Validation of a 5-year risk score of hip fracture in postmenopausal women. The Danish Nurse Cohort Study
- First Online:
- Cite this article as:
- Hundrup, Y.A., Jacobsen, R.K., Andreasen, A.H. et al. Osteoporos Int (2010) 21: 2135. doi:10.1007/s00198-010-1176-7
We evaluated the Women’s Health Initiative (WHI) hip fracture risk score in 15,648 postmenopausal Danish nurses. The algorithm was well calibrated for Denmark. However, the sensitivity was poor at common decision making thresholds. Obtaining sensitivity better than 80% led to a low specificity of 61.4%.
A new score based on data from the WHI has been designed to predict 5-year risk of hip fracture in postmenopausal women. The performance of the algorithm has not been validated in populations with different lifestyle characteristics and ethnicity. The aim of this study was to test the clinical performance of the algorithm in a large Danish cohort of postmenopausal Caucasian women against hip fracture.
The Danish Nurse Cohort is a prospective risk factor and hormone therapy (HT) study established in 1993. Participants in the present analysis were 15,648 postmenopausal nurses. The calibration and diagnostic performance of the WHI algorithm was evaluated using fracture events captured in the Danish National Hospital Registry.
During 5 years of follow-up, 122 participants suffered a hip fracture (1.8/1,000 person years). The WHI algorithm predicted that 107 hip fractures would occur, indicating an underestimation of the number of fractures by 12%. To obtain sensitivity better than 80%, the cut-off value for 5-year risk was set to 0.5%, which was accompanied by a low positive predictive value of 1.9% and a low specificity of 61.4%. The algorithm predicted too many fractures in HT-users (12 observed, 22 expected) and too few in non HT-users (107 observed, 81 expected).
While the WHI algorithm was well calibrated on the Danish population, the clinical utility of the WHI algorithm in Danish postmenopausal women was limited by poor sensitivity at common decision-making thresholds and suboptimal in non-HT-users.