FRAX for fracture prediction shorter and longer than 10 years: the Manitoba BMD registry

Abstract

Summary

In a large clinical registry for the province of Manitoba, Canada, FRAX predicted incident MOF and hip fracture from 1 to 15 years following baseline assessment. A simple linear rescaling of FRAX outputs seems useful for predicting both short- and long-term fracture risk in this population.

Introduction

FRAX® estimates 10-year probability of major osteoporotic fracture (MOF) and hip fracture. We examined FRAX predictions over intervals shorter and longer than 10 years.

Methods

Using a population-based clinical registry for Manitoba, Canada, we identified 62,275 women and 6455 men 40 years and older with baseline dual-energy X-ray absorptiometry scans and FRAX scores. Incident MOF and hip fracture were assessed up to 15 years from population-based data. We assessed agreement between estimated fracture probability from 1 to 15 years using linearly rescaled FRAX scores and observed cumulative fracture probability. The gradient of risk for FRAX probability and incident fracture was examined overall and for 5-year intervals.

Results

FRAX predicted incident MOF and hip fracture for all time intervals. There was no attenuation in the gradient of risk for MOF even for years >10. Gradient of risk was slightly lower for hip fracture prediction in years >10 vs years <5, though HRs remained high. Linear agreement was seen in the relationships between observed vs predicted (rescaled) FRAX probabilities (R ² 0.95–1.00). Among women, there was near-perfect linearity in MOF predictions. Deviations from linearity, with a slightly higher observed than predicted MOF probability, were most evident in the first years following a fracture event and after 10 years for hip fracture prediction in women using FRAX with BMD. Simulations showed that results were robust to large differences in fracture rates and moderate differences in mortality rates.

Conclusions

FRAX predicts incident MOF and hip fracture up to 15 years and could be adapted to predict fracture over time periods shorter and longer term than 10 years in populations with fracture and mortality epidemiology similar to Canada.

http://ift.tt/2r3Qan4