It is recommended that intervention thresholds should be based on absolute fracture risk, but there is a large variation in hip fracture incidence from different regions of the world. The aim of this study was to examine heterogeneity of hip fracture probability in different regions from recent estimates of hip fracture incidence and mortality to adjust intervention thresholds. Ten-year probabilities of hip fracture were computed in men and women at 10-year intervals from the age of 50 years and lifetime risks at the age of 50 years from the hazard functions of hip fracture and death. Lifetime risk at the age of 50 years varied from 1% in women from Turkey to 28.5% in women from Sweden. High lifetime risks in women were associated with high lifetime risks in men (r = 0.83). There also were significant correlations of 10-year risk at any age between men and women. Ten-year probability was standardized to that of men and women from Sweden (set at 1.0). There was a 15-fold range in 10-year probability from 1.24 in Norway to 0.08 in Chile. Countries were categorized by 10-year probabilities comprising very high risk (Norway, Iceland, Sweden, Denmark, and the United States), high risk (China [Taiwan {TW}], Germany, Switzerland, Finland, Greece, Canada, The Netherlands, Hungary, Singapore, Italy, United Kingdom, Kuwait, Australia, and Portugal), medium risk (China [Hong Kong {HK}], France, Japan, Spain, Argentina, and China), and low risk (Turkey, Korea, Venezuela, and Chile). The categorization of hip fracture probabilities can be used to adjust intervention thresholds based on age, sex, and relative risk from a reference population such as Sweden.

THERE ARE now many effective treatments available for osteoporosis so that a major focus of interest has been to develop methods to identify individuals who might best benefit from interventions. Practice guidelines for the assessment of individuals have focused on the use of risk factors to identify patients in whom further assessment (e.g., bone mineral density [BMD] measurements) or treatment should be considered.^{(1–3)} Such guidelines are appropriate for the intended target populations but not necessarily applicable worldwide.

A major problem in the development of a global approach is that the risk of fracture varies markedly in different countries, best documented in the case of hip fracture risk. The incidence of hip fracture worldwide varies >10-fold.^{(4–7)} Many of these are register studies, but prospective studies also show similar variations in risk, for example, a prospective study of 14 centers in six countries of southern Europe. Countries with a catchment of more than 3 million aged 50 years or more in southern Europe showed a >10-fold range between centers.^{(4)} There also are variations in the risk of other osteoporotic fractures, although these have been less well studied. In general, where the risk of hip fracture is high, so too is the risk of other fragility fractures.^{(8–10)} Over the long-term, the probability of fracture depends not only on fracture hazards but also on mortality. This also varies from country to country.

There is a growing view that the decision to intervene should depend on absolute risk of fracture^{(11)} because risk depends not only on BMD (the diagnostic parameter) but also on other factors including age and sex, prior fracture, family history, body weight, or body mass index (BMI). Fracture probabilities have been described for Sweden, not only for hip fracture but also for other osteoporotic fractures^{(12)} and provides a basis for setting intervention thresholds.^{(10)} On an international basis, there appears to be a consistent relationship between hip fracture risk and the risk of other fragility fractures.^{(10)} Thus, intervention thresholds can be computed from the probability of hip fracture. The aim of this study was to document the probability of hip fracture in different countries to provide a basis for adjusting intervention thresholds in these countries compared with those characterized in the population of Sweden.

MATERIALS AND METHODS

Incidence

Hip fracture rates in men and women aged 50 years or more were obtained in 5-year age bands from published sources by literature searches, where possible, within the last 10 years. For some countries regional estimates only were used from a single survey. The majority were hospital register studies, except where indicated. These included Argentina (La Plata^{(13)}), Australia (Geelong^{(14)}), Hungary (Budapest^{(6)}), Iceland (Reykjavic^{(6)}), and South Korea (Honam^{(15)}). Single estimates were used for national data from Chile,^{(5)} Denmark,^{(16)} Germany,^{(16)} Kuwait,^{(17)} Singapore,^{(18)} The Netherlands,^{(19)} and Venezuela.^{(5)} For Canada, we used both national data and data from Quebec.^{(5,20)} China was separated into Hong Kong HK; two studies^{(5,6)}) Taiwan^{(21)} (TW), and four regional estimates from Beijing, Chenyang, and Tangshan.^{(6,22–24)} Rates in mainland China were consistently lower than the estimates in HK and TW. For Finland, three sources were used^{(5,16,25)} that had comparable estimates of incidence. For France we used prospective data from the MEDOS data from Toulouse and Paris, as well as an estimate from Picardy.^{(4,26)} Rates were comparable in the three regions. For Greece we used national data^{(27)} and prospective estimates in Crete from the MEDOS study.^{(4)} For Italy we used five regional estimates from Verona, Venezia (S. Adami, personal communication, 2000), and prospective data from Siena, Rome, and Parma.^{(4)} For Japan we used the average rates described in a national survey^{(28)} and a regional estimate from Nigata,^{(29)} both of which gave comparable rates. For Norway we used two regional estimates, one taken from Oslo and the other from rural Norway.^{(30)} Rates were higher in the urban community than in the rural community. In Portugal two regional estimates from the Azores^{(31)} and a prospective study from Porto^{(4)} were available, all of which had similar incidence rates. For Spain we used information from five regions comprising Barcelona,^{(32)} the Canaries,^{(33)} and Zamora^{(34)} and prospective studies from Seville,^{(4)} and Madrid.^{(4)} Rates were comparable. For Sweden we used data from Malmo and national figures.^{(5,12)} Incidence rates in Malmo were higher than national figures. For Switzerland three comparable estimates were used.^{(5,35,36)} In Turkey we used prospective data from the MEDOS study^{(4)} in rural Turkey and Istanbul. Rates were higher in the urban estimate. For the United Kingdom we used three estimates comprising national data from Scotland, England, and Wales^{(5,16)} and a regional estimate from Edinburgh^{(37)} For the United States we used two surveys from national data and from Rochester.^{(5,38)}

The increase in risk with age was computed from the logarithm of incidence by age (slope) as was the intercept and the risk computed at the age of 80 years.^{(7)} The incidence at 80 years was ranked and divided into tertiles of hip fracture risk.

Mortality hazard

Mortality estimates were obtained from the World Health Organization (WHO) for each country as the number of deaths for the year 1999 in 5-year age bands in men and women together with the population size. Expected survival was computed from the age of 50 years and 75 years in men and women using a Poisson model described in the Appendix.

Probability of hip fracture

Lifetime and 10-year probabilities of hip fracture were computed from the hazard functions of hip fracture and death (see appendix). Hip fracture rates included individuals with a first or a second hip fracture. Thus, hip fracture rates are likely to overestimate the incidence of first fracture. In Sweden the overestimate increases with age from 0% at the age of 50 years to 14% and 20% in men and women, respectively, aged 85-89 years.^{(12)} However, the overestimate of hip fracture probability is offset at least to some degree by the decreasing death rate in most countries.

Ten-year probabilities were computed from the age of 50, 60, 70, and 80 years in men and women. For each country, 10-year probabilities were computed for each age (from 50, 60, 70, and 80 years) and each sex. Where more than one estimate was available for a country, the average probability was taken in each sex for each age. For most countries eight estimates were available and each estimate was compared with the estimate in Sweden for the same age and sex. For each country an average ratio was taken together with the 95% CI. Swedish probabilities were set at 1.00, so that values below this denoted a lower 10-year risk. Countries were arbitrarily divided into four categories of risk: very high risk—ratios in excess of 0.75; high risk—ratios of 0.50-0.75; moderate risk—ratios of 0.25-0.50; low risk—ratios <0.25.

RESULTS

In all regions, the incidence of hip fracture increased exponentially with age. In both men and women there was a significant correlation between the logarithm of the slope and incidence with age. The correlation coefficients exceeded 0.95 except for men and women in Chile (0.688 and 0.943, respectively, and women in Kuwait (0.790). In both men and women there was a significant correlation between the slope and the intercept (r = −0.916 and −0.891, respectively). Incidence at the age of 80 years varied 20-fold in women, with the highest in Singapore and the lowest in Turkey (1896 vs. 91/100,000, respectively; Table 1). In men there was a 22-fold range with the highest in Norway (960/100,000) and the lowest in Chile (44/100,000). There was a similarly large range in remaining lifetime risk at the age of 50 years that varied from 1% in women from Turkey to 24.5% in women from Norway and from 1.8% in men from Turkey and 13.1% in men from Sweden (Table 1). There were significant correlations between lifetime risks between sexes and from each sex between lifetime risk and incidence at 80 years (r > 0.83).

Table Table 1.. Incidence at 80 Years, Life Expectancy, and Life Time Risk of Hip Fracture in Men and Women by Country

Variations in life expectancy were much less than variations in risk of hip fracture. Life expectancy in women from the age of 75 years ranged from 8.1 (Turkey) to 14.4 years (Japan). There was a broad correlation between life expectancy in women with those in men (r = 0.80), although women outlived men at this age by 1-4 years (Table 1). The contribution of variations in incidence to lifetime risk was quantitatively greater than variations in life expectancy (data not shown).

Ten-year probability of hip fracture at the ages of 50, 60, 70, and 80 years are shown in Table 2. In all instances there were significant interage correlations. For example, the 10-year risk at the age of 50 years correlated with that at the age of 60 years in both men and women (r = 0.77 and r = 0.93, respectively). There also were significant correlations between sexes. For example, the 10-year probability in men aged 70 years correlated closely with that in women of the same age (r = 0.832). As expected, there also were significant correlations between the 10-year probabilities at any age with lifetime risk. These became progressively closer with increasing age. For example, in women the correlation between lifetime risk and 10-year probability at the age of 50 years was 0.65 but at the age of 80 years was 0.94.

Table Table 2.. Ten-Year Probability of Hip Fracture (%) in Men and Women According to Age and Country

For each age and sex, 10-year probabilities were compared with those from Sweden and expressed as a proportion of the probability of Sweden. The probability ratio was lower than unity for all countries with the exception of Norway (Fig. 1).

There was a 15-fold range in 10-year probabilities that ranged from 1.07 (Norway) to 0.08 (Chile and Korea) compared with Sweden (=1.00). From the distribution of probability ratios, countries were categorized according to risk:

Very high (>0.75)—Norway, Iceland, Sweden, Denmark, and the United States High (0.5-0.75)—China (TW), Germany, Switzerland, Finland, Greece, Canada, The Netherlands, Hungary, Singapore, Italy, United Kingdom, Kuwait, Australia, and PortugalMedium (0.25-0.50)—China (HK), France, Japan, Spain, Argentina, and China

Low (<0.25)—Turkey, Korea, Venezuela, and Chile

Long life expectancy contributed markedly to risks in Japan, Spain, and France because incidences in those countries were low. At the other extreme, a high probability was found in Iceland despite low life expectancy.

DISCUSSION

The principal finding from this study is the large variation in 10-year probability of hip fracture that was found in the different countries surveyed, which varied >15-fold. This accords with the wide range in annual incidence from both prospective and register studies.^{(4–7)} In this study we have chosen to examine this heterogeneity in hip fracture risk in terms of absolute risk, that is, hip fracture probability over a lifetime (from the age of 50 years) or as 10-year probabilities.

It should be acknowledged that the methodology used to determine fracture rates has varied between studies used in this report. The majority were register studies from hospital discharges. Only a few excluded double counting.^{(4,12)} This will inevitably give rise to errors but the magnitude of the errors is likely to be small compared with the heterogeneity of fracture rates between countries. For example, double counting overestimates hip fracture incidence in Sweden by ∼10%,^{(12)} which is a small error compared with the wide international variation in hip fracture rates. In several countries, more than one estimate was available. The variation in apparent incidence was up to 2-fold different. However, this accords with the variations seen within countries using identical methodologies,^{(30)} including four prospective studies in Italy, Spain, France, and Turkey. However, the variations are substantially less than the variance in rates worldwide. Nevertheless, greater reliance should be placed on countries where national data are available than those based solely on a single regional estimate (e.g., Argentina,^{(13)} Australia,^{(14)} Hungary,^{(6)} Iceland,^{(6)} and South Korea^{(15)}). It is of interest that multinational studies show a similar wide range of international variations irrespective of whether the studies are register-based or prospective.^{(4–7)} These various considerations indicate that unquantifiable errors exist, but they are unlikely to alter our principal conclusions.

In addition to these limitations, further errors arise in computing a single probability over both sexes and all ages. As shown in this study, there were significant correlations in fracture probability between men and women and between each age group studied. The estimate of the ratio varied by 10-40% expressed as a CV. Overall, the errors are still much smaller than the variations in probabilities observed worldwide that varied 15-fold. Nevertheless, caution is required in assuming accuracy for the point estimate in each country.

Lifetime risks are appropriate for characterizing the burden of disease. However, they are less useful for individual assessment. This is because treatments cannot be feasibly given today over a lifetime. Moreover, the predictive value of diagnostic or prognostic tests is attenuated over a lifetime. For example, an early menopause is a significant risk factor for fractures in women just after menopause but is of less value to predict hip fractures in later life.^{(39)} The predictive value of BMD is attenuated also with time^{(40)} and so too may be that of the biochemical markers of bone turnover.^{(41)} Indeed, after 10 years, the predictive value is all but lost. For individual assessment shorter-term probabilities are appropriate. A 10-year time frame is used commonly in health economic analyses, because it covers the period of interventions proven over a 3- to 5-year period and the variable offset time, that is, the period after treatment is stopped, that benefits still occur. The 10-year time frame is consistent with the view of the WHO and International Osteoporosis Foundation that interventions should be based on absolute risk rather than a given BMD or relative risk.^{(11,42)}

A great deal of information is available on 10-year probabilities of hip and other fractures but this is derived mainly from data from the Western World, particularly from Sweden.^{(12)} In the future, these are likely to provide the basis for intervention thresholds. In considering hip fracture alone, an intervention threshold for hip fracture 10-year probability of 7.5% is justifiable from a health economic perspective.^{(43)} Where other osteoporotic fractures are taken into account, lower thresholds would become worthwhile, particularly in younger individuals.

Table 3 shows 10-year probabilities of hip fracture in Malmo, Sweden, compared with other regions with lower risks. If, for example, a threshold of risk of 5% 10-year probability were assumed to be an intervention threshold, then this is exceeded in women from areas of very high risk at the age of 70 years in the general population and in women aged 60 years with a population relative risk of 2.0. In high and medium risk areas the same probability is only exceeded in 80-year-old patients with an average risk. In low risk areas the threshold is exceeded only in 80-year-old patients with very high relative risks.

Table Table 3.. Ten-Year Probability of Hip Fracture in Women According to Age and Relative Risk in Four Countries With Different Absolute Risks

It is likely that absolute risk will be used increasingly as intervention thresholds. This will demand greater information on hip fracture probabilities than is available to date. Indeed, in this study probabilities are from 29 regions only. However, mortality hazards are available worldwide so that data on incidence from other regions is required to enlarge this database. Where probabilities are required for other countries, a surrogate region would need to be chosen that best represented the local mortality and hip fracture risks.

Acknowledgements

We are grateful to the International Osteoporosis Foundation, Lilly, Hologic, Inc., Novartis, and Roche for their support of this work.

Note in proof

Recently, further estimates of incidence have become available from Malaysia, HK, Singapore, Thailand (Chiang Mai),^{(44)} Australia (Tasmania),^{(45)} Germany,^{(46)} and the Czech Republic (source: UZIS, Prague, courtesy of Dr. Jan Stepan). The revised ratios of probabilities to Sweden is 0.52 for HK, 0.62 for Singapore, 0.49 for Australia, and 0.50 for Germany. In the three countries where no prior information was assessed, the probability ratios were 0.38 for Malaysia, 0.45 for Thailand, and 1.15 for the Czech Republic.

APPENDIX

Estimation of the hazard function of death in different countries

At low ages (up to 20 years) the hazard function h(t) was estimated as the quotient between the number of deaths and the population figure of the age interval minus half of the number of deaths.

For ages >20 years, a Poisson model was used:

The data used for the estimation was divided into age interval of the length of 5 years except for the last interval. The age of an interval was put equal to the age at the middle of the interval. The observation time was equal to the population figure minus half of the number of dead individuals.

For those above the largest limit of age (85 years), we put the age provisionally to 88 years. By an iterative procedure described in the following paragraphs, the provisional age was successively adjusted until there was no further change of importance.

Let f_{b}(t) denote the frequency function of the present age of the population and let S_{L,s}(u) be the survival function of the remaining length of life L after the age s. If the refilling of new individuals at lower ages is constant (the same amount at different ages), it can be shown that

Generally, the following relation holds: S_{L,s}(u) = exp(−∫_{s}^{u}h(v)dv). Thus, we can calculate f_{b}(t) (approximately) by use of the hazard function.

The death hazard (number of deaths per observation year) for those above the age G is

where the constant c_{2} is

The group of individuals above the age G has the same risk of death as those at the age

(1)

where h^{−1} is the inverse of h and where c_{2} is given in the foregoing equations (notice that the age at which the risk of death is equal to that of the age group is not equal to the mean age of the group; the first mentioned age is somewhat higher than the mean [that can be proved by Jensens inequality E[h(T)] > (E[T]), when h is convex]).

We first gave a provisional value of the age 1 year and thereafter we calculated h. Then, we calculated the quantity in (Eq. (1) again and used it for a new determination of h, which again was used for a calculation of (Eq. (1), etc. The procedure was repeated until consecutive values of (Eq. (1) differed slightly (<0.1 year).

Calculated probability of hip fracture

For the calculations, the rate of first-time hip fracture should be used. However, such rates are not available in general. No attempt to adjust for that has been performed here. That makes the probabilities too high. In most countries, the death rates are decreasing with calendar time. No adjustment for that has been performed. That makes the probabilities too low.

Let h(t) and d(t) denote the rates (hazard functions) of hip fracture and death, respectively. The probability p of fracture within the time period t was calculated as

In many cases, the information about fracture rates was incomplete, in that the youngest or the oldest age groups were missing. Therefore. we did not extrapolate the rates and accounts for missing probabilities.