Optimization of fracture risk assessment using machine learning

Osteoporotic fracture is associated with mortality and morbidity. Populations are becoming more elderly, increasing the number of fractures as a result of age-related loss of bone mass.

Historically, risk of fracture has been assessed and defined by bone densitometry, although specificity is high, sensitivity is only about 30%.

With the advent of the FRAX risk calculator, the individualised risk of fracture can be calculated, based on clinical risk factors which, together with BMD increases sensitivity to about 50%. FRAX has been incorporated into more than 80 guidelines worldwide.

Although the predicted and observed fracture cases are highly correlated, this approach still does not identify all individuals at risk for fracture, suggesting that further refinement may increase predictive value.

The aim of this work is to investigate whether more than 40 proposed risk factors from cohorts all around the world, together with machine learning, may usefully add to information about future fracture.

The associations will be explored by analyses of primary data in ~60 international population-based cohorts, included in a collaborative network of which the applicant is a leading coordinator.

This project has the potential to improve the accurate identification of individuals at high risk of fracture, and therefore aid the targeting of preventive treatment.

In turn, this will decrease the number of fractures with their attendant morbidity and mortality and lower costs for society.