The search for "risk factors" for various diseases is the primary goal of much observational research. The value in examining risk factors may be identifying causes of disease, understanding relationships among various factors reported to be involved in the disease process, and defining subgroups of patients at unusual risk who might require special attention.

In this issue of Annals, Bauer and colleagues [1] report on risk factors for low bone mass in women older than 65 years. Cauley and colleagues [2] examine the association between thiazide treatment, bone mass, and fractures in women older than 65 years. The Study of Osteoporotic Fractures [1, 2] is the largest completed investigation of factors associated with fractures. Because of the enormous statistical power of the study, the authors were able to consider simultaneously a larger number of potential risk factors than had been possible in earlier studies. How should the results of this study and others dealing with risk factors [3-5] be used in clinical practice?

First, can risk factors be used to identify patients who may need further study? Many investigators have tried to address this question [3-5]. The answer is a qualified "no". Using the risk factors identified by Bauer and colleagues [1], patients with lower body weight, weaker grip strength, and gastric surgery would be expected to have much lower bone mass than obese patients with non-insulin-dependent diabetes mellitus (type 2 diabetes) who are taking thiazides. Thus, at the extremes of predicted values, physicians could be more secure in estimating risk for some patients compared with others. However, the more important issue is how any combination of risk factors compares with an established diagnostic test. Bone mass measurements are accurate within 1% to 2% and provide good estimates of fracture risk [6-8]. Recently, the Studies of Osteoporotic Fractures group [9] showed that a one standard deviation difference in femoral neck density is associated with a 2.6-fold change in hip fracture risk. In the absence of measurements of bone mass, however, risk factors may provide a basis for estimating bone mass, but these estimates will not be as accurate as direct measurements and their value in fracture prediction remains unproven.

Knowledge of risk factors may be useful in other ways. For patients who are premenopausal (and therefore not likely to need therapeutic interventions), physicians may use this information to allay unfounded fears (for example, to not worry about drinking 2 to 3 cups of coffee per day). To the extent that risk factors might be thought to be associated with rates of bone loss (for example, estrogens or smoking), physicians might use this information to change the risk profiles of their patients. Risk factors may be used to tailor the decision to institute preventive interventions in a specific patient with bone mass at a certain level. Physicians may choose to treat those with femoral neck bone mass more than one standard deviation below the mean, to not treat anyone with bone mass at the mean or above, and to re-evaluate those between these values. A patient near the cutoff for low bone mass with a negative risk factor profile (for example, thin, smoker, positive maternal history of hip fracture, poor strength) might be a more appropriate candidate for estrogen therapy than someone with identical bone mass but who was obese, had type 2 diabetes, and was taking thiazide diuretics.

Physicians might offer bone mass measurements to each patient who would consider a therapy to prevent further bone loss (without regard for risk factors). A detailed description of such an approach has been published by an expert panel of the National Osteoporosis Foundation [10]. However, even in these circumstances an understanding of risk factors might influence the decision to institute aggressive preventive measures. For example, the increasing risk for hip fracture associated with increasing age probably is due to a greater likelihood of falling in older patients. Comparing a 50-year-old woman and an 80-year-old woman with identical, low femoral neck bone mass values, estrogen treatment might be more appropriate for the menopausal woman, who is likely, if not treated, to have substantial bone loss and who will spend many more years at risk and, therefore, has a greater future probability of fracture. The recent results [11] of a clinical trial show that external hip padding might be more appropriate in the prevention of hip fractures for the older patient, whose rate of bone loss is probably quite slow but whose probability of falling is much higher.

The value of risk factors in making therapeutic decisions depends on many issues. First, how strong is the risk factor? This is not a reflection of statistical significance but rather the magnitude of change in an outcome associated with a change in the risk factor. Second, how common is it? Consider gastric surgery in the article by Bauer and colleagues [1]. Although a strong risk factor (its presence was associated with 9% lower bone mass), its prevalence was only 2%, and thus from a clinical perspective it will only rarely influence decisions. In contrast, the effect of weight can be calculated for each patient. Third, can the risk factor be altered or would it be desirable to do so? Despite the apparent skeletal benefits of type 2 diabetes, reported previously by others [12, 13] and confirmed by Bauer and colleagues [1], it is an undesirable characteristic for other reasons. Weight loss since age 50 is associated with decreased bone mass. Should physicians then discourage patients without weight-related health problems who are interested in the health and cosmetic benefits of weight loss?

The results of the Bauer study are particularly valuable for the statistical power available to determine, with greater precision than has been possible so far, which factors do not influence risk for low bone mass. Northern European ancestry, blond hair, breast-feeding, modest alcohol intake, occasional antacid use, and many other characteristics previously considered possible risk factors did not meet even minimal criteria for independent risk. Although the authors recognize the limitations inherent in observational studies (for example, heavy users of alcohol are unlikely to be included), little question exists that within the ranges of exposure of the women in this study, these factors have small or no effects on bone mass. The removal of these elements from the assessment of risk may enable physicians to focus their efforts better and to relieve unfounded concerns.

Some factors not associated with bone mass might still influence fracture risk through other mechanisms. For example, occasional heavy use of alcohol might influence the risk for falls and fractures without influencing bone mass. In contrast, other factors positively associated with bone mass might have further beneficial effects on fracture risk. Greater body weight might be associated not only with increased bone mass but also with reduced trauma to the skeleton during falls [14].

The best assessment of patient risk for future fractures comes from direct assessments of bone mass [6-9] and perhaps from other skeletal characteristics available from these same scans [15]. Data about other risk factors can enable physicians to fine-tune preventive interventions.