Part of this study was presented at the 31st Annual Meeting of the ASBMR, in Denver, CO, USA, on (September 13, 2009).
Fracture risk prediction using BMD and clinical risk factors in early postmenopausal women: Sensitivity of the WHO FRAX tool†
Version of Record online: 8 JAN 2010
Copyright © 2010 American Society for Bone and Mineral Research
Journal of Bone and Mineral Research
Volume 25, Issue 5, pages 1002–1009, May 2010
How to Cite
Trémollieres, F. A., Pouillès, J.-M., Drewniak, N., Laparra, J., Ribot, C. A. and Dargent-Molina, P. (2010), Fracture risk prediction using BMD and clinical risk factors in early postmenopausal women: Sensitivity of the WHO FRAX tool. J Bone Miner Res, 25: 1002–1009. doi: 10.1002/jbmr.12
- Issue online: 30 APR 2010
- Version of Record online: 8 JAN 2010
- Accepted manuscript online: 8 JAN 2010 12:00AM EST
- Manuscript Accepted: 15 DEC 2009
- Manuscript Revised: 20 OCT 2009
- Manuscript Received: 28 JUN 2009
- risk factors;
- bone mineral density;
- prediction of risk;
- frax score;
- early post menopausal women
- Top of page
- Materials and Methods
The aim of this prospective study was (1) to identify significant and independent clinical risk factors (CRFs) for major osteoporotic (OP) fracture among peri- and early postmenopausal women, (2) to assess, in this population, the discriminatory capacity of FRAX and bone mineral density (BMD) for the identification of women at high risk of fracture, and (3) to assess whether adding risk factors to either FRAX or BMD would improve discriminatory capacity. The study population included 2651 peri- and early postmenopausal women [mean age (± SD): 54 ± 4 years] with a mean follow-up period of 13.4 years (±1.4 years). At baseline, a large set of CRFs was recorded, and vertebral BMD was measured (Lunar, DPX) in all women. Femoral neck BMD also was measured in 1399 women in addition to spine BMD. Women with current or past OP treatment for more than 3 months at baseline (n = 454) were excluded from the analyses. Over the follow-up period, 415 women sustained a first low-energy fracture, including 145 major OP fractures (108 wrist, 44 spine, 20 proximal humerus, and 13 hip). In Cox multivariate regression models, only 3 CRFs were significant predictors of a major OP fracture independent of BMD and age: a personal history of fracture, three or more pregnancies, and current postmenopausal hormone therapy. In the subsample of women who had a hip BMD measurement and who were not receiving OP therapy (including hormone-replacement therapy) at baseline, mean FRAX value was 3.8% (±2.4%). The overall discriminative value for fracture, as measured by the area under the Receiver Operating Characteristic (ROC) curve (AUC), was equal to 0.63 [95% confidence interval (CI) 0.56–0.69] and 0.66 (95% CI 0.60–0.73), respectively, for FRAX and hip BMD. Sensitivity of both tools was low (ie, around 50% for 30% of the women classified as the highest risk). Adding parity to the predictive model including FRAX or using a simple risk score based on the best predictive model in our population did not significantly improve the discriminatory capacity over BMD alone. Only a limited number of clinical risk factors were found associated with the risk of major OP fracture in peri- and early postmenopausal women. In this population, the FRAX tool, like other risk scores combining CRFs to either BMD or FRAX, had a poor sensitivity for fracture prediction and did not significantly improve the discriminatory value of hip BMD alone. © 2010 American Society for Bone and Mineral Research
- Top of page
- Materials and Methods
The 1994 World Health Organization (WHO) definition of osteoporosis1—that is, a bone mineral density (BMD) value less than 2.5 standard deviations from the young-adult normal value (T-score < −2.5)—represents the landmark that has been used in most clinical trials as a major inclusion criterion2 and consequently continues to be used frequently as an intervention threshold by a number of practitioners.3 However, even though a low BMD is strongly associated with the risk of fracture, it is well recognized that different risk factors, such as age, history of a prior fragility fracture, steroid use, and many others, are independent contributors to the risk of fracture and improve the sensitivity of BMD measurement in identifying patients at high risk of fracture.4, 5 Current guidelines thus have emphasized the usefulness of combining BMD and clinical risk factors to determine an absolute risk of fracture and decide which patient has to be treated and which patient has a sufficiently low fracture risk to be observed without treatment. In this regard, the WHO's fracture risk assessment tool (FRAX), which has just been launched during this last year, allows one to calculate the 10-year probability of fractures in men and women from several clinical risk factors (CRFs) with or without the measurement of femoral neck BMD.6, 7 The performance characteristics of the FRAX tool have been validated in independent cohorts with over 1 million person-years of observation. However, most, if not all, of these cohorts concern elderly women, usually over the age of 65 and have focused mainly on hip fractures. There is some uncertainty as to whether this screening tool would have the same level of performance in early postmenopausal women, where risk factors for fracture and fracture locations are likely different from those in older women. Using data from a large cohort of peri- and early postmenopausal women, we aimed (1) to identify significant risk factors for fracture, independent of age and BMD, (2) to assess and compare the sensitivity and specificity of FRAX and BMD for fracture prediction, and (3) to assess whether adding new risk factors to FRAX or using a different combination of CRFs and BMD would improve the ability to identify women at high risk of fracture.
Materials and Methods
- Top of page
- Materials and Methods
This study is part of the Menopause et Os (MENOS) cohort study, which was conducted in the Menopause Center of the Toulouse University Hospital.8 The primary objective of that study was to determine whether bone mass measurement at menopause, together with other various clinical characteristics, could be a predictor of different postmenopausal diseases such as osteoporotic fractures, cardiovascular diseases, and gynecologic cancers (eg, breast, uterus, and ovarian cancers). The program was initiated in all women aged more than 45 years who were consecutively referred to the Menopause Center during the years 1988–1991 for a systematic “menopause checkup.” This study was approved by the Ethical Committee (CPP) of the Toulouse University Hospital, and all women gave written informed consent. At baseline, all women were evaluated through the same procedure of the menopause unit, which included personal interview and medical examination, fasting blood sample drawing (for hormonal and biochemical measurements), and bone densitometry measurements. Women were considered postmenopausal if they had not menstruated within the last 12 months before the examination, associated with serum follicle-stimulating hormone (FSH) levels above 30 IU/L and serum estradiol (E2) levels below 20 pg/mL.
Between 2002 and 2004, all women were invited back again for a medical visit (follow-up visit) and further bone mass measurements. Of the 4024 women who were included initially, 2651 (65.9%) attended the follow-up visit. Of the 1373 nonrespondents, 109 had died, 424 refused to participate in the follow-up visit, and 840 were lost to follow-up.
- Top of page
- Materials and Methods
Clinical risk factors data collection
At baseline, all women answered a computer-assisted standardized questionnaire that has been described extensively elsewhere,9–11 recorded by the same trained research nurse. The following clinical and historical data were extracted for each subject: age, weight, height, body mass index (kg/m2), reproductive history (age of menarche, prior use of oral contraceptive pills, number of live births, age and type of menopause, hysterectomy, bilateral oophorectomy, postmenopausal hormone therapy use), self-reported personal history of low-trauma fractures (ie, those occurring after a fall from standing height or less) after age 45, as well as a history of hip fracture either in the mother or in the father. History of medical conditions and use of medications known to interfere with bone mass (including steroids, sodium fluoride, calcitonin, and bisphosphonates) as well as L-thyroxine (L-T4) treatment were recorded. Women with past/current osteoporosis treatment for more than 3 months (with the exception of hormone replacement therapy (HRT) and calcium/vitamin D supplementation) at baseline (n = 455) were excluded from the analyses, which left us with 2196 women.
The smoking and drinking status also was determined. Each woman was asked whether she smoked or had ever smoked and the number of glasses of wine or alcohol she used to consume per day or week. Dietary calcium intake was assessed by using Fardellone's questionnaire.12 The woman's overall level of physical activity (low, moderate, or high) was determined by asking whether she practiced a sport or leisure physical activity regularly (at least once a week), what type of professional activity she had, and how much time she spent in home-based physical activity. Each woman then underwent a medical examination where all recorded data were reviewed and completed when necessary.
At the follow-up visit, anthropometric measurements were taken, and all women answered the MENOS epidemiologic standardized questionnaire. For this analysis, information on prior and current HRT use, total duration of HRT use, years since menopause, and history and age of hysterectomy and/or oophorectomy were used to determine the age at menopause (for those who were not menopausal at baseline), as well as HRT use throughout the course of the study.
At the follow-up visit, women were asked whether they had suffered a fracture after baseline (incident fracture), and the type, mechanism, and circumstances of the fracture were recorded. We only considered incident fractures that occurred with minimal or no trauma (fall from standing height or less) at the level of the spine, hip, distal forearm, and proximal humerus. Fractures of fingers, toes, or skull and face were excluded. All incident fractures were confirmed by radiographs or by medical/surgical reports. Systematic radiographs of the spine were not performed, and we considered only symptomatic (clinical) spine fractures.
BMD (g/cm2) of the lumbar spine (L2–L4) was determined in all women using dual-energy X-ray absorptiometry (DXA; DPX-IQ, Lunar GE, Madison, WI, USA) following conventional procedures, as described previously.13 Owing to changes in the bone mass measurements protocol that occurred in the Menopause Center during the baseline examination period, femoral neck BMD also was measured in the 1399 women who were included in the study from 1989 onward. Therefore, in our sample, 797 women had only vertebral BMD measurements, whereas 1399 women had both vertebral and femoral neck BMD measurements at baseline. T-scores (SD from young normal individuals) for each measurement site were calculated based on the comparison of measured BMD values with our own normative database at ages 25 to 35 years (n = 110, mean spine BMD = 1.18 ± 0.12 g/cm2; mean femoral neck BMD = 0.990 ± 0.11 g/cm2; personal data).
The first major osteoporotic fracture (wrist, clinical spine, upper arm, and hip) during the follow-up period was taken as the endpoint event. We used Student's t tests and chi-square tests to compare baseline characteristics of women with and without incident fracture. Variables significantly associated with fracture (p ≤ .10) were retained for the multivariate analysis. We used Cox proportional hazards models to model age to first fracture. Selection of the most predictive factors took place in two steps. First, we examined each variable separately and selected those which were significantly associated with the risk of fracture (p ≤ .10) after adjustment for time since menopause, which was introduced into the model as a time-dependent variable (time set to zero for women before menopause). We then performed manual backward stepwise Cox regression (p ≤ .05; Wald test) to determine the best set of independent predictors of fracture. Use of hormone replacement therapy (HRT: never, past, current use) was introduced into the model as a time-dependent variable because we had information on the age at which women started HRT and on the number of months they used it during follow-up. We report hazard ratios as relative risks (RRs) with 95% confidence intervals (CIs).
We conducted separate analyses for women who had a spine BMD measurement and for those who had a hip BMD measurement in complement with a spine BMD measurement. Hence we present two separate models: one with spine BMD and the other with hip BMD.
In the second part of the analysis, we worked on the subgroup of women who had a hip BMD measurement to assess and compare the discriminatory capacity of FRAX and BMD in identifying women at high risk of fracture. Although HRT may be prescribed for other reasons than osteoporosis prevention, it is well established that its use results in a significant protection against fractures. Hence we excluded from this part of the analysis women who were current HRT users at baseline, which left a sample of 956 women (and 76 fractures). We used the algorithm available at http://www.shef.ac.uk/FRAX to calculate individual FRAX values based on baseline hip BMD and CRF values. We also assessed whether adding new CRFs to FRAX or constructing a risk score based on the best prediction model in our population would improve the ability to identify women at high risk of fracture. We used Cox models with time to fracture as the endpoint and baseline characteristics (CRFs plus either FRAX or hip BMD) as independent variables to calculate individual score values (parametric part of the risk function) and rank women by their predicted risk of fracture. We assessed the overall discriminative value of the different risk scores considered by calculating the areas under the Received Operating Characteristic (ROC) curve (AUC). We also assessed the sensitivity (proportion of women with fracture who had been classified as high risk) and specificity (proportion of women without fracture who had not been classified as high risk) of each risk score for various definitions of the high-risk group based on percentiles of the score distribution in the study population. All statistical analyses were conducted using SAS 9.1 and STATA Version 9 (Stata Corp., College Station, TX, USA)
- Top of page
- Materials and Methods
During the mean follow-up period of 13.4 (±1.4) years, 415 women sustained a first low-energy fracture, including 145 (6.6%) major OP fractures (108 wrist, 44 spine, 20 proximal humerus, 13 hip). Baseline characteristics of women with and without incident major OP fracture are given in Tables 1 and 2. Women with fracture were significantly older and had lower BMD values. They were more likely to have a history of fracture after the age of 45, a family history of hip fracture, a lower number of pregnancies, and a higher age at puberty. However, they were less likely to have ever used oral contraceptives (OCs) and to be using HRT at baseline.
|Women with fracture (n = 145)||Women without fracture (n = 2,051)||p Value$|
|Age (years)||54.8 ± 4.3||53.4 ± 4.2||<.0001|
|BMI (kg/m2)||22.9 ± 2.99||23.1 ± 3.10||.44|
|Age at menarche (years)||13.3 ± 1.5||13.1 ± 1.5||.08|
|Age at menarche (years)||.31|
|10–15||115 (79.3%)||1712 (83.51%)|
|≥15||30 (20.7%)||332 (16.2%)|
|Ever use of oral contraceptive (n)||36 (24.8%)||707 (34.5%)||.0177|
|Parity (n)||1.8 ± 1.3||2.0 ± 1.2||.013|
|Number of pregnancies||.0248|
|0||23 (15.9%)||212 (10.3%)|
|1||38 (26.2%)||429 (20.9%)|
|2||54 (37.2%)||807 (39.4%)|
|≥3||30 (20.7%)||603 (29.4%)|
|Numbers of breast-fed children||.22|
|0||85 (58.6%)||1069 (52.1%)|
|1||31 (21.4%)||448 (21.8%)|
|≥2||29 (20.0%)||534 (26.1%)|
|Total duration of breast-feeding (months)||7.1 ± 7.2||8.0 ± 7.8||.36|
|Menopause status (yes/no)||129 (89.0%)||1766 (86.1%)||.33|
|Early menopause (<40 years) (n)||5 (2.7%)||56 (3.5%)||.61|
|Never||104 (71.7%)||1296 (63.2%)|
|Past||11 (7.6%)||98 (4.8%)|
|Current||30 (20.7%)||657 (32.0%)|
|Prior breast cancer (n)||5 (3.5%)||37 (1.8%)||.16|
|Women with fracture (n = 145)||Women without fracture (n = 2051)||p Value|
|Vertebral BMD (g/cm2)||0.96 ± 0.126||1.03 ± 0.148||<.0001|
|> −1||32 (22.2%)||858 (42.1%)|
|−2.5 to −1||73 (50.7%)||888 (43.5%)|
|≤ –2.5||39 (27.1%)||293 (14.4%)|
|Femoral neck BMDa (g/cm2)||0.77 ± 0.104||0.84 ± 0.115||<.0001|
|> −1||13 (13.3%)||423 (32.5%)|
|−2.5 to −1||57 (58.2%)||713 (54.8%)|
|≤ −2.5||28 (28.5%)||165 (12.7%)|
|BMI <19 kg/m2||10 (6.9%)||110 (5.4%)||.43|
|Previous fracture history (after 45 years) (y/n)||12 (8.3%)||43 (2.1%)||<.0001|
|Maternal/paternal hip fracture history (y/n)||22 (15.2%)||216 (10.5%)||.08|
|Corticosteroid use (y/n)||2 (1.4%)||27 (1.3%)||.95|
|Never||121 (88.3%)||1565 (80.1%)|
|Past||7 (5.1%)||199 (10.2%)|
|Current||9 (6.6%)||191 (9.7%)|
|1+ drink of wine||21 (14.5%)||392 (19.1%)||.17|
|1+ drink of alcohol||22 (15.2%)||39 (19.3%)||.22|
|Rheumatoid arthritis (y/n)||0||1||.79|
|Secondary osteoporosis (y/n)||2 (1.4%)||48 (2.3%)||.10|
|Calcium daily intake (mg)||764.0 ± 492.14||774.7 ± 472.87||.80|
|Calcium/vitamin D supplements (y/n)||11 (7.6%)||135 (6.6%)||.64|
|Low||69 (47.6%)||817 (39.9%)|
|Moderate||68 (46.9%)||1084 (53.0%)|
|High||8 (5.5%)||146 (7.1%)|
In Cox regression models, adjusting for time since menopause, individual variables that remained significantly associated with the risk of fracture were spine BMD, HRT use throughout follow-up, a personal history of fracture, and the number of pregnancies. In stepwise Cox regression, all these variables were retained into the model; hence they represent the best set of fracture predictors. The hazard ratios and corresponding 95% CIs associated with each of these variables are presented in Table 3.
|Predictive factors||Model including spine BMD,a hazard ratio (95% CI)||Model including hip BMD,b hazard ratio (95% CI)|
|Years since menopause (/5 years)||1.09 (0.91–1.31)||1.11 (0.91–1.36)|
|BMD (/1 SD decrease)||1.41 (1.18–1.69)||1.70 (1.35–2.14)|
|Prior history of fracture (>45 years)||2.54 (1.39–4.62)||2.28 (1.04–5.03)|
|HRT use throughout follow-up:|
|Past||0.87 (0.58–1.32)||0.95 (0.57–1.58)|
|Current||0.49 (0.32–0.76)||0.58 (0.34–0.96)|
|1||0.87 (0.52–1.47)||1.02 (0.54–1.95)|
|2||0.68 (0.42–1.11)||0.66 (0.36–1.22)|
|3+||0.44 (0.26–0.76)||0.52 (0.27–1.00)|
We conducted a separate analysis on the subset of women who also had a hip BMD measurement at baseline and ended up with the same selection of predictors. Hazard ratio estimates in the model with hip BMD are of approximately the same magnitude than those in the model with spine BMD (Table 3).
The second part of the analysis focused on the 956 women who had a hip BMD measurement and were not current HRT users at baseline. Their mean FRAX value (average probability of having a major OP fracture over the next 10 years) was 3.8% (±2.4%). Based on FRAX, the total number of major OP fractures expected during the first 10 years of follow-up (sum of the individual FRAX probabilities) was 36, whereas the observed number of fractures during the same period was 42.
We then compared the sensitivity for fracture of FRAX and hip BMD for various definitions of the high-risk group based on percentile of their distribution in the study population (Table 4). If the cutoff for high risk is set, for instance, at 30% (ie, the 30% of women with the highest FRAX values or with the lowest BMD values are classified as being “at high risk”), the sensitivity is approximately equal to 49% and 55%, respectively, for FRAX and hip BMD. To have a sensitivity around 80%, the cutoff has to be raised to 60% for both measures, which means that 60% of the women have to be classified as being “at high risk.” Noteworthy, given the low incidence of fracture (around 0.5%), the specificity of any measure is approximately equal to the percentage of women who are not classified at high risk. For instance, at the 30% cutoff, the specificity was approximately equal to 70% for both FRAX and hip BMD. The overall discriminative value for fracture, as measured by the area under the ROC curve (AUC), is equal to 0.63 (95% CI 0.56–0.69) and 0.66 (95% CI 0.60–0.73), respectively, for FRAX and hip BMD.
|Percent of women in the high-risk group||Hip T-score||FRAX modela|
Since, the first part of the analysis revealed that parity (which is not taken into account in the FRAX score calculation) was a significant predictor of fracture, independent of age, BMD, and history of fracture, we then examined whether adding parity to a predictive model including FRAX would significantly improve the ability to identify women at high risk of fracture. The AUC of the new score FRAX + parity (0.65; 95% CI 0.58–0.71) was not significantly better than the AUC of FRAX or hip BMD alone. We next examined the discriminative value of a simple risk score including the four factors—age, hip BMD, history of fracture, and parity—that best predicted fracture in our study population. The AUC of this model-based risk score was equal to 0.69 (95% CI 0.63–0.72), which was significantly better than the AUC of FRAX (p = .01) but not better than the AUC of hip BMD alone (p = .16). The ROC curves of the four risk assessment tools considered in this analysis are shown in Figure 1.
Finally, we had to consider that in the study population a significant number of women (357 of 956) started HRT during follow-up (on average, 3.1 years after baseline). Fracture risk may be overestimated in these women, which might affect the overall predictive value of the screening tools. Hence we reran the analysis after exclusion of women who started HRT after baseline. We found the same set of predictors as in the overall population, and the AUC values of the different screening tools were of the same magnitude as before (0.64 for FRAX, 0.69 for hip BMD, 0.63 for the score FRAX + parity, and 0.63 for the model-based risk score). Noteworthy, there were no significant difference in mean hip BMD and percentage of osteoporotic women (T-score < –2.5) between the subgroup of women who started HRT during follow-up and those who did not.
- Top of page
- Materials and Methods
This prospective study involving 2196 early postmenopausal women (mean age at baseline 54 ± 4 years) revealed that there were only a few clinical risk factors for major OP fractures independent of BMD and age. Moreover, the FRAX risk assessment tool did not have a better discriminative value for fracture prediction than BMD alone. Adding new CRFs to FRAX or using a combination of CRFs and BMD based on the best predictive model in our population did not significantly improve fracture prediction compared with BMD alone.
Of the 38 potential CRFs examined, only 3—a history of fracture, current HRT use, and parity—were significant and independent predictors of fracture after age and BMD are taken into account. Using BMD measured either at the spine or at the hip to generate two different multivariate models did not modify the set of predictors, nor their gradient of risk. Even though, the gradient of risk appeared slightly better for femoral neck BMD than for spine BMD, there was an overlap in the confidence intervals. With DXA measured at the femoral neck, the risk of a major fracture increased 1.7-fold per SD decrease in BMD, whereas it increased 1.4-fold per SD when the spine BMD was entered into the model. This difference could be explained by the development of arthritis, which is likely to hamper the validity of the BMD measured at the lumbar spine, although the prevalence of arthritis is likely low in this normal-weight young postmenopausal population. Nevertheless, when considering all OP fractures (and not only the major ones), the risk ratio was similar for hip and spinal BMD with a gradient of risk of 1.4 per SD decrease (data not shown). This gradient of risk is consistent with that reported in the meta-analysis by Marshall and colleagues (RR = 1.6, 95% CI 1.4–1.8)14 and in other studies performed in perimenopausal women.15, 16 However, it was higher than that found in the large meta-analysis by Johnell and colleagues17 in women at age 50 (RR = 1.22, 95% CI 1.07–1.39). The reasons for this discrepancy are not fully understood. It might be related to differences in the study populations, especially with regard to age and the type of incident fractures (hip versus other fractures) or the methods used. As already reported,18–20 and not surprisingly, our study confirms that a history of a fragility fracture is a powerful predictor of future fracture independent of BMD. The multivariable- and BMD-adjusted risk of a major OP fracture was 2.3- to 2.5-fold higher in women who had reported a prior low-trauma fracture after the age of 45 years compared with women without a prior fracture history. In agreement with previous studies,21 we also found that use of postmenopausal HRT resulted in a significant protection against fractures: Current HRT users had a 50% to 60% reduced risk of major OP fractures compared with never users. These results are consistent with those reported in Huopio's study,15 which to our knowledge is one of the few that concerned a large population of 3000 perimenopausal women who had been followed over 3.6 years and whose mean age at baseline (53 years) was very similar to that in our study. In this cohort, only two majors CRFs were identified in addition to low BMD: previous fracture history and no use of HRT.
In this study, parity was found to be an additional significant and independent predictor of fracture: Compared with nulliparous women, women with three children or more had a 48% to 56% decreased risk of major fragility fracture independent of BMD and other clinical risk factors. Interestingly, there was a monotonal relationship between increasing parity and the risk of fracture because the probability of fracture decreased with increasing number of children. Such relationship has been reported before in most (but not all) prior studies that examined the association between parity and fracture risk.22–24 The most frequently discussed hypothesis underlying such association between parity and fracture risk would be that each additional pregnancy contribute to increase the accumulated lifetime exposure to estrogen and thus reduce the subsequent risk of fracture.
Several CRFs included in FRAX (in particular, familial history of fracture, current use of systemic glucocorticoids, low BMI, current smoking, high intake of alcohol, and rheumatoid arthritis) were not significantly associated with the risk of fracture in our population. This may be interpreted as a true lack of predictive value in peri- and early postmenopausal women. Alternatively, the lack of association between some factors and fracture risk may be due to a lack of statistical power. This may be the case, in particular, for a family history of fracture, which showed a nonsignificant trend toward an increased risk of fracture in exposed women. If we hypothesize that the RR of fracture associated with a family history of hip fracture is of the same magnitude as in the FRAX cohorts (ie, 1.3),25 we have calculated that our power to show such an RR is only 20% (risk α set at 5%). Lack of power is also evident for CRFs such as exposure to systemic glucocorticoids, high intake of alcohol, and rheumatoid arthritis, whose prevalence is very low in our sample. For instance, only 29 women (1.3%) were using glucocorticoids in our population of young and relatively healthy postmenopausal women. Hence the predictive value of a family history of fracture, corticosteroid use, and some other infrequent CRFs would need to be further studied in larger cohorts of young postmenopausal women.
One of the main differences between our cohort and the cohorts used to develop the FRAX tool5 is that our cohort is, on average, 10 years younger (53 versus 63 years in the FRAX cohorts). The mix of fractures is likely different in early postmenopausal than in elderly women. Hip fractures represent fewer than 5% of all OP fractures in women aged 50 to 54 years (4.4% in our cohort), whereas about 40% of the fractures are wrist fractures (32% in our cohort).26 Hence it seems logical to think that the nature or strength of risk factors for fracture may vary based on the woman's age and/or the type of fracture. Accordingly, in the meta-analyses using data from the FRAX cohorts, all CRFs except a personal history of fracture were more strongly associated with the risk of hip fracture than with the risk of other OP fractures.5 At the age of 50 years, the gradient of risk with BMD alone was 3.7 per SD, but with the addition of CRFs, it increased to 4.2 per SD. For other major OP fractures, the gradient of risk with BMD alone was lower, and adding CRFs only modestly improved it. At the age of 50 years, it was 1.2 per SD with hip BMD alone, whereas it was 1.4 per SD with the combination of BMD and CRFs, which translates to an AUC of around 0.6, which is very similar to what we found in our population of peri- and early postmenopausal women (0.63 and 0.66, respectively, for FRAX and hip BMD).
The use of FRAX in our study population did not significantly improve the sensitivity of BMD assessment, which remained low under most reasonable assumptions that avoid unnecessary treatment. If the cutoff for high risk is set at 30%, for instance, the sensitivity is equal to 49% and 55%, respectively, for FRAX and hip BMD. To identify the majority (ie, around 80%) of women who will suffer a fracture, the cutoff has to be raised to 60% (ie, 60% of the population has to be classified as being at high risk). Noteworthy, other investigators reported the lack of sensitivity of FRAX in elderly women27–29 and in some male populations.30
We found that parity, which is not included in the FRAX algorithm, was a significant predictor of the risk of major OP fracture, independent of BMD and of the other CRFs. This risk factor may be more important with regard to fracture risk prediction in early postmenopausal women than in older women. Hence we assessed whether adding parity to FRAX would improve the ability to identify young postmenopausal women at high risk of fracture. We found that the new score combining parity and FRAX did not significantly improve the sensitivity of FRAX and did not have a better discriminative value than hip BMD alone. Even a score based on the best predictive equation in our population (best fit of our data) was not significantly more discriminant than hip BMD alone. These findings may seem paradoxical because the factors entering the score are all significant predictors of fracture. However, it has been shown that a risk factor must have a much stronger association with the disease outcome (very large RR) than we ordinarily see in etiologic research for it to be a worthwhile screening test, that is, be capable of efficiently discriminate between persons likely to have the outcome and those who do not (good sensitivity and specificity).31–33
Our study has several strengths and limitations. An extensive amount of data were collected for each woman at the baseline and follow-up visits, including the use of osteoporosis medications and HRT throughout the follow-up, which allowed us to examine a large set of potential CRFs for fracture. All incident fractures were confirmed through either medical records or radiographs, and we have been very careful to record the circumstances of each fracture through medical personal interview. However, asymptomatic vertebral fractures could not be taken into account because there were no systematic X-ray examinations during follow-up. A large number of participants (almost 60%) used HRT at some point of the study, and about 36% were still using HRT by the end of the follow-up. Such a high frequency of HRT was very common in the French population before the publication of the Women's Health Initiative (WHI) results, whatever the underlying reasons, mainly to resume climacteric symptoms but also to prevent osteoporosis. Since the use of FRAX is appropriate only for untreated women, we reran the analysis after exclusion of women who took HRT at some point over the course of the study. We found the same set of CRFs for fracture, and the discriminative value of FRAX and the other risk scores remained unchanged. Women included in the MENOS cohort all were referred to our center by their personal physicians for a “menopause checkup.” Reasons for referral are not known but may include a higher risk of osteoporosis. Accordingly, the prevalence of a T-score of less than −2.5 was slightly higher than expected for this age range, which could suggest a selection bias and thus could not allow us to generalize our results to the general population. Nevertheless, our results may broadly apply to women likely to seek medical care and advice at the time of menopause or in the few years after.
In conclusion, only a limited number of clinical risk factors were found associated with the risk of major OP fracture in peri- and early postmenopausal women. In this population, the FRAX tool, like other risk scores combining CRFs with either BMD or FRAX, had a poor sensitivity for fracture prediction and did not significantly improve the discriminatory value of hip BMD alone. Further studies are needed to determine whether including bone loss–related factors such as, for instance, the early rate of postmenopausal bone loss or biomarkers of bone remodeling could improve the capability of clinical scores to identify women at high risk of fracture.
- Top of page
- Materials and Methods
All the authors state that they have no conflicts of interest.
- Top of page
- Materials and Methods
This work was part of the MENOS study and was supported by an institutional grant from Lilly France and Pierre Fabre Santé Laboratories. We gratefully acknowledge the excellent assistance of Mrs Colette Cauneille and Fabienne Cigagna in maintaining all patient data, and we thank all the women who kindly participated in the MENOS study.
- Top of page
- Materials and Methods
- 3Food and Drug Administration. Guidelines for Preclinical and Clinical Evaluation of Agents Used in the Prevention and Treatment of Postmenopausal Osteoporosis. Rockville, MD: Division of Metabolism and Endocrine Drug Products; 1994.