Abstract
Importance Fragility fractures are a public health problem. Over 70% of women having fractures have osteopenia or normal BMD, but they remain unidentified and untreated because the definition of ‘osteoporosis’, a bone mineral density (BMD) T-Score ≤ -2.5SD, is often used to signal bone fragility.
Objective As deep learning facilitates investigation of bone’s multi-level hierarchical structure and soft tissue, we tested whether this approach might better identify women at risk of fracture before fracture.
Design We pooled data from three French and Swiss prospective population-based cohorts (OFELY, QUALYOR, GERICO) that collected clinical risk factors for fracture, areal BMD and distal radius measurements with high resolution peripheral quantitative tomography (HRpQCT). Using only three-dimensional images of the distal radius, ulna and soft tissue acquired by HRpQCT, an algorithm, a Structural Fragility Score-Artificial Intelligence (SFS-AI), was trained to distinguish 277 women having fractures from 1401 remaining fracture-free during 5 years and then was tested in a validation cohort of 422 women.
Setting European postmenopausal women
Participants We have studied postmenopausal women considered as representative of the general population, who were followed for a median 9.4 years in OFELY, 5.4 years in QUALYOR and 5.7 years in GERICO.
Main outcome and measure All types of incident fragility fractures
Results We used data from 2666 postmenopausal women, with age range of 42-94. In women ≥ 65 years having ‘All Fragility Fractures’ or ‘Major Fragility Fractures’, SFS-AI generated an AUC of 66-70%, sensitivities of 60-68% and specificity of 71%. Sensitivities were greater than achieved by the fracture risk assessment (FRAX) with BMD or BMD (6.7-26.7%) with lower specificities than these diagnostics (∼95%).
Conclusion and relevance The SFS-AI is a holistic surrogate of fracture risk that pre-emptively identifies most women needing prompt treatment to avert a first fracture.
Question Can a deep learning model (DL)° based on high resolution images of the distal forearm predict fragility fractures?
Findings In the setting of 3 pooled population-based cohorts, the DL model predicted fractures substantially better than areal bone mineral density and FRAX, especially in women ≥65 years.
Meaning Our DL model may become an easy to use way to identify postmenopausal women at risk for fracture to improve fracture prevention.
Competing Interest Statement
Ego Seeman is a stockholder of CurvebeamAI Yu Peng and Xiaoxu Li are emplyees of CurvebeamAI
Clinical Trial
OEFLY study: RCB est 2010-A00928-31 QUALYOR study: NCT01150032 GERICO study: ISRCTN11865958
Funding Statement
The OFELY study has been funded by MSD and INSERM. The QUALYOR study has been funded by the PHRC program of the French Ministry of Health and MSD. The GERICO cohort has been funded by Geneva University Hospitals and Faculty of Medicine Clinical Research Center and HUG Private Foundation.
Author Declarations
I confirm all relevant ethical guidelines have been followed, and any necessary IRB and/or ethics committee approvals have been obtained.
Not Applicable
The details of the IRB/oversight body that provided approval or exemption for the research described are given below:
CPP Sud-Est II.
I confirm that all necessary patient/participant consent has been obtained and the appropriate institutional forms have been archived, and that any patient/participant/sample identifiers included were not known to anyone (e.g., hospital staff, patients or participants themselves) outside the research group so cannot be used to identify individuals.
Not Applicable
I understand that all clinical trials and any other prospective interventional studies must be registered with an ICMJE-approved registry, such as ClinicalTrials.gov. I confirm that any such study reported in the manuscript has been registered and the trial registration ID is provided (note: if posting a prospective study registered retrospectively, please provide a statement in the trial ID field explaining why the study was not registered in advance).
Not Applicable
I have followed all appropriate research reporting guidelines, such as any relevant EQUATOR Network research reporting checklist(s) and other pertinent material, if applicable.
Not Applicable
Data Availability
Data may be made available upon reasonable request for further analysis, after examination of an analysis proposal by the investigators.
