# External validation of FRISBEE 2-year and 5-year fracture prediction models in a fracture liaison service cohort

## TL;DR

This study tested the accuracy of two fracture risk prediction models in Norwegian women and found they overestimated the risk, suggesting they need recalibration for local use.

## Contribution

The study provides external validation of the FRISBEE fracture risk models in a Norwegian cohort, highlighting the need for recalibration.

## Key findings

- The FRISBEE models overestimated fracture risk in Norwegian women with recent fractures.
- Recalibration of the models is necessary for accurate risk prediction in this population.
- The AUC values for fracture prediction were moderate, indicating limited model performance.

## Abstract

We externally validated the FRISBEE models of 2-year and 5-year fracture risk prediction in 517 women with index fractures. Both models overestimated the fracture risk. Recalibration of the FRISBEE models are needed before use in Norwegian women with recent fractures.

We externally validated the Fracture Risk Brussels Epidemiological Enquiry (FRISBEE) groups’ 2-year and 5-year fracture risk models.

We included women above 50 years with a recent fracture from the consent-based part of the Norwegian Capture the Fracture Initiative study (NoFRACT). They had bone mineral density assessed and filled in a questionnaire including risk factors for fracture at baseline between October 2015 and December 2017. We calculated and validated the 2-year and 5-year fracture risk using the FRISBEE equation models.

Of 517 women aged 65.5 ± 8.6 years with fractures, 94 (18%), 55 (11%), and 31 (6%) sustained a subsequent fracture of any type, major osteoporotic fractures (MOF), or central fracture, during 4.7 ± 1.3 years mean follow-up. The area under the receiver-operating curve (AUC) (95% confidence interval (CI)) for any type of fracture, MOF, and central fracture was 0.57 (0.51–0.63), 0.57 (0.46–0.67), and 0.65 (0.53–0.77), respectively, for the FRISBEE 2-year risk models and 0.57 (0.51–0.64), 0.58 (0.50–0.67), and 0.67 (0.57–0.76) for the FRISBEE 5-year risk models. The calibration slopes (with 95% CI) that compared observed vs. predicted probabilities for fracture across deciles of risk for any type of fracture, MOF, and central fracture were all low: 0.34 (0.02–0.64), 0.33 (− 0.09–0.74), and 0.61 (0.16–1.06), in the FRISBEE 2-year models, and 0.54 (0.13–0.95), 0.43 (0.05–0.80), and 0.69 (0.31–1.08), in the FRISBEE 5-year models.

Overall, the FRISBEE models overestimated both 2-year and 5-year fracture risk. Recalibration is needed before these models can be used in Norwegian women with recent fractures.

The online version contains supplementary material available at 10.1007/s11657-025-01516-5.