Background: Existing fracture prediction tools are not designed to predict fracture-related consequences. We have developed a risk assessment tool1 to predict fractures and mortality using the Dubbo Osteoporosis Epidemiology Study and the Canadian Multicentre Osteoporosis Study. The tool requires external validation for widespread use.
Objective: to quantify the accuracy of the tool in predicting initial fractures, subsequent fractures and mortality in the MrOS USA cohort.
Methods: 5,994 MrOS men aged 74 (±5.9) years were followed for fractures and mortality. Baseline predictors included age, bone mineral density, prior falls, prior fracture, cardiovascular and respiratory diseases, diabetes, hypertension and cancer. Predicted 5- and 10-year transition risks to initial fractures, subsequent fractures and mortality were estimated. Discrimination ability was assessed using the Harrell’s C index, and calibration assessed by dividing the cohort into centiles of predicted risk and comparing the observed and predicted risk.
Results: During a median follow-up time of 14 years (IQR:8-17), there were 1,085 initial incident fractures (~15.5 fractures/1,000 person-years; 95%CI: 15.6-16.5) followed by 236 subsequent fractures (46.7; 40.9-53.1). Mortality rates among subjects with no fracture, an initial and subsequent fracture were 4.8 deaths/100 person-years (95%CI: 4.6-4.9), 11.8 (10.4-12.7), and 19.4 (16.5-22.6), respectively. Baseline predictors remained significant in the MrOS cohort. The tool had moderate discrimination ability with the highest concordance documented for predicting deaths (C index: 0.75), any initial fracture (0.70) or hip fracture (0.74). Importantly, the tool accurately predicted transition risks to initial fractures, subsequent fractures and post-fracture deaths. However, it overestimated the risk to death from those not suffering fracture, possibly due to substantial between-cohort differences in baseline mortality risk (Fig).
Conclusions: The risk assessment tool provided moderate discrimination and accurate calibration for predicting fractures and fracture-related mortality in men. Subsequent validation in women is necessary prior to consideration of its use in clinical practice settings.