Introduction

The clinical manifestation of osteoporosis is in the fractures that arise. Hip fractures are a useful surrogate for determining the international burden of osteoporosis. Although they account for less than 20% of all osteoporotic fractures [1, 2], they account for the majority of fracture-related health care expenditure and mortality in men and women over the age of 50 years [14]. In addition, the vast majority of hip fracture cases come to medical attention and require hospital facilities. As a result, much more is known of the epidemiology of hip fracture than for other fractures associated with osteoporosis.

A variety of studies have examined hip fracture rates in different regions of the world [511]. Greater than 10-fold differences have been found, largely on the basis of register studies undertaken on a regional or national level and at different calendar years. The aim of the present study was to provide the most accurate assessment of hip fracture risk in all countries for which data were available. In addition, we wished to examine the heterogeneity of major fracture probability in those countries where a FRAX model was available.

Methods

Literature survey

We updated a systematic search conducted by Cauley et al. on behalf of the International Task Force for the ISCD IOF FRAX Initiative [12, 13]. This was a Medline OVID search covered between 1 January 1950 and 10 May 2010. Details regarding the search strategy and MeSH terms used are provided in Cauley et al. [12, 13]. The three primary concepts were: fracture, incidence and the country or their related terms. The three concepts were searched singly, and then merged together through the AND term. The information base was updated by the International Osteoporosis Foundation using the same search terms with a cut-off date of 7 November 2011. Additional sources were reviews by Kanis et al. [14] and Cheng et al. [5]. We also supplemented this search by hand-searching the references of all papers to identify any additional articles of interest. In several instances additional information was provided by the authors of papers to aid in the assessment of study quality or to provide additional detail not reported in the original publication.

Exclusion and inclusion criteria

Abstracts and full papers identified by the search were reviewed. We included non-English articles. All papers that reported age- and sex-specific incidence rates of hip fracture in a general population were eligible for a more detailed review. Further exclusion criteria comprised data that could not be standardised to the world population (age categories incomplete from the age of 50 years or age categories >10 years), an uncertain population base or ill-defined cases.

For the remaining studies, a quality assessment, originally developed by Cauley et al. [13], was adapted to provide three grades:

  • Good: Evidence includes consistent results from well-designed, well-conducted studies in representative populations. Selection of hip fracture cases was based on health care records, and the methodology was well described. At least four of the following criteria should be met: prospective study, study population representative of the entire population, study duration of 1 year or more and adequate definition of fracture or use of ICD codes, ethnicities defined when applicable.

  • Fair: Evidence is sufficient to determine effects on outcomes, but the strength of the evidence is limited by the number, quality or consistency of the individual studies, i.e. studies that did not meet the criteria for either good or poor and met some but not all quality criteria.

  • Poor: Evidence is insufficient to assess the effects on outcomes because of limited number or power of studies, important flaws in their design or conduct, gaps in the chain of evidence or lack of information. Criteria were: a retrospective study, study duration of less than 1 year, not population based, inadequate definition of fracture and abstract only available or no definition of ethnicities provided where relevant.

Where assessment was not possible, the study was discarded.

Selection criteria

From the publications available, one dataset was chosen to characterise hip fracture risk in that country which could be a single study or the mean of several studies where appropriate. Criteria for selecting a study or studies over others to represent a country are listed below and details are provided in the Appendix.

  1. 1.

    FRAX model available

  2. 2.

    National rather than regional data

  3. 3.

    Higher quality

  4. 4.

    Most recent study

  5. 5.

    Mean of several regional estimates

  6. 6.

    Sole study available

  7. 7.

    Additional details supplied by the author, see notes in tables

Where a FRAX model was available for a particular country, the hip fracture rates used for FRAX were selected since these used recent data were available and had been vetted previously for quality or consistency [13, 14]. Notwithstanding, recent publications, appearing between May 2010 and November 2011 (search cut-off dates) were reviewed to determine the adequacy of the data used for the FRAX models. In the case of China, more recent regional data had been published [15] and were preferentially selected for this report. For Belgium, we used more extensive national estimates (2005–2007 rather than 2006) supplied by the same author [16, 17], M Hiligsmann 2011, personal communication]. For Italy, we used recent national data for 2007 [18] rather than the four regional estimates used in FRAX (version 3.4) [14].

In the absence of a FRAX model, national studies were preferred over regional estimates. For regional estimates, the most recent and higher quality studies were preferred. In four instances, several studies of comparable quality were available and we used the mean value of these estimates (Brazil, four estimates [10, 1921]; Croatia, three estimates in two reports [22, 23]; Greece, three estimates [9, 24, 25]; Russia, two studies [26] Olga Yershova and Olga Lesnyak 2010, personal communication]; Poland, three studies [27, 28] Edward Czerwinski and Roman Lorenc 2011, personal communication] and Spain, five estimates [14]). For Argentina, China, Iran, Malaysia, Morocco, Norway, Poland, Sweden, Thailand and the UK, we used a single regional estimate but in each instance a review of alternate sources showed that the estimate we chose did not substantially differ from alternate estimates. Sources of information are given in Table 3 of the Appendix.

Data analysis

Incidence (rates/100,000) was assembled by age and by sex. Where possible, 5-year age categories were used. Where 5-year age intervals were not available, 10-year intervals were used (intervals of greater than 10 years were an exclusion criterion). For each country, age- and sex-specific rates were used to compute age-adjusted incidence of hip fracture in men, women and men and women combined adjusted to the world population. UN data were used for population demography in 5-year groups for the year 2010 [29]. In the case of Singapore and USA, hip fracture rates were available by ethnic origin. For the purposes of this study, population-weighted means were used and applied to the total population on the recommendation of the Working Group of the IOF Committee of Scientific Advisors. For Israel, incidence was available by race in a single study and a population-weighted mean was used [30]. A total of 72 studies from 63 countries were selected for the calculation of standardised incidence. Details of each study are given in Table 3 of the Appendix.

Probability estimates

For those countries where a FRAX model was available, we computed the 10-year probability of a major osteoporotic fracture (hip, clinical vertebral, forearm or humeral fracture) using version 3.5 of FRAX (http://www.shef.ac.uk/FRAX/). Since FRAX provides individual rather than population-based probabilities, we chose the clinical scenario of an individual aged 65 years with a prior fragility fracture (and no other clinical risk factors) at the threshold of osteoporosis as judged by BMD at the femoral neck (i.e. a T-score of −2.5 SD). The body mass index was set at 24 kg/m2. Estimates were made for men and women. Note that the T-score in men is calculated using the same reference range as that used in women.

As of November 2011, 45 FRAX calculators were available for the 40 countries listed in Table 1. Note that five models (flagged in Table 1) were not yet published at the cut-off date but were released online at the beginning of 2012. One of these was a surrogate model (Sri Lanka) derived from the fracture hazard of expatriate Indians living in Singapore and the death hazard for Sri Lanka. The models for Belgium, Czech Republic, Lebanon and Italy were updated with improved or more recent epidemiology and were also released online at the start of 2012. For USA and Singapore, fracture probabilities were available by ethnic origin. For the purposes of this study, means were used weighted by population size in addition to ethnic-specific probabilities.

Table 1 FRAX models available

Colour coding

For the purposes of cartography, we colour coded hip fracture incidence according to categories of risk designated as high, medium or low (red, orange or green, respectively) in men, women and men and women combined. The risk categories were arbitrary but selected to approximate tertiles of the distribution (Table 2). For categories of fracture probability, we used the same thresholds for men and women.

Table 2 Categorisation and colour coding for world standardised annual hip fracture rates (/100,000) in men, women and both sexes combined

Results

Hip fracture risk

A total of 72 studies from 63 countries were selected for inclusion into the hip fracture resource. Studies selected are summarised in Tables 4, 5 and 6 of the Appendix together with the selection criteria and quality grades. There was a marked heterogeneity in hip fracture risk between countries. In women (Fig. 1), the lowest annual age-standardised incidences were found in Nigeria (2/100,000), South Africa (20), Tunisia (58) and Ecuador (73). The highest rates were observed in Denmark (574/100,000), Norway (563), Sweden (539) and Austria (501). Numerical data for other countries are given in Tables 4, 5 and 6 of the Appendix. Discounting the estimates from Nigeria (poor quality) and South Africa (rates for 1957–1963), there was approximately a 10-fold range in hip fracture incidence.

Fig. 1
figure 1

Age-standardised annual incidence of hip fractures in women (/100,000) according to country together with the colour codes

Within countries, the age-standardised incidence of hip fracture in men was approximately half that noted in women (Fig. 2). Thus where higher rates were observed in women, higher rates were found in men and vice versa. Omitting the studies from Africa, the highest annual incidence in men was seen in Denmark (290/100,000) and the lowest in Ecuador (35/100,000). There was a significant correlation between the rates in men and women (r = 0.82; p < 0.001). The correlation was similar when only high quality studies or only national studies were considered (data not shown).

Fig. 2
figure 2

World standardised hip fracture rates (/100,000/year) in men and women

The geographic distribution by fracture risk is shown for men, women and men and women combined in Figs. 3, 4 and 5, respectively. In men (Fig. 3), there was a swathe of high-risk countries extending from North Western Europe (Iceland, Ireland, Finland, Denmark, Sweden and Norway), both eastwards to the Russian Federation and downwards through to central Europe (Belgium, Germany, Austria and Switzerland) and thereafter to the south west (Greece, Hungary, Czech Republic and Slovakia) and onwards to Iran, Kuwait and Oman. Other high-risk countries for men were Singapore, Malta, Japan, Korea and Taiwan.

Fig. 3
figure 3

Hip fracture rates for men in different countries of the world categorised by risk. Where estimates are available, countries are colour coded red (annual incidence >150/100,000), orange (100–150/100,000) or green (<100/100,000)

Fig. 4
figure 4

Hip fracture rates for women in different countries of the world categorised by risk. Where estimates are available, countries are colour coded red (annual incidence >300/100,000), orange (200–300/100,000) or green (<200/100,000)

Fig. 5
figure 5

Hip fracture rates for men and women combined in different countries of the world categorised by risk. Where estimates are available, countries are colour coded red (annual incidence >250/100,000), orange (150–250/100,000) or green (<150/100,000)

Regions of moderate risk included Oceania, China and India, Argentina and the countries of North America. If ethnic-specific rates were considered in USA, then the Hispanic, Asian and Black populations of men would be colour coded green. Low-risk countries included Latin America with the exception of Argentina, Africa and Saudi Arabia, the Iberian Peninsula and two countries in South East Asia (Indonesia and Thailand).

In women there was a broadly similar pattern as that seen in men. A notable difference in the distribution of high risk was that Russia was represented as moderate risk in women rather than high risk (in men). Also, the swathe of high-risk countries in Europe and beyond was more consolidated extending from North Western Europe (Iceland, UK, Ireland, Denmark, Sweden and Norway) through to central Europe (Belgium, Germany, Austria and Switzerland Italy) and thereafter to the south west (Greece, Hungary, Czech Republic, Slovakia, Slovenia) and onwards to Lebanon, Oman and Iran. Other high-risk countries for women were Hong Kong, Singapore, Malta and Taiwan. If ethnic-specific rates were considered in USA, then Hispanic, Asian and Black populations would be colour coded green but Caucasian women coded at high risk.

Regions of moderate risk included Oceania, the Russian Federation, the southern countries of Latin America and the countries of North America. Low-risk regions included the northern regions of Latin America, Africa, Jordan and Saudi Arabia, India, China, Indonesia and the Philippines. It is notable that in Europe, the majority of countries were categorised at high or moderate risk. Low risk was identified only in Croatia and Romania.

The consolidated map using age- and sex-standardised hip fracture rates is shown in Fig. 5. Note that the thresholds for categories of risk differ from those used in men and those used in women (which also differ from each other—see Table 1). With this proviso, the general pattern remained similar. Discordances in classification were relatively few. In the consolidated map, two countries coded low risk had been previously coded at intermediate risk (men in India and China). At the other extreme, one country coded as high risk had been previously coded at intermediate risk (men and women in Argentina).

As might be expected, there were more discordances in the moderate risk category. Six countries coded at moderate risk had been previously coded at low risk (men in Portugal, Thailand and Spain; women in Croatia, Jordan and Romania). Twelve countries coded at moderate risk had been previously coded at high risk (women in Hong Kong, Turkey, Italy, Lebanon and the UK; men in Kuwait, Japan, Russia, South Korea and Finland; men and women from Greece and Singapore).

FRAX

A total of 45 country and/or ethnic models were available for inclusion into the distribution of fracture probability. The FRAX models used are summarised in Table 7 of the Appendix. There was a marked heterogeneity in the 10-year probability of a major fracture between countries. In men (Fig. 6), the lowest probabilities were found in Tunisia (1.9%), Ecuador (2.5%), Philippines (4.8%) and China (5.4%). The highest rates were observed in Denmark (23%), Sweden (21%), Norway (19%) and Switzerland (18%). Numerical data for other countries is given in Table 7 of the Appendix. Thus, there was a greater than 10-fold range in fracture probability.

Fig. 6
figure 6

Ten-year probability of a major fracture (in percent) in men and women aged 65 years with a prior fragility fracture (and no other clinical risk factors) at the threshold of osteoporosis as judged by BMD at the femoral neck (i.e. a T-score of −2.5 SD). The body mass index was set at 24 kg/m2

Fracture probabilities were consistently higher in women than in men but the difference was relatively modest. On average, probabilities were 23% higher in women than in men. This contrasts, therefore, with hip fracture incidence which was twofold higher in women than in men. As expected, there was a close correlation between probabilities in men and those in women (r = 0.88; p < 0.001).

The geographic distribution by fracture risk is shown in men and women in Figs. 7 and 8, respectively. High-risk regions for men were Taiwan, Austria, USA (Caucasian), Switzerland, Norway, Sweden and Denmark. Those at low risk included Africa (Tunisia), Oceania, the Latin American countries of Ecuador and Colombia and several European countries (Spain, Poland, Romania, France and Turkey). Other countries at low risk were China, Lebanon, Philippines and the US Black population.

Fig. 7
figure 7

Ten year probability of a major osteoporotic fracture for a man aged 65 years with a prior fragility fracture (and no other clinical risk factors) at the threshold of osteoporosis as judged by BMD at the femoral neck (i.e. a T-score of −2.5 SD). Probability in different countries is categorised as high (red, >15%), moderate (orange, 10–15%) and low (green, <10%)

Fig. 8
figure 8

Ten-year probability of a major osteoporotic fracture for a woman aged 65 years with a prior fragility fracture (and no other clinical risk factors) at the threshold of osteoporosis as judged by BMD at the femoral neck (i.e. a T-score of −2.5 SD). Probability in different countries is categorised as high (red, >15%), moderate (orange, 10–15%) and low (green, <10%)

The general pattern of fracture probability in women was similar to that in men (Fig. 8). Discordances in classification were relatively few. Five countries coded as low risk in men were at intermediate risk for women (Poland, New Zealand, Romania, France and Turkey). Seven countries coded as moderate risk in men were coded at high risk in women (Japan, Belgium, Singapore, Canada, Malta, UK and Slovakia).

Discussion

The principal finding of the present study is that there is a remarkable variation in the risk of hip fracture worldwide. Age-standardised rates varied approximately 10-fold in both men and women. The difference in incidence between countries was much greater than the differences in incidence between sexes within a country. These findings confirm conclusions derived from earlier work [510, 31] but extend the information base considerably. Whereas a recently published structured review provided information on 32 countries [5], the present systematic review identified 62 countries for which hip fracture rates were available. The greater capture of information provides a more detailed map on which to place ecological patterns. In the case of age- and sex-standardised rates for example (see Fig. 5), there appears to be a crescent of high-risk countries beginning in Northern Europe (Iceland, Ireland, Norway and Sweden) that runs through middle Europe (Denmark Belgium, Germany, Switzerland and Austria) and then extends south-eastwards through eastern Europe (Hungary, Czech Republic and Slovakia) and beyond (Oman and Iran). Other high-risk countries (Malta, Argentina and Taiwan) escape this pattern. Hypotheses to explain the heterogeneity in risk will need to take these patterns into account.

The present study also reports the heterogeneity in fracture probability for 45 countries and/or ethnic groups with a FRAX model available. Probability is computed from the hazards of death and fracture and differs fundamentally from incidence—a point often unrecognised [32]. FRAX computes probabilities for individuals and not (normally) for a nation so that, for the expression of fracture probability, we chose a clinical scenario of an individual with a prior fragility fracture and a femoral neck T-score for BMD of −2.5 SD. The choice of scenario is somewhat arbitrary but of clinical relevance. We chose the age of 65 years in order to avoid a marked effect of the death risk on the estimate of probability. The 10-year probability of a ‘major osteoporotic fracture’ (hip, clinical spine, forearm and humerus) varied markedly in the different countries. As in the case of hip fracture incidence, there was a greater than 10-fold range in fracture probability. There was some, though not complete, concordance between FRAX-based probabilities and hip fracture incidence reflecting, in part, the effect of the heterogeneity of mortality in different regions [3, 14].

Although probability estimates were lower in men than in women, the difference was modest (lower by 23%) compared to the twofold difference in age-standardised hip fracture risk. The closer approximation between sexes for the probability estimate arises because the risk of hip and other osteoporotic fractures is more or less identical in men and women of the same age and femoral neck BMD [3335]. The clinical scenario chosen incorporated a BMD (as well as a prior fragility fracture). The somewhat higher probability estimates in women reflects mainly the lower death risk in women compared with men.

There are many well-recognised limitations in this type of analysis, particularly for register studies that include selection bias, the over identification of cases (double counting), inaccurate reporting or coding of fractures and errors in the denominator catchment population, particularly in regional rather than national studies. The question arises to what extent might heterogeneity of risk be accounted for by these artefacts. Several considerations suggest that these errors, though significant, have a minor effect in explaining the heterogeneity in a worldwide perspective. For example, a large prospective study undertaken in 14 regions in six different countries in Europe using standardised methodology demonstrated variability in hip fracture incidence of the same magnitude as that reported in the present study [9]. Analysis of the potential errors of any one estimate was ±10%, which pales into insignificance against the 1,000% differences in fracture risk. This study was a regional study but national register studies in Europe have shown similar findings [31].

Another limitation is the assumption that regional estimates of hip fracture risk are representative of the country in question. In addition to large variations in fracture rates around the world, fracture rates may vary within countries. In addition to ethnic-specific differences [3, 12, 13, 30], up to twofold differences in hip fracture incidence have been reported using common methodology with higher rates in urban communities than rural areas in Argentina [36], Turkey [9], Sweden [37], Norway [3840], Switzerland [41], Croatia [23] and in USA [42, 43]. The concern is perhaps less where several regional estimates have been used. In the present study, the majority of studies chosen (60%) were national rather than regional estimates. In further 10 countries, the regional estimate was consistent with other regional estimates made within a few years (Argentina, China, Iran, Malaysia, Morocco, Norway, Poland, Sweden, Thailand and UK). In five countries with multiple regional surveys, we used a mean value where studies were of comparable quality (Brazil, Croatia, Greece, Spain and Russia). This left 11 regional surveys (18% of countries) where we had to rely on a single regional estimate. The analysis of national rather than regional data did not alter our principal findings.

Notwithstanding, in some regions of the world, not all hip fracture cases come to medical attention. The risk estimate for Russia took this into account [26], but the problem has also been identified in other countries (not included in the present study). The underreporting of hip fracture cases has been observed in Georgia (75% not hospitalised), Kazakhstan (50% not hospitalised), Kyrgyzstan (50% not hospitalised) and Moldova (uncertain proportion) [44]. The likely reason is that facilities for surgical management are limited so that hospital admission is not required. Moreover, patients are required to pay for their prosthesis. Thus substantial errors may arise that lead to underreporting of hip fracture cases.

In addition to the large geographic variation reported in the incidence of hip fracture throughout the world, the age- and sex-specific incidence of fracture is changing. This has been well characterised for hip fracture but also noted at other sites of fracture [45, 46]. Estimates of incidence trends have varied widely and variously reported an increase, plateau and decrease, in age-adjusted incidence rates for hip fracture among both men and women. Studies in Western populations, whether in North America, Europe or Oceania, have generally reported increases in hip fracture incidence through the second half of the last century, but those studies continuing to follow trends over the last two decades have found that rates stabilise, with age-adjusted decreases being observed in certain centres. In contrast, the mortality hazard has continued to decrease in most regions of the world. In other countries (e.g. Japan, China, Turkey, Mexico and Hispanic Americans from California), age-adjusted hip fracture rates continue to rise [15, 4750]. In the majority of countries, there is scanty information available. Thus both national and regional estimates undertaken several years ago may not be representative of current risks. Again, it is useful to place this in perspective. Just over half the studies in the present study (52%) were conducted in 2005 or thereafter and a further 28% at or after the year 2000 (see Tables 4, 5, and 6 of the Appendix). On average, secular changes approximate 1% per annum [44, 46, 47] and if operative are likely to introduce accuracy errors of 10% or less. All these considerations indicate that there are multiple sources of error inherent in this type of analysis but that their magnitude does not undermine the principal finding of 10-fold differences in the risk of hip fracture risk and in 10-year fracture probability worldwide.

The question arises why fracture risk varies so much. The reasons are not known. The trends in incidence strongly suggest environmental rather than genetic factors. This view is supported by changes in risk in immigrant populations. For example, Blacks in USA have lower fracture probabilities than Caucasians, but the incidence of hip fracture in US Blacks is much higher than in African Blacks [12, 13]. A similar ‘acclimatisation’ is seen in the Japanese population of Hawaii [51] and the higher fracture probabilities among Chinese living in Hong Kong and Singapore compared with mainland China (see Table 7 of the Appendix).

Many risk factors for osteoporosis, and in particular for hip fracture, have been identified which include a low body mass index, low BMD, low calcium intake, reduced sunlight exposure, early menopause, smoking, alcohol consumption, physical activity levels, migration status obesity and, somewhat unexpectedly, obesity. These may have important effects within communities but do not explain differences in risk between communities [9]. The factor which best predicts this is socioeconomic prosperity that in turn may be related to low levels of physical activity or an increased probability of falling on hard surfaces [8]. This is plausible, but only a hypothesis. Paradoxically, socioeconomic prosperity may protect against hip fractures within countries [52]. The contrast between ecological and population risk factors is not uncommon and in the context of hip fracture, for example, is noted with calcium nutrition where countries with the higher calcium intakes have the greater hip fracture risk [53, 54]. It will be important to determine whether these and other factors are causally related to the heterogeneity of fracture risk. If such factors can be identified and are reversible, the primordial prevention of hip fracture might be feasible in those communities with presently low rates.