Evaluation of Clinical Decision Rules for Bone Mineral Density Testing among White Women

Background. Osteoporosis is a devastating, insidious disease that causes skeletal fragility. Half of women will suffer osteoporotic fractures during their lifetimes. Many fractures occur needlessly, because of inattentiveness to assessment, diagnosis, prevention, and treatment of osteoporosis. Study Purpose. Study Purpose. To evaluate the discriminatory performance of clinical decision rules to determine the need to undergo bone mineral density testing. Methods. A nationally representative sample from the Third National Health and Nutrition Examination Survey consisted of 14,060 subjects who completed surveys, physical examinations, laboratory tests, and bone mineral density exams. Multivariable linear regression tested the correlation of covariates that composed the clinical decision rules with bone mineral density. Results. Increased age and decreased weight were variables in the final regression models for each gender and race/ethnicity. Among the indices, the Osteoporosis Self-Assessment Tool, which is composed of age and weight, performed best for White women. Study Implications. These results have implications for the prevention, assessment, diagnosis, and treatment of osteoporosis. The Osteoporosis Self-Assessment Tool performed best and is inexpensive and the least time consuming to implement.


Introduction
Osteoporosis is a devastating, insidious chronic disease that causes skeletal fragility. Half of women and one-eighth of men will suffer osteoporotic fractures during their lifetimes [1][2][3]. e occurrence of disability following osteoporotic hip fracture exceeds that of stroke, heart disease, or cancer and oen leads to a profound forfeiture of independence [4][5][6][7][8]. Among people one year aer the occurrence of hip fracture, seven in 10 are unable to walk independently, eight in 10 are unable to perform instrumental functions such as driving or shopping, and one-third of those residing in a nursing home had been living in a residence other than a nursing home prior to hip fracture [9][10][11]. Sequelae of hip fractures, such as pneumonia and pulmonary embolism, are frequently lethal [1]. One in four people die during the year subsequent to hip fracture, and one-third of these deaths are attributable to hip fracture [9,12,13]. Indeed, falls are the leading cause of injury-related death among people aged 65 years and older in the US [14]. ough men have a lower incidence of hip fractures than women, they are twice as likely to die in the year aerward [15].
People are also living longer. e World Health Organi-�ation�s �WHO� health pro�le of the United States reports the life expectancy for women to be 81 years based on 2010 data [16]. Consequently, with the projected growth in the 65 and older population, the number of hip fractures will increase. Osteoporosis plays a role in 90 percent of all hip fractures and 45% of all adults who present with hip fracture have had a prior fracture [17]. Many of these fractures occur in women who have been undiagnosed and untreated for osteoporosis and many of these fractures can be prevented [18].
e dismal rates of morbidity and mortality associated with osteoporosis extract an enormous toll on our economy. Annual direct costs of osteoporotic fractures in the United States were estimated at $17 billion dollars in 2005 with cumulative costs over the next two decades to exceed $474 billion [18,19]. As "baby boomers" age, the osteoporosis "time bomb" may explode. In the Framingham Study, the risk of hip fractures proliferated with each successive generation [20]. Modeling of future incidence and prevalence rates of osteoporotic fractures indicates that by 2025, fractures are projected to grow by more than 48% to greater than 3 million [21].
Osteoporosis screening with bone mineral density (BMD) testing can lead to timely diagnosis, effective medical management, and prevention of fractures [22][23][24][25]. e National Osteoporosis Foundation established guidelines for patient selection for BMD testing based on identi�cation of major risk factors [26] and re�ned them in 2008 [27]. Some family physicians were aware of but did not use clinical practice guidelines for osteoporosis [28]. ese clinicians criticized existing guidelines as too complex. Hence, the transformation of these guidelines from theory into practice was lacking. Primary care providers oen fail to recognize patients at risk for this insidious disease and thereby fail to prescribe treatment [29,30]. Alternately, family physicians appealed for more succinct, practical guidelines and expressed enthusiasm for a clinical decision rule for BMD testing [28].
Several clinical decision rules for referral for BMD testing have been proposed [31][32][33][34][35][36]. Among these, the Osteoporosis Self-Assessment Tool (OST), comprised of age and weight as the only variables, is the most succinct [33]. Studies have demonstrated the selectivity of the OST as a clinical decision rule for referral for BMD measurement, and the Surgeon General's report on bone health and osteoporosis recommended that clinicians consider using a chart version of the OST, or "Chart OST" [33,[37][38][39][40][41][42]. Two other relatively simple clinical decision rules for referral for BMD testing are the Osteoporosis Risk Assessment Instrument (ORAI), which is comprised of age, weight, and estrogen use, and the Age, Body Size, No Estrogen (ABONE), which is comprised of age, weight, and estrogen or oral contraception use [31,36]. Cadarette, et al. [43] reported that the ORAI outperformed the National Osteoporosis Foundation guidelines, but the ABONE was less sensitive. e purpose of this study was to evaluate the discriminatory performance of the OST, ORAI, and ABONE as clinical decision rules for referral for BMD testing in a nationally representative sample.
eoretical Framework. Clinical decision rules are evidencebased criteria to assist health care providers in making decisions about screening, diagnosis, treatment, or prognosis [44]. Such rules are based on patient history, physical examination, and diagnostic tests. Essential characteristics of valid clinical decision rules are generalizability and practicability [45]. Sensitivity, speci�city, positive and negative predictive values, likelihood ratios, and the area under the receiver operator characteristic curve (ROC) are measures of predictive power.
Even a clinical decision rule with high predictive power will be ineffective if it lacks acceptance by health care providers. Simplicity and practicability are essential characteristics. e OST includes only two variables, age and weight; it is practical, because primary care providers routinely collect these data and a chart of the index is available for women. e ORAI and the ABONE include age, weight, and estrogen use.

Methods
e Department of Health and Human Services (DHHS), Centers for Disease Control and Prevention (CDC), designed and conducted the ird National Health and Nutrition Examination Survey (NHANES III). Data collection methods included structured surveys by trained interviewers, physical examinations, and diagnostic tests. Participants aged 2 through 17 years completed the Household Youth questionnaire; participants aged 17 or older completed the Household Adult questionnaire. Interviews were conducted in either English or Spanish. Survey questions elicited demographic, socioeconomic, dietary, and medical history data. Physical examination teams operated in mobile examination centers and consisted of physicians, dentists, dieticians, medical technologists, radiological technologists, sonographers, health interviewers, home examiners, and coordinators. e DHHS implemented exhaustive quality control standards that minimized threats to internal validity.
e DHHS employed a strati�ed, multistage random sampling design. Primary sampling units consisted of 2,812 counties, parishes, and cities in the US [46]. Aer strat-i�cation, the sample included 81 primary sampling units at 89 survey locations widespread across the US In three stages, the survey questionnaire was completed by or for 13,944 individuals aged two months to 16 years and 20,050 individuals aged 17 years or greater [46]. e overall random sample was 33,994 from 39,696 eligible individuals for a response rate of 85.6 percent, which minimized nonresponse bias [46,47]. Of these, 30,818, or 90.7 percent, completed a follow-up physical examination [46,47]. e DHHS oversampled minorities, older people, and children, to increase the statistical power for analyzing these data, and provided sampling weights for each stratum. Exclusion criteria were (a) institutionalized civilians, (b) noncivilians, and (c) residency in the nonconterminous states, Alaska, and Hawaii.
ese data collection and sampling methods resulted in cross-sectional data that were valuable for both descriptive and analytical purposes.  , 1997), were exported to respective Excel �les. e data were then (1) cross-referenced, (2) matched by subjects using their unique identifying numbers, (3) merged into a single Excel �le, and (4) imported into an S�SS �le. e investigator stored the Excel and S�SS �les on a drive that resided on a secure server. e Information Technology Department at the university archived this server on a daily basis. Only the primary investigator had access to the data �les. e study employed S�SS, version 13.0 for Windows, statistical soware to calculate and present statistics.

Results
Demographic and anthropometric data were calculated and presented as means and standard deviations or proportions. e investigator tabulated the prevalence rates of the BMD -scores ≤ −2.5, ≤ −2.0, and ≤ −1.0 by OST risk categories, ethnicity, and gender.
Using a liberal selection threshold of .25 in the univariable logistic regressions, the variables, "lack of physical activity" and "calcium intake, " were excluded from the full model multivariable logistic regression (see Tables 1 and 2). At an alpha =.01, the variables, "age, " "lack of estrogen therapy, " and "current smoker" increased the odds of a BMD the NOF treatment threshold, and "weight" decreased the odds in the full and reduced models; alcohol was retained in the �nal model, because when it was subtracted the difference in the beta for lack of estrogen was greater than 20 percent. Variables excluded from the full model were retested in the reduced model and lacked signi�cance. e Hosmer-�emeshow goodness-of-�t test, .2 , demonstrated adequate model �t. Interaction and collinearity were absent.

Clinical Decision Rules in Older White Women.
In older White women the OST and ORAI were strong predictors of osteoporosis; the ABONE was less effective (see Figure 1). e clinical decision rules demonstrated a similar pattern of performance in the prediction of the NOF treatment threshold. e performance of the three clinical decision rules declined and demonstrated less variance in the prediction of osteopenia. e development of each of the OST, ORAI, and ABONE was the result of multivariable regression analyses. In the development of the ABONE clinical decision rule, Weinstein et al. [48] employed multivariable logistic regressions for osteoporosis of the total hip, femur neck, and spine. eir results indicated that increased age was a risk factor for osteoporosis; increased weight and a history of estrogen intake of least six months, in the form of either estrogen therapy or birth control pills, were protective factors. Age > 65, weight 63.5 kg, and history of estrogen use each account for one point in the ABONE score; the cut score for referral for BMD testing is a score greater than or equal to two. us, estrogen therapy accounts for a one-third of the possible points in the ABONE indices. In comparison to the OST and the ORAI, the relatively lower contribution of age and weight in the scoring system likely accounts for the lower diagnostic accuracy of the ABONE demonstrated in these study results.

Discussion
Cadarette et al. [31] developed the ORAI using multivariable logistic regression, with the NOF treatment threshold, −2.0 standard deviations below the mean for young adult women, at either the femur neck or the spine as the dependent variables. eir study results demonstrated that age, weight, and current estrogen therapy were associated with osteoporosis at both measurement sites. In the ORAI scoring system, age and weight account for 93 percent of the possible points, and either an age of 65 years or greater or weight less than 60 kg would result in a recommendation for referral for a BMD test, regardless of current estrogen therapy. is heavy weighting of age and weight helps to explain the relatively close approximation of the diagnostic accuracy of the ORAI to that of the OST demonstrated in the study results.
Koh et al. [33] developed the OST using linear regression, with femur neck BMD -score as the dependent variable in Asian women. Age and weight were among several independent variables in the �nal regression model. However, the researchers dropped the other variables from the clinical decision rule due to their relatively minor contribution to the diagnostic accuracy of predicting osteoporosis. e minor contribution of hormone replacement therapy may be a result of a low rate of estrogen use among Asian women [49]. Results of the current study corroborated that age and weight were the predominant risk factors for a BMD at or below the NOF threshold for treatment and correlates of BMD. Hence, these study results supported the development method of the OST clinical decision rule as appropriate.
Among the logistic regression �nal models for the NOF treatment threshold among White women, only the �nal model included independent variables other than age and weight. In addition to age and weight the �nal model included lack of current estrogen therapy and current smoker as statistically signi�cant independent variables. e inclusion of current estrogen corroborated the �nal model of the logistic regression for the development of the ORAI by Cadarette et al. [31], in which white women composed 95 percent of the sample. In a sample of Asian women, the study results of Koh et al. [33], too, included current estrogen in the �nal regression model. However, in their development of the OST, the researchers omitted current estrogen as a variable, because it made only a minor contribution to the diagnostic accuracy and applied only to a subset of women [33]. e inclusion of "current smoker" in the �nal model corroborated the determination that smoking is a major risk factor for osteoporosis in the evidence-based NOF clinical practice guidelines, which was based on studies with samples of predominantly Caucasian women [26]. e present study provides further evidence of validation of the OST in various samples of Caucasian women. ese studies of the OST included the following samples: (a) Geusens et al. [40] studied samples of predominantly white women in the US and the Netherlands; (b) Richy et al. [41] studied a sample of white Belgian women; (c) Cadarette et al. [37] studied a sample of Canadian women, which, presumably, predominantly consisted of Caucasian women; and (d) Rud et al. [50] studied a sample of predominantly Caucasian Danish women.
e current study corroborated the �ndings of previous studies of the discriminatory performance of the OST in Caucasian women [37,40,41,50] e results helped to address each of the three essential components of a valid clinical decision rule, as outlined by McGinn et al. [45]: (a) the development method by regression analyses appeared to be appropriate; (b) it appeared to be applicable in a wide range of samples, perhaps even universally applicable to women 50 years of age and older, and; (c) it demonstrated strong and excellent predictive power.
Furthermore, the OST is relatively simple and practical, making it attractive for adoption in clinical practice. Only age and weight, routinely available data, are needed. Graphic representation of the OST calculation on a chart is therefore relatively easy. Such a chart can enable health care providers and patients alike to assess the risk of osteoporosis.
An evidence-based systematic review suggested that clinical decision rules improved physician performance [51]. Hence, with broad generalizability and robust predictive power in combination with simplicity, the OST can potentially improve physicians' assessment of patients' need to undergo BMD testing. e US Surgeon General punctuated the critical need for health care providers to adequately assess osteoporosis [38]. With adequate assessment, health providers can in turn initiate appropriate therapy and educate patients about preventive measures to avoid unnecessary devastating fractures and the associated degradation of healthrelated quality of life.
4.�. �i�ni�canc� o� t�� �t���. e study had implications for the prevention, assessment, diagnosis, and treatment of osteoporosis. e Surgeon General highlighted the development of strategies to identify the need to undergo BMD tests as a key research agenda for the future [38]. Osteoporosis is an insidious, underdiagnosed, and undertreated disease. e National Osteoporosis Foundation guidelines for referral for BMD testing are complex and difficult to transform into clinical practice [26,27]. Succinct clinical decision rules have practical applications. However, clinical decision rules require validation among a wide range of samples. To enhance generalizability of the OST, the study sought to validate these clinical decision rules in a representative national sample.