Prediction Model of Hip Fragility Fracture Using Explainable Artificial Intelligence and Realistic Data From a Traumatology Department and Orthogeriatric Ward

Osteoporosis and associated fragility fractures remain an increasing worldwide burden for both health systems and families, in the context of ageing populations. Hip fractures are particularly severe due to the hospital stay, operations, arduous recovery and risk of subsequent fractures.

Thus, it is of significant importance to detect patients at high risk of femoral fragility fractures and to anticipate their recovery capacities in order to take appropriate medical decisions. Early detection of bone deterioration would be ideal for better prevention and bone reconstruction.

The current gold standard for osteoporosis remains the Dual-energy X-ray absorptiometry (DXA),however one the one hand, a majority of fractured patients are not classified as osteoporotic using the WMO definition and on the other hand, DXA is not widely available in numerous places.

Different alternative devices, such as 3D X-Rays, MRI or ultrasound, with different costs and availability, have been proposed. Moreover, online forms, such as FRAX, Garvan or Qfracture, propose to calculate the fracture risk from a limited number of clinical factors.

Nowadays, growing accessibility to clinical data, processing methods and computing power, opened the way to novel data driven prediction models using a large number of biomarkers or parameters, opening perspective towards personalised precision medicine. However a few challenges arise: