Prediction of fracture risk in men: a cohort study

Abstract

FRAX is a tool that identifies individuals with high fracture risk who will benefit from pharmacological treatment of osteoporosis. However, a majority of fractures among elderly occur in people without osteoporosis and most occur after a fall. Our aim was to accurately identify men with a high future risk of fracture, independent of cause. In the population-based Uppsala Longitudinal Study of Adult Men (ULSAM) and using survival analysis we studied different models' prognostic values (R(2)) for any fracture and hip fracture within 10 years from age 50 (n = 2322), 60 (n = 1852), 71 (n = 1221), and 82 (n = 526) years. During the total follow-up period from age 50 years, 897 fractures occurred in 585 individuals. Of these, 281 were hip fractures occurring in 189 individuals. The rates of any fracture were 5.7/1000 person-years at risk from age 50 years and 25.9/1000 person-years at risk from age 82 years. Corresponding hip fractures rates were 2.9 and 11.7/1000 person-years at risk. The FRAX model included all variables in FRAX except bone mineral density. The full model combining FRAX variables, comorbidity, medications, and behavioral factors explained 25% to 45% of all fractures and 80% to 92% of hip fractures, depending on age. The corresponding prognostic values of the FRAX model were 7% to 17% for all fractures and 41% to 60% for hip fractures. Net reclassification improvement (NRI) comparing the full model with the FRAX model ranged between 40% and 53% for any fracture and between 40% and 87% for hip fracture. Within the highest quintile of predicted fracture risk with the full model, one-third of the men will have a fracture within 10 years after age 71 years and two-thirds after age 82 years. We conclude that the addition of comorbidity, medication, and behavioral factors to the clinical components of FRAX can substantially improve the ability to identify men at high risk of fracture, especially hip fracture.

Figure 1

Flow chart describing the present study. Deaths are presented as cumulative mortality from start of survey 1. Numbers not available represent those who were not living in the Uppsala region at time of invitation. They did not contribute risk factor information at that survey but they could return for a later survey if they had moved back to Uppsala. All men were traced in patient registers for fracture data, including those “not available.” Men not participating in the clinical investigation only completed questionnaires and were not included in our analysis.

Figure 2

Fracture incidence in 50-year-old men. Kaplan-Meier failure estimates for a first fracture of any type, a first hip fracture, and first two consecutive fractures from age 50 years in Swedish men in the Uppsala Longitudinal Study of Adult Men (ULSAM).

Figure 3

Nested models explained variation in fracture risk. Explained variation (R², 95% confidence intervals) of risk of any fracture, two fractures, and hip fracture, for nested models at different ages and for different follow-up times.

Figure 4

Net reclassification improvement (NRI). NRI for prediction of any fracture (left panel) and hip fracture (right panel) when comparing the full model with FRAX variables (V_FRAX). Triangles (▴) represent NRI among events, squares (▪) NRI among nonevents, and circles (•) the combined NRI with 95% confidence interval error bars.