Do Additional Clinical Risk Factors Improve the Performance of Fracture Risk Assessment Tool (FRAX) Among Postmenopausal Women? Findings From the Women's Health Initiative Observational Study and Clinical Trials

Abstract

The ability of the fracture risk assessment tool (FRAX) to discriminate between women who do and do not experience major osteoporotic fractures (MOFs) is suboptimal. Adding common clinical risk factors may improve discrimination. We used data from the Women's Health Initiative, a prospective study of women aged 50 to 79 years at baseline (n = 99,413; n = 5722 in BMD subset) enrolled at 40 US clinical centers. The primary outcome was incident MOFs assessed annually during 10 years' follow‐up. For prediction of incident MOF, we examined the area under the receiver operatic characteristic curve (AUC) and net reclassification index (NRI) of the FRAX model alone and FRAX plus additional risk factors (singly or together: type 2 diabetes mellitus, frequent falls [≥2 falls in the past year], vasomotor symptoms, self‐reported physical function score [RAND 36‐item Health Survey subscale), and lumbar spine BMD). For NRI calculations, high risk was defined as predicted MOF risk ≥20%. We also assessed calibration as observed MOF events/expected MOF events. The AUC value for FRAX without BMD information was 0.65 (95% CI, 0.65 to 0.66). Compared with the FRAX model (without BMD), the AUC value was not improved by the addition of vasomotor symptoms, diabetes, or frequent falls, but was minimally increased by adding physical function score (AUC 0.66, 95% CI, 0.66 to 0.67). FRAX was well‐calibrated for MOF prediction. The NRI of FRAX + additional variables versus FRAX alone was 5.7% (p < 0.001) among MOF cases and −1.7% among noncases (p > 0.99). Additional variables (diabetes, frequent falls, vasomotor symptoms, physical function score, or lumbar spine BMD) did not yield meaningful improvements in NRI or discrimination of FRAX for MOFs. Future studies should assess whether tools other than FRAX provide superior discrimination for prediction of MOFs. © 2019 The Authors. JBMR Plus published by Wiley Periodicals, Inc. on behalf of American Society for Bone and Mineral Research.

Machine Learning to Predict Long and Short Term Fracture Risk in Postmenopausal Women

“…The algorithms demonstrated better performance for prediction of major osteoporotic fractures in comparison to the AUCs that are reported for FRAX by Crandall et al 21 and met or exceeded previously reported performance of the FRAX tool in other studies, including those conducted on comparable populations. [21][22][23]28,31 In addition to traditional risk assessment tools, machine learning has been explored for its utility as a risk strati cation tool for osteoporotic fractures. 26,28,29 ML-based tools for this purpose hold the potential for individualized fracture prediction.…”

“…13 FRAX has also been found to have varying performance across different geographic locations and patient characteristics (eg, age), with AUROCs ranging from 0.65 to 0.81. [21][22][23] Crandall et al 21 reported a FRAX AUROC of 0.65-066 in a US study on women aged 50-79 using both observational and prospectively collected data, whereas a study on Spanish women aged 50-90 reported an FRAX AUROC of 0.812 for predicting major osteoporotic fractures using data obtained from bone density testing and a participant questionnaire. 24 Using data extracted from electronic medical records to calculate FRAX without BMD measurements, Goldshtein et al demonstrated AUROCs of 0.65 and 0.82 for major osteoporotic fracture and hip fractures, respectively.…”

Menopausal osteoporosis: screening, prevention and treatment

“…Secondary causes of osteoporosis, such as diabetes mellitus, rheumatoid arthritis or parathyroid disease, account for up to 45% of cases of osteoporosis in postmenopausal women. (22) OSTA has a sensitivity and specificity of 91% and 45%, respectively, and compares well to other more complicated tools such as SCORE, (16) ORAI (17) and OSIRIS. (18) Clinical risk assessment for osteoporosis using the Fracture Risk Assessment Tool Although osteoporosis can be diagnosed with clarity, it is fractures that are of public health concern.…”

“…Application of United States (US) and Canadian osteoporosis screening strategies indicate that the FRAX and Garvan fracture risk calculators without BMD have specificity of < 50% and both failed to identify the majority of women who actually experienced major osteoporotic fractures during ten years of follow-up in the large Women's Health Initiative (WHI) study. (22) Nevertheless, determining fracture risk through FRAX can assist in clinical decision-making regarding treatment. Singapore-specific thresholds for treatment have been calculated at 14.0% and 3.5% for major and hip osteoporotic fractures, respectively.…”

Comparison between real-world practice and application of the FRAX algorithm in the treatment of osteoporosis

“…In Switzerland, half of the patients analyzed were classified by FRAX as low risk the day before developing a fragility fracture [ 22 ]. According to Crandall et al [ 23 ], the ability of FRAX to discriminate between women who will or will not experience a major osteoporotic fracture is no better than chance for postmenopausal women aged 50–64 [ 24 , 25 ], and aged ≥ 65 years [ 26 ].…”