Bone density and risk of hip fracture in men and women: cross sectional analysisBMJ 1997; 315 doi: http://dx.doi.org/10.1136/bmj.315.7102.221 (Published 26 July 1997) Cite this as: BMJ 1997;315:221
- Chris E D H De Laet, resident in epidemiologya,
- Ben A van Hout, senior econometristb,
- Huibert Burger, resident in epidemiologya,
- Albert Hofman, professor of epidemiologya,
- Huibert A P Pols, senior endocrinologistc
- a Department of Epidemiology and Biostatistics, Erasmus University Medical School, Rotterdam, Netherlands
- b Institute for Medical Technology Assessment, Erasmus University Rotterdam, Netherlands
- c Department of Internal Medicine III, Erasmus University Medical School, PO Box 1738, 3000 DR Rotterdam, Netherlands
- Correspondence to: Dr Pols
- Accepted 25 April 1997
Objective: To determine the relative contribution of decline in bone density to the increase in risk of hip fracture with age in men and women.
Design: Incidence data of hip fracture from the general population were combined with the bone density distribution in a sample from the same population and with a risk estimate of low bone density known from literature.
Setting: The Netherlands.
Subjects: All people with a hospital admission for a hip fracture in 1993, and bone density measured in a sample of 5814 men and women aged 55 years and over in a district of Rotterdam.
Main outcome measure: One year cumulative risk of hip fracture by age, sex, and bone density measured at the femoral neck.
Results: A quarter of all hip fractures occurred in men. Men reached the same incidence as women at five years older. Controlled for age, the risk of hip fracture by bone density was similar in men and women. The risk of hip fracture increased 13-fold from age 60 to 80; decrease in bone density associated with age contributed 1.9 (95% confidence interval 1.5 to 2.4) in women and 1.6 (1.3 to 1.8) in men.
Conclusions: The risk of hip fracture by age and bone density is similar in men and women. The decrease in bone density associated with age makes a limited contribution to the exponential increase of the risk of hip fracture with age.
The risk of hip fracture increases exponentially with age in both men and women
Men have about the same risk of hip fracture five years later than women
The risk of hip fracture by age and bone density is similar in men and women
The difference in age specific incidence is explained completely by the different bone density in men and women
The contribution of decline in bone density to the exponential increase in risk of hip fracture with age is relatively small
The number of people with fracture of the hip is increasing rapidly and by the year 2050 may exceed 6 million a year worldwide, up from 1.6 million in 1990.1 The aging of the population is the most important reason for this increase. In addition, the age specific incidence of hip fractures has also increased in several countries, including the Netherlands.1 2 Hip fractures are a major cause of mortality and disability in elderly people and an important burden for the health services in many countries.3
As most hip fractures occur in women, most attention has focused on bone loss in women, predominantly around the menopause. Less is known about the relation of hip fractures with bone loss later in life, and the high incidence of hip fracture in older men is largely neglected.4
Detailed quantitive knowledge about the effect of age and bone density on the absolute risk of hip fracture is necessary to evaluate the potential benefit of interventions aimed exclusively at bone density. The association of low bone mass with an increased risk of hip fracture is well documented.5 The strong increase of risk with age and the bone loss associated with age are also evident,6 but the effect of both determinants together is poorly understood. This information could be obtained directly from follow up studies, but the numbers and time required make those studies difficult to accomplish. Combination of data can, however, lead to indirect estimates of the absolute risk comparable with the approach used previously to estimate the lifetime risk of hip fracture.7
In the present study we combined cross sectional data on bone mineral density from a population based sample of elderly men and women living independently with incidence data on hip fracture from a national registry in the Netherlands. In combination with data from the literature, this allowed us to estimate the effect of age and bone density on the risk of hip fracture in men and women.
Distribution of bone mineral density
The Rotterdam study, started in 1991, is a prospective follow up study of the occurrence and determinants of disease and disability in elderly people. The design of this study has been described.8 The study focuses on four primary topics of research: neurogeriatric diseases, cardiovascular diseases, locomotor diseases, and ophthalmological diseases. All 10 275 men and women aged 55 and over living in a district of Rotterdam were invited to participate. The study was approved by the appropriate medical ethics committee, and participants provided written informed consent. From those eligible, 7983 participated, bringing the overall response rate of this study to 78%.
The baseline survey included an initial home interview followed by two visits to the research centre for a series of clinical examinations and laboratory assessments. Those baseline assessments included dual energy x ray absorptiometry scans of the femoral neck.
Methods of measuring bone mineral density and data on bone density in a subsample of 1762 subjects have been reported.9 The present study used the data on femoral neck bone density from the total study population. This site was chosen because of the growing consensus that prediction of fractures is best done with site specific measurements.3 People in nursing homes (11%) did not visit the research centre and thus were not eligible for bone density measurements.
We present the results for men and women separately, using the age on the day of the bone density measurement. The bone density distribution by age and sex is presented in 5 year age classes; additionally it was analysed continuously by linear regression. This regression model was extended with quadratic and cubic terms to detect a possible non-linear association between age and bone density. As obesity is well known to affect bone density,9 10 and as in this study body mass index seemed to be related to age, it was added to the regression model as a potential confounder. The results are presented with 95% confidence intervals.
Distribution of hip fractures
The SIG (Foundation for Health Care Information) is a national registry that collects various data related to health care.11 All admissions to hospital in the Netherlands are included in this registration as is most of the information from nursing homes. In the Netherlands virtually all patients with a hip fracture are treated clinically. Therefore, hospital data give accurate information about the incidence of hip fractures.
Data for hip fractures in 1993 (International Classification of Diseases, ninth revision (ICD-9) code 820xx) were collected from the detailed SIG hospital registration data. They were combined with Dutch demographic data for 1993 from the Dutch Central Bureau for Statistics.12 The data were aggregated in one year age classes and a best fitting function estimated with the SPSS curve fitting facility.13
Probability of hip fracture
The relative risk for hip fractures, expressed as relative risk per SD decrease in bone density measured at the femoral neck, was estimated by Cummings et al to be 2.6 (95% confidence interval 1.9 to 3.6) in women.14 Combining this relative risk with data on incidence and bone density made it possible to estimate the probabilities of hip fracture by age, sex, and bone density. The mathematical details are given in the Appendix 1. We used the same relative risk estimate for men. We also estimated the isolated effects of aging and decline in bone density related to age and calculated confidence intervals for these separate effects by using the 95% confidence intervals of the relative risk per SD decrease in bone density.
Distribution of bone mineral density
Table 1 shows the overall characteristics of the study population. From the 7086 people eligible, bone density data were obtained for 5814 (82%). This response rate remained above 70% up to the age of 85 years; in people aged over 85 the response dropped to 54%. Men were slightly younger than women (mean 67.6 (SD 7.6) years v 68.5 (8.3) years). The age at menopause was the same in all age groups (48.9 (5.2) years).
The bone density values, stratified by age and sex, were normally distributed, and the SD was almost constant over the age categories. Bone denisty declined linearly, and introducing quadratic and cubic terms did not improve the model. The apparent decrease in bone density at the femoral neck was 0.0046 (95% confidence interval 0.0040 to 0.0051) g/cm2/year for women and 0.0031 (0.0024 to 0.0038) g/cm2/year for men. Correction for body mass index changed those values only slightly (0.0050 g/cm2/year for women and 0.0028 g/cm2/year for men).
Distribution of hip fractures
In the Netherlands in 1993 there were 15 107 registered hospital admissions for hip fractures in a population of 15 million, a quarter of which occurred in men. The one year incidence of hip fracture (per 100 000) increased from around 40 at age 55-59 to about 3150 over age 95 in men and from around 40 to about 4450 in women. In each age group, the incidence of hip fracture in men was equivalent to that in women approximately five years younger. Figure 1) shows the one year cumulative incidence of hip fractures by age and sex with the fitted curves; details of these functions are given in Appendix 2.
Probability of hip fracture
From the preceding results we estimated the probability of hip fracture by age, sex, and bone density (Appendix 2). Figure 2 represents the association of the incidence of hip fracture with bone density at the femoral neck for different ages in men and women. Comparing an 80 year old woman with average bone density with a 60 year old woman, we found a relative risk for hip fracture of 13.6. When we separated the effects, age contributed 7.1 (5.7 to 8.8) to this relative risk, and age related decline in bone density contributed 1.9 (1.5 to 2.4). For men the relative risk was 12.7; the contribution of age was 8.2 (7.1 to 9.5) and of age related decline in bone density was 1.6 (1.3 to 1.8).
The magnitude of the relative risk per SD change in bone density affected the slope of the risk function (fig 3), which shows the curves for the central estimate (2.6) together with the curves at the lower and upper limits of the 95% confidence interval (1.9 to 3.6). As the incidence of hip fracture specific for age remains constant the risk of low bone density becomes higher, and the risk of high bone density becomes smaller when we assume a higher relative risk. The opposite happens at the lower confidence limit.
After the age of 60 the incidence of hip fracture is consistently lower in men than in women of the same age. Men have about the same risk of hip fracture five years later than women. Though the age related decline in bone density is larger in women, the risk of hip fracture when age and bone density are considered together is remarkably similar in men and women. The five year difference in the age specific incidence of hip fracture between men and women can, in this study, be explained by the different bone density distributions at those ages.
Though the risk for hip fracture increased 13-fold from age 60 to age 80 in both men and women, the age related decline in bone density explained merely a doubling of this risk. The rest of the increased risk is explained by other determinants of risk that have been accounted for by using age as a surrogate and that are approximately equal in men and women. Previous research identified several skeletal and extraskeletal determinants.15 Though bone density is not the main component of the increased risk of hip fracture in old age, risk would be substantially reduced if the age associated decline in bone density between the ages 60 and 80 could be excluded: a 36% reduction in men and 48% in women. Recent clinical trials indicate that part of this risk reduction might be achievable.16 17
The stronger effect of age on risk of fracture in general and of hip fracture in particular has been observed in women,18 but previous studies were based on relatively small numbers of fractures. Our design allowed us to use the information from more than 15 000 hip fractures in the analysis. The data presented here, however, were derived from several sources, which involves some assumptions that need to be examined. We assumed that the distribution of femoral neck bone density in the Netherlands corresponds to the distribution in this study. Even though our sample was population based, it could have been influenced by selection bias: healthy people could have been overrepresented. The high response rates indicate that this effect was probably small.
More importantly, the sample included only people who were living independently. The more frail patients in nursing homes, presumably with lower average bone density, were excluded, resulting in an underestimation of the age associated decline of bone density. Fewer than 9% of elderly people aged under 80 were in nursing homes, but this proportion rose greatly at higher ages. This means that the validity of the data seems assured up to the age of 80, but that the effect of the age associated decline in bone density will probably be somewhat higher than estimated in the older age categories. The age associated decline in bone density that we found was of the same magnitude as in other cross sectional, population based studies, although the absolute levels are slightly higher.19 20 21
Finally, cohort effects cannot be excluded as the bone density data used in this study are cross sectional. If present, these cohort effects would affect the estimated rate of bone loss but not the risk function.
Other risk indicators
In the analysis, age was used as a surrogate marker for several risk indicators, including propensity to fall, types of fall, muscle strength, and bone quality. We used the distribution of bone density and the age related decline in bone density without correction for height, weight, or for the age at menopause as we were interested in the combined effect of these determinants. Moreover, the confounding effect of body mass index was small, and in women the age at menopause was unrelated to age at fracture.
Choice of relative risks
We assumed the relative risk of 2.6 per SD to be the same at all ages, and we also assumed this relative risk applies to the Netherlands. This relative risk estimate influences the slope of the association between incidence of hip fracture and bone density but it does not alter the level of those curves, as is clear from figure 3. It could, however, influence the contribution of the age related decline in bone density to the risk of hip fracture. But, even at the upper confidence limit, this merely doubles the risk over 20 years of aging. It was previously shown in women that bone mineral density predicted fractures equally well at different ages up to the age of 80.22 Additionally, in a recent meta-analysis the relative risk estimate remained at 2.6 while the confidence interval narrowed slightly (2.0 to 3.5).5 This same study also indicated that the estimates of relative risk for different measurement and fracture sites seem to be comparable in different parts of the world.
As no relative risk based on large samples was available for men we assumed the same relative risk in men and women, as others have suggested23 24 and as was confirmed in a recent follow up study of bone density measurements in 752 men in Australia.25 That study estimated the relative risk for hip fractures per SD lower bone density at the femoral neck at 2.9 (1.7 to 5.0). This seems compatible with our a priori assumption of no difference. When we applied this point estimate of 2.9 the results changed only slightly; the contribution of aging (age 60 to age 80) became 7.8 (6.1 to 10) and that of bone density decline 1.6 (1.3 to 2.1), supporting our conclusions.
The risk of hip fracture, when expressed as a function of bone density and age, is remarkably similar in men and women, and the difference in age specific incidence of hip fracture can be explained completely by the different distribution of bone density in men and women. Our results also show that the contribution of age associated decrease in bone density to the exponential increase of the risk of hip fracture with age is limited.
We are grateful to the participants of the Rotterdam study. We also thank the DXA technicians, L Buist and MB IJsselstijn, and all the other field workers in the research centre in Ommoord, Rotterdam.
Funding: This study is part of the research programme of the Erasmus Centre for Research on Aging of the Erasmus University Rotterdam and the University Hospital Rotterdam Dijkzigt, Netherlands.
Conflict of interest: None.
To obtain the incidence of hip fracture specific for age, sex, and bone mineral density (BMD) we combined the BMD distribution in this population based sample, the observed incidence of hip fracture specific for age and sex in the Netherlands, and the relative risk for hip fractures per SD decrease in the femoral neck bone density as described by Cummings et al.14 When we assume a constant relative risk per SD decrease of BMD, the age, sex and BMD specific incidence for hip fractures is given by:
where page,sex denotes the incidence for people with mean BMD for that age and sex, where a is the relative risk per SD decrease in BMD, and where z is the BMD difference from the age and sex specific mean BMD expressed in SDs. The distribution of hip fractures by BMD in people of the same age and sex will then be given by the product of the risk of hip fracture specific for BMD given above and the BMD distribution in this same population, which we know is normal. The distribution of cases is therefore given by:
The total incidence of hip fracture can be calculated by the integration of this distribution over the whole range of z. The incidence of hip fracture specific for age and sex is thus given by:
As the age and sex specific hip fracture incidence is known from population data, we can calculate page,sex.
From (1) and (2) it follows that:
In practice this means that, to obtain the incidence of hip fracture for people with mean BMD, the observed incidence specific for age and sex needs to be divided by a correction factor C, given by:
C equals 1 when a equals 1. In all other cases C is larger than 1. When a=2.6, as is the central estimate in this article, the correction factor C equals 1.578541. (At the lower confidence limit (1.9) the correction factor is 1.228739 and at the upper limit (3.6) it is 2.271399.)
The one year cumulative incidence of hip fracture (per 100 000) in this article is estimated from population data. The best fitting curves are power functions given by:
The relation of femoral neck BMD with age is best described by a linear function:
With the conditions of normality and homoscedasticity fulfilled, and by assuming a relative risk of 2.6 per SD decrease of BMD at the femoral neck, the one year incidence of hip fracture (per 100 000) is thus given by (3):