Spine age derived from DXA VFA images predicts incident fractures and mortality: the Manitoba Bone Mineral Density Registry.
Authors
Affiliations (8)
Affiliations (8)
- Department of Biomedical Systems Informatics, Yonsei University College of Medicine, Seoul, South Korea.
- Institute for Innovation in Digital Healthcare (IIDH), Yonsei University Health System, Seoul, Korea.
- Department of Internal Medicine, Endocrine Research Institute, Severance Hospital, Yonsei University College of Medicine, Seoul, South Korea.
- George & Fay Yee Centre for Healthcare Innovation, University of Manitoba, Winnipeg, Canada.
- Department of Radiology, University of Manitoba, Winnipeg, Canada.
- San Francisco Coordinating Center, California Pacific Medical Center Research Institute, San Francisco, CA, USA.
- Department of Epidemiology and Biostatistics, University of California, San Francisco, CA, USA.
- Department of Internal Medicine, University of Manitoba, Winnipeg, Canada.
Abstract
Biological age may better predict health outcomes than chronological age by capturing individual heterogeneity in aging. We investigated whether accelerated spine aging, estimated from DXA vertebral fracture assessment (VFA) using deep learning, predicts fracture and mortality independently of age, vertebral fracture (VF), and bone mineral density (BMD). A convolutional neural network model to estimate age from lateral spine radiographs was trained in a Korean cohort (VERTE-X, n=10,341). Among 27,601 adults aged ≥50 who underwent DXA VFA in Manitoba, Canada (2010-2023), the pre-trained model was fine-tuned to DXA VFA images using 20% randomly sampled subset. Among remaining 80% set, test set included 8,810 individuals who completed DXA before 2017 as the outcomes were ascertained through 2018. Predicted spine age difference (PAD=spine age-chronological age) was calculated in the test set. During a mean follow-up of 3.9 years, 899 incident fractures and 969 deaths occurred. Spine age positively correlated with chronological age (r=0.89), with a mean difference of 0.0 years (SD=3.4). Factors associated with higher PAD include VFs (+1.02 years), nonvertebral fracture history (+0.22), generalized spine structural artifacts (+1.45), smoking (+1.20), and lower femoral neck BMD (+0.60 per T-score decrement), collectively explaining 66% of PAD variance. Each SD increase in PAD was associated with higher risk of any (adjusted hazard ratio=1.11), nonvertebral (1.10), major osteoporotic (1.12), and hip fracture (1.25), and mortality (1.12), independent of covariates (all p<0.05). In summary, accelerated spine aging detected from DXA VFA predicts fracture and mortality risk independently of age, clinical risk factors, VF, spine structural artifacts, and BMD in individuals at high risk of fracture, supporting its potential to enhance fracture risk assessment.