Biases in the performance of FRAX without BMD in predicting fracture risk in a multiethnic population with diabetes: the Diabetes & Aging Study

Abstract Fracture risk calculators, such as the Fracture Risk Assessment Tool (FRAX), calculate the risk of major osteoporotic (MOF) and hip fracture, but do not account for the excess risk of fracture in people with diabetes. We examined the predictive performance of FRAX without BMD in ethnically diverse, older patients with diabetes. Patients included were between ages 65-89 from the Kaiser Permanente Northern California Diabetes Registry and not already taking osteoporosis medications. Race and ethnicity were self-identified. We calculated FRAX without BMD based on baseline characteristics and assessed how well FRAX predicted MOF and hip fracture over follow-up. Predictive performance was based on measures of discrimination (area under the receiver operator curve, AUC) and calibration (observed-to-predicted ratio, O/P). We identified 96 914 patients (47.0% female), of whom 5383 (5.6%) and 1767 (1.8%) had MOF and hip fracture, respectively, over a mean follow-up of 4.3 years. The AUC for MOF and hip fracture were 0.72 and 0.77, respectively. FRAX mildly underestimated MOF and hip fracture rates (O/P 1.2 for both) overall. Discrimination was similar by race and ethnicity and diabetes duration but was worse in those over age 75 (AUC < 0.7). In some groups, there were substantial calibration errors, such as Hispanic women (O/P: 1.8 and 1.5), Black men (O/P: 1.5 and 1.8), those with duration of diabetes ≥20 years (O/P: 1.6 and 1.5), and those over the age of 80 (O/P: 1.4 and 1.2) for MOF and hip fracture, respectively. While the discriminatory performance of FRAX without BMD was good overall in patients with diabetes, it underestimated risk in Hispanic women, Black men, those with long duration of diabetes, and in the oldest patients with diabetes. These algorithmic biases suggest that diabetes-specific tools may be needed to stratify fracture risk in patients with diabetes.