Table 2 Association of predicted age difference with incident fracture outcome in the derivation test set (spine radiograph cohort, n = 2063) and external test set (DXA VFA cohort, n = 3508)

From: Spine age estimation using deep learning in lateral spine radiographs and DXA VFA to predict incident fracture and mortality

 

Model 1 (age- and sex-adjusted)

Model 2 (model 1 + prevalent vertebral fracture and clinical risk factorsa)

Model 3 (model 2 + osteoporosisc)

Predictor: predicted age difference, per 1 SD increment

HR (95% CI)

p value

HR (95% CI)

p value

HR (95% CI)b

p value

Outcome: incident clinical fracture

 Derivation test set (lateral spine radiograph, VERTE-X cohort)

  Overall

1.74 (1.46–2.07)

<0.001

1.54 (1.29–1.84)

<0.001

1.71 (1.35–2.18)

<0.001

  Vertebral

1.63 (1.31–2.02)

<0.001

1.49 (1.19–1.86)

<0.001

1.55 (1.15–2.08)

0.004

  Nonvertebral

1.92 (1.49–2.47)

<0.001

1.61 (1.24–2.09)

<0.001

1.89 (1.31–2.71)

0.001

 External test set (DXA VFA, KURE cohort)

  Overall

1.28 (1.16–1.42)

<0.001

1.26 (1.14–1.40)

<0.001

1.22 (1.10–1.35)

<0.001

  Vertebral

1.41 (1.21–1.61)

<0.001

1.37 (1.19–1.59)

<0.001

1.34 (1.16–1.54)

<0.001

  Nonvertebral

1.20 (1.05–1.38)

0.006

1.19 (1.04–1.36)

0.011

1.15 (1.01–1.32)

0.042

  1. aHeight, previous history of clinical fracture or morphometric vertebral fracture in spine images, surgical prosthesis in spine images, rheumatoid arthritis, and chronic glucocorticoid use.
  2. bIn derivation test set, BMD data were available in a subset (n = 1441).
  3. cDXA areal BMD T-score −2.5 or lower at lumbar spine, femoral neck, or total hip.
Back to article page