As in other studies, we found a low rate of BMD testing in high risk patients. Only 11% (13% of women and 5% of men) of patients having suffered from a fragility fracture were tested for osteoporosis or were tested to monitor response to therapy in the two years following their fracture. These results are comparable to the 0 to 32%  and 1 to 32%  frequencies of testing post-fragility fracture reported in large review papers. In our study, we found that BMD testing is not completely explained by a patient's physical access to densitometers. Gender, age and social status also seem to influence the rate of BMD testing.
Further, we found a moderate rate of OP treatment after a fragility fracture. A total of 25% (30% of women and 10% of men) of patients received an OP treatment in the year following the fracture. This is comparable with results from a review paper, which reported rates of bisphosphonate use ranging from 0.5% to 38% .
Finally, we found regional disparities in BMD testing according to the socio-demographic variables examined. For instance, age-adjusted BMD testing rates found in our study were statistically different among regions, varying from 0.3 to 16.1% for women and from 0 to 6.1% for men.
Hajcsar et al  had previously depicted gender differences in the investigation frequencies of osteoporosis. They reported a 1-year BMD testing rate of 24% for women and only 8.3% for men. These differences between our results and those reported by Hajcsar et al can be attributed to population and study design differences.
As reported in Jaglal , the accessibility to BMD testing seems to be strongly related to the use of a BMD device. Our results however show that the distance from residence to the nearest BMD device is neither the only nor the most important predictor of BMD testing. Younger age, previous BMD testing, and previous OP treatment use increased the likelihood of BMD testing, whereas social status, such as living in a long term care facility or being recipient of a guaranteed income supplement, decreased the likelihood of BMD testing. This last observation, not reported in other papers, was obtained by using a unique governmental data base that permits analysis of the above-mentioned social variables.
In the review papers cited above, patient characteristics were reported to influence BMD testing. In a study at two primary care practices affiliated with an academic medical center, Solomon et al found that some patient variables significantly lowered the probability of a physician's adherence to local guidelines (including BMD testing and OP treatment). Old age and young age as opposed to middle age, male sex, black race, and having more than one comorbid condition were associated with a decreased likelihood of undergoing a BMD testing or receiving an OP medication . Our results are concordant with these reported findings concerning old age and gender but not regarding comorbidity. Comorbidity, as defined by the number of distinct medications taken in the year preceding the fracture, was negatively correlated with BMD testing for women, but was positively correlated with future OP treatment. In Solomon et al , these two outcomes were merged, although the authors argued that they had observed similar results when they were examined separately. With regards to comorbidity, the difference between our results and those reported by Solomon et al are likely due to differences in the methodology used and populations studied. For instance, comorbidity was defined differently in the two studies. Moreover, our analysis was population-based whereas Solomon et al studied patients seen in primary care practices affiliated with an academic medical center. Health care provided in academic centers may not be representative of the care provided to the general population.
Regarding therapy used to prevent recurrent fractures, our results showed that positive predictors for receiving treatment for both men and women were vertebral fracture, prior BMD testing, prior OP treatment, and long-term glucocorticoid use, as defined in the method section. Men with wrist fractures and women with hip or humerus fractures were less likely to be treated than individuals with fractures at other sites. These results are concordant with those of other studies [21, 29]. The very low BMD testing rate in patients living in long-term care facilities suggests an important care gap in the management of OP in those facilities.
One might argue that patients may have died before receiving BMD testing or OP treatment. This would explain the lower rate of care observed in long term care facilities patients. In order to verify the stability of the associations between predictors and outcomes, we performed analyses on surviving patients during the two years following their fracture. These models showed similar results in predictors of BMD testing and OP treatment, except for site of fracture, for which we observed lower odds ratios. Exclusion of other predictors, such as prior BMD testing and prior OP treatment showed similar results between predictors and outcomes, except for long-term glucocorticoid use. Indeed, since patients who had received a long-term glucocorticoid therapy may have been already tested and/or treated for OP, the removal of these variables increased the associated OR.
We observed high death rates of 23% in women and 35% in men 2 years following fragility fracture. This is a 3-fold increase in death rates compared to those observed in the general elderly population for the year 2001 . The 1-year death rate of the population aged 65 or older in Quebec in 2001 was 3.9% for women and 4.9% for men. An increase in mortality rates following fragility fracture was also reported in other studies [11, 12].
A major strength of this study is its population based design. Moreover, to obtain a more comprehensive model explaining BMD use, we included a geographical distance variable in the modeling process. The major limitation of this study is inherent to the use of administrative databases. On the one hand, we may have underestimated the incidence of fragility fractures as some of these, such as vertebral fractures, are notoriously under-diagnosed by physicians . We may also have overestimated incidence of fragility fractures since we inferred that all fractures not associated with a high velocity trauma were fragility fractures. However, in an unpublished pilot study, we found a positive predictive value of 79% for fragility fracture diagnosis based on this definition . One of the limitations regarding the use of the guaranteed income supplement as a social status indicator includes a possible under-estimation of the number of beneficiaries in this category, as nearly 31% of individuals in Canada who are admissible for a guaranteed income supplement do not benefit from it . Furthermore, as we have excluded fragility fractures within the 2 years prior to the index date we may have excluded at risk patients. Unfortunately, we are limited by the fact that data before the year 1997 are unavailable. It is well known that physician characteristics may affect the incidence of BMD testing [27, 28]. Although we acknowledge the importance of studying physician characteristics in the modeling process, we were unable to retrieve such data for technical reasons.