Long-Term Corticosteroid Therapy Induces Mild Changes in Trabecular Bone Texture



The relative roles of bone mineral density (BMD) decrease and of microarchitectural changes in corticosteroid-induced osteoporosis (CIOP) are debated. Our objective has been to evaluate both bone microarchitecture (by a fractal analysis of texture on radiographs) and BMD in corticosteroid (CS)-treated patients. In this study, 60 patients from a rheumatology unit with a mean age of 60.6 ± 14.8 years taking CS therapy for more than 6 months and a cumulative dose of prednisone over 1 g and 57 controls among age-matched patients and hospital staff were recruited. Bone diseases and bone-modifying drugs (except calcium, vitamin D, and hormonal replacement therapy [HRT]) were considered as exclusion criteria. A fractal analysis of trabecular bone texture was performed on calcaneus radiographs after an oriented analysis in 18 directions. The fractal analysis was based on the fractional Brownian motion model Results were expressed by H parameter (H = 2 – fractal dimension) in each direction, Hmean being the average of 18 directions, Hmini the minimum, and Hmaxi the maximum. BMD was measured by double-energy X-ray absorptiometry (DEXA) at the femoral neck (FN) and lumbar spine (LS). The odds ratios (OR) were calculated for a variation of 1 SD. The mean duration and dose of CS therapy was 5.6 ± 6.6 years and 16.9 ± 19.7 g. CS therapy was significantly correlated to a decrease in FN or LSBMD: OR = 1.95,95% confidence interval (CI, 1.29–2.97) and OR = 3.19 (CI, 1.80–5.66), respectively. The Hmean and Hmaxi were significantly lower in the cases than in the controls: P = 0.03 and P = 0.02; OR = 1.67 (CI, 1.10–2.54) and OR = 1.75 (CI, 1.05–2.37). A similar trend was observed with Hmini but the difference did not reach the level of statistical significance: P = 0.06, OR = 1.57 (CI, 1.05–2.37). This study was repeated among cases and controls who had never taken HRT (respectively, n = 40 and n = 39). The results were similar. Among patients taking CS therapy, the presence of nontraumatic fractures was inversely related to BMD values but not to texture parameters. These data have shown that long-term CS therapy induces both BMD decrease and trabecular bone texture changes. The effect of CS therapy was much stronger on BMD than on the fractal H parameter. These results are in accordance with previous studies showing a lower effect of CS therapy on bone microarchitecture than on bone mass. These results can be contrasted with those observed in women with postmenopausal osteoporosis and vertebral crush fractures in which the variations in the fractal parameters are more significant than the BMD variations.


Corticosteroid-induced osteoporosis (CIOP) is the first cause of secondary osteoporosis. The use of corticosteroids (CS) is rapidly increasing and has become common mainly in patients over 55 years old.(1) According to a study by Walsh et al., it was concluded that over 250,000 people in the United Kingdom are taking continuous oral steroids.(2) The influence of CS therapy on the trabecular bone is considered predominant versus cortical bone.(3) Fractures are especially more frequent at trabecular site such as the spine.(4) The bone mineral density BMD decrease during CS therapy is found both at trabecular and cortical sites.(3,5–7) In some studies, the BMD decrease does not completely explain the fractures.(8,9) For instance in CS-treated rheumatoid arthritis, Peel et al. found the decrease in lumbar spine BMD (LSBMD) to be less than expected considering the observed prevalence of vertebral fractures. They noticed that cases with vertebral fracture did not have lower LSBMD values than those without vertebral fracture.(8) Luengo et al. studied CS-dependent asthma by comparison with involutional osteoporosis.(9) In cases of vertebral fracture BMD was significantly lower in involutional osteoporosis than in CS-treated patients. These data suggest an additional mechanism of bone fragility, for instance qualitative changes in trabecular bone. Some histomorphometric studies have shown that CS therapy induces bone loss by trabecular thinning rather than by perforations or disconnections within the trabecular network.(10–12)

Therefore, our aim has been to characterize trabecular bone changes in CS-treated patients by a fractal analysis of texture on radiographs. We previously developed and validated a fractal analysis of trabecular bone texture on plain radiographic images.(13) Fractal mathematics are particularly suited to this kind of analysis because they permit a numerical characterization of complex images. Both a high degree of complexity and a repetition of self-similar patterns characterize fractal images over different scales. Concerning grey level images, their characterization can be achieved by the use of a fractal model derived from the fractional Gaussian noise, which is the increment of the fractional Brownian motion. This model has been shown to be well suited to the fractal analysis of trabecular bone texture on calcaneus radiographs, using the maximum likelihood estimator as an indicator applied to the fractional Gaussian noise.(13,14) The result is expressed by the H parameter, which is linked to the fractal dimension (D) by H = 2 – D.

Our method has been shown to be reproducible concerning intra- and interobserver reproducibility as well as for long-term reproducibility, with the following coefficients of variation: 0.61, 0.68, and 2.07%, respectively.(13) In a pilot study on a small number of subjects including several types of osteoporotic fractures, we showed that this analysis was able to distinguish between an osteoporotic fracture group and a control group.(15) Recently, we confirmed these results in a study comparing postmenopausal controls to women with osteoporotic vertebral crush fractures.(16)

The aim of the present work was to evaluate the ability of the fractal analysis of texture on radiographs to distinguish a group of patients receiving CS therapy from an age-matched control group in parallel with a BMD study in these same groups, to determine the relative roles of density and microarchitecture bone changes in CIOP.



We studied 60 female patients from our Rheumatology Unit, taking CS for more than 6 months, with a cumulative dose over 1 g of prednisone. Their age ranged between 26 and 84 years (mean, 60.6 ± 14.8 years). All other bone diseases (cancer, Paget's disease, hyperparathyroidism, and hyperthyroidism) and long-term immobilization were considered as exclusion criteria. Patients who were taking bone-modifying drugs except for calcium, vitamin D, and hormonal replacement therapy (HRT) were excluded. Eighteen patients were taking CS for rheumatoid arthritis, 11 for polymyalgia rheumatica and temporal arteritis, 9 for vasculitis or lupus, 10 for other types of inflammatory rheumatism, 6 for asthma, and 6 for miscellaneous disorders. Thirty-two fractures were reported in 25 patients: 12 vertebral fractures, 9 rib fractures, 7 wrist fractures, 3 hip fractures, and 1 ankle fracture. The duration and cumulative dose of CS were calculated by physicians according to patients' medical records.

Age-matched controls were recruited among patients from the Rheumatology Unit and hospital staff (n = 57). Exclusion criteria for controls were those previously defined for cases but also included present or past CS therapy.

Evaluation of bone densitometry

BMD was measured at the hip and LS by double-energy X-ray absorptiometry (DEXA; HOLOGIC QDR 1000/W; Hologic, Inc., Waltham, MA, U.S.A.). The manufacturer's recommended standard analysis procedure was performed on the hip of all patients and controls. Results were expressed in grams per centimeter squared. LSBMD measurement was not possible in all the subjects because of osteoarthritis, scoliosis, or vertebral crush fracture. Therefore, LSBMD was measured for 48 cases and 47 controls.

Evaluation of bone texture

The fractal analysis of trabecular bone texture on radiographs was performed after the procedure reported by Benhamou et al.(13) A standardized procedure was defined to take the radiographic images. The calcaneus was placed in contact with the film and we fixed a distance of 1 m between the film and the X-ray focal source. Gridless cassettes (Plastilix) and a Kodak MIN-R film (Kodak, Paris, France) were used. We used a 48-kV voltage of the X-ray tube and exposure conditions were fixed at 18 mAs for an exposure time of 0.08 s. A region of interest in an area of trabecular bone defined by anatomic marks was selected on the radiographs and digitized with a scanner (AGFA Duoscan II; AGFA Gevaert N.V., Mortsel, Belgium). Representative radiographs of a CS-treated patient and a control case with their digitized counterparts are presented in Fig. 1. An oriented texture analysis was performed to study the anisotropy of the trabecular network projected on the texture of the trabecular bone image.(17) The value of H was calculated in 18 directions on the image. For each image, the mean, the minimum, and the maximum of the 18 values were calculated: results were Hmean, Hmini, and Hmaxi, respectively.

Statistical analysis

The level of significance was chosen at P < 0.05 and all statistic tests were two-tailed. For quantitative values, the mean ± SD is given. The Student's t-test was used to compare averages between cases and controls when the Fisher test for equality of SD was not significant. When there was a small heterogeneity of SD between the two groups (P between 0.001 and 0.05) the Satterthwaite method was applied, and when the heterogeneity was strong (P < 0.001) the nonparametric test of Mann and Whitney was used to compare both groups. The odds ratios (OR) were estimated by logistic regression for a variation of BMD (femoral neck [FN] or LS) of 1 SD and a variation of H of 0.05. The relationship between age CS therapy duration or dose and BMD or H parameters was studied by simple linear regression analysis: y = ax + b where y is the dependent variable and x the predictor variable. When the examination of the graphs showed a nonlinear relationship between two variables, the logarithmic relationship was studied (model, y = a × log[x] + b). When necessary, adjustment for age has been performed using a variance analysis with two factors. The SAS (Statistical Analysis System) statistical software package was used (SAS Institute, Inc., NC, U.S.A.).

Figure FIG. 1..

Representative radiographs (A and C) of a corticosteroid-treated patient (left) and a control case (right) with their digitized counterparts (B and D).


Patients with CS therapy are described in Table 1. They had received a mean cumulative dose of 16.9 g (between 1 g and 86.4 g) for 5.6 years (between 6 months and 30 years). The presence of fractures was significantly related to age: 55.3 years in patients without fractures versus 68.1 years among patients with fractures (P < 0.001). Patients who had fractures had received a longer and stronger CS therapy compared with patients without fractures (7.5 years vs. 4.3 years and 19.7 g vs. 14.8 g), but these differences were not significant (respectively, P = 0.07 for duration and P = 0.35 for dose). Significant heterogeneity in age and CS therapy duration were observed between groups of diseases (P < 0.001). Concerning the CS cumulative dose, differences between diseases were not significant (P = 0.60). Patients with asthma received the longest and the highest dose of CS therapy (17.1 years and 26.4 g). Patients treated for vasculitis or lupus were the youngest and had received a milder CS course (11.8 g for 1.9 years). After adjusting for age, the CS duration remained significantly associated with diseases (P < 0.001) and differences in cumulative doses remained nonsignificant (P = 0.36).

Table Table 1.. Description of the 60 Female Patients with Corticosteroid Therapy
 Age (years)CS therapy duration (years)CS therapy dose (grams prednisone equivalent)
 nMean ± SDMinimumMaximumMean ± SDMinimumMaximumMean ± SDMinimumMaximum
All cases6060.6 ± 14.826845.6 ±6.60.530.016.9 ±
 No fractures3555.3 ± 4.626844.3 ±4.80.515.214.8 ±
 At least one fracture2568.1 ± 1.729837.5 ± 8.30.530.019.7 ± 19.32.479.2
 Rheumatoid arthritis1860.5 ± 16.231817.3 ±60.10.520.019.6 ±
 Polymyalgia rheumatica and temporal arteritis1176.6 ± 5.270841.9 ± ± 22.43.379.4
 Inflammatory rheumatism1052.3 ± 11.439723.0 ±2.90.510.09.8 ± 11.11.536.0
 Vasculitis or lupus948.7 ± 18.626751.9 ± ± 8.53.328.8
 Asthma661.2 ±6.6557017.1 ± ± 30.12.479.2
 Miscellaneous diseases663.2 ± 13.450816.2 ±6.30.515.220.2 ±

The bone texture parameters and BMD of CS-treated patients were compared with those of controls (Table 2). CS therapy was significantly associated with a decrease in FN and LSBMD: odds ratio (OR) = 1.95 (confidence interval [CI], 1.29–2.97) and 3.19 (CI, 1.80–5.66), respectively. The Hmean and Hmaxi were significantly lower in patients than in controls: OR = 1.67 (CI, 1.10–2.54) and 1.75 (CI, 1.13–2.71). A similar trend was observed for Hmini, but the difference did not reach the level of statistical significance: OR = 1.57 (CI, 1.05–2.37) and P = 0.06. The OR for bone texture parameters decreased slightly after being adjusted for LS or FNBMD and became nonsignificant (Table 2). Nevertheless, associations were borderline of the significant threshold, particularly for Hmean and Hmaxi when FNBMD was taken into account (OR = 1.53, [CI, 0.97–2.42] and 1.56 [CI, 0.97–2.50], respectively). This study was repeated among patients and controls who had never taken HRT (n = 40 and n = 39). The OR of bone texture parameters and BMD were significant and greater than previous ORs, but CIs were larger because of the smaller number of subjects (Table 3). The Hmean, Hmini, and Hmaxi became nonsignificantly correlated with CS therapy after being adjusted for LS or FNBMD.

The association between the presence of fractures and bone texture parameters or BMD was studied among patients treated by CS (Table 4). LS and FNBMD were significantly lower among women with fractures compared with women without fracture: 0.792 vs. 0.915 and P = 0.002 for LSBMD and 0.555 vs. 0.700 and P < 0.001 for FNBMD, respectively. The Hmean, Hmini, and Hmaxi were not related to the presence of fracture. Because women with fractures were significantly older than control women the study was repeated after adjustment for age. This adjustment has been performed using a variance analysis with two factors. Bone texture parameters remained nonsignificantly correlated to fracture. The difference in LSBMD was lower (P = 0.07), and the difference in FNBMD remained highly significant (P < 0.001). Among patients with fractures, 7 were taking HRT and 18 were not; χ2 test, P = 0.463; OR = 0.658 (CI, 0.216–1.99).

The relationship between the duration or dose of CS therapy and BMD or H parameters was studied in patients taking CS. No linear and log-linear trends were observed on graphs between variables. Age was studied as a potential confounding factor. An inverse relationship between BMD (FN or LS) and age was found (P < 0.0001). The H parameters were not significantly correlated with age. We have not found correlations (P > 0.40) between age and dose or duration of CS therapy. The population was divided into two groups: one for women over 60 years old (n = 31) and the other for those under 60 years old (n = 29). The duration and cumulative dose of CS was slightly higher among the youngest patients compared with the oldest, but the correlations were not significant (6.4 ± 5.9 vs. 7.9 ± 7.2, P = 0.39 for duration and 19.7 ± 22.0 vs 14.2 ± 17.9, P = 0.29 for dose). In conclusion, the absence of correlations between the duration and dose of CS therapy with BMD and H parameters could not be explained by age.

Table Table 2.. Comparison of Texture Parameters and BMD Between Patients with CS Therapy and Controls
 Controls (n = 57)Patients (n = 60)PCrude OR (CI)OR adjusted for LSBMD (CI)OR adjusted for FNBMD (CI)
Age (mean ± SD)60.3 ± 13.860.6 ± 14.80.90   
Hmean (mean ± SD)0.626 ± 0.060.591 ± (1.10–2.54)1.57 (0.94–2.63)1.53 (0.97–2.42)
Hmini (mean ± SD)0.494 ± 0.040.473 ± (1.05–2.37)1.39 (0.84–2.28)1.43 (0.91–2.25)
Hmaxi (mean ± SD)0.851 ± 0.090.793 ± (1.13–2.71)1.58 (0.93–2.68)1.56 (0.97–2.50)
LSBMD (g/cm2, mean ± SD)1.031 ± 0.170.877 ± 0.13<0.0013.19 (1.80–5.66)  
FNBMD (g/cm2, mean ± SD)0.730 ± 0.130.644 ± 0.14<0.0011.95 (1.29–2.97)  
Table Table 3.. Comparison of Texture Parameters and BMD Between Patients with CS Therapy, in a Population Without HRT
 Controls (n = 39)Patients (n = 40)POR (CI)
Age (mean ± SD)58.7 ± 13.361.8 ± 17.40.38 
Hmean (mean ± SD)0.629 ± 0.060.586 ± (1.08–3.17)
Hmini (mean ± SD)0.498 ± 0.040.469 ± (1.07–3.20)
Hmaxi (mean ± SD)0.856 ± 0.080.786 ± (1.13–3.64)
LSBMD (g/cm2, mean ± SD)1.069 ± 0.180.869 ± 0.14<0.0014.92 (2.12–11.40)
FNBMD (g/cm2, mean ± SD)0.760 ± 0.120.631 ± 0.15<0.0012.83 (1.58–5.08)
Table Table 4.. Comparison of Texture Parameters and BMD Between Patients with CS Therapy and Controls After the Presence or Absence of Fractures
 Crude analysisAnalysis after adjusting for age
 Cases without fractures (n = 35)Cases with fractures (n = 25)PCases without fracturesCases with fracturesP
Age (mean ± SD)55.34 ± 14.5668.1 ± 11.7<0.001   
Hmean (mean ± SD)0.588 ± 0.090.594 ± 0.080.810.5840.6000.52
Hmini (mean ± SD)0.475 ± 0.060.470 ± 0.050.780.4710.4740.87
Hmaxi (mean ± SD)0.786 ± 0.1420.803 ± 0.130.640.7790.8130.40
LSBMD (g/cm2, mean ± SD)0.915 ± 0.130.792 ± 0.090.0020.8980.8300.07
FNBMD (g/cm2, mean ± SD)0.700 ± 0.130.555 ± 0.11<0.0010.6730.597<0.001


These data have shown that a fractal analysis of bone texture on calcaneus radiographs can distinguish a group of patients receiving long-term CS therapy from an age-matched control group.

The BMD decrease in CS long-term therapy is considered to be more marked in trabecular bone than in cortical bone.(4,18,19) However, a significant reduction in bone mineral content has been observed at both the distal and the proximal radius showing that cortical bone is not spared from the effects of CS.(8,20,21) At the proximal femur, it has been shown that CS therapy induces a similar bone mass reduction at the FN, at the trochanter, and at the Ward's triangle.(5) The FN and the trochanter are known to contain, respectively, 75% and 50% of cortical bone whereas the Ward's triangle is considered to be a trabecular site.(22–24) In our study we also found a significantly lower BMD value at the FN in the CS-treated patients than in the controls; the OR for a decrease of 1 SD induced by CS therapy concerning the FNBMD was evaluated: OR = 1.95 (CI, 1.29–2.97). LSBMD was more significantly lowered: OR = 3.19 (CI, 1.80–5.66). The relative contributions of the cortical and trabecular bones to BMD measurement by DEXA at the lumbar site is about 50:50.(23) Nevertheless, the incidence of fractures in CS-treated patients was greater at trabecular bone sites than at cortical sites.(25) In our study we observed 12 vertebral crush fractures, 9 rib fractures, 7 Colles fractures, 3 hip fractures, and 1 ankle fracture confirming the trabecular predominance.

The relationship between bone loss and dose or duration of CS is generally admitted but is not always easy to show.(25–27) Significant relationships between the duration of previous steroid therapy, total cumulative CS dose, and BMD have been reported.(3,7) The total duration of CS has been found to be correlated negatively with BMD at the proximal femur but not at the LS.(28) Other studies have failed to show such an association.(5,8,9) In the present study we did not observe any relationship between duration or dose of CS therapy and BMD or H parameter values. The assessment of the cumulative dose of CS is difficult and involves some approximation whereas the duration of CS therapy can be measured accurately.

In the present study the fractal parameter H was lower in CS-treated patients than in controls but HRT was likely to constitute a confounding factor. We compared the texture parameter and BMD values in subgroups having received no HRT. The results showed that the discrimination of H parameter and BMD values remained similarly significant or slightly increased in comparison with the whole population (Tables 2 and 3). For instance, CS therapy was associated with OR = 1.67 (CI, 1.10–2.54) for a decrease of 1 SD of the H mean parameter in the whole population versus OR = 1.85 (CI, 1.08–3.17) in the subgroups without HRT (Tables 2 and 3). This influence of HRT was more marked on BMD than on H parameter; at the FN and LS we observed OR of decreasing BMD of 1 SD by CS, respectively, at 1.95 (CI, 1.29–2.97) and 3.19 (CI, 1.80–5.66) in the whole population versus 2.83 (CI, 1.58–2.83) and 4.92 (CI, 2.12–11.40) in the subgroups without HRT (Tables 2 and 3).

When comparing CS-treated patients with fractures to those without fractures, the H parameter was not discriminant (Table 4). Both LS and FNBMD values were significantly lower in patients with fractures (Table 4). Considering that patients with CS treatment and fractures were older than patients without fractures, we have envisaged this difference as being a potential confounding factor. This hypothesis has been excluded (Table 4).

In our study, the influence of CS therapy on the fractal parameter of texture H was just above the statistical significance: P = 0.03. In a previous study in postmenopausal osteoporosis, comparing women with vertebral crush fractures to age-matched controls, the mean difference for H parameter was far more significant (P < 0.0001).(16) In the same study using receiver operating characteristic curves the H parameter had better discriminant value than BMD. Furthermore in a group of patients and controls whose BMD values overlapped, the H parameter was still able to separate crush fracture cases from control cases.(16) Our data showing no significant difference in H value between fracture cases and no fracture cases, the bone texture changes present in CS-treated patients did not seem to explain the bone fragility induced by CS treatment.

The H parameter has shown correlations with trabecular microarchitectural parameters (derived from histomorphometry) and with bone strength evaluation.(29,30) The role of trabecular microarchitecture alterations could be less important in CS-induced fragility than in postmenopausal osteoporosis. This remark is consistent with histomorphometric studies in CS osteoporosis.(10,11) Some authors found in male CIOP a decrease in trabecular thickness (Tb. Th.) whereas connectivity indices like the star volume or the interconnectivity index were not statistically different between patients and controls.(12) When bone volume/tissue volume was higher than 11%, there were mild changes in these indices.(12) In this study, microarchitectural alterations were mainly observed when Tb. Th. was ≤ 70 μm. In CIOP, Aaron et al. also found a predominant trabecular thinning process, contrasting with a predominant trabecular number decline in idiopathic osteoporosis.(10) Dempster et al. also considered that CS-induced bone loss mechanism was primarily a trabecular thinning without loss of complete trabeculae and linked these data to the depressive effect of CS therapy on osteoblastic function.(11) Thus, these studies concluded that trabecular microarchitecture alterations were present but that a simple trabecular thinning was predominant, in accordance with our data showing significant but mild texture changes.

However, one should be cautious to not overinterpret our findings. Our study design may not allow accurate evaluation of the respective role of the CS therapy and the disease on bone loss and bone microarchitecture changes. Several patients with active inflammatory rheumatic disease, in whom disorders of bone metabolism have been described, have been included in our study. For instance, the disease activity and the functional disability that are associated with rheumatoid arthritis may lead, by themselves, to osteopenia.(31) We do not completely exclude the possibility that underlying disease of the patients may have played a role in bone microarchitecture changes for instance by juxta-articular changes.

In conclusion, our data suggest that the CIOP is characterized by both mineral bone loss and trabecular microarchitecture changes. These microarchitecture changes exist but do not seem to be the predominant factor inducing bone fragility.


The image scanner device is due to the help of the Regional Council of Region Centre (France). We acknowledge Hologic Europe Company for its highly appreciated help.