Hip structural analysis: a comparison of DXA with CT in postmenopausal Japanese women

Japan is facing the major challenge of meeting the burgeoning health care needs of a rapidly aging population. Osteoporotic fracture is one of the common causes of elderly persons becoming bed-ridden in Japan, and is an issue that needs to be urgently addressed. Among common sites of osteoporotic fractures, the proximal femur can be one the most devastating. In the first year following hip fracture the mortality rate rises to 20-24% (Cooper et al. 1993; Leibson et al. 2002), and survivors of hip fracture often live with chronic pain, disability, and increasing dependence on caretakers (Keene et al. 1993). In Japan, approximately 150,000 hip fractures occur per year, and this number has continued to rise over the past 20 years (Orimo et al. 2009).

Assessment of bone strength is important in the diagnosis of osteoporosis. Although bone mineral density (BMD) correlates with bone strength and bone densitometry, especially dual-energy X-ray absorptiometry (DXA), is widely used to assess BMD in clinical practice (Cummings et al. 2002), it is well known that BMD itself is inadequate for accurate estimation of bone strength (Cody et al. 1999; Esses et al. 1989; Beck et al. 1990). DXA has limitations owing to the complex three-dimensional (3D) geometry of the hip, low spatial resolution, and the two-dimensional (2D) nature of this imaging approach. To overcome these limitations, Beck et al. have worked to derive more complex biomechanical indices based on hip structural analysis (HSA) (Beck et al. 1990). DXA-based HSA (DXA-HSA) is a simple and easy method that can be used to assess proximal femoral geometry. Although various investigations have been made on the clinical relevance of DXA-HSA, consensus has not been obtained because of the variations between races, evaluation methods, et al. Furthermore, limitations imposed by the 2-dimensional nature of DXA have not been fully addressed (Bouxsein and Karasik 2006). From the currently available data, the HSA structural parameters by DXA are highly correlated with areal BMD and while predictive of fracture risk, they have not shown much improvement in the fracture prediction compared to areal BMD. Nevertheless, HSA with DXA has provided unique insights into the mechanisms of both the pathophysiology of osteoporotic fracture and the therapeutic efficacy of bone-active agents.

The major limitations of HSA with DXA primarily reflect limitations imposed by the 2-dimensional nature of DXA. As an alternative to DXA, quantitative computed tomography (QCT) offers complete 3D information, high in-plane spatial resolution, bone geometry, and separate assessment of the cortical and trabecular bones of the femur (Bauer et al. 2006; Duchemin et al. 2008). Recently, CT technology has dramatically advanced and broadened the range of CT applications from the diagnostic purpose to the navigation system in orthopedic surgery. These CT data can be used for HSA without additional radiation exposure to patients. While some studies have suggested the superior precision in HSA with QCT compared to DXA-HSA (Ramamurthi et al. 2012), it remains to be seen whether this is true for the analysis by utilizing CT data of general clinical practice. Thus, one of the major research questions in this study is to compare the performance of HSA between DXA and CT that has been scanned for the purpose other than HSA.

In this paper, in vivo comparison of DXA-HSA to HSA based on QCT (CT-HSA) is reported in a population of postmenopausal Japanese women. Additionally, the biomechanical significance of these two kinds of HSA were evaluated by comparing results to the CT-based finite-element method (CT/FEM), which is considered one of the gold standards for noninvasive assessment of bone strength (Bessho et al. 2007; Keyak et al. 1998; Keller 1994).

DXA-HSA indices were significantly correlated (p<0.01) with CT-HSA both in the NN and IT regions (Table 3, Figures 3, and 4). Correlation coefficients were high (>0.8) especially in CSA_NN, CTh_NN, CSA_IT, CTh_IT, and SM_IT.

Region	CSA	CTh	SM	BR
NN	0.90*	0.85*	0.60*	0.74*
IT	0.86*	0.85*	0.82*	0.32*

Values are Pearson's correlation coefficient.

CSA: bone cross-sectional area, CTh: average cortical thickness, SM: section modulus, BR: buckling ratio, NN: narrow neck region, IT intertrochanteric region.

*p<0.001.

Total hip BMD was significantly correlated with FL and the correlation coefficient was 0.66 and 0.77 in stance and fall configuration, respectively (p<0.01).

Although BMD (in grams per square centimeter as measured by DXA) is the current gold standard for clinical assessment of bone fragility (Cummings et al. 2002; Marshall et al. 1996), it is well known that BMD itself is inadequate for accurate estimation of bone strength (Cody et al. 1999; Esses et al. 1989; Beck et al. 1990). It is necessary to evaluate strength considering not only BMD but also bone quality. Among several factors contributing to bone quality, the geometry of the proximal femur can be assessed using an HSA program in routine hip DXA measurement. This technique is expected to have a supplemental role in assessment of bone strength.

It has been reported that Asians generally have a lower BMD than Caucasians, but hip fracture rates are lower in Asians (Villa et al. 2001; Xu et al. 1996; Lau et al. 2001; Ross et al. 1991). The size and geometry of proximal femur could partially account for this difference across ethnicities. Indeed, a comparative study between Japanese and Caucasian American women showed that Japanese women had shorter femoral neck and smaller femoral neck angle compared to Caucasian women, which was associated with a lower risk of structural failure (Nakamura et al. 1994). This characteristic of a shorter femoral neck in Japanese women can sometimes lead to inaccurate hip DXA measurement. Before the present study, there had been no report on the accuracy of HSA assessment in Japanese subjects.

First, HSA indices were compared based on DXA, with the equivalent indices calculated on QCT. Overall, DXA-HSA was highly correlated with CT-HSA. The correlation had an r>0.8, except for SM in the NN region and for BR in the NN and IT regions. This is consistent with the recent report by Ramamurthi et al. (Ramamurthi et al. 2012), in which DXA-HSA was compared with measurements obtained by high-resolution QCT. They showed that DXA-HSA correlates strongly with CT-HSA both in the NN and IT regions (r=0.89-95). Although the correlation in the present study was slightly lower than that reported by Ramamurthi et al., this difference could be attributed to the difference in the registration technique. They used a sophisticated method to ensure that really the same regions were analyzed between CT and DXA. Another possible cause of the difference would be the difference in slice thickness in CT. In the present study, CT data with a 2.5 mm slice thickness was used, parameters which are routinely adopted in preoperative examination of joint replacement surgery in our institute. On the other hand, Ramamurthi et al. used a 1 mm slice thickness. Because even high-resolution QCT cannot accurately measure cortical thickness below 1.0-1.5 mm (Prevrhal et al. 1999), CTh estimated with CT in the present study would have been affected by partial volume averaging, especially in subjects with thin cortices.

Compared to the overall high correlation of HSA between DXA and CT, the correlation of BR_IT was significantly lower than the other parameters. Since the mean value of BR was calculated in the present study by averaging the BR of each sector, sectors with a thin cortex could be largely affected by the partial volume effect and result in an error in estimating cortical thickness. Error in cortical thickness is augmented by calculating BR.

When comparing DXA-HSA with CT-HSA, the correlation of HSA with FL estimated by FEM based on QCT (which is one of the current gold standards of in vivo assessment of bone strength) was also evaluated. The correlation coefficients between total hip BMD and FL were 0.66 and 0.77 in stance and fall configuration, respectively and were equivalent to the results by Danielson et al. (Danielson et al. 2013) and Orwoll et al. (Orwoll et al. 2009). Both DXA-HSA and CT-HSA were significantly correlated with FL, and the correlation was similar or even higher in DXA-HSA compared to CT-HSA. However, the correlation itself was not very high and did not suggest equivalence. This is consistent with recent reports by Danielson et al. (Danielson et al. 2013), who found that femoral neck geometry computed by HSA from DXA data corresponds well to that from QCT for analysis of load stress in a large cohort of postmenopausal women. They also showed that proximal femur breaking strength estimated from 2D DXA data was not as well correlated with femur breaking strength derived by 3D FEM using QCT data (Danielson et al. 2013). The localized measurement in the proximal femur in the neck or trochanter may not be representative for the total proximal femur.

Several groups have recently published prospective studies assessing the efficacy of HSA to predict hip fractures. Kaptoge et al. analyzed DXA-HSA data from 7,474 women in the prospective population-based Study of Osteoporotic Fractures (SOF) (Kaptoge et al. 2008). In the SOF, 635 women suffered incident hip fractures over 13 years of follow-up. They concluded that for hip fracture prediction using NN region parameters, CTh, areal BMD, and BR were equivalent, but SM performed less well. LaCroix et al. studied 10,290 postmenopausal women from the Women's Health Initiative (LaCroix et al. 2010). They concluded that two hip geometry parameters, outer diameter and BR in the IT region, can predict incident hip fracture after accounting for clinical risk factors and BMD (Ross et al. 1991). In the present study, significant predictors of FL_SC in DXA-HSA were CSA_NN, SM_NN, SM_IT, and age; and significant predictors of FL_SC in CT-HSA were CSA_NN, SM_NN, SM_IT, age, and height. Significant predictors of FL_FC in DXA-HSA were CSA_IT, age, and weight and significant predictors of FL_FC in CT-HSA were CSA_IT, CSA_NN, SM_IT, SM_NN, total hip BMD, and weight. Thus the results in this study do not correspond well with those of the prospective studies. Although it is difficult to compare many variables that are closely correlated with each other in multiple regression analysis, the discrepancy might suggest the biomechanical assumption does not completely describe our FEM model.

This study had a number of limitations. First, all subjects were postmenopausal women before artificial joint replacement surgery for osteoarthrosis of the hip or knee joint. Because CT has been adopted as a routine preoperative examination for artificial joint replacement surgery in our institute, this method of subject recruitment solved the problem of additional radiation exposure. On the other hand, the subjects may not represent the general postmenopausal Japanese population. It has been reported that subjects with osteoarthrosis may have increased BMD (Dequeker et al. 2003). Alternatively, osteoarthrosis may also be related to increased bone resorption, which results in decreased bone mass (Henrotin et al. 2009) and increased fracture risk. In this study, lumbar and total hip BMDs were higher, whereas femoral neck BMD was close to previous reports (Orimo et al. 2001; Orimo et al. 1998). Second, the methods used to calculate HSA indices are inherently different and limit strict comparison. Finally, QCT data scanned with a slice thickness of 2.5 mm was used, potentially negatively affecting the precision of HSA measurement. However, since reduced slice thickness is associated with a corresponding increase in radiation exposure, reduction of slice thickness is often limited in general clinical practice. The present data suggest that the CT-HSA using the scan condition such as 2.5 mm slice thickness would provide little information over DXA-HSA and the DXA-HSA could be sufficiently accurate compared to such CT methods.

In conclusion, there was high correlation between DXA-HSA and CT-HSA for CSA, CTh, and SM in postmenopausal Japanese women. Moreover, the correlation of HSA with FL was similar between DXA-HSA and CT-HSA (and was even slightly higher in DXA-HSA). These results suggest that the geometry of proximal femoral cross sections is reasonably well characterized by DXA.