SEARCH

SEARCH BY CITATION

Keywords:

  • Aging;
  • Proximal Femur;
  • Computed Tomography;
  • Cortical Bone;
  • HIP Fractures

Abstract

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Disclosures
  8. Acknowledgements
  9. References

The anatomic distribution of cortical and cancellous bone in the femoral neck may be critical in determining resistance to fracture. We investigated the effects of aging on femoral neck bone in women. In this cross-sectional study, we used clinical multidetector computed tomography (MDCT) of the hips to investigate aging effects in 100 female volunteers aged 20 to 90 years. We developed a clinically efficient protocol to measure cortical thickness (C.Th) and cortical, trabecular, and integral bone mineral density (CtBMD, TrBMD, and iBMD in mg/cm3) in anatomic quadrants of the femoral neck. We used a nested ANOVA to evaluate their associations with height, weight, location in the femoral neck, and age of the subject. Age was the principal determinant of both cortical thickness and BMD. Age had significantly different effects within the anatomic quadrants; compared with young women, elderly subjects had relative preservation of the inferoanterior (IA) quadrant but strikingly reduced C.Th and BMD superiorly. A model including height, weight, and region of interest (and their interactions) explained 83% of the measurement variance (p < .0001). There were marked C.Th and BMD differences between age 25 and age 85 in the already thin superior quadrants. At 25 years the predicted C.Th of the superoposterior quadrant was 1.63 mm, whereas at 85 years it was 0.33 mm [−1.33 mm, 95% confidence interval (CI) of difference over 60 years −1.69 to −0.95]. By contrast, at 25 years mean C.Th of the IA quadrant was 3.9 mm, whereas at 85 years it was 3.3 mm (−0.6 mm, 95% CI −0.83 to −0.10). CtBMD of the IA region was equivalent at 25 and 85 years. In conclusion, elderly women had relative preservation of IA femoral neck bone over seven decades compared with young women but markedly lower C.Th and BMD in the other three quadrants. The IA quadrant transmits mechanical load from walking. Mechanical theory and laboratory tests on cadaveric femurs suggest that localized bone loss may increase the risk of fracture in elderly fallers. It remains to be determined whether this MDCT technique can provide better prediction of hip fracture than conventional clinical dual X-ray absorptiometry (DXA). © 2010 American Society for Bone and Mineral Research


Introduction

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Disclosures
  8. Acknowledgements
  9. References

Hip fractures are projected to rise worldwide from 1.7 million in 1990 to 6.3 million in 2050.1 Most occur at the femoral neck or trochanteric region of the hip.2, 3 Because these fractures increase exponentially with age, there is long-standing interest in bone tissue loss at the proximal femur with aging. While intertrochanteric hip fractures are clearly associated with systemically low bone mass,4, 5 femoral neck fractures may have a somewhat different etiology. Fall orientation may influence the pattern of femoral neck injury,6 as may localized bone loss in the proximal femur. We investigated age-related cortical and trabecular bone loss in the midfemoral neck noninvasively using multidetector computed tomography (MDCT).

Human femurs have thinner superior than inferior cortices at the midfemoral neck region (Fig. 1). In a previous study, we found that elderly cadaveric femurs had marked thinning in the superior regions but had thicker inferior cortices compared with young cadavers.7–9 Furthermore, elderly femoral neck fracture specimens differed from control specimens by additional thinning, especially of the inferoanterior (IA) cortex.9–11 Maximal compressive strain from a sideways fall onto the greater trochanter occurs in the superior femoral neck6, 9, 12–15 with maximal tensile strain in the inferior cortex.15 In load-to-failure testing, resected cadaveric femurs often fractured (in a sideways fall simulation) at the thin superior cortex,16 as predicted.6, 9, 17 This prompted us to study a region of interest (ROI) in the midfemoral neck.

thumbnail image

Figure 1. 3D CT rendering (left) showing ROIs and slice position. Shape of femoral neck at max/min ratio 1.4 (middle) showing anatomic quadrants (note the clockwise shift of one sector owing to sagittal positioning). A 10 µm thick histologic section of the femoral neck (right) from a 50-year-old woman at postmortem (bar = 250 µm; Goldner's stain; methacrylate embedded).

Download figure to PowerPoint

Our principal aims were to investigate age-related differences in cortical thickness of the midfemoral neck in women aged 20 to 90 years and to provide a technique that in future could advance the clinical study of the mechanics of hip fracture beyond what is possible with conventional dual X-ray absorptiometry (DXA). We used our experience from evaluating biopsy specimens9, 11 to develop a computed tomographic (CT) protocol for extracting and measuring a region of interest at the midfemoral neck that was accurate and reproducible using MDCT. We then studied age effects on regional cortical thickness and cortical and trabecular bone mineral density (BMD) separately as well as integral (i.e., cortical plus trabecular) BMD in relation to anatomic location within the femoral neck.

Methods

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Disclosures
  8. Acknowledgements
  9. References

Study participants

Women volunteers aged 20 to 90 were recruited to investigate hip volumetric CT parameters using whole-body multidetector CT (Siemens SOMATOM Sensation 64, Siemens AG, Erlangen, Germany) and QCT PRO software (Mindways, Austin, TX, USA). Participants were recruited from May 2006 to October 2007 and stratified aiming for 15 patients per decade from 20 to 90 years. Study participants were volunteers attending Addenbrooke's Hospital for a routine clinical CT scan (including the abdomen and pelvis) for purposes unrelated to skeletal disease (renal/abdominal/pelvic pain or altered bowel habit) but who were otherwise healthy. They consented to a hip quantitative computed tomographic (QCT) scan involving both proximal femurs that was an extension of the pelvic scan to 2 cm below the lesser trochanters (for details, see Table 1). Exclusion criteria were a clinical diagnosis of cancer at any time during the study period (because of the possibility of bone metastases), use of oral corticosteroids, current anemia, and recent weight loss. The study was approved by the Cambridgeshire Regional Ethics Committee, and subjects were recruited according to the Declaration of Helsinki.

Image processing

Bone/soft tissue segmentation and density/shape evaluation

The 3D hip image was reconstructed, and volumetric BMD (vBMD over all voxels in the bone volume), areal BMD (aBMD, DXA equivalent), and bone architecture were assessed using QCT PRO CTXA software (Mindways, Austin, TX, USA). CT images were calibrated from Hounsfield units to equivalent concentrations (g/cm3) of mineral content against a phantom (present in all scans) and calibrated to aqueous K2HPO4-equivalent mineral standard.18 Automated segmentation of bone from soft tissue voxels generated axial, coronal, and sagittal reconstructions of the hip followed by 3D rendering and femoral neck axis placement. A region-growing algorithm classified each pixel as either “bone” or “not bone” using an adaptive classifier. Operation of the classifier was controlled by three parameters. The first parameter defined a threshold below which a pixel could not be classified as bone (default −250 mg/cm3). The second parameter defined the size of the neighborhood of pixels considered when adaptively modifying the local threshold (range between 0, no neighbors considered, and 1, all pixels in the axial image considered, default 0.085). The third parameter was the initial threshold used to seed the adaptive algorithm, with pixels exceeding the initial threshold tentatively classified as “bone” and the remainder as “not bone” (default 100 mg/cm3). A graphical user interface allowed the operator to optimize the reconstructed images in three planes of section: to trim excessive soft tissue pixels, to fill artefactual bone holes by varying these parameters, and to optimize the femoral neck axis on the 3D surface-rendered image.

Cortical thickness in quadrants

Cross-sectional placement: A scripting command was written in QCT PRO Bone Investigational Toolkit software (v2, Mindways, Austin, TX, USA) to direct automatic midfemoral neck cross-sectional placement perpendicular to the femoral neck axis, where the approximate ratio of maximum to minimum diameters (max/min ratio) equaled 1.4 (9–11, 19; slice 6 in Fig. 1). Five further slices were extracted at 1 mm intervals medially. The max/min ratio was estimated as the square root of the ratio of the eigenvalues of the matrix of second-order mass moments associated with the principal axes of the cross-sectional image, equivalent to the ratio of the largest diameter and its perpendicular diameter for the elliptical cross section. The extracted image had a 0.589 mm pixel size. The location of the most medial femoral neck cross section (slice 1 in Fig. 1) was measured as a percentage of the hip axis length (%HAL) using a Mitutoyo Digimatic Calliper on color printed images. Hip axis length was measured from 2D projection screen images using the QCT PRO Slicepick measurement tool (Mindways, Austin, TX, USA).

Quadrant boundaries: The center of area was the internal reference point, with 16 equal sectors defined by equal angles (22.5 degrees; see Fig. 1) and the first sector boundary defined by a vertical line on the image. In QCT PRO analysis, the anterior surface of the femur on the sagittal plane image is positioned vertically (to ensure reproducibility in a limited scan field). However, the anterior surface of the whole femur is curved anteriorly in stance (giving a mean angle of incidence of 22.2 degrees).20 Thus quadrant division boundaries on cross sections of the midfemoral neck required rotation (by 22.5 degrees, or one sector) to ensure anatomic relevance when considering the femurs in “stance” (see Fig. 1, center image). This resulted in four anatomic quadrants (see Fig. 1): superoanterior (SA, from sectors 2, 3, 4, and 5), inferoanterior (IA, from sectors 6, 7, 8, and 9), inferoposterior (IP, from sectors 10, 11, 12, and 13), and superoposterior (SP, from sectors 14,15,16, and 1).

Determining a fixed cortical/trabecular threshold for MDCT: Cortical thickness (C.Th) measurements were made in 22 cadaveric femurs to evaluate different Bone Investigational Toolkit version 2 (BIT-2) threshold subclassifiers for cortical and trabecular bone within cross sections. Samples were scanned in a water bath to simulate soft tissue (Siemens MDCT, 0.589 µm/pixel) with higher-resolution pQCT (0.275 µm/pixel) measurements as described by Mayhew and colleagues9 as the standard.

Cortical thickness measurement: C.Th estimates were calculated for each of the 16 sectors as follows: The cortical area was measured automatically (by pixel counting) on each cross-sectional image. Cortical bone mass for a sector was taken to be homogeneously distributed between two surface boundaries (one periosteal and the other endosteal). The boundaries were approximated as concentric arcs of constant curvature that enclosed the point of the cortical center of mass (relative to the intersection point of the sector lines, as estimated by density-weighted pixel counting). The software automatically selected two radii of curvature (one periosteal, one endosteal) that matched the pixel-counted cortical area. The difference between these radii was reported as the C.Th for the sector. Integral, cortical, and trabecular vBMD in quadrants, cross-sectional area, and the distances from the center of mass to the center of area also were measured.

Precision studies: Thin-walled SIMAX borosilicate tubes of varying known thickness were filled with polyvinyl alcohol (to simulate soft tissue), configured on the CT table, and surrounded by fluid bolus bags to establish the accuracy and precision of average-thickness measurements using two CT convolution kernels and clinical study parameters (Table 1). CT-derived thickness measurements were compared with measures of actual thickness made using callipers. Short term precision: Repeat analysis of MDCT scans from 12 hips generated measurement pairs to estimate C.Th measurement precision for each quadrant (Table 2). Longitudinal drift was measured by scanning a quality assurance (QA) phantom weekly.

Table 1. Image Acquisition and Analysis Parameters
Patient Positioning for Hip QCT
 Supine on Siemens Somatom Sensation 64 scanner table
 Ergonomic solid-state phantom positioned under the hips (calibrated to aqueous K2HPO4 density)
 Separate QA phantom (containing aqueous K2HPO4) imaged prior to study
 Scout view from iliac crest to lesser trochanters
Acquisition parameters
 120 kV
 Tube current target 160 mAs (Siemens CARE dosing)
 0.6 mm detectors
 Pitch 1.4
 64 overlapping 0.6 mm slices per rotation
 Table height 155 cm
Reconstruction: To capture one hip and the phantom in each reconstruction
 1 mm slice thickness
 0.5 mm increment
 DFOV 300 mm (512 × 512 pixel matrix) = pixel size 0.5859 mm
 B20f convolution kernel
 CT DICOM format images
Image processing
 QCT PRO CTXA software (v4.1.3): reconstruct 3D image, extract vBMD, aBMD, bone shape
 Soft tissue/bone segmentation
 Bone Investigational Toolkit (BIT version 2)
 Scripting
 Automated femoral neck axis placement
 Identify where femoral neck max-min diameter ratio is 1.4 (midfemoral neck),
 Extract a single cross-sectional image at this location and five further medial slices 1 mm apart.
 Fixed threshold of 450 mg/cm3 to delineate cortical from trabecular bone
Table 2. Precision Studies (see Methods Section)
Thickness and diameter of Simax tubes versus callipers (B20f convolution kernel)
Regression equation r2p value
  • a

    Proportional measurement of sixth slice (max/min diameter ratio 1.4) from point a to point b in Fig. 1 (left),

True tube thickness = (0.96 × MDCT thickness) + 0.036 0.987<.0001
True tube diameter = (1.003 × MDCT diameter) + 0.002 0.978<.0001
Precision of paired measurements of C.Th, CSA, and slice position from 12 hips analyzed twice   
QuadrantMean of DistributionUnitsPrecision as SD/√2
Superoposterior1.36mm0.08
Superoanterior1.57mm0.11
Inferoanterior2.03mm0.13
Inferoposterior4.63mm0.09
Cross-sectional area7.99cm20.102
Slice positiona51%4.1
DXA-equivalent femoral neck aBMD, femoral neck volume, and vBMD

The software-determined regions of interest (ROIs) from the proximal femur on each side were the 3D equivalents of standard 2D axial DXA “femoral neck” and “total hip” ROIs (see Fig. 1). For each region, volume (cm3) and vBMD (mg/cm3) were calculated. To compare with expected age-related changes in aBMD, projected aBMD (g/cm2) values were calculated.

Statistical analysis

Linear regression was used to compare MDCT thickness of tubes with BIT-2-measured thickness. The dispersion-of-differences method was used to assess the precision (as 1 SD√2) of repeated measurement pairs of C.Th by quadrant and cross-sectional area. A nested ANOVA regression model was used to evaluate the relationships between C.Th (or other index of interest) at the quadrant and “16 sector” levels plus age, height, weight, slice position, and all significant interactions between them using JMP version 5.0.1 (SAS Institute, Cary, NC, USA). Means of results for the outcome variables were adjusted to ages 25 and 85 and height (1.6 m), weight (70 kg), and slice position.

Results

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Disclosures
  8. Acknowledgements
  9. References

Demographics

One-hundred and twenty-five women fulfilled the inclusion criteria, but 20 declined to participate (84% compliance rate) (Table 3). Five further participants were excluded after scanning [three based on the predefined medical exclusion criteria (carcinoma), two because of technical failure of the measurement]. One hundred scans underwent clinical reporting: Fifty-seven were classified as normal, and the others suggested benign pathology in the upper GI region (n = 4), lower GI region (n = 7), urinary system (n = 15), gynecologic system (n = 13), endocrine organs (n = 3), and liver (n = 1). The effective dose from the additional hip CT slices alone (to a 60 kg participant) was 1.8 mSV.

Table 3. Demographics of Participants, Mean ± SD
DecadeNo.AgeHeight (m)Weight (kg)BMI (kg/m2)aBMD T scorea
  • a

    Projection DXA equivalent (see Methods section). Regression equations:

  • b

    Height (m) = 1.68 – (0.001 × age), p < .0001.

  • c

    Weight (kg) = 76.8 – (0.11 × age), p < .0001.

  • d

    BMI; p = 0.82.

20–29824.5 ± 3.21.61 ± 0.0762.0 ± 8.826.9 ± 6.01.0 ± 1.1
30–392035.4 ± 2.81.61 ± 0.0675.5 ± 20.227.7 ± 7.70.3 ± 0.6
40–491444.4 ± 2.91.60 ± 0.1173.6 ± 18.126.6 ± 3.90.4 ± 0.9
50–591754.2 ± 3.01.63 ± 0.0570.3 ± 15.026.4 ± 6.0–0.5 ± 1.0
60–691463.5 ± 2.81.60 ± 0.0669.6 ± 14.027.1 ± 4.6–1.2 ± 0.7
70–791374.8 ± 2.31.65 ± 0.0872.1 ± 10.226.4 ± 4.8–1.3 ± 0.7
80–891483.1 ± 2.11.63 ± 0.0668.0 ± 12.726.7 ± 6.2–1.5 ± 0.8
r2 (age)0.08b0.02c0.0005d0.48

Clinical study

One hundred left hips and 97 right hips were analyzed (three right hips failed image processing; in three patients slight alterations in the display field of view (DFOV) led to slightly different pixel widths) (Table 4).

Table 4. Differences in C.Th and BMD by Quadrant (n = 100)
Variable, unitsMean ± SD 20–29 years raw dataMean ± SD 80–89 years raw data60-year difference 25–85 yearsa (95% CI of difference)
  • a

    A model evaluating age, quadrant, weight, height, and slice position explained 83% of the variance in C.Th (p < .0001), 86% of the variance in integral BMD (p < .0001), 79% of cortical BMD, and 51% of trabecular BMD.

  • b

    Slice position of starting slice no.1.

Superoposterior    
 Cortical thickness, mm2.1 ± 1.20.5 ± 0.4−1.32−1.69 to −0.96
 Integral BMD, mg/cm3273 ± 62103 ± 55−162−192 to −132
 Cortical BMD, mg/cm3569 ± 57361 ± 123−212−257 to −166
 Trabecular BMD, mg/cm3182 ± 1880 ± 46−104−121 to −88
Superoanterior    
 Cortical thickness, mm2.3 ± 0.90.9 ± 0.5−1.36−1.72 to −0.99
 Integral BMD, mg/cm3264 ± 60136 ± 32−126−156 to −96
 Cortical BMD, mg/cm3602 ± 70461 ± 84−154−199 to −108)
 Trabecular BMD, mg/cm3146 ± 1792 ± 26−54−70 to −37
Inferoanterior    
 Cortical thickness, mm3.9 ± 0.83.3 ± 0.5−0.47−0.83 to −0.10
 Integral BMD, mg/cm3408 ± 67368 ± 54−38−68 to −8
 Cortical BMD, mg/cm3745 ± 44740 ± 46−3−43 to −48
 Trabecular BMD, mg/cm3143 ± 22100 ± 27−50−67 to −32
Inferoposterior    
 Cortical thickness, mm3.7 ± 0.62.2 ± 0.8−1.21−1.6 to −0.84
 Integral BMD, g/cm3410 ± 54263 ± 62−137−167 to −107
 Cortical BMD, mg/cm3747 ± 41622 ± 137−103−149 to −58
 Trabecular BMD, g/cm3153 ± 1988 ± 33−74−91 to −57
Other variables    
 Femoral neck aBMD, g/cm20.91 ± 0.10.63 ± 0.1−0.29−0.34 to −0.23
 Femoral neck vBMD, mg/cm3435 ± 76282 ± 39−149−179 to −119
 Femoral neck volume, cm36.8 ± 1.07.8 ± 1.4+0.89+0.22 to +1.56
 Hip axis length (HAL), mm104.1 ± 6.1101.6 ± 5.2+2.4−1.01 to +5.78
 Slice position (%HAL), %b55.0 ± 0.0253.0 ± 0.02−1.20−0.025 to 0.004
Determining a fixed cortical/trabecular threshold for MDCT

In cadaveric femurs, MDCT C.Th was uniformly higher than pQCT thicknesses at the default threshold of 350 mg/cm3. This difference was statistically dependent on age and quadrant. Analysis of the MDCT data over a range of thresholds demonstrated that the overestimates declined as threshold increased and were nearest zero at 450 mg/cm3. This threshold therefore was used for all clinical analyses.21

Precision study using Simax tubes (see Table 2)

The correlation between the thickness of Simax tubes as measured by callipers and by the software was satisfactory (r2 = 0.987; p < .0001) at thicknesses as low as 1 mm using both convolution kernels and the relationship between CT and calliper measurements of wall thickness approximated the line of unity . For the B20 convolution kernel at a true thickness of 1 mm, the software underestimated thickness by 0.004 mm, or less than 0.5%. The B20 kernel was used for all clinical reconstructions.

Short-term precision (see Table 2)

The precision (SD/√2) of paired C.Th measurements ranged from 0.08 mm in SP to 0.13 mm in IA, but the SP region measurements included zero values. Longitudinal drift by QA phantom measurement was less than 2%. The location of the slice at max/min ratio 1.4 was reproducible, being a mean 51% distance along an extended femoral neck axis of the projected coronal image (indicated as a to b in Fig. 1), precision 4.1%.

Main outcome measures

Cortical thickness in quadrants

There were no side-to-side differences, so results for the 100 left hips are presented (see Table 4). The main statistical determinant of cortical thickness, as of the other outcome variables, was age. The model r2 improved from 0.77 to 0.81 with inclusion of weight, height, and starting slice position alongside age, indicating the small but significant contribution of these variables. A model including these factors at the sector level (16 sectors) explained 81% of the variance in C.Th (p < .0001). The model using the mean of C.Th for each quadrant as the outcome measure similarly explained 83% of the variance (p < .0001) (see Table 4 and Fig. 2). The results for the other outcome measures were similar in this respect except that the trabecular bone model explained only 51% of the variance.

thumbnail image

Figure 2. Primary outcome measure: regional cortical thickness with age by region. Linear regression lines for the mean and upper and lower 95% CI by quadrant.

Download figure to PowerPoint

The mean outcome variables predicted by the models at 25 and 85 years are shown in Table 4 and for C.Th in Fig. 3. Aging effects differed by quadrant (p < .0001). In particular, the differences between age 25 and age 85 were of greatest magnitude in the superior quadrants for all outcome variables. At 85 years, the apparent cortical thickness was just 0.33 mm [−1.3 mm; 95% confidence interval (CI) of difference over 60 years −1.69 to −0.95]. In the IA region mean C.Th at 25 years was 3.9 mm (predicted), whereas at 85 years it was 3.3 mm (−0.6 mm, 95% CI of difference over 60 years −0.83 to −0.10). Cortical BMD (CtBMD) of the IA region was similar at 25 (756 mg/cm3) and 85 years (753 mg/cm3; 95%CI of change −43 to 48). Femoral neck C.Th was expected to decrease when moving medially along the neck axis from its midpoint,11, 22 which was confirmed, particularly in the posterior quadrants. However, there was no interaction between age and starting slice position (mean 53% ± 2%), suggesting that there was no significant drift with age in the automatic ROI positioning method.

thumbnail image

Figure 3. (Left) The location of the midfemoral neck in stance from the “third-generation composite femur” (Day R, Sivandran S, Abeywickrema S, Third Generation Composite Femur with Intramedullary Canal. From: The BEL Repository, http//www.tecno.ior.it/VRLAB) rendered with Oracle AutoVue (Oracle Corporation, Redwood Shores, CA, USA). (Above, right) Schematic diagram showing cortical thickness at the midfemoral neck at 25 and 85 years, adjusted for weight (70 kg), height (1.6 m), and slice position (0.54): (a) 188 degree angle of thickest sector at age 25; (b) 173 degree angle of thickest sector at 85 years. (Below, right) Bone loss in millimeters per sector over 60 years (in millimeters of arrow length), all significant except sector 7 IA.*

Download figure to PowerPoint

Height was associated with HAL and starting slice position. The angle of the thickest sector relative to vertical (derived from the first-order Fourier coefficient) was 188 degrees at age 25 but was 173 degrees at 85 years (see Fig. 3). As age increased, the separation of the geometric center from the center of mineral mass increased from 2.5 ± 0.1 mm in the 20 to 30 age group to 4.3 ± 0.1 mm in those aged 80 to 90 years (p < .0001), as reported previously in studies using DXA.23, 24 These results reflect relative preservation of bone inferoanteriorly with loss superiorly. Models evaluating age, quadrant, weight, height, and slice position explained 86% of the variance in integral BMD (iBMD; p < .0001), 79% of CtBMD, and 51% of trabecular BMD (TrBMD) (see Table 4). ANOVA identified significant (p < .0001) within-subject differences in CtBMD by quadrant, being significantly higher in the IP than in the SP region (HSD). Such differences predictably reflect the partial-volume effect. Age had no significant association with apparent hip axis length (p = .47, mean 103.7 ± 6.0 mm).

DXA-equivalent femoral neck aBMD, femoral neck volume, and vBMD

Age alone explained 49% of the variance in femoral neck aBMD in g/cm2 (p < .0001), with height and weight effects being nonsignificant (aBMD = 1.006 – [age × 0.005]) (see Table 4). vBMD (in mg/cm3) demonstrated a similarly strong negative linear association with age. Femoral neck volume was +0.89 cm3 larger at 85 years than at 25 years (95% CI of difference +0.22 to +1.56).

Discussion

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Disclosures
  8. Acknowledgements
  9. References

We estimated the anatomic distribution of cortical and cancellous bone in the midfemoral neck in a cross-sectional study of 100 women with MDCT and identified strong aging effects. We found a thinner cortex in the femoral necks of older compared with younger women, with relative preservation of the IA region. The affected superior quadrants were markedly thinner at age 25 than the inferior quadrants, and the apparent rate of decline in cortical thickness with age was faster, especially in comparison with the thicker IA quadrant that is mechanically loaded by walking.25 TrBMD was lower in all regions in the eighth decade, but as with cortical thickness, the elderly women showed the greatest apparent reduction in the SP quadrant.

Endosteal resorption (with associated “trabecularization” of the cortex) is the most likely mechanism responsible for the targeted cortical bone loss associated with aging in the proximal femur.26 Referring to the lower right panel of Fig. 3, it is noteworthy that the regions of maximal bone loss coincided with areas of low mechanical strain during gait, heel strike, and stance,13 whereas the best preserved IA region was that experiencing higher strain in stance and heel strike (see Fig. 3, lower right panel). Generally, these new in vivo results are in agreement with our study of 77 male and female cadaveric femurs.9 In those deceased subjects, the IA quadrant appeared to increase in thickness by about 8 µm per year,9 but this apparent increase was not statistically different from zero (p = .11).

Recent studies using high-resolution CT scanning have given similar results for cortical bone at peripheral sites. Thus, in the radius and tibia, the apparent effect of age in a population cohort was to reduce cortical thickness, especially in women.27 Further, Vico and colleagues28 found that subjects with prior hip and forearm fractures had lost more cortical bone at these sites than their age-matched peers, as we found in our biopsy study of the femoral neck.9 Using an approximate 2D estimating method based on DXA, Kaptoge and colleagues demonstrated that apparent cortical thickness performed similarly to DXA BMD in predicting hip fracture.24 It seems possible that with a more accurate method for measuring the cortical thickness in locations experiencing the maximal mechanical loading during a fall, the predictive ability of DXA BMD eventually might be exceeded, and work is ongoing to test this possibility. These results are also concordant with previous ex vivo case-control studies.9–11

Injurious falls, like hip fractures increase with age, and fall risk and type of fall must be considered when interpreting our findings with regard to fracture mechanisms. Whatever the mechanical explanation or the type of fracture experienced,29 loss of cortical and trabecular bone must reduce resistance to fracture in a fall unless the mechanical load experienced is reduced in proportion. We have argued elsewhere that the importance of trabecular bone for fracture resistance is increased if the femoral neck cortex fails through local buckling rather than through materials failure in which the superior cortex is crushed.17 This study has shown intriguingly that TrBMD is considerably less predictable from the age of the subject than cortical thickness, so this should be a focus of future prospective investigations on differences in hip fracture susceptibility between elderly subjects with similar aBMD values at the femoral neck. Recent MDCT studies of volunteers and proximal femoral samples have identified femoral CtBMD and TrBMD not only as correlates of failure load in cadaveric specimens30–33 but also as predictors of hip fracture in life,34 although like DXA-based hip structural analysis,24 a CT-based prediction method for hip fracture has yet to outperform conventional DXA-based BMD measurement. Very recently, however, it has been shown that in a sideways fall simulation, femoral neck fractures can initiate in the superior cortex, at least in load-to-failure testing of resected cadaveric bone specimens.16

These results have relevance for prevention. The very strong inverse linear association between superior region bone indices and age requires explanation. While walking generally continues into advanced old age, the femoral necks of older women are rarely subjected to the directionally variable “odd impact” and “high impact” increased loads that are associated in physically active young women with thickened femoral neck cortices.35 The ossification pattern of the relatively long human femoral neck engenders a thin superior cortex during growth.36 However, it also has been postulated that an adaptive bone remodeling response to the typical strains from bipedal locomotion causes increasing asymmetry of the femoral neck cross section during life.37 A recent finite-element modeling study infers that this asymmetry helps to resist stress fracture resulting from running or jumping.25 The low-strain environment in the superior region of the femoral neck might result, through targeted remodeling,26 in thinning and trabecularization of cortical bone, so helping the development of asymmetry of the cortex.38 Currey and colleagues39 argued from an evolutionary perspective that lifelong periosteal bone formation and endosteal resorption of the cortex,26, 40 which increase bending resistance and simultaneously maintain skeletal lightness in the femur, are beneficial in young adulthood for their effects on fracture risk, well before fragility is reached. Enlargement of the femoral neck cross section with asymmetric cortical thinning continues into old age, as it appears to do in other limb bone cortices,27, 41, 42 so the relative preservation of the IA cortices loaded by walking is more unusual than the observed loss of superior cortical bone.

Our compliance rate of 84% was good for a study of this nature. However, our study has limitations. First are those associated with all cross-sectional studies, such as the difficulty of differentiating age-related effects from cohort effects, for example, the gradual increase in height and other physical dimensions of the UK population. Previous prospective studies have shown that aging is associated with a gradual increase in femoral neck volume43 owing to periosteal apposition.40 Our population was Caucasian, so our results may not necessarily apply to other racial groups. For simplicity in application, we used a single threshold to delineate cortex from trabecular bone. The partial-volume effect tends to make trabecular bone close to the endosteal boundary appear more dense and cortical bone less dense than they really are.44, 45 We suspect that these effects led to cortical thickness overestimation, particularly in the inferior region of the femoral neck cortex in the younger women. Downward bias also can occur when true cortical thickness (in this case just 0.6 mm10 in the SP quadrant) approaches the pixel size, especially for highly porous cortices.45 Higher apparent cortical density in the inferior regions compared with the superior regions appeared exaggerated compared with direct measurements made with a scanning electron microscopy (SEM) technique.46 Zero values accounted for 6.8% of the 9600 sector measurements we made in 100 left hips. While partial-volume errors clearly affect true estimates, they are unlikely to mask large trends, such as that seen with aging. Also, while the protocol used was appropriate for our scanner, it likely needs adjusting for different CT machines and convolution kernels.

In conclusion, elderly women had relative preservation of IA femoral neck bone (the region that transmits load from walking) over seven decades but markedly lower cortical thickness and BMD in the other three quadrants of the femoral neck. The superior quadrants were most affected by aging. Although it is possible that structural changes such as these in the proximal femur predispose to hip fracture in elderly fallers, it remains to be determined whether this MDCT technique can reliably identify individuals at risk of hip fracture.

Disclosures

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Disclosures
  8. Acknowledgements
  9. References

KESP acknowledges the support of the arthritis research campaign (Clinician Scientist Fellowship). This work also was supported by a project grant from the Medical Research Council (UK). PMM and CR acknowledge the support of the National Institute of Health Research Biomedical Research Centre Musculoskeletal Theme, Cambridge.

The funding bodies had no role in study design; in the collection, analysis, and interpretation of data; in the writing of the report; nor in the decision to submit the article for publication.

JKB developed QCT PRO and BIT software and has a proprietary interest in Mindways Software. KESP, PMM, CMR, PJB, NL, and JR purchased the software and receive free software upgrades and technical support. They certify that they have no affiliations or involvement in any organization or entity with a direct financial interest in the subject matter or materials discussed in this article.

Acknowledgements

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Disclosures
  8. Acknowledgements
  9. References

We thank Stephen Kaptoge, Department of Public Health and Primary Care, for expert statistical advice and Stuart Yates, who measured radiation from the CT scanner used in the study and estimated the whole-body equivalent exposure received from the scanner with IMPACT dosimetry software.

References

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Disclosures
  8. Acknowledgements
  9. References