ, BME/ME 456 Biomechanics - Bone Structure. 30. Allen MR, Hock JM, Burr DB. Periosteum: biology, regulation, and response to osteoporosis therapies. Bone 2004; 35(5): 1003-12. 31. Einhorn T, The bone organ system: Form and function., in Osteoporosis, R. Marcus, D. Feldman, and J. Kelsey, Editors. 1996, Academic Press: New York. 32. Duncan RL and Turner CH. Mechanotransduction and the functional response of bone to mechanical strain. Calcif Tissue Int 1995; 57(5): 344-58. 33. Currey JD, Bones: Structure and Mechanics. 2002, Princeton: Princeton University Press. 34. Bouxsein ML, Myers ER, Hayes WC, Biomechanics of Age-Related Fractures, in Osteoporosis, R. Marcus, D. Feldman, and J. Kelsey, Editors. 1996, Academic Press, Inc.: San Diego, California. 35. Turner CH and Burr DB. Basic biomechanical measurements of bone: a tutorial. Bone 1993; 14(4): 595-608. 36. Carter DR and Hayes WC. Bone compressive strength: the influence of density and strain rate. Science 1976; 194(4270): 1174-6. 37. Martens M, Van Audekercke R, Delport P, De Meester P, Mulier JC. The mechanical characteristics of cancellous bone at the upper femoral region. J Biomech 1983; 16(12): 971-83. 128 38. Lotz JC, Cheal EJ, Hayes WC. Stress distributions within the proximal femur during gait and falls: implications for osteoporotic fracture. Osteoporos Int 1995; 5(4): 252-61. 39. Bell KL, Loveridge N, Power J, Garrahan N, Stanton M, Lunt M, Meggitt BF, Reeve J. Structure of the femoral neck in hip fracture: cortical bone loss in the inferoanterior to superoposterior axis. J Bone Miner Res 1999; 14(1): 111-9. 40. Ferretti JL, Frost HM, Gasser JA, High WB, Jee WS, Jerome C, Mosekilde L, Thompson DD. Perspectives on osteoporosis research: its focus and some insights from a new paradigm. Calcif Tissue Int 1995; 57(6): 399-404. 41. Cooper C, Campion G, Melton LJ, 3rd. Hip fractures in the elderly: a world-wide projection. Osteoporos Int 1992; 2(6): 285-9. 42. Cummings SR and Nevitt MC. A hypothesis: the causes of hip fractures. J Gerontol 1989; 44(4): M107-11. 43. Max W, Sinnot P, Kao C, Sung HY, Rice DP. The burden of osteoporosis in California, 1998. Osteoporos Int 2002; 13(6): 493-500. 44. Jaglal SB, Weller I, Mamdani M, Hawker G, Kreder H, Jaakkimainen L, Adachi JD. Population Trends in BMD Testing, Treatment, and Hip and Wrist Fracture Rates: Are the Hip Fracture Projections Wrong? J Bone Miner Res 2005; 20(6): 898-905. 45. Melton LJ, 3rd, Lane AW, Cooper C, Eastell R, O'Fallon WM, Riggs BL. Prevalence and incidence of vertebral deformities. Osteoporos Int 1993; 3(3): 113-9. 46. Cooper C and Melton LJ, 3rd, Magnitude and impact of osteoporosis and fractures, in Osteoporosis, R. Marcus, D. Feldman, and J. Kelsey, Editors. 1996, Academic Press, Inc.: San Diego, California. 47. Cummings SR and Melton LJ. Epidemiology and outcomes of osteoporotic fractures. Lancet 2002; 359(9319): 1761-7. 48. Adachi JD, Ioannidis G, Pickard L, Berger C, Prior JC, Joseph L, Hanley DA, Olszynski WP, Murray TM, Anastassiades T, Hopman W, Brown JP, Kirkland S, Joyce C, Papaioannou A, Poliquin S, Tenenhouse A, Papadimitropoulos EA. The association between osteoporotic fractures and health-related quality of life as measured by the Health Utilities Index in the Canadian Multicentre Osteoporosis Study (CaMos). Osteoporos Int 2003; 14(11): 895-904. 49. Heggeness MH and Mathis KB, An orthopedic perspective of osteoporosis, in Osteoporosis, R. Marcus, D. Feldman, and J. Kelsey, Editors. 1996, Academic Press, Inc.: San Diego, p. 1097-1111. 50. Saarenpaa I, Partanen J, Jalovaara P. Basicervical fracture—a rare type of hip fracture. Arch Orthop Trauma Surg 2002; 122(2): 69-72. 51. Blair B, Koval KJ, Kummer F, Zuckerman JD. Basicervical fractures of the proximal femur. A biomechanical study of 3 internal fixation techniques. Clin Orthop Relat Res 1994;(306): 256-63. 52. Merlino AF and Nixon JE. Isolated fractures of the greater trochanter. Report of twelve cases. Int Surg 1969; 52(2): 117-24. 53. Zuckerman JD. Hip fracture. N Engl J Med 1996; 334(23): 1519-25. 54. Michelson JD, Myers A, Jinnah R, Cox Q, Van Natta M. Epidemiology of hip fractures among the elderly. Risk factors for fracture type. Clin Orthop 1995;(311): 129-35. 55. Marks R, Allegrante JP, Ronald MacKenzie C, Lane JM. Hip fractures among the elderly: causes, consequences and control. Ageing Res Rev 2003; 2(1): 57-93. 56. Koval KJ, Aharonoff GB, Rokito AS, Lyon T, Zuckerman JD. Patients with femoral neck and intertrochanteric fractures. Are they the same? Clin Orthop 1996;(330): 166-72. 129 57. Karagas MR, Lu-Yao GL, Barrett JA, Beach M L , Baron JA. Heterogeneity of hip fracture: age, race, sex, and geographic patterns of femoral neck and trochanteric fractures among the US elderly. Am J Epidemiol 1996; 143(7): 677-82. 58. Nevitt M C and Cummings SR. Type of fall and risk of hip and wrist fractures: the study of osteoporotic fractures. The Study of Osteoporotic Fractures Research Group. J A m Geriatr Soc 1993; 41(11): 1226-34. 59. Greenspan SL, Myers ER, Maitland L A , Resnick N M , Hayes WC. Fall severity and bone mineral density as risk factors for hip fracture in ambulatory elderly. Jama 1994; 271(2): 128-33. 60. Greenspan SL, Myers ER, Kiel DP, Parker RA, Hayes WC, Resnick N M . Fall direction, bone mineral density, and function: risk factors for hip fracture in frail nursing home elderly. A m J Med 1998; 104(6): 539-45. 61. Dargent-Molina P, Favier F, Grandjean H, Baudoin C, Schott A M , Hausherr E, Meunier PJ, Breart G. Fall-related factors and risk of hip fracture: the EPIDOS prospective study. The Lancet 1996; 348(9021): 145-149. 62. Bergmann G, Graichen F, Rohlmann A. Hip joint loading during walking and running, measured in two patients. J Biomech 1993; 26(8): 969-90. 63. Bergmann G, Deuretzbacher G, Heller M , Graichen F, Rohlmann A , Strauss J, Duda GN. Hip contact forces and gait patterns from routine activities. Journal of Biomechanics 2001; 34(7): 859-871. 64. Beck TJ, Ruff CB, Warden K E , Scott WW, Jr., Rao GU. Predicting femoral neck strength from bone mineral data. A structural approach. Invest Radiol 1990; 25(1): 6-18. 65. Courtney A C , Wachtel EF, Myers ER, Hayes WC. Effects of loading rate on strength of the proximal femur. Calcif Tissue Int 1994; 55(1): 53-8. 66. Backman S. The proximal end of the femur: investigations with special reference to the etiology of femoral neck fractures; anatomical studies; roentgen projections; theoretical stress calculations; experimental production of fractures. Acta Radiol 1957;(Suppl 146): 1-166. 67. van den Kroonenberg AJ , Hayes WC, McMahon TA. Hip impact velocities and body configurations for voluntary falls from standing height J Biomech 1996; 29(6): 807-11. 68. Pinilla TP, Boardman K C , Bouxsein M L , Myers ER, Hayes WC. Impact direction from a fall influences the failure load of the proximal femur as much as age-related bone loss. Calcif Tissue Int 1996; 58(4): 231-5. 69. Eckstein F, Wunderer C, Boehm H, Kuhn V, Priemel M , Link T M , Lochmuller E M . Reproducibility and side differences of mechanical tests for determining the structural strength of the proximal femur. J Bone Miner Res 2004; 19(3): 379-85. 70. Bouxsein M L , Courtney A C , Hayes WC. Ultrasound and densitometry of the calcaneus correlate with the failure loads of cadaveric femurs. Calcif Tissue Int 1995; 56(2): 99-103. 71. Beck TJ, Mourtada FA, Ruff CB, Scott WW, Jr., Kao G. Experimental testing of a DEXA-derived curved beam model of the proximal femur. J Orthop Res 1998; 16(3): 394-8. 72. Keyak JH, Rossi SA, Jones K A , Les C M , Skinner HB. Prediction of fracture location in the proximal femur using finite element models. Med Eng Phys 2001; 23(9): 657-64. 73. Cadarette SM, Jaglal SB, Kreiger N , Mclsaac WJ, Darlington GA, Tu JV. Development and validation of the Osteoporosis Risk Assessment Instrument to facilitate selection of women for bone densitometry. Cmaj 2000; 162(9): 1289-94. 130 74. Committee MPS. Protocol for Bone Density Measurement. November 27, 2001 [cited 2005 May 16]; Available from: http://www.healthservices.gov.bc.ca/msp/protoguides/gps/bone/bonederisitv.html. 75. Njeh CF, Fuerst T, Hans D, Blake G M , Genant HK. Radiation exposure in bone mineral density assessment. Applied Radiation and Isotopes 1999; 50(1): 215-236. 76. Cummings SR, Black D M , Nevitt M C , Browner W, Cauley J, Ensrud K, Genant HK, Palermo L, Scott J, Vogt T M . Bone density at various sites for prediction of hip fractures. The Study of Osteoporotic Fractures Research Group. Lancet 1993; 341(8837): 72-5. 77. Stone K L , Seeley DG, Lui L Y , Cauley JA, Ensrud K, Browner WS, Nevitt M C , Cummings SR. B M D at multiple sites and risk of fracture of multiple types: long-term results from the Study of Osteoporotic Fractures. J Bone Miner Res 2003; 18(11): 1947-54. 78. De Laet CEDH, van Hout B A , Burger H, Hofman A, Pols HAP. Bone density and risk of hip fracture in men and women: cross sectional analysis. B M J 1997; 315(7102): 221-225. 79. Hui SL, Slemenda CW, Johnston CC, Jr. Age and bone mass as predictors of fracture in a prospective study. J Clin Invest 1988; 81(6): 1804-9. 80. Bolotin HH, Sievanen H, Grashuis JL. Patient-specific D X A bone mineral density inaccuracies: quantitative effects of nonuniform extraosseous fat distributions. J Bone Miner Res 2003; 18(6): 1020-7. 81. Svendsen OL, Hassager C, Skodt V , Christiansen C. Impact of soft tissue on in vivo accuracy of bone mineral measurements in the spine, hip, and forearm: a human cadaver study. J Bone Miner Res 1995; 10(6): 868-73. 82. Bolotin HH. The significant effects of bone structure on inherent patient-specific D X A in vivo bone mineral density measurement inaccuracies. Med Phys 2004; 31(4): 774-88. 83. Delmas PD and Seeman E. Changes in bone mineral density explain little of the reduction in vertebral or nonvertebral fracture risk with anti-resorptive therapy. Bone 2004; 34(4): 599-604. 84. Black D M , Pearson J, Harris F, LaCroix A, Cummings SR. Predicting the effect of antiresorptive treatments on risk of vertebral fractures: a meta-analysis. J Bone Miner Res 1999; 14(Suppll): S137. 85. Melton LJ , 3rd, Beck TJ, Amin S, Khosla S, Achenbach SJ, Oberg A L , Riggs BL. Contributions of bone density and structure to fracture risk assessment in men and women. Osteoporos Int 2005. 86. Khoo BC, Beck TJ, Qiao QH, Parakh P, Semanick L , Prince RL, Singer KP, Price RI. In vivo short-term precision of hip structure analysis variables in comparison with bone mineral density using paired dual-energy X-ray absorptiometry scans from multi-center clinical trials. Bone 2005. 87. Nicholson PH, Lowet G, Cheng X G , Boonen S, van der Perre G, Dequeker J. Assessment of the strength of the proximal femur in vitro: relationship with ultrasonic measurements of the calcaneus. Bone 1997; 20(3): 219-24. 88. Robinovitch SN, Keaveny T M , Nevitt M C , Cummings SR, Denianke KS, Somsnith D. Hip fracture risk in elderly females associates more strongly with bone density than with a biomechanical factor-of-risk. in American Geriatrics Society. 1997. 89. Pulkkinen P, Partanen J, Jalovaara P, Jamsa T. Combination of bone mineral density and upper femur geometry improves the prediction of hip fracture. Osteoporos Int 2004; 15(4): 274-80. 90. Guglielmi G and Lang TF. Quantitative computed tomography. Semin Musculoskelet Radiol 2002; 6(3): 219-27. 131 91. Link T M and Majumdar S. Osteoporosis imaging. Radiol Clin North A m 2003; 41(4): 813-39. 92. Lotz JC, Cheal EJ, Hayes WC. Fracture prediction for the proximal femur using finite element models: Part I—Linear analysis. J Biomech Eng 1991; 113(4): 353-60. 93. Lotz JC, Cheal EJ, Hayes WC. Fracture prediction for the proximal femur using finite element models: Part II~Nonlinear analysis. J Biomech Eng 1991; 113(4): 361-5. 94. Cheng X G , Lowet G, Boonen S, Nicholson PHF, Brys P, Nijs J, Dequeker J. Assessment of the Strength of Proximal Femur In Vitro: Relationship to Femoral Bone Mineral Density and Femoral Geometry. Bone 1997; 20(3): 213-218. 95. Lang T, Augat P, Majumdar S, Ouyang X , Genant HK. Noninvasive Assessment of Bone Density and Structure Using Computed Tomography and Magnetic Resonance. Bone 1998; 22(5, Supplement 1): 149S-153S. 96. Augat P, Gordon C L , Lang TF, Iida H, Genant HK. Accuracy of cortical and trabecular bone measurements with peripheral quantitative computed tomography (pQCT). Physics in Medicine and Biology 1998; 43(10): 2873. 97. Aldrich J. Sources of Radiation Exposure. 2000 March 2000 [cited 2004 August 30]; Available from: http://web.ucs.ubc.ca/aldrich/radpublc.htm. 98. Aldrich J, S. Manske, Editor. 2005: Vancouver. 99. Horikoshi T, Endo N , Uchiyama T, Tanizawa T, Takahashi HE. Peripheral quantitative computed tomography of the femoral neck in 60 Japanese women. Calcif Tissue Int 1999; 65(6): 447-53. 100. Augat P, Fan B, Lane NE, Lang TF, LeHir P, Lu Y , Uffmann M , Genant HK. Assessment of bone mineral at appendicular sites in females with fractures of the proximal femur. Bone 1998; 22(4): 395-402. 101. Beck TJ. On measuring bone to predict osteoporotic fracture: moving beyond statistical inference. Radiology 1996; 199(3): 612-4. 102. Hashemi RH, Bradley WJ, Lisanti CL, MRI: The Basics. Second ed. 2004, Philadelphia, PA: Lippincott Williams & Williams. 103. Pooley RA, Felmlee JP, Morin RL, Basic principles and terminology of magnetic resonance imaging, in MRI of the musculoskeletal system, T.H. Berquist, Editor. 2001, Lippincott Williams & Wilkins: Philadelphia. 104. Heinonen A, McKay HA, Whittall KP, Forster BB, Khan K M . Muscle cross-sectional area is associated with specific site of bone in prepubertal girls: a quantitative magnetic resonance imaging study. Bone 2001; 29(4): 388-92. 105. FDA. Magnetic Resonance Diagnostic Device: Panel Recommendation and Report on Petitions for M R Reclassification. Federal Register 1988; 53: 7575-9. 106. Hogler W, Blimkie CJ, Cowell CT, Kemp AF, Briody J, Wiebe P, Farpour-Lambert N , Duncan CS, Woodhead HJ. A comparison of bone geometry and cortical density at the mid-femur between prepuberty and young adulthood using magnetic resonance imaging. Bone 2003; 33(5): 771-8. 107. Duncan CS, Blimkie CJ, Kemp A, Higgs W, Cowell CT, Woodhead H, Briody JN, Howman-Giles R. Mid-femur geometry and biomechanical properties in 15- to 18-yr-old female athletes. Med Sci Sports Exerc 2002; 34(4): 673-81. 108. Bass SL, Saxon L, Daly R M , Turner CH, Robling A G , Seeman E, Stuckey S. The effect of mechanical loading on the size and shape of bone in pre-, peri-, and postpubertal girls: a study in tennis players. J Bone Miner Res 2002; 17(12): 2274-80. 109. Arokoski M H , Arokoski JP, Vainio P, Niemitukia L H , Kroger H, Jurvelin JS. Comparison of D X A and MRI methods for interpreting femoral neck bone mineral density. J Clin Densitom 2002; 5(3): 289-96. 132 110. Kontulainen S, Liu D, Jamieson M, Manske S, Macdonald H, Whittall KP, McKay HA. Optimizing the accuracy of pQCT measurements of bone geometry. In press 2005. 111. Majumdar S, Newitt D, Mathur A, Osman D, Gies A, Chiu E, Lotz J, Kinney J, Genant H. Magnetic resonance imaging of trabecular bone structure in the distal radius: relationship with X-ray tomographic microscopy and biomechanics. Osteoporos Int 1996; 6(5): 376-85. 112. Link TM, Vieth V, Langenberg R, Meier N, Lotter A, Newitt D, Majumdar S. Structure analysis of high resolution magnetic resonance imaging of the proximal femur: in vitro correlation with biomechanical strength and BMD. Calcif Tissue Int 2003; 72(2): 156-65. 113. Majumdar S. Magnetic resonance imaging of trabecular bone structure. Top Magn Reson Imaging 2002; 13(5): 323-34. 114. Beck TJ, Looker AC, Ruff CB, Sievanen H, Wahner HW. Structural trends in the aging femoral neck and proximal shaft: analysis of the Third National Health and Nutrition Examination Survey dual-energy X-ray absorptiometry data J Bone Miner Res 2000; 15(12): 2297-304. 115. Beck TJ, Ruff CB, Bissessur K. Age-related changes in female femoral neck geometry: implications for bone strength. Calcif Tissue Int 1993; 53 Suppl 1: S41-6. 116. Allen MR and Burr DB. Human femoral neck has less cellular periosteum, and more mineralized periosteum, than femoral diaphyseal bone. Bone 2005; 36(2): 311-316. 117. Burstein AH, Reilly DT, Martens M. Aging of bone tissue: mechanical properties. J Bone Joint Surg Am 1976; 58(1): 82-6. 118. Zioupos P and Currey JD. Changes in the stiffness, strength, and toughness of human cortical bone with age. Bone 1998; 22(1): 57-66. 119. Thomsen JS, Ebbesen EN, Mosekilde Li. Age-related differences between thinning of horizontal and vertical trabeculae in human lumbar bone as assessed by a new computerized method. Bone 2002; 31(1): 136-142. 120. Kiebzak GM. Radiolucent Casting Tape Allows for Accurate Measurement of Forearm Bone Mineral Density Using Dual-Energy X-ray Absorptiometry. J Clin Densitom 1998; 1(4): 369-74. 121. Hologic, Hologic QDR 4500 User's Guide, Waltham, MA: Hologic, Inc. 122. Beck TJ, Oreskovic TL, Stone KL, Ruff CB, Ensrud K, Nevitt MC, Genant HK, Cummings SR. Structural adaptation to changing skeletal load in the progression toward hip fragility: the study of osteoporotic fractures. J Bone Miner Res 2001; 16(6): 1108-19. 123. Eckstein F, Kuhn V, Lochmuller EM. Strength prediction of the distal radius by bone densitometry—evaluation using biomechanical tests. Ann Biomed Eng 2004; 32(3): 487-503. 124. Robinovitch SN, Hayes WC, McMahon TA. Prediction of femoral impact forces in falls on the hip. J Biomech Eng 1991; 113(4): 366-74. 125. Robinovitch SN, McMahon TA, Hayes WC. Force attenuation in trochanteric soft tissues during impact from a fall. J Orthop Res 1995; 13(6): 956-62. 126. Robinovitch SN, Hayes WC, McMahon TA. Predicting the impact response of a nonlinear single-degree-of-freedom shock-absorbing system from the measured step response. J Biomech Eng 1997; 119(3): 221-7. 127. Evans EM. The treatment of trochanteric fractures of the femur. J Bone Joint Surg Br 1949;31B(2): 190-203. 128. Steiger JH. Tests for comparing elements of a correlation matrix. Psychological Bulletin 1980; 87(2): 245-51. 129. Bland JM and Altman DG. Comparing two methods of clinical measurement: a personal history. Int J Epidemiol 1995; 24 Suppl 1: S7-14. 133 130. Sievanen H, Koskue V , Rauhio A, Kannus P, Heinonen A, Vuori I. Peripheral quantitative computed tomography in human long bones: evaluation of in vitro and in vivo precision. J Bone Miner Res 1998; 13(5): 871-82. 131. Beer FP and Johnston RJ, Jr., Mechanics of Materials. Vol . 2. 1992, New York: McGraw-Hill, Inc. 132. Melton LJ, 3rd, Beck TJ, Amin S, Khosla S, Achenbach SJ, Oberg A L , Riggs B L . Contributions of bone density and structure to fracture risk assessment in men and women. Osteoporos Int 2005; 16(5): 460-7. 133. Kroger H, Kotaniemi A , Vainio P, Alhava E. Bone densitometry of the spine and femur in children by dual-energy x-ray absorptiometry. Bone Miner 1992; 17(1): 75-85. 134. Beck TJ, Ruff CB, Scott WW, Jr., Plato CC, Tobin JD, Quan CA. Sex differences in geometry of the femoral neck with aging: a structural analysis of bone mineral data Calcif Tissue Int 1992; 50(1): 24-9. 135. Gnudi S, Ripamonti C, Lisi L, Fini M , Giardino R, Giavaresi G. Proximal femur geometry to detect and distinguish femoral neck fractures from trochanteric fractures in postmenopausal women. Osteoporos Int 2002; 13(1): 69-73. 136. Duboeuf F, Hans D, Schott A M , Kotzki PO, Favier F, Marcelli C, Meunier PJ, Delmas PD. Different morphometric and densitometric parameters predict cervical and trochanteric hip fracture: the EPIDOS Study. J Bone Miner Res 1997; 12(11): 1895-902. 137. Greenspan SL, Myers ER, Maitland L A , Kido TH, Krasnow M B , Hayes WC. Trochanteric bone mineral density is associated with type of hip fracture in the elderly. J Bone Miner Res 1994; 9(12): 1889-94. 138. Weber TG, Yang K H , Woo R, Fitzgerald RH. Proximal femur strength: correlation of the rate of loading and bone mineral density, in ASME Advances in Bioengineering. 1992: American Society of Mechanical Engineers, New York. 139. Hannan MT, Felson DT, Dawson-Hughes B, Tucker K L , Cupples L A , Wilson PW, Kiel DP. Risk factors for longitudinal bone loss in elderly men and women: the Framingham Osteoporosis Study. J Bone Miner Res 2000; 15(4): 710-20. 140. Looker A C , Orwoll ES, Johnston CC, Jr., Lindsay RL, Wahner HW, Dunn WL, Calvo MS, Harris TB, Heyse SP. Prevalence of low femoral bone density in older U.S. adults fromNHANES III. J Bone Miner Res 1997; 12(11): 1761-8. 141. Erman J and Ott SM. Accuracy of dual photon absorptiometry in excised femurs. J Nucl Med 1988; 29(11): 1853-5. 142. Ruff CB and Hayes WC. Cross-sectional geometry of Pecos Pueblo femora and tibiae—a biomechanical investigation: I. Method and general patterns of variation. Am J Phys Anthropol 1983; 60(3): 359-81. 143. Sedlin ED. A rheologic model for cortical bone. A study of the physical properties of human femoral samples. Acta Orthop Scand 1965: Suppl 83:1-77. 144. Pelker RR, Friedlaender GE, Markham TC, Panjabi M M , Moen CJ. Effects of freezing and freeze-drying on the biomechanical properties of rat bone. J Orthop Res 1984; 1(4): 405-11. 145. Borchers RE, Gibson LJ , Burchardt H, Hayes WC. Effects of selected thermal variables on the mechanical properties of trabecular bone. Biomaterials 1995; 16(7): 545-51. 146. Boutros CP, Trout DR, Kasra M , Grynpas MD. The effect of repeated freeze-thaw cycles on the biomechanical properties of canine cortical bone. Vet Comp Orthop Traumatol 2000; 13: 59-64. 147. Tabachnick B G and Fidell LS, Using multivariate statistics. Fourth ed. 2001, Needham Hieghts, M A . 134 148. Neele SJ, Evertz R, De Valk-De Roo G, Roos JC, Netelenbos JC. Effect of 1 year of discontinuation of raloxifene or estrogen therapy on bone mineral density after 5 years of treatment in healthy postmenopausal women. Bone 2002; 30(4): 599-603. 149. Weinstein RS, Parfitt A M , Marcus R, Greenwald M , Crans G, Muchmore DB. Effects of raloxifene, hormone replacement therapy, and placebo on bone turnover in postmenopausal women. Osteoporos Int 2003; 14(10): 814-22. 150. Liu-Ambrose T, Eng JJ, Khan K M , Mallinson A, Carter ND, McKay HA. The influence of back pain on balance and functional mobility in 65- to 75-year-old women with osteoporosis. Osteoporos Int 2002; 13(11): 868-73. 151. Link T M , Majumdar S, Grampp S, Guglielmi G, van Kuijk C, Imhof H, Glueer C, Adams JE. Imaging of trabecular bone structure in osteoporosis. Eur Radiol 1999; 9(9): 1781-8. 152. Robinovitch SN, Hayes WC, McMahon TA. Energy-shunting hip padding system attenuates femoral impact force in a simulated fall. J Biomech Eng 1995; 117(4): 409-13. 135 APPENDIX 1: CALCULATION OF DISPLACEMENT R A T E Loading Rate dF - Estimated Failure load of bone in a fall = 5000 N [124] dt - time to develop peak force at the hip in a fall from standing height = 0.05 s [124] dF _ failure load _ 500(W dt time to peak load 0.05.S = 100 OWN Is However, load-controlled testing is not possible in this configuration. Therefore the displacement rate will be calculated based on several assumptions, to be used in displacement-control testing. Displacement Rate At impact of the hip with the floor, the effective stiffness of the body has been calculated to be 23 000 N/m. The deflection of the body that occurs on impact will be contained mostly in the soft tissue surrounding the greater trochanter, and some will occur in the femur and some in the pelvis. Therefore the total effective stiffness (K, = 23 kN/m) is composed of: 1. Soft tissue stiffness over trochanter (Ks= 30 - 35 kN/m) [152] 2. Proximal femur stiffness (Kf= unknown) 3. Pelvis stiffness (Kp= unknown) To determine the proximal femur stiffness, pelvis and proximal femur stiffness (Kjp) are first combined with the springs in series with the soft tissue: K = KsxK K.+K fp /P Kfp =98,570 N/m There is no known data on the distribution of the femuropelvis stiffness between the femur and pelvis. If the femur is assumed to be twice as stiff as the pelvis (2Kp=Kj), and again the springs are in series: K _ 2Kp x Kp Pf 2Kp+Kp Kp =147,900 N/m Kf =295,800 N/m 136 Then the loading rate is calculated dx _dFldt _ 100,000 N/s dt~ dFldx~ 295,800 N/m = 0.338 m/s APPENDIX 2: CORRELATIONS BETWEEN INDEPENDENT VARIABLES Table 18. Correlation coefficients (R) between independent variables (N = 27). T R = trochanteric region, N N = narrow neck region, IT = intertrochanteric region. *p < 0.05, * *p < 0.01 MRI DXA ToCSA CoCSA Ix Sx TR aBMD TR BMC Neck aBMD Neck BMC MRI ToCSA 1.00 0.45* 0.75** 0.49** 0.41* 0.68** 0.24 0.51** CoCSA 0.45* 1.00 0.82** 0.79** 0.73** 0.72** 0.64** 0.70** Ix 0.75** 0.82** 1.00 0.91** 0.67** 0.78** 0.57** 0.73** Sx 0.49** 0.79** 0.91** 1.00 0.63** 0.66** 0.58** 0.65** DXA TR aBMD 0.41* 0.73** 0.67** 0.63** 1.00 0.90** 0.91** 0.91** TR BMC 0.68** 0.72** 0.78** 0.66** 0.90** 1.00 0.79** 0.92** Neck aBMD 0.24 0.64** 0.57** 0.58** 0.91** 0.79** 1.00 0.93** Neck BMC 0.51** 0.70** 0.73** 0.65** 0.91** 0.92** 0.93** 1.00 HSA NN aBMD 0.02 0.50** 0.43* 0.52** 0.76** 0.60** 0.92** 0.81** NNCSA 0.35 0.59** 0.65** 0.65** 0.84** 0.79** 0.92** 0.92** NN CoTh -0.01 0.48* 0.41* 0.50** 0.75** 0.59** 0.91** 0.80** NN CSMI 0.78** 0.53** 0.77** 0.62** 0.71** 0.85** 0.63** 0.82** NN_S 0.62** 0.51** 0.72** 0.62** 0.77** 0.84** 0.76** 0.88** IT aBMD 0.23 0.65** 0.57** 0.61** 0.92** 0.75** 0.90** 0.86** IT CSA 0.49** 0.74** 0.73** 0.68** 0.93** 0.86** 0.87** 0.92** IT CoTh 0.04 0.60** 0.45* 0.54** 0.89** 0.69** 0.87** 0.82** IT CSMI 0.78** 0.73** 0.82** 0.66** 0.78** 0.87** 0.63** 0.80** ITS 0.69** 0.75** 0.79** 0.68** 0.86** 0.88** 0.71** 0.83** 138 Table 18 continued. Correlation coefficients (R) between independent variables (N = 27). *p < 0.05, * *p < 0.01 HSA NN aBMD NN CSA NN CoTh NN I NN S IT aBMD IT CSA IT CoTh IT I ITS MRI ToCSA 0.02 0.35 -0.01 0.78** 0.62** 0.23 0.49** 0.04 0.78** 0.69** CoCSA 0.50** 0.59** 0.48* 0.53** 0.51** 0.65** 0.74** 0.60** 0.73** 0.75** Ix 0.43* 0.65** 0.41* 0.77** 0.72** 0.57** 0.73** 0.45* 0.82** 0.79** Sx 0.52** 0.65** 0.50** 0.62** 0.62** 0.61** 0.68** 0.54** 0.66** 0.68** DXA TR aBMD 0.76** 0.84** 0.75** 0.71** 0.77** 0.92** 0.93** 0.89** 0.78** 0.86** TR BMC 0.60** 0.79** 0.59** 0.85** 0.84** 0.75** 0.86** 0.69** 0.87** 0.88** Neck aBMD 0.92** 0.92** 0.91** 0.63** 0.76** 0.90** 0.87** 0.87** 0.63** 0.71** Neck BMC 0.81** 0.92** 0.80** 0.82** 0.88** 0.86** 0.92** 0.82** 0.80** 0.83** HSA NN aBMD 1.00 0.93** 0.99** 0.53** 0.71** 0.88** 0.80** 0.90** 0.49** 0.58** NN CSA 0.93** 1.00 0.92** 0.79** 0.90** 0.90** 0.92** 0.87** 0.74** 0.79** NN CoTh 0.99** 0.92** 1.00 0.51** 0.69** 0.87** 0.78** 0.89** 0.47* 0.56** NN I 0.53** 0.79** 0.51** 1.00 0.96** 0.65** 0.83** 0.61** 0.91** 0.88** NN S 0.71** 0.90** 0.69** 0.96** 1.00 0.77** 0.88** 0.74** 0.84** 0.84** IT aBMD 0.88** 0.90** 0.87** 0.65** 0.77** 1.00 0.95** 0.99** 0.69** 0.80** IT CSA 0.80** 0.92** 0.78** 0.83** 0.88** 0.95** 1.00 0.93** 0.89** 0.94** IT CoTh 0.90** 0.87** 0.89** 0.61** 0.74** 0.99** 0.93** 1.00 0.66** 0.78** IT I 0.49** 0.74** 0.47* 0.91** 0.84** 0.69** 0.89** 0.66** 1.00 0.98** ITS 0.58** 0.79** 0.56** 0.88** 0.84** 0.80** 0.94** 0.78** 0.98** 1.00 139 APPENDIX 3: LINEAR REGRESSION ANALYSES WITH M U L T I - L E V E L MODELLING After excluding specimens based on fracture type and age, there were 27 specimens remaining. Of those specimens, 12 were paired (i.e. 6 left and 6 right limbs from 6 donors). I utilized a multilevel (random effects) modelling approach to determine whether treating the paired specimens as independent samples biased the correlations between the independent variables and failure load. Methods Linear regression models were constructed using a multilevel, random effects, modelling approach (MLwiN version 2.0 Centre for Multilevel Modelling, Institute of Education, University of London). The data were modelled hierarchically, with the individual limbs at level 1 and the donor at level 2 to account for the possible effects of the donor, rather than treating each limb as an independent sample. The pairs were modelled with a random intercept and constant slope. Level 2 Level 1 Donor 1 Donor 2 Donor 3 \ Left Left Right Right Figure 43. Schematic showing the hierarchical structure of the data. For cortical cross-sectional area (CoCSA), I provide the regression line when all specimens are treated as independent samples (Figure 44). I highlighted the paired samples. In addition, I plotted the paired samples only, calculating the regression line for each of the pairs (Figure 45). For several independent variables of interest, I present the regression coefficients and standard error of the estimates when modelling the data as independent samples and when modelling the data with random effects, accounting for the pairs (Table 19 to Table 21). 140 Results 10000 9000 H 8000 7000 H 6000 5000 4000 -I 3000 2000 H 1000 50 • » 100 150 CoCSA (mm2) 200 • All specimen • 1067 1082 x 1091 • 1093 • 1095 + 1159 y=35.702x- 378.03 R 2 = 0.4665 250 Figure 44. Failure load vs. CoCSA, measured with MRI. Each specimen is treated as an independent sample. The donor pairs are highlighted by the colour indicated in the legend. The regression equation treats each specimen as an independent sample. 141 10000 -, 9000 8000 _ 7000 -| 1 6000 3 | 5000 to 4000 3000 -| 2000 1000 50 100 150 CoCSA (mm2) • 1067 • 1082 1091 X 1093 X 1095 • 1159 Linear (1067) Linear (1095) Linear (1091) Linear (1093) Linear (1082) Linear (1159) 200 250 y = -1.4391x+ 1979.7 y =23.408x+617.06 y =-23.769x+6665.3 y = -422.06x+54295 y =-53.25 lx+ 12804 y - -83.876x+ 17557 Figure 45. Failure load vs. CoCSA for paired samples only. Regression lines are plotted for each donor. Table 19. Regression coefficients and standard errors for the relationship between trochanteric aBMD and failure load. Two analysis models are presented, 1) treating the specimens as independent samples, and 2) treating the data as pairs, where appropriate, in a random effects model (N = 27). Trochanteric aBMD Analysis Procedure Independent samples Intercept B SE of B SEE -2292.05 10945.23 1372.91 1010.48 Random effects -2860.89 + U d o n o r 11983.34 1460.77 289.71 Table 20. Regression coefficients and standard errors for the relationship between femoral neck aBMD and failure load. Two analysis models are presented, 1) treating the specimens as independent samples, and 2) treating the data as pairs, where appropriate, in a random effects model (N = 27). Femoral neck aBMD Analysis Procedure Intercept B SE ofB SEE Independent samples -251.61 6997.52 1681.96 1444.59 Random effects -857.24 + Udonor 8104.23 1823.15 414.33 142 Table 21. Regression coefficients and standard errors for the relationship between C o C S A measured by MRI and failure load. Two analysis models are presented, 1) treating the specimens as independent samples, and 2) treating the data as pairs, where appropriate, in a random effects model (N = 27). CoCSA Analysis Procedure Intercept B SE ofB SEE Independent samples -378.025 35.70 7.35 1351.63 Random effects -121.40 + Pdonor 33.95 7.83 1026.94 Discussion For each of the independent variables examined, by accounting for the paired specimens as paired rather than independent samples, the residual error is reduced. Thus, I am able to explain more of the variance in failure load when using a multi-level, random effects model. The B coefficients are not greatly affected by adding random effects, suggesting that the effect of paired samples is not substantial. 143 

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.831.1-0092401/manifest

Comment

Related Items