Performance of FRAX in women with breast cancer initiating aromatase inhibitor therapy: A Registry-Based Cohort Study

FRAX was developed to predict 10-year probability of major osteoporotic fracture (MOF) and hip fracture in the general population. Aromatase inhibitors (AI) used in breast cancer induce loss in bone mineral density (BMD) and are reported to increase fracture risk. AI exposure is not a direct input to FRAX but is captured under “secondary osteoporosis”. To inform use of FRAX in women treated with AI, we used a population-based registry for the Province of Manitoba, Canada, to identify women aged ≥40 years initiating AI for breast cancer with at least 12 months’ AI exposure (n = 1775), women with breast cancer not receiving AI (n = 1016), and women from the general population (n = 34,205). Among AI users, fracture probability estimated without BMD (AI use coded as secondary osteoporosis) significantly overestimated risk (10-year observed/predicted ratio 0.56, 95% confidence interval [CI] 0.45–0.68; 10-year hip fracture observed/predicted ratio 0.33, 95% CI 0.18–0.49). However, when BMD was included in the fracture probability, there was no significant difference between observed and predicted fracture risk. In Cox proportional hazards models, FRAX stratified risk of MOF, hip, and any fracture equally well in all subgroups (p-interaction >0.1). When adjusted for FRAX score without BMD, with AI use coded as secondary osteoporosis, AI users were at significantly lower risk for MOF (hazard ratio [HR] = 0.78, 95% CI 0.64–0.95), hip fracture (HR = 0.46, 95% CI 0.29–0.73) and any fracture (HR = 0.75, 95% CI 0.63–0.89). AI use was no longer significantly associated with fractures when AI use was not entered as secondary osteoporosis in FRAX without BMD or when BMD was included in the FRAX calculation. In conclusion, FRAX scores stratify fracture risk equally well in women receiving AI therapy as in non-users, but including secondary osteoporosis as a risk factor for AI users overestimates fracture risk. Our results call this practice into question.