Calcif Tissue Int (2015) 96:500–509 DOI 10.1007/s00223-015-9980-x

ORIGINAL RESEARCH

Adjusting Fracture Probability by Trabecular Bone Score Eugene V. McCloskey1 • Anders Ode´n1 • Nicholas C. Harvey2 • William D. Leslie3 Didier Hans4 • Helena Johansson1 • John A. Kanis1

•

Received: 30 October 2014 / Accepted: 3 March 2015 / Published online: 22 March 2015 Ó Springer Science+Business Media New York 2015

Abstract The aim of the present study was to determine the impact of trabecular bone score on the probability of fracture above that provided by the clinical risk factors utilized in FRAX. We performed a retrospective cohort study of 33,352 women aged 40–99 years from the province of Manitoba, Canada, with baseline measurements of lumbar spine trabecular bone score (TBS) and FRAX risk variables. The analysis was cohort-specific rather than based on the Canadian version of FRAX. The associations between trabecular bone score, the FRAX risk factors and the risk of fracture or death were examined using an extension of the Poisson regression model and used to calculate 10-year probabilities of fracture with and without TBS and to derive an algorithm to adjust fracture probability to take account of the independent contribution of TBS to fracture and mortality risk. During a mean followup of 4.7 years, 1754 women died and 1639 sustained one or more major osteoporotic fractures excluding hip fracture and 306 women sustained one or more hip fracture. When fully adjusted for FRAX risk variables, TBS remained a statistically significant predictor of major osteoporotic fractures excluding hip fracture (HR/SD 1.18, 95 % CI 1.12–1.24), death (HR/SD 1.20, 95 % CI 1.14–1.26) and & John A. Kanis [email protected] 1

Centre for Metabolic Bone Diseases, University of Sheffield Medical School, University of Sheffield, Beech Hill Road, Sheffield S10 2RX, UK

2

MRC Lifecourse Epidemiology Unit, University of Southampton, Southampton, UK

3

University of Manitoba, Winnipeg, Canada

4

Center of Bone Diseases, Lausanne University Hospital, Lausanne, Switzerland

123

hip fracture (HR/SD 1.23, 95 % CI 1.09–1.38). Models adjusting major osteoporotic fracture and hip fracture probability were derived, accounting for age and trabecular bone score with death considered as a competing event. Lumbar spine texture analysis using TBS is a risk factor for osteoporotic fracture and a risk factor for death. The predictive ability of TBS is independent of FRAX clinical risk factors and femoral neck BMD. Adjustment of fracture probability to take account of the independent contribution of TBS to fracture and mortality risk requires validation in independent cohorts. Keywords Epidemiology  Fracture probability  FRAX  Osteoporosis  Trabecular bone score

Introduction The measurement of bone mineral density (BMD) by dual X-ray absorptiometry (DXA) remains the cornerstone for the assessment of skeletal strength and fracture risk in the clinical setting [1]. The wide recognition that BMD has limitations in terms of capturing other skeletal determinants of bone strength (for example, bone structure, geometry, cortical and trabecular architecture, and bone turnover), and has led to many attempts to provide supplementary information to improve the predictive ability of skeletal assessments. Such approaches include indices such as hip axis length, hip structural analysis including buckling ratios, cortical thickness, cortical porosity, and more complex techniques such as finite element analysis [2–9]. The gain from such developments has been disappointing and none has been incorporated into routine clinical practice. More recently, a novel gray-scale textural analysis of antero-posterior lumbar spine DXA images has produced

E. V. McCloskey et al.: Fracture Probability and Trabecular Bone Score

an index, the trabecular bone score (TBS), that shows promise as a BMD-independent predictor of skeletal strength [10]. TBS appears to be an index of bone microarchitecture that provides skeletal information additional to the standard BMD measurement. Higher values of TBS correlate with better skeletal microstructure. A recent literature review determined that TBS consistently discriminated between fracture cases and non-fracture controls in both cross-sectional and longitudinal studies, and that the discrimination is complementary to BMD derived from the same images [10]. TBS can therefore be regarded as another potential clinical risk factor for fracture and, as such, its ability to predict fracture independently of other well-established risk factors needs to be determined. Risk factors that are partly independent of both age and BMD have been incorporated into FRAXÒ. FRAX estimates the 10-year probability of hip and major osteoporotic fracture based on the individual’s risk factor profile [11, 12], and is now the leading risk assessment model used worldwide [13]. The probability of fracture is calculated from age, body mass index (BMI), a number of dichotomised risk variables, and additionally femoral neck BMD. The aim of the present analysis was to determine whether TBS could predict fracture risk independently of the risk factors incorporated in FRAX, and if so, to derive adjustment algorithms to take account of the independent contribution of TBS to fracture risk.

501

Measurement of Bone Mineral Density (BMD) All lumbar spine and proximal femoral DXA scans were performed using Prodigy scanners (GE-Healthcare, Madison, WI, USA) and analyzed (enCore Software 12.4, GE-Healthcare, Madison, WI, USA) in accordance with manufacturer recommendations. Femoral neck T scores and Z scores were calculated using the NHANES III white female reference values [20]. Instruments were crosscalibrated using anthropomorphic phantoms. All three instruments used for this study exhibited stable long-term performance [coefficient of variation (CV) \0.5 %]. Other Clinical Risk Factors for Fracture

Materials and Methods

Data sources previously described [16] were accessed for the presence of specific FRAX variables required for the calculation of fracture probability and included prior fracture (hip, clinical spine, forearm, and humerus fractures), a diagnosis of rheumatoid arthritis, chronic obstructive pulmonary disease (COPD; a proxy for smoking), alcohol, or other substance abuse (a proxy for high alcohol intake), and[90 days of systemic glucocorticoid use in the last year. Secondary osteoporosis was operationalized as previously diagnosed hyperthyroidism or diabetes. Selfreported parental hip fracture information was collected at the time of BMD testing and was available from 2005 onwards, but not in the earlier years. Anthropomorphic data (height and weight) were measured at the time of DXA and body mass index (BMI) was calculated.

Patient Population

Measurement of Trabecular Bone Score (TBS)

We used a large population-based cohort of women from the Canadian province of Manitoba referred to the Manitoba Bone Density Program for assessment. The Program is a targeted case-finding clinical program with a database of all testing results that has been shown to exceed 99 % in terms of completeness and accuracy [14–16]. All women aged 40–100 years who had undergone BMD measurement of the spine and hip by DXA were eligible for inclusion providing they were Manitoba residents with medical coverage during the observation period. We excluded non-residents and men. For women with more than one eligible set of BMD measurements, only the first record available was included in analysis. Women on osteoporosis treatment were not excluded as previous work had shown that this does not interfere with the use of FRAX for osteoporotic fracture prediction [17]. Fracture outcomes were based upon non-traumatic vertebral (clinical), hip, wrist, and humeral fracture codes recorded up to March 31, 2008 using previously validated algorithms [18, 19].

TBS measurements were performed in the Bone Disease Center at the Lausanne University Hospital (CHUV), Lausanne, Switzerland (TBS iNsightÒ Software version 2.1, Med-Imaps, Pessac, France), using anonymized spine DXA files from the Manitoba database to ensure blinding of the Swiss investigators to all clinical parameters and outcomes. The average short-term reproducibility (CV) for TBS calculated from all three instruments and technicians used for this study was 2.1 % in 92 individuals with repeat spine DXA scans performed within 28 days (51 same day, 41 different day) [21]. Analytic Approach and Statistics The analytic framework was based on the construct of FRAX but used cohort-specific data rather than the Canadian version of FRAX. FRAX is a computer-based algorithm (http://www.shef.ac.uk/FRAX) that provides models for the assessment of fracture probability in men and women [11, 12, 22, 23]. The approach uses easily obtained

123

502

clinical risk factors to estimate 10-year fracture probability. The estimate can be used alone or with BMD to enhance fracture risk prediction. In addition to fracture risk, FRAX uses an extension of the Poisson regression model to derive hazard functions of death. These hazard functions are continuous as a function of time which permits the calculation of the 10-year probability of hip, clinical spine, humerus or wrist fracture, and the 10-year probability of hip fracture. In FRAX, two fracture outcomes are provided comprising the 10-year probability of hip fracture and the 10-year probability of a major osteoporotic fracture (hip, clinical vertebral, humerus, and forearm). The FRAX model itself is based, however, on two probability outcomes namely hip fracture on one hand and major osteoporotic fracture (clinical vertebral, humerus, and forearm) without hip fracture (OWH fracture) on the other. The reason that the FRAX models are built with the two models is because several of the risk factors have different weights for the two fracture outcomes [22]. In the present study, the 10-year probability of fracture was estimated for each subject using the previously defined variables and femoral neck BMD to generate a Manitobaspecific model (i.e., not the Canadian FRAX model). In line with the construct of FRAX, this was derived from the hazard function of hip fracture, major osteoporotic fracture without hip fracture (OWH fracture), and death (i.e., three hazard functions). The models were integrated to provide a FRAX-like probability model for a major osteoporotic fracture and for hip fracture alone [23]. The next step was to create similar models that incorporated TBS as a risk variable. Thus, six hazard functions were computed, two for OWH fracture with and without TBS, two for hip fracture with and without TBS, and two for death with and without TBS. The hazard functions were modeled using an extension of the Poisson regression model. From the six hazard functions, the 10-year probability of fracture was calculated without and with TBS [23]. For each hazard function, interactions between potential risk variables and age, time since baseline and the variable itself (for continuous variables) were tested for each endpoint. When the interaction between the risk variable itself was significant, a piecewise linear function with a knot at the mean was constructed to determine whether the predictive effect was the same over the whole range of the variable. The associations between risk variables and fracture or death were described as the hazard ratio (HR) for per 1 unit or 1 SD change together with 95 % confidence intervals (CI). The final variables and interactions included in the hazard functions were selected in a multivariate Poisson model using the forward selection method based on p value thresholds of 0.05.

123

E. V. McCloskey et al.: Fracture Probability and Trabecular Bone Score

Linear regression was used to study the association between 10-year probability calculated with TBS on one hand, and 10-year probability calculated without TBS, age, and the interaction between age and TBS on the other. Probabilities (p) were transformed (-log(100/p - 1) to avoid a regression model that generated probabilities outside the range 0–100, since the result of the regression model (Z) then is transformed as 100/(1 ? exp(-Z)) (Appendix 1). In an ancillary analysis, we examined the relationship between TBS and lumbar spine BMD.

Results Characteristics of Participants During an average 4.7 years of follow-up (range 0–18 years), 1754 (5.3 %) women died, 1639 (4.9 %) women had one or more osteoporotic fracture without hip fracture (OWH fracture), and 306 (0.9 %) women sustained one or more hip fracture (with or without OWH). Table 1 shows the baseline characteristics of the study population in those with and without incident fractures and/ or mortality. TBS was significantly lower in women who sustained a fracture or died than in those that did not (Tables 1, 2). Women who sustained a fracture or died were also older, had lower femoral neck BMD T score, and a higher prevalence of prior fractures, smoking, alcohol exposure, glucocorticoid use, rheumatoid arthritis, and secondary osteoporosis. There was no statistically significant difference between the groups for family history of hip fracture. Women who sustained a fracture had also a lower BMI than women without incident fracture (p = 0.020 for OWH fractures and p \ 0.001 for hip fractures). Women who died during follow-up had a lower BMI than those women who did not (p = 0.041). Risk Factors, TBS, and Fracture Outcomes In Poisson regression analysis, adjusted for age and time since baseline, TBS and all of the risk factors except family history of hip fracture were significantly associated with OWH fractures, hip fractures, and mortality (Table 2). In subsequent analyses, family history of hip fracture was excluded. For TBS, a 1 SD decrease was associated with a 35 % increase in OWH, a 48 % increase in hip fracture, and a 32 % increase in mortality. These associations with TBS were attenuated but remained significant following adjustment for femoral neck BMD alone, the clinical risk factors alone and both combined (Table 3). For example, when adjusted for BMD and the risk factors, each SD decrease in TBS was associated with an 18 % increase in

E. V. McCloskey et al.: Fracture Probability and Trabecular Bone Score

503

Table 1 Baseline characteristics of 33,352 women according to whether they sustained an incident osteoporotic fracture without hip fracture (OWH fracture), hip fracture, or death during follow-up Variable

No fracture or death during follow-up (n = 29,962) Mean ± SD

Age (years)

OWH fracture during followup (n = 1639) Mean ± SD

Hip fracture during followup (n = 306) Mean ± SD

62.5 ± 10.5

67.3 ± 11.0

2

26.7 ± 5.3 -1.14 ± 1.07

26.2 ± 5.0 -1.75 ± 1.04

24.1 ± 4.4 -2.38 ± 0.97

26.0 ± 5.4 -1.78 ± 1.13

TBS L1–L4

1.33 ± 0.12

1.27 ± 0.12

1.23 ± 0.13

1.25 ± 0.13

BMI (kg/m ) Femoral neck BMD T score

74.4 ± 9.5

Death during follow-up (n = 1754) Mean ± SD

% (n)a

71.9 ± 10.9

% (n)b

% (n)b

% (n)c

6 (29,962) 6 (29,962)

10 (1639) 10 (1639)

12 (306) 14 (306)

17 (1754) 20 (1754)

Previous osteoporotic fracture

12 (29,962)

30 (1639)

30 (306)

20 (1754)

Parental history of hip fracture

13 (8815)

16 (200)

4 (24)

9 (149)

Glucocorticoid use Smoking proxy

Rheumatoid arthritis Diabetes or hyperthyroidism

3 (29,962)

5 (1639)

6 (306)

5 (1754)

10 (29,962)

12 (1639)

17 (306)

16 (1754)

2 (29,962)

4 (1639)

5 (306)

5 (1754)

Alcohol proxy

The two fracture groups are not mutually exclusive; 73 of the women with hip fracture had also sustained OWH fractures. Furthermore, 236 of the women who died had sustained a prior incident fracture a Percent of total number of women (n) with a positive response b

Percent of total number of women with a fracture (n) with a positive response

c

Percent of total number of women who died (n) with a positive response

Table 2 The association between each potential risk factor and osteoporotic fracture without hip fracture (OWH fracture), hip fracture, and death adjusted for time since baseline and age

Variable

HR per 1 unit (95 % CI) OWH fracture

Hip fracture

Mortality

BMI (kg/m2)

0.94 (0.89, 0.99)a

0.56 (0.48, 0.64)a

0.95 (0.90, 1.00)a

Previous osteoporotic fracture

2.62 (2.35, 2.92)

1.84 (1.43, 2.35)

1.18 (1.05, 1.33)

Parental history of hip fracture Smoking proxy

1.23 (0.84, 1.81) 1.30 (1.10, 1.53)

0.35 (0.05, 2.61) 1.59 (1.15, 2.19)

0.77 (0.44, 1.34) 2.50 (2.22, 2.81)

Glucocorticoid use

1.37 (1.17, 1.62)

1.74 (1.24, 2.44)

2.60 (2.30, 2.95)

Rheumatoid arthritis

1.59 (1.29, 1.97)

1.90 (1.19, 3.02)

1.42 (1.14, 1.76)

Diabetes or hyperthyroidism

1.23 (1.06, 1.43)

1.72 (1.27, 2.33)

1.73 (1.53, 1.97)

Alcohol proxy

1.83 (1.42, 2.35)

3.05 (1.78, 5.23)

2.73 (2.19, 3.41)

Femoral neck BMD T score

1.60 (1.51, 1.69)a

2.58 (2.22, 2.99)a

1.23 (1.16, 1.30)a

TBS L1–L4

1.35 (1.29, 1.42)a

1.48 (1.33, 1.66)a

1.32 (1.26, 1.39)a

a

HR per SD

OWH, a 23 % increase in hip fracture and a 20 % increase in mortality. Hazard Functions in Multivariate Models The variables and interactions included in the final hazard functions for OWH fracture, hip fracture, and mortality were selected in multivariate models using the forward selection method based on p values and are shown in Tables 4, 5, and 6. Models were constructed with and without the inclusion of TBS.

OWH Fracture Glucocorticoid use and the smoking proxy were not statistically significant independent predictors of OWH in this model (p = 0.12 and p = 0.051, respectively). As expected, the HRs for most risk variables included were attenuated in the multivariate model (Table 4) than when simply adjusted for just age (see Table 2). In general, there was further attenuation when TBS was included in the model. Importantly, the predictive ability of prior fracture waned with both age (p = 0.016) and time since baseline

123

504

E. V. McCloskey et al.: Fracture Probability and Trabecular Bone Score

Table 3 The association between TBS and osteoporotic fracture without hip fracture (OWH fracture), hip fracture, and death adjusted for time since baseline and age and additionally adjusted for clinical risk factors (CRFs) and BMD TBS adjusted for

OWH fracture HR per 1 SD (95 % CI)

Hip fracture HR per 1 SD (95 % CI)

Mortality HR per 1 SD (95 % CI)

Time since baseline and age

1.35 (1.29–1.42)

1.48 (1.33–1.66)

1.32 (1.26–1.39) 1.23 (1.17–1.29)

a

1.27 (1.20–1.33)

1.40 (1.25–1.57)

BMD

1.25 (1.18–1.31)

1.26 (1.12–1.42)

1.29 (1.23–1.35)

CRFs ? BMD

1.18 (1.12–1.24)

1.23 (1.09–1.38)

1.20 (1.14–1.26)

CRFs

a

BMI, Previous fracture, smoking, glucocorticoids, RA, secondary osteoporosis and alcohol use

Table 4 HR for the hazard function for significant predictive variables and interactions for osteoporotic fracture without hip fracture (OWH fracture) in models with and without TBS included

Without TBS HR per 1 unit (95 % CI)

With TBS HR per 1 unit (95 % CI)

BMI (kg/m2)

1.05 (1.00–1.11)a

1.06 (1.01–1.12)a

Rheumatoid arthritis

1.45 (1.17–1.80)

1.43 (1.15–1.77)

Secondary OP proxy

1.27 (1.09–1.48)

1.21 (1.04–1.41)

Alcohol proxy

1.64 (1.27–2.10)

1.55 (1.20–1.99)

Previous osteoporotic fracture 50 years at baseline

4.31 (3.31–5.61)

4.07 (3.12–5.30)

90 years 6 years from baselinec

1.09 (0.82–1.45)

1.08 (0.82–1.43)

Femoral neck BMD T score BMD \ -1.20b

1.63 (1.49–1.78)a

1.54 (1.41–1.69)a

b

a

1.29 (1.14–1.46)a

BMD [ -1.20

1.36 (1.20–1.54)

TBS TBS \ 1.30b TBS [ 1.30

1.08 (1.00–1.17)a

b

1.38 (1.23–1.56)a

a

HR per 1 SD

b

piecewise linear function with knot at the mean

c

The example is provided since there are interactions of TBS both with previous fracture 9 age and previous fracture 9 time since baseline

(p \ 0.001) (Table 4). In addition, the HR per 1 SD for BMD was somewhat higher below the mean of BMD than above the mean (e.g., 1.6, 95 % CI 1.5–1.8 vs. 1.4, 95 % CI 1.2–1.6 in the model without TBS) (Table 4). The HR per 1 SD for TBS was somewhat higher above the mean of TBS (1.4, 95 % CI 1.2–1.6) than below the mean (1.1, 95 % CI 1.0–1.2; p = 0.044) (Table 4). Hip Fracture The hazard function for hip fracture excluded the smoking proxy and rheumatoid arthritis (p = 0.19 and p = 0.28, respectively) (Table 5). Again, the predictive ability of prior fracture waned with age (p = 0.0071) but, unlike for OWH, did not wane with time since baseline. In contrast to OWH, the HR per 1 SD of BMD for hip fracture was similar above and below the mean values but the HR waned with time since baseline (p = 0.0068). Finally, the predictive ability of TBS also waned with age (p = 0.0011) such that differences in TBS no longer contributed to fracture risk from the age of 88 years.

123

Mortality The hazard function of death excluded BMI, prior fracture, and rheumatoid arthritis (Table 6) (p = 0.23, p = 0.11 and p [ 0.30, respectively). The HR per 1 SD for BMD was higher below the mean of BMD than above the mean. The HR per 1 SD for TBS was 1.18 (95 % CI 1.13–1.24). TBS and Probability of Major Osteoporotic Fractures Hazard functions in Tables 4, 5, and 6 were used to calculate 10-year probabilities of major osteoporotic fractures with and without the use of TBS. The potential impact of TBS on the calculated osteoporotic fracture probabilities across different ages is shown in Fig. 1. For example, in a woman aged 50 years with a BMI of 27 kg/m2 and femoral neck BMD T score of -2SD but no other risk factors, the 10-year probability was 8.4 %. If TBS were found to lie at the 10th percentile for the current study population (equivalent to a TBS of 1.16) and was included in the risk

E. V. McCloskey et al.: Fracture Probability and Trabecular Bone Score Table 5 HR for the hazard function for significant predictive variables and interactions for hip fracture in models with and without TBS included

505

Without TBS HR per 1 unit (95 % CI)

With TBS HR per 1 unit (95 % CI)

0.69 (0.55–0.87)a

0.67 (0.53–0.84)a

BMI (kg/m2) BMI \ 26.6b

a

0.7 (0.56–1.05)a

BMI [ 26.6

0.79 (0.59–1.08)

Glucocorticoids

1.65 (1.17–2.32)

1.61 (1.14–2.26)

Secondary OP proxy

2.10 (1.55–2.84)

1.97 (1.46–2.68)

Alcohol proxy

2.44 (1.42–4.19)

2.23 (1.29–3.84)

Previous osteoporotic fracture 50 years

3.93 (1.82–8.47)

2.96 (1.33–6.60)

90 years

0.93 (0.62–1.40)

1.00 (0.66–1.51)

Femoral neck BMD T score At baseline 6 years from baseline

2.84 (2.29–3.67)a

2.62 (2.01–3.40)a

a

1.59 (1.26–2.01)a

1.71 (1.36–2.15)

TBS

Table 6 HR for the hazard function for significant predictive variables and interactions for death in models with and without TBS included

50 years

2.08 (1.49–2.91)a

90 years

0.94 (0.77–1.15)a

a

HR per 1 SD

b

Piecewise linear function with knot at the mean

Without TBS HR per 1 unit (95 % CI)

With TBS HR per 1 unit (95 % CI)

2.13 (1.88–2.40)

2.01 (1.78–2.27)

50 years

5.34 (4.02–7.10)

5.15 (3.87–6.84)

90 years

1.24 (0.99–1.56)

1.23 (0.98–1.54)

2.60 (1.90–3.58) 1.34 (1.07–1.69)

2.39 (1.73–3.29) 1.33 (1.05–1.67)

Smoking proxy Glucocorticoid

Secondary OP proxy 50 years 90 years Alcohol proxy 50 years at baseline

5.23 (3.55–7.70)

5.08 (3.94–7.50)

90 years

0.69 (0.90–1.21)

0.67 (0.38–1.17)

Femoral neck BMD T score BMD \ -1.20b At baseline 6 years after baseline

1.27 (1.13–1.41)a

1.20 (1.07–1.34)a

a

1.41 (1.28–1.56)a

0.88 (0.78–1.00)a

0.85 (0.75–0.96)a

a

0.99 (0.88–1.13)a

1.49 (1.35–1.63)

b

BMD [ -1.20 At baseline

6 years after baseline

1.03 (0.91–1.17)

1.18 (1.13–1.24)a

TBS a

HR per 1 SD

b

Piecewise linear function with knot at the mean

assessment, the 10-year probability increased from 8.4 to 10.7 %. In contrast, if the TBS value lay at the 90th percentile for the population (equivalent to a TBS of 1.47), the probability decreased to 6.1 %. A similar exercise in a woman at the age of 80 years gave 10-year probabilities of 16.5, 17.6, and 12.9 % for each of the three scenarios, respectively.

As might be expected, the impact of TBS on the potential variability in 10-year probability was somewhat less at lower fracture probabilities e.g., at a higher femoral neck BMD. This is also illustrated in Fig. 1 where the patient characteristics are maintained from the examples above but the femoral neck BMD T score is ?1. For example, in a woman aged 50 years with a BMI of 27 kg/m2 and femoral

123

506

E. V. McCloskey et al.: Fracture Probability and Trabecular Bone Score

neck BMD T score of ?1 but no other risk factors, the 10-year probability was estimated to be 3.2 %. If TBS were subsequently found to lie at the 10th percentile or the 90th percentile, as above, the 10-year probability would be 4.7 and 2.7 %, respectively. In a woman at the age of 80 years, the same scenarios would have resulted in 10-year probabilities of 6.4, 7.8, and 5.5 %, respectively. The impact of the interactions between probability and TBS (p \ 0.001) is shown in Fig. 2 with scenarios at the age of 50 and 80 years. The adjustment of fracture probability at low (or high) values for TBS was less marked at the age of 80 than at the age of 50 years. The adjustment was less marked, the lower the unadjusted fracture probability.

10-year probability (%) 40

Age 50 years and probability without TBS 21.0% Age 80 years and probability without TBS 4.6%

30

Age 80 years and probability without TBS 21.0%

20

10

0

TBS and Probability of Hip Fractures

Age 50 years and probability without TBS 4.6%

1.0

1.1

1.2

1.3

1.4

1.5

TBS

The potential impact of TBS on the estimated hip fracture probabilities across different ages showed a similar pattern. For example, in a woman aged 50 years with a BMI of 27 kg/m2 and femoral neck BMD T score of -2SD but no other risk factors, the 10-year probability was estimated to be 0.4 %. If TBS was found to lie at the 10th percentile for the current study population (equivalent to a TBS of 1.16) and was included in the risk assessment, the 10-year probability increased from 0.4 to 0.8 %. Conversely, if the TBS value lay at the 90th percentile for the population (equivalent to a TBS of 1.47), the probability decreased to 0.2 %. A similar calculation in a woman at the age of 80 years gave 10-year probabilities of 4.4, 4.6, and 4.7 %

10-year fracture probability (%) 20 15

T-score -2.0 T-score +1.0

10 5 0

50

60

70

80

Age (years) Fig. 1 10-year probability of a major osteoporotic fracture in women according to age and a T score for femoral neck BMD set at -2 or ?1 SD in the absence of clinical risk factors. The symbols denote probabilities calculated without TBS and the lines the range of probabilities with high and low values for TBS using the model incorporating TBS [TBS values set at the 10th percentile (TBS = 1.16) and the 90th percentile (TBS = 1.47)]. BMI is set to 27 kg/m2

123

Fig. 2 The relationship between 10-year probability of a major osteoporotic fracture and trabecular bone score (TBS) at the ages of 50 and 80 years. Fracture probabilities without TBS are set at 4.7 and 21.0 % (10th and 90th percentiles) indicated by the dashed horizontal lines

for each of the three scenarios, respectively, indicating little or no change in probability over the range of TBS. The mathematical functions to describe the manner whereby fracture probabilities can be adjusted to take account of TBS is given in Appendix 2. Lumbar Spine BMD The ancillary analysis examined the dependence of the impact of TBS on fracture risk and fracture probability on lumbar spine BMD. As expected lumbar spine BMD was a significant predictor of hip fracture with a gradient of risk [GR or HR/SD 1.32 (95 % CI 1.17–1.48)] and of other major fractures (GR 1.51; 95 % CI 1.43–1.60). The predictive value was less than in the case of femoral neck BMD (see Table 2). TBS adjusted for BMD at the lumbar spine remained a significant risk factor for hip fracture (GR 1.42; 95 % CI 1.26–1.59), other major fractures (GR 1.22; 95 % CI 1.16–1.28), and mortality (GR 1.35; 95 % CI 1.29–1.42). These adjusted risks were similar to TBS adjusted for femoral neck BMD (see Table 3). When lumbar spine BMD was substituted for femoral neck BMD, high TBS remained a significant predictor of a major osteoporotic fracture (HR/SD 1.33; 95 % CI 1.18–1.50) and TBS remained a significant predictor of hip fracture (HR/ SD 2.37; 95 % CI 1.73–3.23). When lumbar spine BMD was added to the fully adjusted model of fracture probability with TBS, lumbar spine BMD remained a significant risk factor for fracture probability (GR 1.11; 95 % CI 1.04–1.09).

E. V. McCloskey et al.: Fracture Probability and Trabecular Bone Score

Discussion A previous analysis from the Manitoba cohort [16] confirmed that TBS, derived from DXA scans of the lumbar spine, is an age and BMD-independent risk factor for incident fractures; the same report showed that it was an independent predictor in a fracture probability model and additionally independent of lumbar spine BMD. The present multivariate analysis differs from the earlier report using the more complex structure of FRAX models that are used in clinical practice. The analysis confirms that TBS is a significant fracture risk variable that is independent of the clinical variables included in FRAX. Furthermore, a low TBS is also associated with an increase in mortality, again independently of the other FRAX variables. These results suggest that the addition of TBS to risk assessment is likely to be useful in the risk stratification of patients in clinical practice. A particular feature is that the gradient of risk of TBS for OWH fracture was higher at higher values of TBS and a higher gradient of risk with lower age for the outcome of hip fracture. Conversely, the clinical utility of TBS is likely to be less in patients with the greatest deficits in BMD. If confirmed in independent studies, such characteristics of TBS might be of value for risk assessment in younger women traditionally considered to be at low risk. The interplay between a risk factor, other risk factors, and different fracture outcomes is complex. For example, in multivariate models, femoral neck BMD itself showed diverse relationships with fracture and mortality. For OWH, the BMD showed a stronger relationship at lower baseline values (below the median value), whereas for hip fracture the relationship was constant across the range of BMD values but waned with time; for mortality, the relationship with BMD was both value and time dependent. Similar complexities were observed for the relationship between TBS and outcomes in these multivariate models containing all the significant predictors. For OWH, in contrast to BMD, TBS showed a stronger relationship at higher baseline values (above the median value), whereas for hip fracture the relationship was weaker at older ages; for mortality, the relationship with TBS was stable across values, ages, and time with an 18 % increase in death risk for each 1SD decrease in TBS. The complexity of combining multiple risk factors into FRAX-like probability algorithms and their interactions are clearly illustrated by the examples of the relationships between major osteoporotic fracture (shown in Figs. 1, 2) and hip fracture probabilities. As others have concluded, our results suggest a role for combining TBS with FRAX variables in those at intermediate risk where reclassification above or below an intervention threshold is likely. Prior to that, a number of other developmental steps are

507

required before TBS could be adopted more widely in assessing fracture risk. It is important to note that while a FRAX-like approach has been utilized here, it differs from the FRAX algorithms as accessed through the website (www.shef.ac.uk/FRAX). It will be important to determine whether TBS remains a significant independent predictor when the original algorithms are used and to establish whether the observed relationships are maintained across and within other international cohorts. Such an approach has already been undertaken to examine and establish a simple adjustment to the FRAX probabilities for discordances between femoral neck and lumbar spine T scores [24, 25]. The adjustment is of a similar order of magnitude as in the present study, but it is unlikely that the complexities of the combinations reported here will enable the derivation of a simple mathematical adjustment. One possible approach is to undertake an automated adjustment of the FRAX probabilities outside the setting of FRAX itself, as undertaken to adjust FRAX probabilities for glucocorticoid dose [26] in the transition between the UK FRAX calculation tool and the National Osteoporosis Guideline Group website [27]. Several limitations should be acknowledged in the present study. The population studied was a referral cohort rather than a random sample of the population with an increased prevalence of risk factors for osteoporosis, though this is unlikely to affect the relationships between TBS and these risk factors. Additional limitations include incomplete information on parental history of hip fracture and use of proxy variables for smoking and high alcohol intake, although the prevalence of diagnosed COPD and alcohol/substance abuse in this population was close to that for current smoking and high alcohol intake seen in the population-based Canadian Multicentre Osteoporosis Study [28]. The present study population was largely Caucasian, and our results may not be applicable to other ethnic groups. Similarly, our findings should not be applied to men as the TBS–fracture risk relationship has not yet been well characterized. Additionally, it is assumed, but not proven, that low lumbar spine TBS identifies a fracture risk that is responsive to pharmacologic intervention. Available data indicate that lumbar spine TBS increases following treatment for osteoporosis, although the increment is less than that seen with BMD [29, 30]. A final limitation of this study is that the TBS adjustment algorithms were derived from a single cohort and require validation in different settings. In conclusion, we have demonstrated that lumbar spine texture analysis using TBS is a risk factor for osteoporotic fracture and also for risk of death. The predictive ability of TBS is independent of FRAX clinical risk factors and femoral neck BMD; finally we have derived an algorithm

123

508

to adjust probability of fracture from clinical risk factors and BMD to account for TBS, an approach that will require validation in further independent cohorts. Acknowledgments The authors acknowledge the Manitoba Centre for Health Policy (MCHP) for use of data contained in the Population Health Research Data Repository (HIPC Project Number 2012/201318). The results and conclusions are those of the authors, and no official endorsement by the MCHP, Manitoba Health, or other data providers is intended or should be inferred. The results and conclusions are those of the authors, and no official endorsement by Manitoba Health is intended or should be inferred. This article has been reviewed and approved by the members of the Manitoba Bone Density Program Committee. Conflict of interest E McCloskey: Nothing to declare for FRAX and the context of this paper, but numerous ad hoc consultancies/ speaking honoraria and/or research funding from Amgen, Bayer, General Electric, GSK, Hologic, Lilly, Merck Research Labs, Novartis, Novo Nordisk, Nycomed, Ono, Pfizer, ProStrakan, Roche, Sanofi-Aventis, Servier, Tethys, UBS and Warner-Chilcott. H. Johansson and A. Oden declare that they have no conflict of interest. N. C. Harvey: Nothing to declare for FRAX and the context of this paper, but consultancy, lecture fees and honoraria from Alliance for Better Bone Health, AMGEN, MSD, Eli Lilly, Servier, Shire, Consilient Healthcare and Internis Pharma. W. D. Leslie: Speaker bureau: Amgen, Eli Lilly, Novartis. Research grants: Amgen, Genzyme. Didier Hans: Co-ownership in the TBS patent. Stock options or royalties: Med-Imaps. Research grants: Amgen, Eli Lilly. JA. Kanis: Nothing to declare for FRAX and the context of this paper, but numerous ad hoc consultancies for: Industry: Abiogen, Italy; Amgen, USA, Switzerland and Belgium; Bayer, Germany; Besins-Iscovesco, France; Biosintetica, Brazil; Boehringer Ingelheim, UK; Celtrix, USA; D3A, France; Gador, Argentina; General Electric, USA; GSK, UK, USA; Hologic, Belgium and USA; Kissei, Japan; Leiras, Finland; Leo Pharma, Denmark; Lilly, USA, Canada, Japan, Australia and UK; Merck Research Labs, USA; Merlin Ventures, UK; MRL, China; Novartis, Switzerland and USA; Novo Nordisk, Denmark; Nycomed, Norway; Ono, UK and Japan; Parke-Davis, USA; Pfizer USA; Pharmexa, Denmark; Roche, Germany, Australia, Switzerland, USA; Rotta Research, Italy; Sanofi-Aventis, USA; Schering, Germany and Finland; Servier, France and UK; Shire, UK; Solvay, France and Germany; Strathmann, Germany; Tethys, USA; Teijin, Japan; Teva, Israel; UBS, Belgium; Unigene, USA; Warburg-Pincus, UK; Warner-Lambert, USA; Wyeth, USA Governmental and NGOs: National Institute for health and clinical Excellence (NICE), UK; International Osteoporosis Foundation; INSERM, France; Ministry of Public Health, China; Ministry of Health, Australia; National Osteoporosis Society (UK); WHO. Human and Animal Rights and Informed Consent The study was approved by the Research Ethics Board for the University of Manitoba and the Health Information Privacy Committee of Manitoba Health.

Appendix 1 The following method estimating the difference between probability of fracture with and without TBS avoids probability values outside the range 0–100. The inverse of the logistic function was used for each probability, i.e., -log(100/p - 1), where p is the 10-year probability in %.

123

E. V. McCloskey et al.: Fracture Probability and Trabecular Bone Score

Let us call this quantity for W1 when calculated for the 10-year probability including TBS and W0 for the probability without TBS. Now W1 will be the dependent variable of a multivariable regression analysis. Independent variables are age, TBS, ageTBS, and W0. We will find an estimated mean of W1 as a function of TBS, age, and W0. Let us call this estimated mean Z(W1, TBS, age) or shortly Z. Now we calculate 100/(1 ? exp(-Z)), which is always a number in the interval 0–100. The number is our estimate of the 10-year probability including TBS. Again we can replace W0 for the corresponding number calculated (by the inverse of the logistic function) from the FRAX probability.

Appendix 2 The adjustment of fracture probabilities according to TBS is given for the 10-year probabilities of hip fracture and major osteoporotic fracture are given below Outcome: Hip fracture 100 The 10-year probability calculated with TBS is 1þe w ; where W = 15.420 - 12.627 9 TBS - 0.194 9 age ? 0.157 9 TBS 9 age ? 0.920 9 L, L = -ln(100/p - 1), p is the 10-year probability calculated without TBS

Outcome: Major Osteoporotic Fracture 100 The 10-year probability calculated with TBS is 1þe w ; where W = 5.340 - 4.213 9 TBS - 0.0521 9 age ? 0.0393 9 TBS 9 age ? 0.897 9 L, L = -ln(100/p - 1), p is the 10-year probability calculated without TBS

References 1. World Health Organization (1994) Assessment of fracture risk and its application to screening for postmenopausal osteoporosis. Report of a WHO Study Group. World Health Organ Tech Rep Ser 843:1–129 2. Bala Y, Zebaze R, Ghasem-Zadeh A et al (2014) Cortical porosity identifies women with osteopenia at increased risk for forearm fractures. J Bone Miner Res 26:1356–1362 3. Diez-Perez A, Gonzalez-Macias J, Marin F et al (2007) Prediction of absolute risk of non-spinal fractures using clinical risk factors and heel quantitative ultrasound. Osteoporos Int 18:629–639 4. Faulkner KG, Cummings SR, Black D, Palermo L, Gluer CC, Genant HK (1993) Simple measurement of femoral geometry predicts hip fracture: the study of osteoporotic fractures. J Bone Miner Res 8:1211–1217 5. Gluer CC (2007) Quantitative Ultrasound—it is time to focus research efforts. Bone 40:9–13 6. Guerri-Fernandez RC, Nogues X, Quesada Gomez JM et al (2013) Microindentation for in vivo measurement of bone tissue

E. V. McCloskey et al.: Fracture Probability and Trabecular Bone Score

7.

8.

9.

10.

11.

12.

13.

14.

15.

16.

17.

18.

19.

material properties in atypical femoral fracture patients and controls. J Bone Miner Res 28:162–168 Yang L, Peel N, Clowes JA, McCloskey EV, Eastell R (2009) Use of DXA-based structural engineering models of the proximal femur to discriminate hip fracture. J Bone Miner Res 24:33–42 Keaveny TM, Kopperdahl D, Melton L et al (2010) Age-dependence of femoral strength in white women and men. J Bone Miner Res 25:994–1001 Randall C, Bridges D, Guerri R et al (2013) Applications of a new handheld reference point indentation instrument measuring bone material strength. J Med Device 7:410051–410056 Silva BC, Leslie WD, Resch H et al (2014) Trabecular bone score: a noninvasive analytical method based upon the DXA image. J Bone Miner Res 29:518–530 Kanis JA on behalf of the World Health Organization Scientific Group. 2007 Assessment of osteoporosis at the primary healthcare level. Technical Report. World Health Organization Collaborating Centre for Metabolic Bone Diseases, University of Sheffield, UK Kanis JA, Johnell O, Oden A, Johansson H, McCloskey E (2008) FRAX and the assessment of fracture probability in men and women from the UK. Osteoporos Int 19:385–397 Kanis JA, Johansson H, Oden A, Cooper C, McCloskey EV, Epidemiology and Quality of Life Working Group of IOF (2013) Worldwide uptake of FRAX. Arch Osteoporos 8:166 Leslie WD, Metge C (2003) Establishing a regional bone density program: lessons from the Manitoba experience. J Clin Densitom 6:275–282 Leslie WD, Caetano PA, Macwilliam LR, Finlayson GS (2005) Construction and validation of a population-based bone densitometry database. J Clin Densitom 8:25–30 Leslie WD, Johansson H, Kanis JA et al (2014) Lumbar spine texture enhances ten-year fracture probability assessment. Osteoporos Int 25:2271–2277 Leslie WD, Lix LM, Johansson H, Oden A, McCloskey E, Kanis JA (2012) Does osteoporosis therapy invalidate FRAX for fracture prediction? J Bone Miner Res 27:1243–1251 Leslie WD, Tsang JF, Caetano PA, Lix LM (2007) Effectiveness of bone density measurement for predicting osteoporotic fractures in clinical practice. J Clin Endocrinol Metab 92:77–81 Lix LM, Azimaee M, Osman BA et al (2012) Osteoporosis-related fracture case definitions for population-based administrative data. BMC Public Health 12:301

509 20. Looker AC, Wahner HW, Dunn WL et al (1998) Updated data on proximal femur bone mineral levels of US adults. Osteoporos Int 8:468–489 21. Hans D, Goertzen AL, Krieg MA, Leslie WD (2011) Bone microarchitecture assessed by TBS predicts osteoporotic fractures independent of bone density: the Manitoba study. J Bone Miner Res 26:2762–2769 22. Kanis JA, Oden A, Johnell O et al (2007) The use of clinical risk factors enhances the performance of BMD in the prediction of hip and osteoporotic fractures in men and women. Osteoporos Int 18:1033–1046 23. Kanis JA, McCloskey EV, Johansson H, etal. 2008 Case finding for the management of osteoporosis with FRAXÒ—Assessment and intervention thresholds for the UK. Osteoporos Int 19: 1395–1408 Erratum 2009 Osteoporos Int 20, 499–502 24. Leslie WD, Lix LM, Johansson H, Oden A, McCloskey E, Kanis JA (2011) Spine-hip discordance and fracture risk assessment: a physician-friendly FRAX enhancement. Osteoporos Int 22:839–847 25. Johansson H, Kanis JA, Ode´n A et al (2014) Impact of femoral neck and lumbar spine BMD discordances on FRAX probabilities in women: a meta-analysis of international cohorts. Calcif Tissue Int 95:428–435 26. Kanis JA, Johansson H, Oden A, McCloskey EV (2011) Guidance for the adjustment of FRAX according to the dose of glucocorticoids. Osteoporos Int 22:809–816 27. National Osteoporosis Guideline Group. http://www.shef.ac. uk/NOGG/result-nobmd.html?age=65&fracture1=14&glucocorti coids=1. Accessed 7 Oct 2014 28. Leslie WD, Lix LM, Johansson H et al (2010) Independent clinical validation of a Canadian FRAX tool: fracture prediction and model calibration. J Bone Miner Res 25:2350–2358 29. Krieg MA, Aubry-Rozier B, Hans D et al (2013) Effects of antiresorptive agents on trabecular bone score (TBS) in older women. Osteoporos Int 24:1073–1078 30. Popp AW, Guler S, Lamy O et al (2013) Effects of zoledronate versus placebo on spine bone mineral density and microarchitecture assessed by the trabecular bone score in postmenopausal women with osteoporosis: a three-year study. J Bone Miner Res 28:449–454

123