Osteoporosis and its consequent increase in fracture risk are major health concerns for postmenopausal women and older men. The diagnosis of osteoporosis is based on bone mineral density (BMD) because a low BMD has been shown in epidemiologic studies to be a strong predictor of osteoporotic fracture. There is, however, a wide overlap of BMD values between fracture cases and controls because of the multiple determinants of skeletal fragility.1 The incorporation of the non-BMD risk factors has been demonstrated to improve the accuracy of fracture risk prediction.2, 3 Then BMD is recognized to be one among other factors to predict fracture risk.
Until recently, the assessment of fracture risk has been based largely on the relative risk (RR) measure, which is a population measure of risk that is clinically useless in individuals. For this reason, there has been interest in the development of algorithms that express absolute risk, or the probability of fracture within a given time period.4–7 The World Health Organization (WHO) Collaborating Center for Metabolic Bone Diseases conducted several metaanalyses of data from large-scale prospective studies in various countries to identify and quantify the risk of fracture associated with several clinical factors (personal history of fracture8, 9 or parental history of fracture,10 body mass index,11 smoking,12 and glucocorticoid treatment13) independent from BMD and age. These extensive studies have contributed to the development of a standardized method—aimed at use in clinical practice—for expressing absolute fracture risk in men and women. Thus a risk assessment tool—FRAX—has been established from nine population-based studies.14, 15 The FRAX tool estimates an individual's probability of both hip fracture and a major osteoporosis-related fracture (hip, clinical spine, shoulder, or wrist) in the next 10 years from clinical risk factors, with or without BMD measurement. Several clinical risk factors (eg, prior fragility fracture, parental history of hip fracture, current smoking, use of systemic corticosteroids, excess alcohol intake, low body mass index, rheumatoid arthritis, and other causes of secondary osteoporosis) are used in addition to age and sex.
The FRAX tool has been validated in several cohorts from various countries, including Epidémiologie des Ostéoporoses (EPIDOS) from France.2 The EPIDOS study was a population-based cohort study of 7500 women aged 75 years or more (mean age 81 years). To enhance FRAX tool applicability in the French population, we reasoned that it would be clinically useful to assess the predictive value of the FRAX tool in a younger cohort with well-characterized risk factors and a known incidence of fracture over 10 years. The Os des Femmes de Lyon (OFELY) study is a population-based cohort study of 1039 women aged 31 years or more recruited at the same time as the EPIDOS cohort and includes 867 women 40 years of age or older (mean age 59 years). Thus we compared the 10-year predicted fracture probabilities estimated with the FRAX tool and the observed fracture incidence in French women from the OFELY cohort during a 10-year follow-up.
Material and Methods
The 867 women aged 40 years and over at the inclusion in the OFELY cohort were analyzed. The OFELY cohort has been described elsewhere.16, 17 Briefly, OFELY is an ongoing prospective study of the determinants of bone loss in 1039 volunteer women, recruited between February 1992 and December 1993, 31 to 89 years of age, randomly selected from the affiliates of a large health insurance company (Mutuelle Générale de l'Education Nationale) from the Rhône District (ie, Lyon and its surroundings in France), with an annual follow-up. Written informed consent was obtained from each woman, and the study was approved by the local ethics committee. All the women aged 40 years or more at the inclusion in the study (n = 867) were analyzed, including 680 postmenopausal women (mean age 62.0 ± 9 years) and 187 premenopausal women (47.2 ± 5 years). This analysis was focused on the prediction of fractures sustained over the first 10 years of the follow-up.
The 10-year predicted probability of fracture was calculated among subjects with the FRAX tool (March 2009) with the variables obtained at baseline to determine the predicted probability of osteoporotic (OP) fractures at four major sites (ie, hip, clinical spine, shoulder, and wrist) and at the hip with and without BMD.
Women completed a questionnaire at the initial screening visit, as described previously,18 including all clinical risk factors used in calculation of the predicted fracture probability with the FRAX tool: parental history of hip fracture, prior fragility fracture, current tobacco smoking, daily consumption of alcohol of more than 2 units, ever long-term use of oral glucocorticoids, rheumatoid arthritis, and other secondary causes of osteoporosis. Weight and height were recorded at the same visit. Prior fragility fractures included wrist, humeral, vertebral, or hip fractures that occurred after the age of 40 years. Only low-trauma fractures (falls from a standing height or less) were recorded.
BMD was measured by dual-energy X-ray absorptiometry (DXA) with a QDR 2000 device (Hologic, Waltham, MA, USA) at the femoral neck and was entered in the tool as a T-score calculated from the National Health and Nutrition Examination Survey (NHANES) III reference values. The in vivo precision error of DXA, expressed as the coefficient of variation (CV), was 1.2% at the femoral neck. A control phantom was scanned every day, and all DXA measurements were performed by the same experienced operator.
Incident fracture evaluation
Incident nonvertebral and clinical vertebral fractures were reported during each annual follow-up. For women who did not come to the clinical center, a letter was sent every year to identify the occurrence of fractures. All fractures were confirmed by radiographs or by a surgical report. Only low-trauma fractures (ie, those occurring as a result of falls from standing height or less) were taken into account, and we excluded fractures of fingers, toes, skull, and face. Vertebral fractures were assessed on lateral X-ray films of the thoracic and lumbar spine obtained at baseline and every 4 years—in women aged 50 years and over at the inclusion in the study—and were identified with the semiquantitative method of Genant by a trained physician. In this analysis, and according to the FRAX tool, we focused on clinical incident vertebral fractures, that is, symptomatic vertebral fractures that came to clinical attention. For prevalent vertebral fractures, both clinical and morphometric vertebral fractures were taken into account.
Chi-squared tests and unpaired t tests were used to compare baseline characteristics between women with and without incident fracture. The observed fracture incidence over 10 years was corrected for mortality and expressed in incidence/1000 person-years. Chi-squared tests were used to compare the predicted fracture probabilities with the observed fracture occurence over 10 years. All women were categorized according to age, below 65 years and at least 65 years, which was the highest quartile of age. Postmenopausal women were categorized into three groups according to their baseline femoral neck BMD values using the WHO criteria: normal (T-score > –1), osteopenia (T-score between –2.5 and –1), and osteoporosis (T-score ≤ –2.5), with T-scores calculated from the NHANES III reference values. Women with low BMD values were those with osteoporosis or osteopenia, that is, with a T-score of –1 or below. The predictive value of the FRAX tool with and without BMD was compared with that of femoral neck BMD and age with a receiver operating curve (ROC). All statistical analyses were performed using SAS statistical analysis software (Version 9.1, SAS Institute, Cary, NC, USA).
During the 10-year follow-up period, 116 women sustained 151 incident clinical fragility fractures at all sites (not counting fingers, toes, skull, and face). Among them, 82 women sustained 95 incident fractures at the four major sites: hip (n = 17), vertebra (n = 25), shoulder (n = 9), and forearm (n = 44). Fifty nonfractured women (mean age 73 ±11 years) died during the 10-year follow-up. Since information about incident fractures was not obtained until 10 years for 16 other women, the latter were excluded from further analysis (mean age 58 ±11 years). After accounting for mortality, the observed incidence of fracture was 9.6 per 1000 person-years for major OP fractures and 2.0 per 1000 person-years for hip fractures.
The characteristics of the 867 women at baseline according to their fracture status at the four major sites after 10 years are shown in Table 1. Women with incident major OP fractures were significantly older, shorter, had a lower BMD, had more often a parental history of hip fracture, and had more prior fractures than with women with no incident fractures. The other clinical risk factors included in the FRAX tool were not significantly different between the groups.
Table 1. Characteristics of Women at Baseline According to Incident Major OP Fracture Status After a 10-Year Follow-up (n = 867)
Information about incident fracture was not obtained until 10 years for 16 women.
Between women who sustained incident major OP fracture and women who did not. Values are mean ± SD unless otherwise specified.
58.8 ± 10.3
67.8 ± 9.6
57.9 ± 10.0
60.2 ± 9.1
58.4 ± 9.0
60.4 ± 9.1
159 ± 6
157 ± 6
159 ± 6
Body mass index (Kg/m2)
23.8 ± 3.5
23.8 ± 3.4
23.8 ± 3.5
Femoral neck BMD (g/cm2)
0.717 ± 0.12
0.625 ± 0.10
0.726 ± 0.11
Femoral neck T-score
−1.2 ± 1.0
−2.0 ± 1.0
−1.1 ± 1.0
Parental history of hip fracture, n (%)
Prior fragility fracture, n (%)
Current smoking, n (%)
Daily consumption of alcohol of >2 units, n (%)
Long-term use of oral glucocorticoids, n (%)
Rheumatoid arthritis, n (%)
Secondary osteoporosis, n (%)
Premature menopause (<45 years), n
Type I Diabetes, n
Chronic malabsorption, n
Chronic liver disease, n
Ten-year predicted fracture probability from FRAX
Among all women, the predicted probabilities calculated without and with BMD were, respectively, 6.6% ± 7.3% and 5.9% ± 6.3% for major OP fracture and 2.4% ± 5.1% and 1.8% ± 4.3% for hip fracture (mean ± SD). They all increased with age (r = 0.7 for major OP fractures, r = 0.7 and 0.6 for hip fractures without and with BMD, respectively, p < .0001). In women aged at least 65 years (n = 229), the mean predicted probabilities were 12.5% and 15% for major OP fractures and 5.4% and 7.6% for hip fractures with and without BMD, respectively. In contrast, in women younger than 65 years (n = 638), the mean predicted fracture probabilities were 3.5%, 3.6%, 0.5%, and 0.6%, respectively (p < .0001). Among postmenopausal women (n = 680), the 10-year predicted probabilities of both major OP and hip fractures were significantly higher in women with osteoporosis (n = 77, 18% and 10%) and osteopenia (n = 390, 6% and 2%) than in women with normal BMD values (n = 208, 3% and <1%, p < .0001). Femoral neck BMD was not available in five postmenopausal women.
Observed fracture incidence over 10 years versus predicted fracture probability
The higher the predicted fracture probability, the higher was the observed fracture incidence. Thus the observed major OP fracture incidence increased significantly with the quartiles of the predicted major OP fracture probability: from 1.9 per 1000 person-years to 24.1 per 1000 person-years and from 2.1 per 1000 person-years to 24.8 per 1000 person-years from the first to the fourth quartile, without and with BMD, respectively (p < .0001). Women who sustained an incident major OP fracture over 10 years had a 2.2 times higher predicted major OP fracture probability compared with women who did not fracture (p < .0001). Women who sustained an incident hip fracture had a 4.9 and 4.6 times higher predicted hip fracture probability compared with women who did not fracture (p < .0001; Fig. 1). Nevertheless, among women who sustained incident major OP fractures, the mean predicted fracture probability was lower than 13%, and among women who had hip fractures, the mean predicted fracture probability was less than 7%. Furthermore, the predicted major OP fracture probabilities with BMD were lower than 9.2% for 50% of women who sustained incident major OP fractures and lower than 24% for 90% of them (Fig. 2).
When the women were categorized according to age, the observed fracture incidence tended to be higher after age 55 compared with the 10-year predicted fracture probability calculated with the FRAX tool, but this did not reach statistical significance (Fig. 3). When the postmenopausal women were categorized according to the presence or absence of parental hip or personal prior fracture or to the WHO classification of BMD, we did not observe any significant difference between predicted fracture probability and observed fracture incidence. Among women aged 65 years and older with low BMD values (n = 199), the predicted major OP fracture probability with BMD was 13.5% ± 9%, that is, 48% lower than the observed fracture incidence (26.2 per 1000 person-years, p < .01; Fig. 4).
Among all postmenopausal women, 127 women (55 ± 4 years) took hormone-replacement therapy (HRT) for 5 years and more—including baseline—and no woman took bisphosphonates. Indeed, the latter—except etidronate—were not yet available in France at the time. When the analysis was focused on the whole group of untreated women at baseline, the predicted fracture probabilities (with BMD) were 6.3% ± 7% for major OP fractures and 2% ± 5% for hip fractures, not significantly lower than the observed fracture incidence (10.9 per 1000 person-years for major OP fractures). Nevertheless, when the analysis was restricted to untreated women aged at least 65 years, the difference between predicted fracture probability and observed fracture incidence were 49% in women with osteoporosis (p < .01) and 59% in women with osteopenia (p < .05) for women aged 65 years and older.
When all women were categorized into deciles according to the predicted fracture probability, the highest decile (probability of major fracture > 12%, n = 86) identified 43% (n = 35) with observed incident fractures, which was better than the prediction provided by BMD in the osteoporotic range (11%, n = 74) that identified 24% of women (n = 20) with incident fractures (p < .05). Nevertheless, the area under the ROC obtained with FRAX was not better than that of BMD alone or BMD and age in predicting major OP fractures (Fig. 5).
The observed incidence of all clinical fragility fractures increased significantly with the quartiles of the predicted major OP fracture probability: from 3.8 per 1000 person-years to 37.5 per 1000 person-years and from 6.7 per 1000 person-years to 36.6 per 1000 person-years from the first to the fourth quartile, without and with BMD, respectively (p < .0001).
In French women from the OFELY cohort, the 10-year predicted fracture probabilities obtained from the FRAX tool increased with age—with or without the use of BMD in the model—following a similar pattern to the observed incidence of fragility fractures over 10 years. However, in women aged at least 65 years with low BMD values, the 10-year predicted fracture probability was substantially lower than the observed incidence of fractures. Half the women who sustained a major fracture had a FRAX probability below 9.2%.
The WHO collaborating center for metabolic bone diseases conducted several metaanalyses of data from large-scale prospective studies in various countries, including the OFELY study in France,9, 10 before developing a risk assessment tool—FRAX—from nine population-based studies—not including OFELY—for expressing absolute fracture risk in men and women in clinical practice. In France, the external validity of the model has been verified with the EPIDOS cohort, but the participants were only elderly and less representative of the French general population than the OFELY cohort. Indeed, the prevalence of osteoporosis in the OFELY study is close to that observed in women 45 years of age and older from a cross-sectional epidemiologic survey conducted in a representative sample of the French general population (INSTANT study).19, 20 In that study, the prevalence of osteoporosis diagnosed with densitometry was 9.7% compared with 9.6% in women over 45 years of age from the OFELY cohort.20 Moreover, the prevalence of some risk factors for fractures was of the same order as in the OFELY cohort (low BMI, respectively, in 4% of women from INSTANT and 4% from OFELY, premature menopause in 10% and 6%, parental history of hip fracture in 9% and 14%, previous peripheral fracture in 10% and 7%, respectively), whereas others were higher than in the OFELY cohort (long-term corticosteroids in 11% of women in INSTANT versus 4% in OFELY, previous vertebral fracture in 21% and 7%, respectively; data not shown).18 However, prior vertebral fractures in the INSTANT study were self-reported, whereas they were confirmed radiographically in the OFELY study, which might explain some discrepancies.
As expected, the predicted fracture probabilities increased with age, with low values in women younger than age 65 years both at the four major sites and at the hip. Moreover, both predicted fracture probabilities and observed fracture incidence were higher in women with parental hip fracture, prior fragility fracture, or low BMD values than in women without those risk factors, indicating that the FRAX tool is reliable. However, in women aged at least 65 years with low BMD values, the observed incidence of major OP fractures over 10 years was twofold higher than the predicted fracture probability given by the FRAX tool. Discrepancies between the observed and predicted rate of fracture could be explained in part by the fact that most countries incorporated into FRAX—including France—have provided hip fracture data only. The rate for the major four fracture areas, however, was estimated using Swedish age-specific ratios of four area fractures to hip fractures, which could differ in France.13 Theorically, this should not result in a serious problem in elderly with low BMD values because the estimated risk is already quite high and likely to exceed the intervention threshold. The WHO makes no specific recommendation concerning intervention thresholds, which should be determined by each country based on the local health care situation and cost-effectiveness in the treatment of osteoporosis. Such thresholds have been formulated in the United Kingdom,21 the United States,22 and Sweden.23 If the cost-effectiveness threshold of 7.5% for major osteoporotic fractures defined in the United Kingdom were applied in France, 44% of women from the French OFELY cohort with incident major OP fractures (n = 36) would not be treated.
In agreement with prior literature, the combination of clinical risk factors and BMD using FRAX improved the prediction of fracture compared with BMD alone.24–26 Nevertheless, in our study, this improvement was observed only in women with the highest values of estimated risk and the lowest values of BMD. Indeed, in the whole group of women, the prediction with FRAX was not better than the age-adjusted femoral neck BMD model. This probably could be explained by the low prevalence of clinical risk factors in the OFELY women, who are healthy and relatively young. Then the effect of these risk factors is lower than that of BMD.
The probability of any OP fracture was underestimated by the FRAX tool. Indeed, if all the clinical fragility fractures (except skull, face, toes, and fingers) were taken into account, the number of women with incident fractures was 1.4-fold higher than in women who sustained a fracture at the four major sites. Their incidence increased according to a similar risk gradient as the predicted major OP fracture probability, confirming that they actually are osteoporotic fractures, but they are not captured by the FRAX tool.
Our study has strengths and limitations. The OFELY study is a population-based cohort study, and all fragility fractures were assessed prospectively and confirmed radiographically. Moreover, all risk factors used in the FRAX tool were collected at baseline, and only a minority of women did not reach 10 years of observation. The observed fracture incidence might fall within confidence intervals of the FRAX tool predicted probabilities, but these data are not available. A limitation of our study is that it is applicable only to a French population of women aged 40 years and older, whereas the FRAX tool has been developed for both men and women. In a study from an independent cohort of Australian men and women, assessing the ability of the Garvan model (including age, sex, prior fractures, falls, and either femoral neck BMD or body weight) and the FRAX algorithm to predict OP fracture, the authors found that the Garvan model had good discriminatory performance for fracture prediction in both men and women, whereas discriminatory performances of the FRAX was better in women than in men.27 In addition, even if the number of hip fracture during follow-up was of the same order as the probability given by the calculation of FRAX, it was too low to permit further interpretable analysis. Finally, some differences between predicted probability and observed incidence of major OP fractures may become significant with a bigger sample size.
In French women from the OFELY cohort, the 10-year predicted fracture probabilities obtained from the FRAX tool increased with age—with or without the use of BMD in the model—following a similar pattern as the observed incidence of fragility fractures over 10 years. A predicted major OP fracture probability higher than 12% identified more women with observed incident fractures over 10 years than a T-score in the osteoporotic range. However, in women aged at least 65 years with low BMD values, the predicted fracture probability was substantially lower than the observed incidence of fractures. Most fractures were observed in women with relatively low probabilities.
All the authors state that they have no conflicts of interest.
We wish to thank Annick Bourgeaud, Betty Vey-Marty, and Wafaa Wirane for excellent technical assistance. This work was supported by an unrestricted research grant from AMGEN to INSERM.