Fracture Risk Prediction Using Phalangeal Bone Mineral Density or FRAX Ò?dA Danish Cohort Study on Men and Women

Transcription

1 Journal of Clinical Densitometry: Assessment & Management of Musculoskeletal Health, vol. 17, no. 1, 7e15, 2014 Ó Copyright 2014 by The International Society for Clinical Densitometry /17:7e15/$ Section I: Fracture Risk Assessment Fracture Risk Prediction Using Phalangeal Bone Mineral Density or FRAX Ò?dA Danish Cohort Study on Men and Women Teresa Friis-Holmberg,*,1 Katrine Hass Rubin, 2 Kim Brixen, 2 Janne Schurmann Tolstrup, 1 and Mickael Bech 3 1 National Institute of Public Health, University of Southern Denmark, Copenhagen, Denmark; 2 Institute of Clinical Research, University of Southern Denmark, Odense, Denmark; and 3 COHERE, Department of Business and Economics, University of Southern Denmark, Odense, Denmark Abstract In this prospective study, we investigated the ability of Fracture Risk Assessment Tool (FRAX), phalangeal bone mineral density (BMD), and age alone to predict fractures using data from a Danish cohort study, Danish Health Examination Survey 2007e2008, including men (n ) and women (n ) aged 40e90 yr. Data were collected using a self-administered questionnaire and by phalangeal BMD measurement. Information on incident and prevalent fractures, rheumatoid arthritis, and secondary osteoporosis was retrieved from the Danish National Patient Registry. Survival analyses were used to examine the association between low, intermediate, and high risk by phalangeal T-score or FRAX and incident fractures, and receiver operating characteristic curves were obtained. Mean follow-up time was 4.3 yr, and a total of 395 persons (3.1%) experienced a fracture during follow-up. The highest rate of major osteoporotic fractures was observed in persons with a high combined risk (FRAX 20% and T-score 2.5; women: 32.7 and men: 27.6 per 1000 person-yr). This group also had the highest risk of hip fractures (women: 8.1 and men: 7.2 per 1000 person-yr). FRAX and T-score in combination analyzed as continuous variables performed overall best in the prediction of major osteoporotic fractures. In predicting hip fractures, there was a tendency of T-score performing worse than the other methods. Key Words: Bone mineral density; follow-up studies; fracture; FRAX; phalangeal bone. Introduction Fractures associated with osteoporosis are very common in the elderly population (1). In Denmark and other countries, a case-finding strategy is adopted recommending general practitioners (GPs) to refer persons with 1 or more risk factors to bone mineral density (BMD) measurement by dual-energy X-ray absorptiometry (DXA); however, a large proportion at high risk of fracture are not diagnosed or treated (2,3). Central DXA is, furthermore, inaccessible in many countries and regions, and longer distances to DXA facilities seem to be associated with lower use of DXA (4e6). Also, other methods Received 01/31/13; Revised 03/15/13; Accepted 03/15/13. *Address correspondence to: Teresa Friis-Holmberg, MSPH, PhD Student, National Institute of Public Health, University of Southern Denmark, Øster Farimagsgade 5A, 2nd Floor, 1353 Copenhagen, Denmark. tho@niph.dk for measurement of BMD are available. Some of these systems have the advantage of portability, low X-ray exposure, and may be delivered as point of care to identify individuals at high risk of fracture with need of medical checkup and central DXA. Both prospective studies (7e13) and a metaanalysis (14) have found low peripheral BMD to be associated with increased risk of fracture. A number of risk factors besides BMD are associated with increased risk of osteoporotic fractures, like age, gender, low body mass index, smoking, excessive alcohol intake, parental hip fracture, and a history of low-energy fracture (1,15,16). The Fracture Risk Assessment Tool (FRAX) that predicts the 10-yr probability of hip and major osteoporotic fractures (17) was applied in the clinical management of osteoporosis in some countries; for example, in the United Kingdom where National Osteoporosis Foundation (NOF) recommends treatment based on FRAX (18). 7

2 8 Friis-Holmberg et al. Nevertheless, to our knowledge, no studies have compared the predictive capability of FRAX (without BMD) with a point-of-care densitometer and the combined use of both methods regarding fracture risk prediction. A method holding both the result of the phalangeal densitometer and the 10-yr fracture risk by FRAX would possibly enhance the preselection of person in need of a DXA. In the present prospective study, we, therefore, aimed to investigate the ability of phalangeal BMD using radiographic absorptiometry (RA), FRAX, and age as well as FRAX and BMD in combination in different risk strata to predict osteoporotic fractures. Material and Methods We used data on a cohort of women and men aged 18e95 yr who participated in the Danish Health Examination Survey 2007e2008 (DANHES 2007e2008) (19). In short, the study was conducted in 13 of 98 Danish municipalities. All adult citizens aged 18þ yr were invited to answer an Internetbased questionnaire comprising more than 100 items on lifestyle, health, and morbidity. Furthermore, a representative sample of the citizens was invited to participate in a health examination. Overall, 180,103 persons were invited to the health examination containing a range of measurements (details described elsewhere (19)) and 18,065 participated. The present study includes data from participants aged 40e90 yr (i.e., the applied age range in FRAX), who had a BMD scan. Participants were excluded if height or weight was missing (n 5 5). A total of 12,758 persons were included in analyses (see Fig. 1). The health examination was initialized by Fig. 1. Flowchart of study population. BMD, bone mineral density; FRAX, Fracture Risk Assessment Tool. verbal information, and written informed consent was obtained. Phalangeal BMD Measurement BMD was measured at the middle phalanges of the second, third, and fourth fingers on the nondominant hand using a compact RA system (Alara MetriScan Ò ; Alara, Inc., Fremont, CA). Gender-specific T-scores (compares measured BMD with the average BMD for a young healthy subject) were calculated using participants from DANHES aged 20e39 yr as reference (BMD g/cm 2 ; n ). Participants were informed on their result, and participants with phalangeal T-scores below 2.5 were advised to consult with their GP. Register Follow-Up of Fractures All persons in Denmark are assigned with a unique personal identification number that can be linked to all public registers at an individual level. In this study, data were merged with information on fracture and surgical procedures from the National Patient Register (NPR) and information on death and emigration from the Civil Registration System. NPR includes discharge diagnoses coded by physicians according to the International Classification of Diseases, Tenth Revision (ICD-10). In the present study, we extracted data on prevalent and incident major osteoporotic fractures; corresponding ICD-10 codes are shown in Table 1. Incident fractures were defined as fractures occurring between the date of BMD measurement in 2007e2008 and end of follow-up (10th of August 2012) and calculated as the number of persons with a fracture during the follow-up period. Prevalent fractures were defined as fractures occurring before the date of BMD measurement. were validated and excluded if no corresponding surgical code of primary hip arthroplasty or osteosynthesis (NFB00e92 and NFJ00e 92, respectively). Follow-up was ended by 10th of August 2012, date of fracture, date of death, or migration as appropriate. Clinical Risk Factors and FRAX As the algorithm for FRAX is unpublished, the 10-yr risk of fracture was calculated by individual risk scoring of the Danish version of FRAX using a programed call of the FRAX Web site (version 3.1). The FRAX algorithm is based on the following risk factors: age, sex, height (cm), weight (kg), history of fracture, parental history of hip fracture, current smoking, 3 or more units of alcohol per day, glucocorticoid use in the last 3 mo, the presence of rheumatoid arthritis, and other types of secondary osteoporosis. The FRAX tool was used without the inclusion of BMD. At baseline, body height and weight was measured. The questionnaire included self-reported information on smoking (daily smokers listed as current smokers in FRAX), alcohol consumption, and parental history of hip fracture after the age of 50 yr. Information on prevalent fractures, the presence of rheumatoid arthritis, and other types of secondary osteoporosis was extracted from NPR (ICD-10 codes are listed in Table 1). Information

3 Fracture Risk Prediction Using Phalangeal BMD or FRAX 9 Table 1 ICD-10 Codes Used in the Analyses and Calculation of FRAX Variable Vertebral fractures Forearm fractures Humerus fractures Rheumatoid arthritis Type 1 diabetes Osteogenesis imperfecta Chronic liver disease Anorexia nervosa Inflammatory bowel disease, e.g., Crohn disease Celiac disease and malabsorption Thyroid disorders (hyperthyroidism, thyrotoxicosis) Premature menopause ICD-10 codes DS720, DS721AeB, DS722 DS120, DS121AeB, DS122AeE, DS220AeL DS320AeE, DT08A DS525AeC, DS526 DS422AeC, DS423A DM05 DE10 DQ780 DK72, DK73, DK74 DF50eDF509 DK50, DK51 DK90 DE05 DE283A Abbr: FRAX, Fracture Risk Assessment Tool; ICD-10, International Classification of Diseases, Tenth Revision. on premature menopause (younger than 45 yr) was extracted from the DANHES questionnaire and incorporated in secondary osteoporosis. We had no information on glucocorticoids use. Statistical Analysis Results of descriptive analyses are presented as mean standard deviation or frequencies. Age was analyzed using the age limit for osteoporosis screening set by the US Preventive Services Task Force (2). With regard to FRAX, we adapted the high-risk threshold used by NOF (18) and defined an intermediate category. T-score risk categories were defined corresponding to the World Health Organization thresholds for osteopenia and osteoporosis (16). Moreover, we constructed a combined risk model including FRAX and T-score: Age: (1) younger than 65 yr; (2) 65 yr and older FRAX 10-yr probability of major osteoporotic fracture: (1) low risk:!10%; (2) intermediate risk: 10e19.99%; and (3) high risk: 20% FRAX 10-yr probability hip fracture: (1) low risk:!1.5%; (2) intermediate risk: 1.5e2.99%; and (3) high risk: 3% T-score: (1) normal: O 1; (2) intermediate: between 1 and 2.5; and (3) low: 2.5 Combined risk of major osteoporotic fractures: (1) low combined risk: FRAX!20% and T-score O 2.5; (2) intermediate combined risk: FRAX 20% or T-score 2.5; and (3) high combined risk: FRAX 20% and T-score 2.5 Combined risk of hip fractures: (1) low combined risk: FRAX!3% and T-score O 2.5; (2) intermediate combined risk: FRAX 3% or T-score 2.5; and (3) high combined risk: FRAX 3% and T-score 2.5. Survival analyses were used to examine the association between T-score, FRAX, and incident fractures, and Cox regression analysis was used to assess hazard ratios (HRs) with 95% confidence intervals (95% CIs). The primary outcome was incident osteoporotic fractures (yes/no) or incident hip fractures (yes/no). Analyses were stratified according to sex (because of general variations in BMD and incidence of fracture (1,15)) and known osteoporosis as answered in the questionnaire. Cox regression analyses were tested for the proportional hazard assumption. Receiver operating characteristic (ROC) curves were used to assess the predictive capability of age, FRAX, and BMD and were obtained using C statistics estimated from Cox regression. We analyzed 4 models: (1) FRAX, (2) T-score, (3) T-score and FRAX in combination, (4) age alone, both as continuous variables and categorical variables holding the previously defined risk categories. Differences in the area under the ROC curve (AUC) were tested for significance using the lincom command in STATA, version 12.1 (StataCorp LP, College Station, TX). Results Complete follow-up information was available on all 12,758 participants, and mean follow-up time was 4.3 yr (range: 0.03e4.9), giving 54,980 person-yr. During followup, a total of 395 (3.1%) participants suffered 1 or more major osteoporotic fractures, 54 (0.42%) a hip fracture, and 226 (1.5%) died (data not shown). Table 2 shows the general characteristic of participants (40.8% men and 59.2% women). With the exception of parental hip fracture and rheumatoid arthritis, the clinical risk factors were all significantly different between the sexes. Only 0.9% among men had a high 10-yr probability of major osteoporotic fractures (20%) compared with 12.6% among women, whereas, 15.0% among men and 25.2% among women had a high probability (3.0%) of hip fractures. Mean BMD was, as expected, lower among women than

4 10 Friis-Holmberg et al. Table 2 General Characteristics of Participants Characteristics Men, n (40.8) Women, n (59.2) p Value Age (yr) ! e (24.7) 2113 (28.0) 50e (27.4) 2372 (31.4) 60e (33.2) 2204 (29.2) 70e (12.5) 745 (9.9) 80e (2.3) 118 (1.6) Weight (kg) !0.001 Height (cm) !0.001 Previous fracture 132 (2.5) 443 (5.9)!0.001 Parental hip fracture 460 (8.8) 753 (10) 0.03 Current daily smoker 691 (13.3) 803 (10.6)!0.001 Rheumatoid arthritis 20 (0.38) 42 (0.56) 0.17 Secondary osteoporosis 71 (1.4) 757 (10.0)!0.001 Alcohol intake O3 units/d 777 (14.9) 255 (3.4)!0.001 FRAX 10-yr probability of major osteoporotic !0.001 fractures (%) Low (!10%) 4436 (85.2) 4625 (61.2) Intermediate (10e19.99%) 723 (13.9) 1977 (26.2) High (20%) 47 (0.90) 950 (12.6) FRAX 10-yr probability of hip fractures (%) !0.001 Low (!1.5%) 3517 (67.6) 4284 (56.7) Intermediate (1.5e2.99%) 907 (17.4) 1368 (18.1) High (3%) 782 (15.0) 1900 (25.2) Phalangeal BMD !0.001 Phalangeal T-score a !0.001 O (72.4) 4328 (57.3) 1.0 to (23.9) 2296 (30.4) (3.7) 928 (12.3) Incident major osteoporotic fractures 86 (1.7) 309 (4.0)!0.001 Incident hip fractures 20 (0.38) 35 (0.45) 0.57 Known osteoporosis 40 (0.82) 371 (5.3)!0.001 Note: Data are shown as mean standard deviation or n (%). Abbr: BMD, bone mineral density; FRAX, Fracture Risk Assessment Tool. a Phalangeal T-scores calculated from the Danish Health Examination Survey 2007e2008 population. men, and a higher proportion of women was categorized with low BMD (T-score 2.5) (12.3% vs 3.7%, p! 0.001). Table 3 shows rates of incident fractures per 1000 personyr according to FRAX, T-score, and the combined risk score. We found increasing rates of major osteoporotic fractures with increasing risk categories. The highest rates were observed in persons with a high combined risk profile (FRAX 20% and T-score 2.5) (women: 32.7 per 1000 personyr and 95% CI: 24.3e43.9; men: 27.6 per 1000 person-yr and 95% CI: 3.9e196) followed by persons with low T-score ( 2.5). With regard to hip fractures, the highest rate was also observed in persons with a high combined risk profile. Table 4 shows HR for the different risk categories. In both genders, T-scores 2.5 were associated with higher risk of fracture when compared with T-scores above 1.0 (women: HR 5 5.5, 95% CI: 4.1e7.4; men: HR 5 4.1, 95% CI: 2.0e8.3). Also, ages 65 yr and older vs younger than 65 yr were associated with higher risk in both genders (women: HR 5 2.9, 95% CI: 2.3e3.6; men: HR 5 1.9, 95% CI: 1.2e2.9). In women, a high FRAX score and a high combined risk were also associated with an increased risk of major osteoporotic fractures compared with lower risk categories of the same variable (HR 5 3.2, 95% CI: 2.4e4.3 and HR 5 4.5, 95% CI: 3.3e6.3, respectively). In men, a high combined risk was associated with a higher risk of fractures when compared with low combined risk (HR 5 7.7, 95% CI: 1.1e55.6). The subgroup analyses including only participants without known osteoporosis at baseline revealed no major differences, but there was a slight tendency of increasing risk estimates in most risk categories. An age older than 65 yr compared with age younger than 65 yr was associated with a 16.2 higher risk of hip fractures

5 Fracture Risk Prediction Using Phalangeal BMD or FRAX 11 Table 3 Event Rate (95% CI) of Major Osteoporotic Fractures (n 5 395) or Hip Fractures (n 5 54) Per 1000 Person-Yr According to Risk Categories of Age, FRAX, and Phalangeal T-Score Measure Men (n ) Women (n ) Major osteoporotic fractures Age (yr)! (2.3e4.0) 6.7 (5.8e7.8) (4.1e7.8) 19.3 (16.4e22.8) FRAX score risk Low (!10%) 3.1 (2.4e4.0) 5.6 (4.7e6.8) Intermediate (10e19.99%) 7.7 (5.2e11.5) 15.0 (12.6e17.9) High (20%) 10.2 (2.6e40.7) 17.8 (14.1e22.5) Phalangeal T-score a Normal (O 1.0) 2.7 (2.0e3.7) 4.4 (3.6e5.5) Intermediate ( 1.0 to 2.49) 5.9 (4.2e8.4) 13.5 (11.4e15.9) Low ( 2.5) 11.1 (5.8e21.4) 24.4 (19.9e29.9) Combined risk Low (FRAX!20% and T-score O 2.5) 3.5 (2.8e4.4) 7.3 (6.3e8.4) Intermediate (FRAX 20% or T-score 2.5) 9.7 (5.0e18.6) 15.0 (12.0e18.7) High (FRAX 20% and T-score 2.5) 27.6 (3.9e196) 32.7 (24.3e43.9) Age (yr)! (0.21e0.92) 0.24 (0.11e0.53) (1.1e3.3) 3.9 (2.7e5.6) FRAX score risk Low (!1.5%) 0.52 (0.26e1.05) 0.05 (0.01e0.38) Intermediate (1.5e2.99%) 0.75 (0.24e2.3) 0.85 (0.35e2.1) High (3%) 2.7 (1.4e5.1) 3.5 (2.4e5.1) Phalangeal T-score a Normal (O 1.0) 0.67 (0.37e1.2) 0.21 (0.08e0.57) Intermediate ( 1.0 to 2.49) 1.1 (0.50e2.5) 0.82 (0.41e1.6) Low ( 2.5) 3.7 (1.20e11.5) 5.7 (3.8e8.7) Combined risk Low (FRAX!20% and T-score O 2.5) 0.53 (0.29e0.99) 0.13 (0.04e0.41) Intermediate (FRAX 20% or T-score 2.5) 2.2 (1.1e4.1) 1.7 (0.96e3.0) High (FRAX 20% and T-score 2.5) 7.2 (1.8e28.9) 8.1 (5.2e12.7) Abbr: 95% CI, 95% confidence interval; FRAX, Fracture Risk Assessment Tool. a Phalangeal T-scores calculated from the Danish Health Examination Survey 2007e2008 population. among women (95% CI: 6.7e39.0) and a 4.37 higher risk among men (95% CI: 1.8e10.9). Also, a high probability of hip fractures (FRAX 3%) compared to a low probability (!1.5%) was associated with an increased risk of hip fractures in both women (HR , 95% CI: 8.8e475) and men (HR 5 6.0, 95% CI: 1.9e13.1). Low T-score ( 2.5) was also associated with a higher risk of hip fractures compared with normal T-scores in both genders (women: HR , 95% CI: 9.2e77.4; men: HR 5 5.6, 95% CI: 1.6e20.2). Finally, both women and men with a high combined risk profile had an increased risk of hip fractures compared with participants with low and intermediate combined risk (women: HR , 95% CI: 18.2e207; men: HR , 95% CI: 3.0e61.4) (Table 4). In participants without osteoporosis, similar risks of hip fractures were seen regarding the different risk categories. In women, however, an intermediate T-score was now significantly associated with a higher risk of hip fracture than normal T-scores, and when holding a high combined risk, the risk of hip fracture was slightly increased. Table 5 shows the predictive value of the models in the identification of participants, who had a major osteoporotic fracture or a hip fracture during follow-up. When analyzed as continuous variables, the combination of FRAX and BMD performed best in prediction of major osteoporotic fractures in both genders (women: AUC , 95% CI: 0.692e0.746; men: AUC , 95% CI: 0.609e0.730); however, in men the combination of FRAX and T-score was not statistically significantly different from FRAX alone. In women, T-score was also significantly better than age and

6 Measure Table 4 HRs for Major Osteoporotic Fractures (n 5 395) or Hip Fractures (n 5 54) According to Risk Categories of Age, FRAX, and Phalangeal T-Score in All Participants and Participants Without Osteoporosis at Baseline All participants (n 5 12,758) Participants without osteoporosis (n 5 12,347) Men Women Men Women HR (95% CI) p Value HR (95% CI) p Value HR (95% CI) p Value HR (95% CI) p Value Major osteoporotic fractures Age (yr)! (1.2e2.9) (2.3e3.6)! (1.9e2.8) (2.4e3.8)!0.001 FRAX score risk Low (!10%) Intermediate (10e19.99%) 2.5 (1.5e3.4)! (2.1e3.5)! (1.6e4.1)! (2.1e3.5)!0.001 High (20%) 3.3 (0.80e13.4) (2.4e4.3)! (0.24e12.8) (2.4e4.5)!0.001 Phalangeal T-score a Normal (O 1.0) Intermediate ( 1.0 to 2.49) 2.2 (1.4e3.4)! (2.3e4.0)! (1.4e3.6) (2.2e3.9)!0.001 Low ( 2.5) 4.1 (1.2e8.3)! (4.1e7.4)! (1.9e8.4)! (4.1e7.6)!0.001 Combined risk Low (FRAX!20% and T-score O 2.5) Intermediate (FRAX 20% or T-score 2.8 (1.4e5.5) (1.6e2.7)! (1.1e4.9) (1.7e2.9)! ) High (FRAX 20% and T-score 2.5) 7.7 (1.1e55.6) (3.3e6.3)! (1.1e57.5) (3.5e7.2)!0.001 Age (yr)! (1.8e11.0) (6.7e39.0)! (1.6e10.4) (7.4e50.3)!0.001 FRAX score risk Low (!1.5%) Intermediate (1.5e2.99%) 1.5 (0.38e5.7) (1.8e135) (0.38e5.4) (1.5e117) High (3%) 5.0 (1.9e13.1) (8.8e475)! (1.7e12.1) (8.8e484)!0.001 Phalangeal T-score a Normal (O 1.0) Intermediate ( 1.0 to 2.49) 1.7 (0.61e4.5) (1.2e12.7) (0.67e5.1) (1.0e11.9) Low ( 2.5) 5.6 (1.6e20.2) (9.2e77.4)! (1.7e22.8) (9.2e79.6)!0.001 Combined risk Low (FRAX!20% and T-score O 2.5) Intermediate (FRAX 20% or T-score 2.5) 4.1 (1.6e10.4) (3.6e45.3)! (1.4e9.5) (3.2e41.7)!0.001 High (FRAX 20% and T-score 2.5) 13.5 (3.0e61.4) (18.2e207)! (3.1e63.6) (20.4e237)!0.001 Abbr: 95% CI, 95% confidence interval; FRAX, Fracture Risk Assessment Tool; HR, hazard ratio. a Phalangeal T-scores calculated from the Danish Health Examination Survey 2007e2008 population. 12 Friis-Holmberg et al.

7 Measure Table 5 AUC Using C Statistics Estimated From Cox Regression Model (Age, FRAX, Phalangeal BMD, and Combined Risk) in All Participants and Participants Without Osteoporosis at Baseline All participants (n 5 12,758) Participants without osteoporosis (n 5 12,347) Men Women Men Women Analyzed as continuous variables Major osteoporotic fractures Age (0.550e0.671) (0.659e0.713) (0.541e0.666) (0.662e0.718) FRAX score risk (0.567e0.688) (0.652e0.706) (0.555e0.681) (0.654e0.711) T-score (0.576e0.701) (0.686e0.739) a,b (0.568e0.666) (0.685e0.741) b FRAX þ T-score (0.609e0.730) a,c (0.692e0.746) a,b,c (0.596e0.722) (0.693e0.749) a,b,c Age (0.689e0.859) (0.816e0.916) (0.678e0.852) (0.817e0.924) FRAX score risk (0.661e0.851) (0.816e0.903) (0.648e0.841) (0.832e0.915) T-score (0.511e0.770) (0.777e0.890) (0.504e0.777) (0.776e0.898) FRAX þ T-score (0.596e0.844) c (0.809e0.916) c (0.561e0.821) (0.804e0.920) c Analyzed as categorical variables Major osteoporotic fractures Age (younger than 65 yr, 65 yr and older) (0.521e0.627) (0.589e0.645) (0.516e0.624) (0.588e0.647) FRAX score risk (low [!10%], intermediate (0.530e0.628) (0.606e0.664) (0.526e0.626) (0.606e0.667) [10e19.99%], high [20%]) T-score (normal [O 1.0), intermediate (0.552e0.663) (0.651e0.708) a,b (0.554e0.667) (0.647e0.707) a,b ( 1.0 to 2.49), low [2.5]) Combined risk (low, intermediate, high) (0.503e0.571) c (0.575e0.632) b,c (0.496e0.561) c (0.574e0.633) b,c Age (younger than 65 yr, 65 yr and older) (0.573e0.782) (0.735e0.863) (0.561e0.778) (0.746e0.880) FRAX score risk (low [!10%], intermediate (0.546e0.786) (0.742e0.884) (0.531e0.777) (0.795e0.884) [10e19.99%], high [20%]) T-score (normal [O 1.0], intermediate (0.483e0.717) (0.742e0.884) (0.493e0.734) (0.734e0.892) [ 1.0 to 2.49], low [ 2.5]) Combined risk (low, intermediate, high) (0.556e0.782) (0.799e0.915) a (0.541e0.773) (0.780e0.928) Abbr: AUC, area under the receiver operating characteristic curve; BMD, bone mineral density; FRAX, Fracture Risk Assessment Tool. a Significance level compared with age ( p! 0.05). b Significance level compared with FRAX ( p! 0.05). c Significance level compared with T-score ( p! 0.05). Fracture Risk Prediction Using Phalangeal BMD or FRAX 13

8 14 Friis-Holmberg et al. FRAX; this was also the tendency for men but not statistically significant. This was also the trend when excluding participant with osteoporosis from analyses. When using the different risk categories, the predictive ability decreased for all variables, and the combined risk score of T-score and FRAX was no longer superior. Now the categorization of T-score most accurately identified persons with fracture (in men, not significantly different from age and FRAX score). The same was found when excluding participants with osteoporosis. The AUCs for hip fractures did not reveal any great differences between the 4 different models. However, T-score analyzed as continuous variables tended to perform worse than the other models in both genders, but the differences were only statistically significant in T-scores alone compared with FRAX and T-scores in combination. Moreover, the predictive ability in women of the combined risk score was significantly superior to age (younger than 65 vs 65 yr and older) and tended to perform better (but not significant statistically) than FRAX score (low, intermediate, and high) and T-score (normal, intermediate, and low). Discussion In this large prospective study, we observed the highest rate of major osteoporotic fractures and hip fracture among persons who had a high 10-yr fracture probability (calculated by FRAX) and a low phalangeal T-score (measured by RA). This was followed by persons only having a low T-score (T-score 2.5). The predictive ability of the different methods showed somewhat inconsistent results depending on what approach we used (analyzed as continuous vs categorical variables based on risk strata). Currently, femoral neck BMD by DXA could be included in FRAX to enhance the calculation of the 10-yr fracture probability, but DXA scanners are often located at hospitals or often unavailable. In contrast, phalangeal BMD could be measured with an RA scanner that is portable, simple to use, and easily accessible. Thus, the combined use of phalangeal BMD and FRAX available on site could improve the selection of person for further work-up by central DXA. Pfister et al (20) looked at a portable peripheral forearm densitometer and the 10-yr probability of major osteoporotic fractures calculated by FRAX for treatment selection (N women aged 60e64 yr). They concluded that an approach included treatment is initiated in women who had a prior fracture or FRAX 20% (no peripheral DXA [pdxa] confirmation) or based on pdxa evaluations in women with a FRAX value between 9.3% and 20%. This approach would significantly reduce the number of pdxa examinations and the cost of screening (20). Also, Gasser et al (21) evaluated phalangeal RA with or without clinical risk factors in a general practice setting among postmenopausal women (21). They found that the model including RA in combination with age, height, and weight performed best. RA alone performed better than 2 different models with clinical risk factors alone (21). In contrast to our study, none of these studies (20,21), however, included osteoporotic fractures as outcome. Additionally, Durosier et al (22) evaluated if the detection of women at low or high risk for hip factors could be improved by combining clinical risk factors and quantitative ultrasound (QUS). They found that the combined risk score (risk factors þ QUS) improved the specificity of detection to 42.4% from 33.8% using clinical risk factors alone and 38.4% using QUS alone (22). AUCs of the combined risk score were statistically significantly better than clinical risk factors alone and QUS alone (22). In the present study, we found that phalangeal T-score predicted major osteoporotic fractures better than hip fractures. This corresponds to the fact that BMD measurement at the hip compared with other measurement sites is superior when identifying hip fracture cases, as well as the fact that age, included in FRAX, is a more important risk factor for hip fractures. Furthermore, we found that the accuracy of the different methods declined when analyzed as risk score categories compared with continuous variables (especially for the combined risk category of FRAX and T-scores in predicting major osteoporotic fractures). This supports that further studies should look into defining the best risk strata for categorizing participants in low- and high-risk groups and obtaining the highest possible sensitivity and specificity. This also applies to the T-score and FRAX risk strata widely used. It is noteworthy that only 0.9% of men in our cohort met or surpassed the NOF-defined FRAX threshold for major osteoporotic fractures (20%). In contrast, 15% of men met or surpassed the hip fracture threshold (3%). Our study has several strengths. It is to our knowledge, the first study that combined FRAX and phalangeal BMD measured with a portable densitometry system in fracture risk prediction. Second, our study comprised 12,758 participants of both genders giving a total of 54,980 person-yr. Third, the linkage to national registers ensured complete follow-up. The register is considered one of the most comprehensive in the world and has a high validity concerning the diagnosis and procedure codes (23); moreover, we validated all hip fractures for corresponding surgery code. In other studies, this information is often self-reported, which could lead to bias. Rheumatoid arthritis, for instance, is often by laymen confused with osteoarthritis resulting in an overestimated incidence (24). Fourth, height and weight were measured as a part of the health examination giving more reliable numbers than self-reports (25). Fifth, we evaluated the use of the FRAX risk score and phalangeal BMD measurement using robust statistical methods including AUCs by C statistic and Cox regressions. There are also potential limitations in our study. Only clinical vertebral fractures were registered as no sequential spine X-rays were performed. Many vertebral fractures are asymptomatic or associated with few uncharacteristic symptoms. Thus, many vertebral fractures remain undiagnosed (16). Moreover, FRAX gives the estimated 10-yr fracture probability, and it have been criticized that FRAX could not be validated in a shorter period than 10 yr (26). However, we used the FRAX score as a predictor and took time to event into

9 Fracture Risk Prediction Using Phalangeal BMD or FRAX 15 account. Studies with longer follow-up than ours have showed similar results independently of the follow-up time (12). As we were unable to validate if a fracture was a lowenergy trauma, we used types most frequently caused by osteoporosis (1) and used by FRAX to define major osteoporotic fractures (hip, vertebral, humerus, and forearm). In the assessment of fractures and secondary causes of osteoporosis, we used ICD-10 codes, which only include diagnoses experienced after In DANHES, we did not obtain information on glucocorticoids use. If information on a risk factor is missing, the calculation of FRAX should be made without this given risk factor, potentially leading to an underestimation of the individual fracture risk assessed by FRAX. In Denmark, it is possible to obtain information on medication use from the Danish National Prescription Registry. This was not done because of delay of information when using this register via Statistic Denmark. Finally, a possible limitation is the representativity and the fact that nonresponse bias may have occurred (as described elsewhere (19)). It is known from DANHES and other studies that the proportions with unhealthy lifestyles like smoking (that affect bone health negatively) are greater among persons with lower socioeconomic status (who was underrepresented in the present study). This probably led to an underestimation of fracture rates. In conclusion, our study supports that persons with a high combined risk (low phalangeal BMD measured with a portable densitometer and a high fracture probability by FRAX) has the highest rate of hip and major osteoporotic fractures. Moreover, FRAX and phalangeal T-score in combination analyzed as continuous variables performed overall best in prediction of major osteoporotic fractures but not in the prediction of hip fracture. However, when using the different risk categories, the overall predictive ability decreased for all variables, and the combined risk score of T-score and FRAX was no longer superior. This requires that we are careful when defining risk strata for categorizing participants in low-and high-risk groups. References 1. Johnell O, Kanis J Epidemiology of osteoporotic fractures. Osteoporos Int 16(2 Suppl):S3eS7. 2. Nguyen TV, Center JR, Eisman JA Osteoporosis: underrated, underdiagnosed and undertreated. Med J Aust 180(5 Suppl): S18eS Vestergaard P, Rejnmark L, Mosekilde L Osteoporosis is markedly underdiagnosed: a nationwide study from Denmark. Osteoporos Int 16(2):134e Curtis JR, Laster A, Becker DJ, et al The geographic availability and associated utilization of dual-energy X-ray absorptiometry (DXA) testing among older persons in the United States. Osteoporos Int 20(9):1553e Frost M, Gudex C, Rubin KH, et al Pattern of use of DXA scans in men: a cross-sectional, population-based study. Osteoporos Int 23(1):183e Rubin KH, Abrahamsen B, Hermann AP, et al Prevalence of risk factors for fractures and use of DXA scanning in Danish women. A regional population-based study. Osteoporos Int 22(5):1401e Duppe H, Gardsell P, Nilsson B, Johnell O A single bone density measurement can predict fractures over 25 years. Calcif Tissue Int 60(2):171e Huang C, Ross PD, Yates AJ, et al Prediction of fracture risk by radiographic absorptiometry and quantitative ultrasound: a prospective study. Calcif Tissue Int 63(5):380e Miller PD, Siris ES, Barrett-Connor E, et al Prediction of fracture risk in postmenopausal white women with peripheral bone densitometry: evidence from the National Osteoporosis Risk Assessment. J Bone Miner Res 17(12):2222e Mussolino ME, Looker AC, Madans JH, et al Phalangeal bone density and hip fracture risk. Arch Intern Med 157(4): 433e Ross P, Huang C, Davis J, et al Predicting vertebral deformity using bone densitometry at various skeletal sites and calcaneus ultrasound. Bone 16(3):325e Stone KL, Seeley DG, Lui LY, et al BMD at multiple sites and risk of fracture of multiple types: long-term results from the Study of Osteoporotic Fractures. J Bone Miner Res 18(11):1947e Friis-Holmberg T, Brixen K, Rubin KH, et al Phalangeal bone mineral density predicts incident fractures; a prospective cohort study on men and women. Results from the Danish Health Examination Survey 2007e2008 (DANHES 2007e2008). Arch Osteoporos 7:291e Marshall D, Johnell O, Wedel H Meta-analysis of how well measures of bone mineral density predict occurrence of osteoporotic fractures. BMJ 312(7041):1254e Kanis JA, Johansson H, Oden A, McCloskey EV Assessment of fracture risk. Eur J Radiol 71(3):392e World Health Organization Assessment of fracture risk and its application to screening for postmenopausal osteoporosis: report of the WHO study group. Geneva, Switzerland: World Health Organization. WHO Technical Report Series; Kanis JA, Johnell O, Oden A, et al FRAX and the assessment of fracture probability in men and women from the UK. Osteoporos Int 19(4):385e National Osteoporosis Foundation Clinician s guide to prevention and treatment of osteoporosis. Washington, DC: National Osteoporosis Foundation. 19. Eriksen L, Gronbaek M, Helge JW, et al The Danish Health Examination Survey (DANHES ). Scand J Public Health 39(2):203e Pfister AK, Welch CA, Emmett MK, Gessford AK An approach to identify rural women aged 60 to 64 for osteoporosis treatment. South Med J 105(1):11e Gasser KM, Mueller C, Zwahlen M, et al Osteoporosis case finding in the general practice: phalangeal radiographic absorptiometry with and without risk factors for osteoporosis to select postmenopausal women eligible for lumbar spine and hip densitometry. Osteoporos Int 16(11):1353e Durosier C, Hans D, Krieg MA, et al Combining clinical factors and quantitative ultrasound improves the detection of women both at low and high risk for hip fracture. Osteoporos Int 18(12):1651e Mosbech J, Jorgensen J, Madsen M, et al The national patient registry. Evaluation of data quality. [Article in Danish]. Ugeskr Laeger 157(26):3741e Sambrook PN, Flahive J, Hooven FH, et al Predicting fractures in an international cohort using risk factor algorithms without BMD. J Bone Miner Res 26(11):2770e Bridges MJ, Ruddick S Can self-reported height and weight be used to calculate 10 year risk of osteoporotic fracture? J Nutr Health Aging 14(8):611e Kanis JA, Oden A, Johansson H, McCloskey E Pitfalls in the external validation of FRAX. Osteoporos Int 23(2):423e431.