Fracture Prediction From Bone Mineral Density in Japanese Men and Women

Authors


  • The authors have no conflict of interest

Abstract

In a cohort of 2356 Japanese elderly, after adjusting for age and prevalent vertebral fracture, baseline BMD predicted the risk of spine and hip fracture with similar RR to that obtained from previous reports in whites. The RR per SD decrease in BMD for fracture declined with age.

Introduction: Low bone mineral density (BMD) is one of the most important predictors of a future fracture. However, we are not aware of any reports among Japanese in Japan.

Materials and Methods: We examined the association of BMD with risk of fracture of the spine or hip among a cohort of 2356 men and women aged 47–95 years, who were followed up by biennial health examinations. Follow-up averaged 4 years after baseline measurements of BMD that were taken with the use of DXA. Vertebral fracture was assessed using semiquantitative methods, and the diagnosis of hip fracture was based on medical records. Poisson and Cox regression analysis were used.

Results: The incidence was twice as high in women as in men, after adjusting for age. After adjusting for baseline BMD and prevalent vertebral fracture, however, the gender difference was no longer significant. Age, baseline BMD of spine and femoral neck, and prior vertebral fracture predicted vertebral fracture and hip fracture. Loss of absolute BMD of the femoral neck predicted spine fracture, after adjusting for baseline BMD; rates of change in percent BMD, weight, height, body mass index, and age at menopause did not. The predictive value of baseline BMD for vertebral fracture risk was similar in men and women. The relative risk (RR) for vertebral fracture and hip fracture per SD decrease in BMD declined with age, after adjustment for prevalent vertebral fractures.

Conclusions: Baseline BMD, loss of femoral neck BMD, and prior vertebral fracture predict the risk of spine and hip fracture in Japanese with similar RR to that obtained from previous reports in whites. The RR per SD decrease in BMD for fracture declined with age, suggesting that factors other than BMD might play a greater role in the elderly.

INTRODUCTION

Bone mineral density (BMD) is widely recognized as one of the main factors predictive of bone fractures. In the United States and Europe, a number of prospective studies have demonstrated that BMD is one of the most important predictors of a future fracture. These studies indicated that measuring any skeletal site could predict future fracture with relative risk (RR) of around 2 per SD decrease in BMD.(1–3) However, there has been no report to our knowledge based on longitudinal study on BMD and the subsequent fracture risk of Japanese in Japan. Epidemiological studies indicate that BMD is lower,(4,5) prevalence of vertebral fracture is higher,(6) but the incidence of hip fracture is lower(7,8) among Japanese than in whites in the United States and Europe. Furthermore, the ability of BMD to predict future fracture was similar among Japanese-Americans in Hawaii to that in whites in the United States and other countries.(1) The aim of this study was, therefore, to assess the ability of BMD measurements and prior vertebral fracture to predict future spine and hip fracture among Japanese men and women using a population-based prospective study with an average follow-up period of ∼4 years.

MATERIALS AND METHODS

The Adult Health Study (AHS) was established in 1958 to document the late health effects of radiation exposure among atomic bomb survivors in Hiroshima and Nagasaki. The original AHS cohort consisted of about 15,000 atomic bomb survivors and about 5000 controls selected from residents in Hiroshima and Nagasaki using the 1950 national census supplementary schedules and the Atomic Bomb Survivors Survey. AHS subjects have been followed through biennial medical examinations since July 1, 1958. The participation rate has been around 80% throughout this period. Details concerning recruitment and examination of participants have been reported.(9–11)

A total of 2613 AHS subjects, aged 47–95 years, underwent physical examinations in Hiroshima in the 1994–1995 examination cycle. Among them, 2356 subjects (763 men and 1593 women) were selected for our analyses. Two hundred fifty-seven were excluded because they had conditions associated with impaired bone metabolism (such as hyperparathyroidism, renal osteodystrophy, and bilateral oophorectomy), they took medication that affects bone metabolism (such as corticosteroids, calcitonin, vitamin D, bisphosphonate, estrogen, and vitamin K), or they were examined only once during the period from 1994 to 2000. The age distribution of the study population is shown in Table 1.

Table Table 1. Age Distribution of the Study Population
original image

BMD at the spine (L2-L4, antero-posterior direction) and proximal femur were measured at each biennial health examination using DXA (QDR-2000; Hologic Inc, Waltham, MA, USA). An anthropomorphic spine phantom was scanned daily to calibrate the instrument. There was no drift in machine performance during the study period. The precision of the instrument was carefully monitored over the study period using anthropomorphic phantoms and was less than 1%. Paired initial and follow-up spine and proximal femoral scans were analyzed on each patient using the automated compare feature of the machine software. All paired images of spine and hip were reviewed by a trained observer (SF) to exclude inadequate measurements. Measurements in which areas of the paired spine images differed more than 5% were excluded. Rates of changes in BMD values and loss of absolute BMD values were divided by the time interval between measurements. Rates of change in BMD were expressed as percentage per year (%/year) relative to the baseline value for each person.

Trained nurses interviewed the subjects about age at menopause, cause of menopause, and history of bone fracture at the biennial AHS examinations. Measurements of height and weight were also made at each examination.

Vertebral fracture was diagnosed by lateral and posterior-anterior chest and spinal X-ray examination. The vertebral fracture was assessed using semiquantitative assessment,(12,13) in which a new radiographic vertebral fracture was defined as a decrease of at least 20% in height of any vertebral body from initial reading to the end the study. The diagnosis of hip fracture was made by a physician based on the history of hip fracture taken at the biennial AHS examinations. Diagnoses of disease were coded using the International Classification of Diseases (ICD). Pathological fractures or fractures caused by traffic accidents or falls from heights were excluded.

Follow-up of all subjects began in the 1994–1995 examination cycle. The accumulation of each subject's person-years of risk ended at the date of first diagnosis of vertebral fracture or hip fracture or the date of the last examination before December 2000. Attrition from 1994 to 2000 of the AHS participants because of death or hospitalization was approximately 7% at each examination cycle.

All AHS participants gave written informed consent for BMD measurements, spine X-ray examination, and all other health examinations.

Incidence rates of vertebral fracture were examined by Poisson regression analysis, assuming the number dij of incident cases of vertebral fracture to be an independent Poisson variable with the expected value of E(dij)= PYij′λij, where PYij and λij are the person-years at risk and the vertebral fracture risk, or hazard, respectively, in the ijth stratum produced by making a cross-classification of all categories of the variables: sex (si), age at baseline (ai), baseline BMD (bi), prevalent vertebral fracture (VFi), and follow-up period (pj). The form of λ in the ijth stratum was assumed to be a log-linear function; namely:

equation image

where ageij is the age that the ijth stratum members attained at the jth follow-up period, defined by ageij= ai+pj. The terms ai and pj are mean values of age at baseline and follow-up period for each stratum, respectively. The parameters were estimated using the maximum likelihood method. The significance of the parameter estimates was based on the likelihood ratio test.

The effect on the future fracture risk of potential baseline covariates (age, weight, height, body mass index [BMI], age at menopause, BMD at baseline, and prevalent vertebral fracture) was investigated using Cox regression models. After considering statistical significance and goodness of fit of the model, age, prevalent vertebral fracture, baseline BMD, and interaction between age and BMD were selected as significant predictors. We determined RR in the model including interaction between age and BMD using “dummy” variables for the different age groups. We calculated T-score and used BMD values of 1 SD based on the published database(14) of the 1089 Japanese young adult women (mean BMD and 1 SD are 1.011 and 0.119 g/cm2 for spine and 0.763 and 0.109 g/cm2 for femoral neck, respectively).

RESULTS

The mean ages in the 1994–1995 examination of the subjects ± SD for men were 62.9 ± 9.8 years, ranging from 47 to 92 years; 65.4 ± 9.8 years (range, 47–95 years) for women. The mean follow-up was approximately 4 years, ranging from 1.2 to 6.7 years. The characteristics of the population taken as “baseline” is shown in Table 2. Approximately 90% of women were postmenopausal, and the average age at menopause was 47.7 years. Twenty-four men (3.2%) and 151 women (9.5%) had prevalent vertebral fractures in the 1994–1995 examination period, and 27 men (3.5%) and 149 women (9.5%) developed new vertebral fractures. Twenty-one women but no men developed a new hip fracture during the 4-year follow-up period. Regarding the 257 people (31 men and 226 women) we excluded from analysis, 5 men (16.1%) and 35 women (15.5%) had prevalent vertebral fractures.

Table Table 2. Characteristics of the Study Population
original image

Incidence of vertebral fracture

The incidence of vertebral fracture increased approximately exponentially with age (Fig. 1). The annual incidence for women in their 70s was 40 per 1000 person-years (PY); 84 per 1000 PY for women in their 80s. The incidence rates of vertebral fracture stratified according to the presence or absence of a baseline prevalent vertebral fracture are shown in Table 3. Women without a vertebral fracture at baseline examination had a lower incidence of vertebral fractures, ranging from 5.2 to 24.5/1000 PY in their 50s and 70s, respectively, compared with 36.2–88.0/1000 PY for women with a prior vertebral fracture (Table 3).

Table Table 3. Incidence of Vertebral Fracture in Men and Women Sorted by Prevalent Vertebral Fracture Detected at the Baseline Examination
original image
Figure FIG. 1..

Incidence of vertebral fracture by attained age among the overall population studied. Solid line shows men, and dotted line shows women.

The incidence was twice as high in women as in men, after adjusting for age. After adjusting for baseline BMD at spine or femoral neck and prevalent vertebral fracture, however, the gender difference in the age-specific incidence of vertebral fracture was no longer significant (Fig. 2).

Figure FIG. 2..

Incidence of vertebral fracture with spine or femoral neck BMD in men and women. Lines display adjusted age of 65 years old and no prevalent vertebral fracture at baseline.

Influence of age, BMD, and prevalent fracture on the incidence of vertebral fracture in men and women

In both men and women, multivariate analyses showed that age, baseline BMD at the spine and hip, and prevalent vertebral fracture independently predicted spine fracture. The RR for vertebral fracture per 1 SD decrease in BMD was 1.54 (95% CI, 1.26–1.88) for spine BMD and 1.78 (95% CI, 1.37–2.33) for femoral neck BMD among women of all ages. Thus, measurements at the spine and femoral neck had similar predictive abilities for spine fracture. Vertebral fracture incidence per decrease of 1 SD in BMD was modified by age in both men and women, after adjusting for age and prevalent vertebral fractures. The RR for vertebral fracture per 1 SD decrease in BMD at spine and femoral neck declined with age in both men and women (Fig. 3), although the RRs for vertebral fracture per 1 SD decrease in BMD were marginally significant at ages over 60 years for men. After adjusting for age and prevalent fracture, the RR per 1 SD decrease in BMD was similar in men and women (sex difference, p = 0.18).

Figure FIG. 3..

RR for vertebral fracture per 1 SD decrease in BMD at spine or femoral neck by age category, after being adjusted for prevalent vertebral fracture.

After adjusting for age and baseline spine BMD, prevalent vertebral fracture was associated with a 4.4-fold (95% CI, 1.5–13.5) increase in future vertebral fracture risk among men and 2.94-fold (95% CI, 2.0–4.3) among women (Fig. 4). Similar RR for future vertebral fractures was obtained among those with prevalent fractures, after adjusting for baseline femoral neck BMD; the RR was 4.5 (95% CI, 1.6–13.0) among men and 3.2 (95% CI, 2.2–4.6) among women (Fig. 4).

Figure FIG. 4..

RR for spine or hip fracture among women with prevalent vertebral fracture compared with those without prevalent vertebral fracture, after being adjusted for age and BMD.

After adjusting for age, baseline BMD, and prevalent vertebral fracture, loss rates of absolute BMD values in the femoral neck predicted vertebral fracture, but loss rates of spine BMD values were not associated. The RR for vertebral fracture for each 0.05-g/cm2 loss from the baseline BMD in femoral neck BMD was 1.50 (95% CI, 0.97–2.32; p = 0.06) for men and 1.22 (95% CI, 1.03–1.45; p = 0.02) for women. However, rate of change in BMD expressed as a percent of baseline BMD was not significantly associated with vertebral fracture risk. Furthermore, baseline body weight, height, BMI, and age at menopause did not predict spine fracture, after adjusting for baseline BMD and prevalent fracture.

Influence of age, BMD, and prevalent fracture on the incidence of hip fracture in women

All of the women who sustained a new hip fracture were at least 66 years old. The average age was 79.1 ± 7.4 years. Their spinal BMDs averaged 0.698 g/cm2 (T-score −2.6), ranging from 0.390 to 1.042 g/cm2, and their femoral neck BMD averaged 0.485 g/cm2 (T-score −2.5), ranging from 0.379 to 0.652 g/cm2 at the baseline examination. Among women, age, baseline BMD at the spine and hip, and prevalent vertebral fracture independently predicted future hip fracture risk. The RR for hip fracture per 1 SD decrease of BMD at both the spine and femoral neck decreased with increasing age, after adjusting for prevalent vertebral fractures (Fig. 5). The RR for hip fracture per 1 SD decrease in femoral neck BMD was much higher than that for spine BMD. The RR per 1 SD decrease of spinal BMD was 2.13, 1.35, and 1.00 among women in their 60s, 70s, and 80s, respectively, while the RR per 1 SD decrease of femoral neck BMD was 5.49, 2.94, and 2.37, respectively.

Figure FIG. 5..

RR for hip fracture per 1 SD decease in BMD by age category, after being adjusted for prevalent vertebral fracture.

Prevalent vertebral fracture was predictive for hip fracture, with an RR of 5.16 (95% CI, 1.94–14.42) after adjusting for age and spine BMD, and with an RR of 2.91 (95% CI, 1.10–7.90) after adjusting for age and femoral neck BMD (Fig. 3).

DISCUSSION

We investigated the association between baseline BMD and fracture incidence as a predictor of vertebral and hip fracture risk in a population-based cohort consisting of 763 Japanese men and 1593 women, aged 47–96 years. There have been several prospective studies investigating the ability of BMD at baseline to predict vertebral fracture in women in the United States and Europe.(1,15–18) Studies on the Japanese are more rare. Ross et al. and Wasnich et al.(15,16) found that a 1 SD decrease in BMD measured at the distal radius, proximal radius, calcaneus, or lumbar spine was associated with an RR of about 2.3 for vertebral fracture among Japanese-Americans in Hawaii. In a meta-analysis,(1) most measurement sites had virtually the same predictive ability for a decrease of 1 SD in BMD with an RR of 1.5–2, except for spine or calcaneus BMD for vertebral fracture and hip BMD for hip fracture. BMD at the spine and calcaneus seemed to have a better predictive ability for spine fractures with RRs of 2.3 and 2.4 compared with other sites, and femoral neck BMD was better for predicting hip fractures, with an RR of 2.6. Previous epidemiological studies reported that BMD was lower, prevalence of vertebral fracture was higher, and incidence of hip fracture was lower among Japanese compared with whites in the United States and Europe.(2–8) The question arises whether the predictive ability for fracture per a 1 SD decrease in BMD is the same for Japanese as for whites. In the present study, the CI of relative risk for vertebral or hip fracture overlapped the values of other studies. We used BMD values of 1 SD from the database of Japanese young adult women.(14) Mean BMD and 1 SD of Japanese young adult women corresponds to 1.011 and 0.119 g/cm2 for spine and 0.763 and 0.109 g/cm2 for femoral neck, respectively. Manufacturer reference values of Hologic QDR for white young adult women show that mean BMD and 1 SD are 1.084 and 0.111 g/cm2 for spine and 0.895 and 0.100 g/cm2 for femoral neck, respectively. Mean BMD values are lower in Japanese women than in white women, but the BMD values of 1 SD are almost same between the two populations. It is known that Japanese and whites obviously differ in body size. Body size provides greater strength and also greater load-bearing forces for a given level of true bone density. This study shows that BMI or body weight was not associated with the incidence of fracture, after adjusting for BMD. Taking together, we conclude that the relative risk for vertebral or hip fracture per decrease in BMD is similar in Japanese to that in other populations, including whites in the United States and Europe and Japanese in Hawaii, even if body size and BMD at the same age differ between these populations.

Nevitt et al.(19) reported that the RRs of fractures associated with a 1 SD decrease in BMD were similar in two age groups consisting of white women aged 65–79 years and women 80 years and older, whereas the EPIDOS study demonstrated that the RR was significantly lower in older women than in women younger than 80 years.(20) The present study indicated that the RR of a 1 SD decrease in BMD for both vertebral and hip fracture decreased with age in both men and women after adjusting for prevalent vertebral fractures. This finding suggests that BMD remains a predictor of fracture risk in the elderly, although the contribution of fracture risk factors such as falls, fall severity, and neuromuscular impairment other than BMD may be greater to develop not only hip fracture but spine fracture with among eldest population of women. One limitation of our analysis is that some of the frailest people with a high risk of fractures, including those with the most severe osteoporosis and lowest BMD, were excluded or are probably unable or unwilling to participate in the study, which may have biased the study. Also, adjusting for prevalent vertebral fractures may have influenced the observed relationship between BMD and fracture risk at different ages. Vertebral fractures become more prevalent with age, and it is likely that people with vertebral fractures also have the lowest BMD values and most severe osteoporosis. Thus, adjusting for prevalent vertebral fractures may have reduced the magnitude of associations between BMD and fracture risk compared with unadjusted analyses. An overall reduction in the association between BMD and fracture risk after adjustment for vertebral fracture was reported earlier.(15)

There have been relatively few studies of association of BMD and fracture risk in men. It has been argued that uncertainties remain about the effect of gender on the BMD-fracture risk relationship and that current thresholds based on absolute levels of BMD used for women may unreasonably minimize the number of men who would be identified at a risk of fracture.(21) In a longitudinal study among the residents in Dubbo, Australia, each decrease in femoral neck BMD of 1 SD was associated with a 2-fold increase for men in risk of osteoporotic fracture, including hip, distal radius, ribs, humerus, ankle, and foot.(22) Melton et al.(23) reported that the predictive ability of BMD measurement for osteoporotic fracture was similar for men and women, although the risk increased more rapidly in women in a population-based cross-sectional study. Prospective studies demonstrated that the increase in risk of hip or vertebral fracture per SD decrease in BMD was similar between men and women.(24,25) Our study also shows that the RR of vertebral fracture per decrease in SD was the same in men and women. However, the magnitude of RR for vertebral fracture per 1 SD decrease in BMD at spine and femoral neck was similar for both men and women but did not attain statistical significance after age 60 in men, perhaps because the sample size for men was too small to attain statistical significance. In addition, errors in measuring BMD at the spine, especially in men aged 60 years and older in men, is confounded by the increasing prevalence with age of osteoarthritis and hypertrophic changes. This also might partly explain the lack of association between spinal BMD and subsequent vertebral fracture risk in elder men. More recent reports demonstrated that the RR of vertebral or hip fracture in men and women is the same at any absolute level of bone density,(24–26) which is compatible with the present study. De Laet et al.(27) and Blake and Fogelman(28) reported that men may fracture at higher BMD values than women because of the different BMD distribution, although fracture risk in men seemed to be similar at the same BMD. They proposed that absolute risk should be used rather than RR for the purpose of intervention decisions.

Numerous studies have reported increased risks of hip, spine, and other fractures among people who had previous clinical diagnosed fractures or radiographic evidence of vertebral fractures. Klotzbuecher et al.(29) summarized these studies, noting that peri- and postmenopausal women with prior fractures had an ∼2-fold risk of subsequent fracture compared with women without prior fractures. An exception to this was that women with prior vertebral fractures had ∼4 times the risk of subsequent vertebral fracture. In our population, the risk of vertebral fracture increased by four times for men and by three times for women with prevalent vertebral fractures, which is similar to the RR reported by others.(29) Prevalent vertebral fracture was more predictive for hip fracture after adjusting for spine BMD compared with adjusting for that of femoral neck BMD. Because the predictive ability of spine BMD for future hip fracture is relatively weak compared with that of femoral neck BMD, the RR for prevalent vertebral fracture may be higher when adjusting for spine BMD than when adjusting for hip BMD (Fig. 5).

Other studies have reported that rates of bone loss or absolute bone mass loss can predict future fracture risk.(26,30,31) Our study also found that loss of femoral neck BMD, but not spine BMD, adjusted for baseline BMD, predicted spine fracture, indicating that accelerated decline in bone density is associated with increased risk of subsequent fractures. As noted earlier, hypertrophic changes may influence spinal BMD measurements and impair their ability to predict fracture risk in older persons.

A few studies have investigated the incidence of vertebral fracture in white populations and Japanese in Hawaii. In women over 70 years of age among Japanese-Americans in Hawaii,(26) the incidence of new vertebral fracture was approximately 22 per 1000 PY in women 65–69 years old. In a cohort study in Rotterdam, Van Der Klift et al.(32) reported that the incidence of vertebral fracture was 7.8, and 19.6/1000 PY at the ages of 55–65 years and over 75 years for women, respectively, and 5.2 and 9.3/1000 PY for men. These incidence rates were about one-half of our incidence rates. Comparison with these studies, however, is hampered by the fact that different methods for defining vertebral fractures were used.

There are some limitations to our findings. First, the subjects included atomic bomb survivors, who may differ from the general population in Japan. There are no indications from earlier studies of this cohort that radiation affected BMD and fracture frequency(4,11,33) so our findings may be relevant to the Japanese population. Second, the prevalence of vertebral fracture at the baseline examination was slightly lower in the present study than that in the same cohort in an earlier report.(6) This is because we excluded people who were taking treatments for osteoporosis. Age-specific BMD values in the present study population may be higher than that in the original AHS population, including those who undertook medication for osteoporosis. Therefore, the age-specific incidence of vertebral fracture may be lower than that of the age-matched general Japanese population.

In summary, we find that baseline BMD can predict vertebral fracture in both men and women and that the predictive ability is similar in both sexes. Furthermore, the RR of vertebral fracture per 1 SD decrease in BMD among Japanese in Japan is similar to that among whites and Japanese in Hawaii as reported previously. BMD is, therefore, a good predictor of fracture risk in the Japanese elderly. On the other hand, the RR of vertebral and hip fracture per 1 SD decrease in BMD declined with age in both men and women, suggesting that factors other than BMD may play an increasingly important role as people age.

Acknowledgements

We thank Dr Philip D Ross, Merck Research Laboratories, for his comments, suggestions, and editorial assistance. This publication is based on research performed at RERF, Hiroshima and Nagasaki, Japan. RERF is a private nonprofit foundation funded equally by the Japanese Ministry of Health, Welfare, and Labor and the U.S. Department of Energy through the National Academy of Sciences.

Ancillary