These investigators used data from nearly 100,000 women participating in the observational component of the WHI to identify eleven risk factors that were strong and significant predictors of hip fractures. These are listed in Table 1 (p. 22) and include older age, worse self-reported health, height above 64 inches, weight less than 200 pounds, fracture after 54 years of age, white race, physical inactivity, current smoking, parental history of hip fracture, corticosteroid use, and use of a hypoglycemic agent. Many of these risk factors have been well described in prior literature. These factors were then combined using a scoring system (available in a calculator form at http://hipcalculator.fhcrc.org) and compared with the five-year fracture rates.
The clinical prediction rule was found to be a good predictor of hip fractures within five years with areas under the ROC curve (AUC) of approximately 0.80. The performance of the clinical prediction rule varied by the predicted five-year risk of hip fracture. For example, at low five-year predicted risk value (i.e., 0.1%), the rule was very sensitive but less specific. At higher five-year predicted risk values (i.e., 1%), the rule was very specific but less sensitive. The predictive ability of the rule was tested against BMD in a subset of women and found to have a slightly lower AUC, 0.71 versus 0.79. As expected, combining the clinical prediction rule with BMD information gave the best predictive information.
This study was well conducted and pursues an important clinical area. There are many non-BMD risk factors that should be assessed when considering how to counsel women (and men) about their risk of a future fracture. This study informs us about some important factors and gives precise information about combining these factors for a precise estimate of risk. The clinical prediction rule and its calculation can easily be accessed and used via the Internet. However, because most drug treatment trials have not used such a risk calculator to define included populations, it is unclear how treatment will affect the calculated risks. I look forward to the day when clinical prediction rules are combined with treatment trials to better guide our therapeutic decision-making.
