Background/Purpose: Current algorithms for identifying patients at high risk of fracture focus on long-term risk. We developed simple 10- and 3-predictor models and points-based risk scores to enable pragmatic identification of patients at near-term (12-24 months) risk of fracture.  Physicians and patients could benefit from such models to ensure timely care.

Methods: This retrospective cohort study included women ≥50 years who had dual-energy x-ray absorptiometry (DXA) bone mineral density (BMD) data and were members of an integrated healthcare delivery system in the United States. Clinic location determined inclusion in the development vs. validation cohort. Using randomly selected outpatient visits (2008- 2012) as index dates, we followed women up to 2 years for the occurrence of “any fracture” – closed fracture other than skull/face/fingers/toes. We adapted our previously validated 10- and 3-predictor models to predict 2-year fracture risks among the subgroup of women with recent BMD T-scores, using Cox regression. We used predictor coefficients from these models to develop risk score points.

Results: 68,589 women in the development cohort (mean±sd age 67.0±9.6 years, 1,816 fractures) and 58,812 women in the validation cohort (age 67.2±9.8 years, 1,440 fractures) had an overall 2-year fracture rate of 2.6% and 2.4%, respectively. We developed a 10-predictor model for “any fracture,” which included age, race/ethnicity, body mass index (BMI), BMD T-score, fracture in the prior year, fall risk, tobacco use, and use of antidepressant, narcotic, and glucocorticoid medications. This model discriminated risk effectively in both the development cohort (c-statistic = 0.75) and the external validation cohort (c-statistic = 0.76). Our reduced, 3-predictor model included age, recent fracture, and BMD T-score, and explained 88.4% of the variation in the 10-variable model. Discrimination was 0.72 in the development and 0.73 in the external validation cohorts. Calibration plots revealed excellent agreement between observed and predicted risks. A 2-year predicted risk ≥10% in the 3-predictor model accounted for 4% of patients and 20% of fractures in the total study population (Table 1). We also developed a points-based risk score for easy implementation of the model to identify patients at high near-term fracture risk (Table 2).

Conclusion: Our data suggest that models using as few as 3 variables can perform well in estimating near-term fracture risk. Our simple models can aid in identification of patients in greatest need of impactful interventions to quickly lower fracture risk.