Low BMD is less predictive than reported falls for future limb fractures in women across Europe: results from the European Prospective Osteoporosis Study

We have previously shown that center- and sex-specific fall rates explained one-third of between-center variation in upper limb fractures across Europe. In this current analysis, our aim was to determine how much of the between-center variation in fractures could be attributed to repeated falling, bone mineral density (BMD), and other risk factors in individuals, and to compare the relative contributions of center-specific BMD vs. center-specific fall rates. A clinical history of fracture was assessed prospectively in 2451 men and 2919 women aged 50-80 from 20 centers participating in the European Prospective Osteoporosis Study (EPOS) using standardized questionnaires (mean follow-up = 3 years). Bone mineral density (BMD, femoral neck, trochanter, and/or spine) was measured in 2103 men and 2565 women at these centers. Cox regression was used to model the risk of incident fracture as a function of the person-specific covariates: age, BMD, personal fracture history (PFH), family hip fracture history (FAMHIP), time spent walking/cycling, number of 'all falls' and falls not causing fracture ('fracture-free') during follow-up, alcohol consumption, and body mass index. Center effects were modeled by inclusion of multiplicative gamma-distributed random effects, termed center-shared frailty (CSF), with mean 1 and finite variance theta (theta) acting on the hazard rate. The relative contributions of center-specific fall risk and center-specific BMD on the incidence of limb fractures were evaluated as components of CSF. In women, the risk of any incident nonspine fracture (n = 190) increased with age, PFH, FAMHIP, > or =1 h/day walking/cycling, and number of 'all falls' during follow-up (all P < 0.074). 'Fracture-free' falls (P = 0.726) and femoral neck BMD did not have a significant effect at the individual level, but there was a significant center-shared frailty effect (theta = 0.271, P = 0.001) that was reduced by 4% after adjusting for mean center BMD and reduced by 19% when adjusted for mean center fall rate. Femoral trochanter BMD was a significant determinant of lower limb fractures (n = 53, P = 0.014) and the center-shared frailty effect was significant for upper limb fractures (theta = 0.271, P = 0.011). This upper limb fracture center effect was unchanged after adjusting for mean center BMD but was reduced by 36% after adjusting for center mean fall rates. In men, risk of any nonspine fracture (n = 75) increased with PFH, fall during follow-up (P < 0.026), and with a decrease in trochanteric BMD [RR 1.38 (1.08, 1.79) per 1 SD decrease]. There was no center effect evident (theta = 0.081, P = 0.096). We conclude that BMD alone cannot be validly used to discriminate between the risk of upper limb fractures across populations without taking account of population-specific variations in fall risk and other factors. These variations might reflect shared environmental or possibly genetic factors that contribute quite substantially to the risk of upper limb fractures in women.