Fractures Attributable to Osteoporosis: Report from the National Osteoporosis Foundation
To assess the cost-effectiveness of interventions to prevent osteoporosis, it is necessary to estimate total health care expenditures for the treatment of osteoporotic fractures. Resources utilized for the treatment of many diseases can be estimated from secondary databases using relevant diagnosis codes, but such codes do not indicate which fractures are osteoporotic in nature. Therefore, a panel of experts was convened to make judgments about the probabilities that fractures of different types might be related to osteoporosis according to patient age, gender, and race. A three-round Delphi process was applied to estimate the proportion of fractures related to osteoporosis (i.e., the osteoporosis attribution probabilities) in 72 categories comprised of four specific fracture types (hip, spine, forearm, all other sites combined) stratified by three age groups (45–64 years, 65–84 years, 85 years and older), three racial groups (white, black, all others), and both genders (female, male). It was estimated that at least 90% of all hip and spine fractures among elderly white women should be attributed to osteoporosis. Much smaller proportions of the other fractures were attributed to osteoporosis. Regardless of fracture type, attribution probabilities were less for men than women and generally less for non-whites than whites. These probabilities will be used to estimate the total direct medical costs associated with osteoporosis-related fractures in the United States.
Estimating the expenditures for osteoporotic fractures has proven to be a difficult task. Medical costs related to the diagnosis and treatment of fractures generally were estimated to total $20 billion in the United States in 1988,1 but the methodological problem is to determine the proportion of the total that might be attributable to osteoporosis. Because most fractures result directly from some traumatic event,2 it would be an overestimate to attribute all fractures in the community solely to the influence of osteoporosis. However, overwhelming trauma (e.g., motor vehicle accidents) and specific pathological processes (e.g., metastatic malignancies) are relatively uncommon, accounting for only about 11% of all hip fractures,3 17% of vertebral fractures,4 and 8% of distal forearm fractures in one community,5 and most fractures are related at least in part to the low bone density6,7 that characterizes osteoporosis in vivo.8 However, to ignore all fractures except those of the hip, as is often done,9 leads to an underestimate of expenditures. To assess the cost of diseases like diabetes mellitus,10 it has been possible to count those adverse events (e.g., hospitalization for myocardial infarction) where diabetes is a second listed diagnosis. This is not possible for osteoporosis, which is rarely listed in conjunction with its associated fractures. For example, only 6% of hospital discharges for hip fracture in the United States in 1992 had an associated diagnosis of osteoporosis (Medical Technology and Practice Patterns Institute, unpublished data from the 1992 National Hospital Discharge Survey) despite the fact that most hip fractures occur among elderly individuals whose bone mass is low enough to be considered osteoporotic.11
Therefore, a panel of expert clinicians with broad experience in the diagnosis and treatment of patients with osteoporotic fractures was convened to assess the contribution of osteoporosis to specific types of fractures among different populations residing in the United States. Using the Delphi approach, they estimated the probability that each of 72 categories consisting of four fracture types (hip, spine, forearm, all other sites combined), three age groups (45–64 years, 65–84 years, 85 years and older), three racial groups (white, black, all others), and both genders (female, male) are associated with osteoporosis. These probability estimates are used to determine the direct medical costs of osteoporotic fractures in the United States in 1995.12
The Delphi technique has been described as “a method for structuring a group communication process so that the process is effective in allowing a group of individuals, as a whole, to deal with a complex problem.”13 Because the expertise and opinion of each member of the Delphi panel is highly valued, the technique was designed to mitigate the impact that particular individuals might have in more traditional face-to-face communication strategies. While the Delphi approach was originally developed by the Rand Corporation to incorporate various opinions from experts in making predictions about national defense needs,14 the technique has been applied extensively in the medical arena where empirical data are insufficient.15,16 For example, the Delphi technique has been used to obtain estimates of influenza epidemic parameters,17 the probabilities of various outcomes among patients with isoniazid-resistant tuberculosis infections,18 the likelihood of HIV exposure from different sexual practices,19 the risks and benefits of measles vaccine for HIV-infected children,20 and the costs and outcomes related to prenatal screening and immunization for hepatitis B virus.21 The Delphi technique has also been widely used to provide guidelines for improving clinical practice, including indications for commonly used medical and surgical procedures,22 as well as assessments of the extent to which current practices are appropriate.23,24 Other applications have involved designing a clinical grading scale to predict hyperthermia,25 improving the interpretation of electrocardiograms26 and radiographic studies,27 and specifying the procedures for brain imaging.28 Finally, the Delphi process has been used to develop guidelines for clinical trials29,30 and to determine resource-based relative value scales.31 Use of the Delphi technique has been validated in several of these applications,13,32,33 including the landmark Rand study in 1964.14 As suggested by its increasingly versatile use in the health care field over the past three decades, the Delphi technique is recognized as a systematic, literature-based, scientific method that utilizes expert group judgment in the absence of adequate data.
Design of the Delphi process
The Delphi method was selected to develop a group judgment on the probabilities that specific fractures in different age, gender, and ethnic groups are associated with osteoporosis. An expert panel was convened under the auspices of the National Osteoporosis Foundation, a nonprofit patient advocacy group established in 1984. The Osteoporosis Delphi process comprised three iterative rounds based on the observation that more than three rounds leads to diminishing returns.13 An Osteoporosis Attribution Probability Response Form was designed to accommodate the 72 age-race-gender-fracture–type categories of interest. The issues to be addressed by the panel and the Response Form to be completed were presented in each round. Round I (initial estimate of osteoporosis attribution probabilities) was completed independently by each panel member by mail, as is typically done with this approach. The six clinicians on the expert panel then met for a day in Washington, DC, to review results from the first round and to hear a presentation of relevant data. Thus, the standard Delphi approach34 was modified to permit face-to-face discussions. Rounds II and III (second and final estimations of osteoporosis attribution probabilities) included a more in-depth discussion of panel assumptions in determining estimations. A detailed description of these three stages is presented below.
To determine the degree of confidence with which the panel estimated each of the 72 attribution probabilities, each panelist was also asked to associate a validity score with each final probability estimate. The scores were based on the following validity scale: 1 = certain, low risk of the attribution probability being wrong (±5%); 2 = reliable, some risk of being wrong (±10%); 3 = risky, substantial risk of being wrong (±20%); or 4 = unreliable, great risk of being wrong (> ±20%). The group validity scores are important in using the attribution probabilities in future health services research applications.
Selection of the expert panel
The size and composition of a Delphi panel are crucial to its success since the results are based on the combined expert knowledge of its members. The panel must reflect relevant perspectives on an issue while permitting a complete and free exchange of views among all concerned in a relatively short period of time.13 With these limitations in mind, the six-member panel of expert physicians was selected on the basis of the following criteria: extensive experience in treating patients with osteoporosis, prominence in research or health policy issues relevant to osteoporosis, geographic balance, and representation from the medical and surgical specialties primarily involved. The final panel, listed at the end of the report, was composed of clinicians in the fields of internal medicine, endocrinology, rheumatology, orthopedic surgery, and nuclear medicine. The work of the six-member expert panel was facilitated by a chairman, with a background in epidemiology and health services research, and by three senior members of a nonprofit health services research organization who were responsible for designing the necessary forms and analyzing the results.
Description of the three Delphi rounds
The Delphi process was carried out in three stages: Round I was completed in advance of and Rounds II and III were completed during the Delphi committee meeting in Washington, DC, on June 15, 1995.
Round I—Initial estimation of osteoporosis attribution probabilities: The first round was conducted by mail prior to the Delphi Committee meeting. Based upon their clinical knowledge and expertise, each member of the expert panel was asked to estimate the proportion of fractures related to osteoporosis for each of the 72 fracture-age-race-gender categories specified. These attribution probabilities were recorded in the appropriate boxes on an Osteoporosis Attribution Probability Response Form in increments of 0.05 (i.e., 0.05, 0.10, 0.15, etc.). In addition, participants outlined the key assumptions underlying their attribution probabilities on a separate form. To assist in the initial estimates, the panel members were provided with the rates of hospitalization and outpatient physician visits in the United States for treatment of fractures of the hip, spine, forearm, and all other sites combined in 1992 by age, race, and gender.
Round II—Second estimation of osteoporosis attribution probabilities: Prior to the second round, the modal probabilities and associated ranges were calculated for each of the 72 fracture-age-race-gender categories, and the underlying assumptions of the panel members were summarized. Each panelist had an opportunity to review the modal probabilities, the range of probabilities, and the underlying assumptions. At this point, published data on fracture incidence by age, gender, and ethnic group, as well as the excess of fractures in each group over and above the rates seen in young individuals (i.e., an assessment of the extent to which fractures at different skeletal sites are age-related) were presented.35–39 Because this approach underestimates the contribution of osteoporosis to some limb fractures,6 data were also presented on the association of different types of fracture with bone density after adjustment for age (i.e., an assessment of the extent to which different fractures are related to osteoporosis). The latter data were only available for fractures among elderly white women.6,40 Discussion then focused on the attribution probabilities from Round I, where differences in opinion seemed to be the greatest. Upon completion of this discussion, each member of the expert panel developed a new set of attribution probabilities by again completing the Osteoporosis Attribution Probability Response Form.
Round III—Final estimation of osteoporosis attribution probabilities: Prior to the third round, the modal probabilities and associated ranges were again calculated for each of the 72 fracture-age-race-gender categories using the estimates assigned by each panel member during Round II. Another discussion ensued regarding the areas where most disagreement on the probabilities remained. To complete Round III, each panelist assigned a final probability value for the 72 different categories. The median attribution probabilities from this final stage were calculated and redistributed to panel members who in turn ranked each of the final 72 probabilities according to the numeric validity scale described above. The goal of this task was to document the degree of certainty associated with the final attribution probability for each of the 72 fracture-age-race-gender categories.
Panelists were given an opportunity to list the major assumptions they used to estimate the attribution probabilities for each subpopulation and fracture type of interest. The assumptions were used in two important ways: first, to assist each panel member in estimating the probabilities in as systematic a manner as possible and, second, to provide an organized approach for discussing disagreements between members in order to reduce the discrepancy in probability estimates at each stage. The various assumptions that were recorded, organized by general topic area, are presented in Table 1. It must be noted that, although the assumptions were shared by the majority of the panel, not all members agreed with every assumption. Panelists generally assumed that women have more fractures than men after age 45 and that a greater proportion of them are osteoporotic in nature. It was agreed that elderly patients have more osteoporotic fractures than middle-aged patients, although differences exist by fracture type. The panel overwhelmingly believed that whites have higher rates of osteoporotic fractures than non-whites, but an attempt was made to distinguish the lower fracture rates among non-white populations from the proportion of the observed fractures that might be attributable to osteoporosis in these groups. However, the paucity of relevant data for non-white populations was recognized. The last set of assumptions made by the panel addressed issues related to fracture type.
Table Table 1. Initial Assumptions Consideredin the Determination of Osteoporosis Attribution Probabilities
OSTEOPOROSIS ATTRIBUTION PROBABILITIES
The expert panel's final osteoporosis attribution probabilities (median value and range) for white, black, and other populations are shown in Tables 2, 3, 4. For example, it was estimated that 90% of proximal femur fractures among white women 65–84 years of age are related to osteoporosis. A somewhat lower proportion of hip fractures among white men was counted, (80%), presuming that most of the hip fractures that do occur among these elderly men are related to osteoporosis even though the actual risk of a hip fracture is much less in men than in women. However, there was less certainty in the estimate for men compared with women in this age group (validity score 1.8 vs. 1.2). At least 90% of vertebral fractures were attributed to osteoporosis among older white women, and the proportion among elderly white men was judged to be similar since spine fractures were assumed to be more closely linked with osteoporosis than hip fractures. Consequently, the attribution probabilities for spine fractures among white women and men were all ≥70% and were also in this range among elderly women and men of other races. Much smaller proportions of forearm fractures and all other fracture types combined were considered to be related to osteoporosis. Regardless of fracture type or race or gender, the panel members agreed that attribution probabilities generally increase with age. As expected also, the attribution probabilities were less for men than women in almost every instance, and were less for non-whites than for whites in most categories. However, the confidence assigned to these estimates was less in men and non-whites. Indeed, because of the dearth of data on some of these populations, many of the attribution probabilities were judged to have a potential error factor of ±20% or more.
Table Table 2. Final Osteoporosis Attribution Probabilities by Fracture Type, Gender, and Age: White Population
Table Table 3. Final Osteoporosis Attribution Probabilities by Fracture Type, Gender, and Age: Black Population
Table Table 4. Final Osteoporosis Attribution Probabilities by Fracture Type, Gender, and Age: Other Race* Population
It is difficult to specify the fractures attributable to osteoporosis, and therefore the cost of treating the condition, because every fracture results from the interplay of bone strength with skeletal loading.41 Bone strength is influenced by a variety of factors2 but is strongly correlated with bone mineral density,42–46 and numerous studies demonstrate that bone density measurements predict fracture risk.47–52 Fractures do not occur, though, until the loads encountered in the course of everyday activities or with specific episodes of trauma, mostly falls, exceed the breaking strength of bone.41 Thus, it is impossible to assign responsibility for a given fracture specifically to insufficient bone strength (osteoporosis) or to excessive skeletal loading because both factors are at work in each case. Even when the degree of trauma is considerable,53 fractures generally occur in a setting of low bone density, and detailed investigation indicates that bone density is an important determinant of risk.11 Indeed, recent studies indicate that most limb fractures are related to bone density among elderly white women.6,7 Comparable data are not available for non-white women or for men, and the absence of empirical data precludes an objective assessment of the contribution of osteoporosis to the fractures that occur among these diverse populations.
In the absence of empirical data, we attempted to address this issue by using the Delphi method to estimate the probability that fractures of a given type are related to osteoporosis. As noted above, the Delphi approach is a widely used and validated method to make such estimations. An earlier osteoporosis expert panel used a similar approach and judged that the osteoporosis attribution probabilities for hip fractures among white women aged 45–59, 60–74, and ≥75 years were 0.51, 0.71, and 0.91, respectively.54 Our estimates of 0.80, 0.90, and 0.95, respectively, for hip fractures among white women were higher, but this can probably be accounted for by the older age groups (e.g., 45–64, 65–84, and ≥85 years) that we used. Our estimates for spine fractures were also greater than those from the previous panel of 0.72, 0.75, and 0.75 for white women aged 45–59, 60–74, and ≥75 years, respectively. Their estimates for distal forearm fractures (0.70, 0.78, and 0.84, respectively) were very close to those for white women in this report (0.70, 0.70, and 0.80, respectively, for age groups 45–64, 65–84, and ≥85 years) because forearm fracture rates change little with age. The previous panel made no estimates for non-white women or for men. Moreover, validity scores were not used by the earlier expert panel, which further hinders direct comparison with our results. In the absence of a gold standard, it is not possible to rigorously validate any of these osteoporosis attribution probabilities. Nevertheless, they are needed to provide a basis for cost-effectiveness analyses which might provide a health care policy perspective on interventions designed to prevent osteoporotic fractures.