DOI 1.17/s198-1-152- ORIGINAL ARTICLE Evaluation of FRAX to characterise fracture risk in Poland E. Czerwinski & J. A. Kanis & J. Osieleniec & A. Kumorek & A. Milert & H. Johansson & E. V. McCloskey & M. Gorkiewicz Received: 13 August 21 / Accepted: 21 October 21 # International Osteoporosis Foundation and National Osteoporosis Foundation 21 Funding None of the authors has received or will receive any compensation in relation to this study. E. Czerwinski (*) : A. Milert Department of Bone and Joint Diseases, FHS, Jagiellonian University Medical College, Kopernika 32, 31-51, Krakow, Poland e-mail: czerwinski@kcm.pl J. A. Kanis : H. Johansson : E. V. McCloskey WHO Collaborating Centre for Metabolic Bone Disease, University of Sheffield Medical School, Sheffield, UK E. Czerwinski : J. Osieleniec : A. Kumorek Krakow Medical Centre, Krakow, Poland M. Gorkiewicz Epidemiology and Population Studies Department, Jagiellonian University Medical College, Grzegórzecka 2 31-531, Krakow, Poland Abstract Summary The UK FRAX model was evaluated retrospectively in Polish women assessed 11 years previously for fracture risk. Results were compared with fracture risk observed during follow-up. The UK model can be used to stratify risk, but caution is required in interpretation of absolute fracture risk. Introduction In the absence of a FRAX model for Poland, the UK FRAX tool has been widely used. The aim of this study was to evaluate the validity of the surrogate model in a Polish setting. Methods We studied a convenience sample of 51 women who had been referred for the assessment of bone mineral density and clinical risk factors 9 12 years previously. Incident fractures in the intervening period were selfreported by telephone interview. Fracture probabilities, calculated using the UK FRAX tool, were compared to the incidence of new fractures during follow-up. Results Incident fractures were reported in 16 women. Incident fractures of the major osteoporotic fractures were reported in 89 women. The observed incidence of fractures rose progressively in women according to percentile of fracture probability. Between the 1th and 9th percentiles, hip fracture probability computed with bone mineral density (BMD) differed 49-fold. The range was fivefold in the case of a major osteoporotic fracture. The observed/ expected ratio for fracture was significantly greater than unity when the expected number was calculated without BMD (1.79; 95% confidence interval=1.44 2.21) and when BMD was included in the FRAX calculation (1.94; 95% confidence interval=1.45 2.54). Conclusion The UK FRAX tool categorised fracture risk well in this Polish cohort but significantly overestimated fracture risk. The UK model can be used to stratify risk in the population, but caution is required in interpretation of absolute risk. Keywords BMD. Fracture probability. FRAX. Osteoporosis Introduction Fractures are the most important complication of osteoporosis, and thus, the identification of patients at risk is a priority to direct effective treatments. This, in turn, demands the assessment of fracture risk and a view of what represents a risk sufficiently high that intervention can be considered worthwhile. The assessment of bone mineral density (BMD) provides the cornerstone for the diagnosis
of osteoporosis [1, 2], and intervention thresholds have traditionally been based on densitometric criteria [3 8]. Despite the widespread use of BMD testing, it should be recognised that, just because bone mineral density is normal, there is no guarantee that a fracture will not occur only that the risk is decreased. Conversely, if bone mineral density is in the osteoporotic range, then fractures are more likely, but not invariable. At the age of 5 years, the proportion of women with osteoporosis who will fracture their hip, spine or forearm or proximal humerus in the next 1 years (i.e. positive predictive value) is approximately 45%. The detection rate for these fractures (sensitivity) is, however, low and 96% of such fractures would occur in women without osteoporosis [9, 1]. The low sensitivity is one of the reasons why widespread population-based screening is not generally recommended in women at the time of the menopause [3 5, 7]. Although bone mass is an important determinant of the risk of fracture, other abnormalities occur in the skeleton that contribute to fragility. In addition, a variety of nonskeletal factors, such as the liability to fall and force of impact, contribute to fracture risk. Since BMD forms but one component of fracture risk, accurate assessment of fracture risk should ideally take into account other readily measured indices of fracture risk, particularly those that add information to that provided by BMD. Over the past few years, a series of meta-analyses has been undertaken to identify clinical risk factors that could be used in case finding strategies with or without the use of BMD [11]. Algorithms that integrate the weight of clinical risk factors for fracture risk, with or without information on BMD, have been developed from these analyses by the WHO Collaborating Centre for Metabolic Bone Diseases at Sheffield, UK [2, 12]. The FRAX tool (www.shef.ac.uk/ FRAX) computes the 1-year probability of hip fracture or a major osteoporotic fracture. A major osteoporotic fracture is a clinical spine, hip, forearm and humerus fracture. There is a marked variation in fracture probability in different regions of the world, particularly well documented for hip fracture [13]. There are also differences in mortality. As a result, probability models need to be calibrated to the epidemiology of fracture and death of any particular region. FRAX algorithms are available for several index countries (including Japan, China, the US, UK, Sweden, Turkey, France and Spain), and several more are being developed. Where a country is not represented (because of the lack of epidemiological data), a surrogate may be chosen. At present, there is no FRAX model available for Poland largely because of marked differences in fracture rates reported in different studies. For example, the crude incidence reported in women over the age of 5 years varies from 138 to 283/1, [14 16]. For this reason, the UK model has been used as a surrogate model for Poland [17]. The aim of the present study was to determine the applicability of the UK model in a Polish setting. Patients and methods Approximately 9, men and women have had a densitometric examination performed in Krakowskie Centrum Medyczne since 1994. The vast majority applied for testing following newspaper advertisements offering densitometry examination free of charge between 1994 and 1997. Of these, 37, had central dual-energy x-ray absorptiometry (DXA) measurements. From these, we selected women aged 5 years of age or more who had been examined on average 11 years previously (between 9 and 12 years), and who, in 29, were not older than 75 years. Of the 2,75 women who fulfilled these criteria, 1,214 women could not be recontacted by telephone (59%), and in 249, the original documentation was incomplete with regard to the FRAX risk factors (12%). Of the remaining 612 women, telephone interviews were completed in 51 women who were not receiving treatment with bone active medication. At baseline, all FRAX risk factors were documented by face to face interviews with medical staff using a structured questionnaire. In addition to glucocorticoid exposure and rheumatoid arthritis, secondary osteoporosis was defined as a history of hyperparathyroidism, diabetes, asthma, and early menopause. At follow-up, particular attention was given to the occurrence of a new fracture during the follow-up period. Information was available on the site of fracture but not on the date of fracture. The duration of follow-up was 9 years (79 patients), 1 years (143 patients), 11 years (161 patients) or 12 years (118 patients). BMD was measured by DXA (GE Lunar) at the lumbar spine (L2 L4) in all patients and at the femoral neck in 269 patients. Ten-year fracture probability was assessed with the UK version of the FRAX tool (version 3.) [12]. For each patient, we estimated the 1-year probability of a major osteoporotic fracture (hip, clinical spine, forearm or humerus) or of a hip fracture. The estimate of probability was made with clinical risk factors alone, and, where available, with femoral neck BMD. The estimates of 1- year probability were compared with the fracture outcomes reported in the follow-up. In addition, probabilities of a major osteoporotic fracture were compared with the reported incidence at these fracture sites (hip, clinical spine, forearm or humerus). Because of a skewed distribution, FRAX probabilities were shown as probabilities at the 1th, 25th, 5th, 75th and 9th percentiles of the distribution. The relationship between the FRAX probability of a major fracture and the
incidence of fractures at these sites was examined by logistic regression. Fisher s permutation test was used to compare women with and without fractures during follow-up. The expected number of fractures in 1 years was computed from the probabilities of a major osteoporotic fracture (clinical vertebral, hip forearm and humeral fractures). The observed fractures included fractures only at these sites. No account was taken of variations in the duration of follow-up. The relationship between observed and expected probabilities was examined using logistic regression. Results The baseline characteristics of the patients are shown in Table 1. The mean age was 61 years and the mean BMD T- score at the spine and femoral neck was 2.13 and 1.27 standard deviation (SD), respectively. Thirty percent of patients (n=147) reported a history of fracture. The probability of a major fracture was 9.9% when calculated without BMD and 1.2% with the inclusion of BMD. Hip fracture probabilities were approximately five times lower than the probability of a major fracture. Hip fracture probabilities were similar with or without the inclusion of BMD in the FRAX model. The similarity indicates that the cohort studied was not preferentially enriched by women with low BMD for age. At follow-up, the mean age of women was 71.8 years. Incident fractures were reported in 16 women (21%). Incident fractures of the forearm, humerus, spine and hip were reported in 89 women. Women who fractured after the baseline assessment were more likely to have had a prior fracture, had a lower T-score at the femoral neck and higher FRAX probabilities (Table 2). The distribution of fracture probabilities is shown at different percentiles of the distribution in Fig. 1. For hip fracture probability, there was an 11-fold difference between the 1th and 9th percentiles when the FRAX model excluded BMD. The range was fourfold in the case of probabilities of a major fracture. This dynamic range increased with the inclusion of BMD in the FRAX models to 49 and five, respectively indicating the more accurate characterisation of risk by the inclusion of BMD. In the case of major osteoporotic fractures, FRAX would predict 49.6 women with fractures in 1 years in the 51 women studied without BMD and 27.3 fracture cases in the 269 women in whom FRAX was calculated with the inclusion of BMD. The observed number of women with one or more major osteoporotic fracture was however greater (89 and 53, respectively). The observed/expected ratio was significantly greater than unity when the expected number was calculated without BMD (1.79; 95% confidence interval=1.44 2.21) and when BMD was included in the FRAX calculation (1.94; 95% confidence interval=1.45 2.54). The relationship between observed and expected probabilities is shown in Fig. 2 using logistic regression. Note that the figure shows data between approximately the 1th to 9th percentiles of FRAX probabilities. The discrepancy between the observed fracture probability and that provided by the FRAX model using the UK version of FRAX differed according to fracture probability. At high fracture probabilities, there was a good agreement between observed and expected fractures. Table 1 Baseline characteristics of 51 women Mean/prevalence SD Range a Available in 269 women N 51 Age (years) 61. 5.9 5 73 BMI (kg/m 2 ) 26.8 4.2 17.6 45.8 Previous fracture (%) 3 Parental history of hip fracture (%) 9 Current smoker (%) 14 Glucocorticoids (%) 3 Rheumatoid arthritis (%) 2 Alcohol intake 3 units daily (%) Secondary osteoporosis (%) 13 Femoral neck T-score (SD) a 1.29 1.1 3.6 to 1.7 Lumbar spine T-score (SD) 2.13 1.43 5.9 to 2. 1-year probability of major fracture without BMD (%) 9.9 6.1 2.5 59. 1-year probability of major fracture with BMD (%) a 1.2 6.5 2.8 38.7 1-year probability of hip fracture without BMD (%) 1.9 2.2.1 3.1 1-year probability of hip fracture with BMD (%) a 2. 2.6. 15.
Table 2 Comparison of baseline characteristics in women with and without an incident fracture Variable No incident fracture (n=395) Incident fracture (n=16) Two-sided p value b Age (years) 6.8±6. 61.6±5.5.23 BMI (kg/m 2 ) 27.±4.3 26.4±3.9.21 Previous fracture (%) 27% 41%.57 Parental history of hip fracture (%) 8% 12%.3 Current smoker (%) 12% 2%.67 Glucocorticoids (%) 3% 3% >.3 Rheumatoid arthritis (%) 2% 3% >.3 Alcohol intake 3 units daily (%) % % >.3 Secondary osteoporosis (%) 13% 14% >.3 Femoral neck T-score (SD) a 1.18±1.12 1.67±.97.19 1-year probability of major fracture without BMD (%) 9.5±6. 11.4±6.1.69 1-year probability of major fracture with BMD (%) a 9.6±6.4 12.±6.4.16 1-year probability of hip fracture without BMD (%) 1.8±2.2 2.4±2.1.19 1-year probability of hip fracture with BMD (%) a 1.7±2.4 2.7±2.9.16 a Available in 27 and 62 women Fisher s permutation test. Discussion The present study, undertaken in a convenience sample of women referred for BMD assessment, indicates that the UK version of FRAX can stratify fracture risk in a Polish setting. When fracture risk was estimated by percentile of FRAX probability, observed incidence rose progressively with increasing FRAX probabilities, irrespective of the model used (hip fracture or major osteoporotic fracture) or the use of the BMD result. Notwithstanding, FRAX significantly overestimated observed fracture risk by 79% and 94% when FRAX was calculated without or with BMD in the FRAX model, respectively. The overestimate was not linear in the sense that it was greater at low fracture probabilities and negligible at the higher fracture probabilities. These findings might lead one to the view that the use of the UK FRAX model in Poland gives misleading information due to errors of accuracy. Errors of accuracy may reside in either the fracture hazards or the death hazards, both of which contribute to fracture probability and both of which vary between countries [13]. In the case of the death hazard, the average life expectancy in Poland is 76.4 years, which is 3.7 years less than in the UK. Thus, the use of the UK FRAX model will overestimate fracture probability when used in a Polish setting. There are, however, a number of caveats to consider. Firstly, the sample of women studied is from an urban region of Poland. In the case of hip fracture, it is well established that fracture rates may vary twofold within a country. Higher fracture rates are reported from urban regions than in rural communities [18 2], and there Fig. 1 Ten-year probability of fracture computed with the UK FRAX tools in women from Krakow at given percentiles of the distribution Fracture probability (%) 2 Major fracture 15 6 4 Hip fracture Percentile 1 th 1 25 th 5 2 5 th 75 th 9 th no BMD BMD no BMD BMD
Fig. 2 The relationship between the 1-year probability (%) of a major osteoporotic fracture using FRAX and the observed incidence of fracture at these sites together with 95% confidence intervals. a FRAX computed without BMD and b with BMD. The diagonal dashed line shows the line of identity Observed fracture incidence.4 a Without BMD.3.2.5.4.3.2 b With BMD.1.1 5 1 15 2 5 1 15 2 Ten year probability of a major osteoporotic fracture (%) appears to be similar regional differences reported in Poland [14 16]. For this reason, it would be unwise to conclude that the UK version of FRAX is inappropriate to use in Poland without more national representation. A second consideration relates to the study design. The sampling frame was retrospective and subject to unquantified selection biases. The number of fractures is limited (n= 89) and even fewer in those with a BMD test (n=53). Incident fractures were documented by patient recall at the end of follow-up and thus subject to recall bias. However, recall bias might be expected to underreport incident fractures and thus increase rather than decrease the disparity between the observed and expected fracture rates. In addition, patients were not consistently followed up for 1 years, and the majority (56%) were followed up after 11 or 12 years. Again, this would tend to increase rather than decrease the disparity between the observed and expected fracture rates. A more important methodological factor is that FRAX takes account of the competing death hazard. In contrast, the present study documented fractures only in survivors. Patients with a major fracture (forearm fractures excepted) have a higher-than-average risk of death [3]. Thus, the study design would be expected to underreport incident fractures. It is not possible to estimate the quantum of this effect. In this general context, it is also relevant to recognise accuracy errors in the construct of FRAX. For example, major fracture rates should ideally be based on the incidence of a first fracture at any one site whereas current estimates are based on the incidence of fracture irrespective of whether this is a first or subsequent fracture at the same skeletal site. The overestimate that arises has been characterised for Sweden [21], but not for other countries. Incomplete epidemiologic data for fractures other than hip fracture are also likely to give errors of accuracy. Importantly, these accuracy errors have little impact on the efficiency with which the FRAX tool categorises risk. This is because they do not change markedly the rank order of fracture probability in any population. By way of illustration, the FRAX tool for the US was revised recently to take account of changes in hip fracture and mortality risks [22]. A white female aged 6 years at the 9th percentile of risk had a 1-year fracture probability of 53% using the old version (version 2.) and the revised estimate (version 3.) was 44%. Despite the difference in absolute risk, the correlation coefficients between range probabilities derived from versions 2. and 3. exceeded.99, so that the one could be accurately predicted from the other. In other words, an individual at the 9th percentile of risk would remain at the 9th percentile of risk with a revision of FRAX tool of the hip fracture or death risk. Thus, the consequences of improving accuracy reside more in the absolute number generated rather than in the rank order of risk. This is of little consequence to the management of patients or the interpretation of clinical studies, provided that the same FRAX tool is consistently used. There is a useful analogy with the different DXA devices available, where a substantial difference in femoral neck BMD is seen between Hologic and Lunar machines, but the T-score derived from these is more or less identical [23]. An exception arises when fracture probabilities are used in health economic analysis to inform practice guidelines. The analyses are dependent on the absolute values for probabilities, since for any given efficacy, the number of fractures saved is less, the lower the fracture probability. We conclude that the UK FRAX tool categorises fracture risk appropriately in this Polish cohort but significantly overestimated fracture risk, an effect more marked at lower fracture probabilities. It is not known whether the overestimate is due to errors in the fracture or death hazards used in the UK model or variations in the fracture risk within Poland. The UK model can be used to stratify risk, but caution is required in interpretation of absolute risk. Ultimately, many of these difficulties would be overcome with the development of a country-specific FRAX model
for Poland. Further epidemiologic studies are in progress to remedy the deficit. Acknowledgements The authors thank students of Jagiellonian University Medical College for collecting data: Barłowska M., Domagała M., Górna A., Jewiarz J., Jurgiełan A., and Król D. Conflicts of interest References None. 1. Kanis JA, Melton LJ, Christiansen C, Johnston CC, Khaltaev N (1994) The diagnosis of osteoporosis. J Bone Miner Res 9(8):1137 1141 2. Kanis JA, Gluer CC (2) An update on the diagnosis and assessment of osteoporosis with densitometry. Osteoporos Int 11:192 22 3. World Health Organization (24) Assessment of fracture risk and its application to screening for postmenopausal osteoporosis. Technical Report Series 843. WHO, Geneva, p 129 4. European Community (1998) Report on osteoporosis in the European community. EC, Strasbourg 5. Kanis JA, Delmas P, Burckhardt P, Cooper C, Torgerson D, on behalf of the European Foundation for Osteoporosis and Bone Disease (1977) Guidelines for diagnosis and management of osteoporosis. Osteoporos Int 7:39 46 6. Royal College of Physicians, Bone and Tooth Society of Great Britain, National Osteoporosis Society (22) Glucocorticoidinduced osteoporosis: guidelines on prevention and treatment. Royal College of Physicians, London 7. Royal College of Physicians, Bone and Tooth Society of Great Britain (2) Update on pharmacological interventions and an algorithm for management. Royal College of Physicians, London 8. National Osteoporosis Foundation (NOF) (23) Physician s guide to prevention and treatment of osteoporosis. National Osteoporosis Foundation, Washington, DC 9. Kanis JA, Johnell O, Oden A, De Laet C, Jonsson B, Dawson (21) Ten year risk of osteoporotic fracture and the effect of risk factors on screening strategies. Bone 3:251 258 1. Siris ES, Miller PD, Barrett-Connor E, Faulkner KG, Wehren LE, Abbott TA et al (21) Identification and fracture outcomes of undiagnosed low bone mineral density in postmenopausal women: results from the National Osteoporosis Risk Assessment. JAMA 286:2815 2822 11. Kanis JA, Borgstrom F, De Laet C et al (25) Assessment of fracture risk. Osteoporos Int 16:581 589 12. Kanis JA, Johnell O, Oden A, Johansson H, McCloskey EV (28) FRAX and the assessment of fracture probability in men and women from the UK. Osteoporos Int 19:385 397 13. Kanis JA, Johnell O, De Laet C, Jonnson B, Oden A, Ogelsby AK (22) International variations in hip fracture probability: implications for risk assessment. J Bone Miner Res 17:1237 1244 14. Roszkowska H, Goryński P, Wojtyniak B (1998) Hospitalizacja z powodu złamania szyjki kości udowej w Polsce w latach 1979 1995 wg płci, wieku i miejsca zamieszkania (Polish). [Hospitalisation of patients with femoral neck fracture in Poland in years 1979 1995 according to gender, age and place of residence]. Postepy Osteoartrol 1:15 156 15. Jaworski M, Lorenc RS (27) Risk of hip fracture in Poland. Med Sci Monit 13:26 21 16. Czerwiński E, Kanis JA, Trybulec B, Johansson H, Borowy P, Osieleniec J (29) The incidence and risk of hip fracture in Poland. Osteoporos Int 2:1363 1368 17. Czerwiński E, Badurski J, Lorenc R, Osieleniec J (21) Wytyczne w sprawie diagnostyki osteoporozy i oceny ryzyka złamania w Polsce (Polish) Guidelines of osteoporosis diagnosis and risk fracture assessment in Poland. Medycyna po Dyplomie 3:2 6 18. Kaastad TS, Meyer HE, Falch JA (1998) Incidence of hip fracture in Oslo, Norway: differences within the city. Bone 22(2):175 178 19. Sanders KM, Nicholson GC, Ugoni AM, Seeman E, Pasco JA, Kotowicz MA (22) Fracture rates lower in rural than urban communities: the Geelong Osteoporosis Study. J Epidemiol Community Health 56(6):466 47 2. Elffors L, Allander E, Kanis JA et al (1994) The variable incidence of hip fracture in Southern Europe. The MEDOS Study. Osteoporosis Intl 4:253 263 21. Kanis JA, Johnell O, Oden A et al (2) Long-term risk of osteoporotic fracture in Malmo. Osteoporos Int 11:669 674 22. Kanis JA, Johansson H, Oden A, Dawson-Hughes B, Melton LJ 3rd, McCloskey EV (21) The effects of a FRAX((R)) revision for the USA. Osteoporosis Int 21:35 4 23. Binkley N, Kiebzak GM, Lewiecki EM et al (25) Recalculation of the NHANES database SD improves T-score agreement and reduces osteoporosis prevalence. J Bone Miner Res 2:195 21