Previous Fractures at Multiple Sites Increase the Risk for Subsequent Fractures: The Global Longitudinal Study of Osteoporosis in Women

Abstract

Previous fractures of the hip, spine, or wrist are well-recognized predictors of future fracture, but the role of other fracture sites is less clear. We sought to assess the relationship between prior fracture at 10 skeletal locations and incident fracture. The Global Longitudinal Study of Osteoporosis in Women (GLOW) is an observational cohort study being conducted in 17 physician practices in 10 countries. Women ≥ 55 years answered questionnaires at baseline and at 1 and/or 2 years (fractures in previous year). Of 60,393 women enrolled, follow-up data were available for 51,762. Of these, 17.6%, 4.0%, and 1.6% had suffered 1, 2, or ≥3 fractures since age 45. During the first 2 years of follow-up, 3149 women suffered 3683 incident fractures. Compared with women with no prior fractures, women with 1, 2, or ≥ 3 prior fractures were 1.8-, 3.0-, and 4.8-fold more likely to have any incident fracture; those with ≥3 prior fractures were 9.1-fold more likely to sustain a new vertebral fracture. Nine of 10 prior fracture locations were associated with an incident fracture. The strongest predictors of incident spine and hip fractures were prior spine fracture (hazard ratio 7.3) and hip (hazard ratio 3.5). Prior rib fractures were associated with a 2.3-fold risk of subsequent vertebral fracture, previous upper leg fracture predicted a 2.2-fold increased risk of hip fracture; women with a history of ankle fracture were at 1.8-fold risk of future fracture of a weight-bearing bone. Our findings suggest that a broad range of prior fracture sites are associated with an increased risk of incident fractures, with important implications for clinical assessments and risk model development.

Introduction

The history of a previous fracture is a potent predictor of future fracture among postmenopausal women.^(1,2) Although there is general agreement on the predictive value of prior fractures of the hip, spine, and wrist, there is less clarity on the role of other fracture sites in conferring risk. Frequently referenced clinical guidelines and risk assessment systems use terms such as “osteoporotic fracture”, or “fragility fracture” when referring to other such fractures, but offer little guidance on which specific bone locations are associated with increased risk.^(3–5) The definition of previous fracture in the FRAX^® fracture risk assessment ⁽⁵⁾ generally considers all fracture sites as potential risk factors but supposes that each site confers the same added risk of future fracture. Because substantial numbers of fractures occur at other sites, neglecting their possible predictive importance and variable influence may lead to an under-appreciation of an individual’s risk, with missed opportunities for intervention and an underestimation of the burden that fractures place on the healthcare system.

The Global Longitudinal Study of Osteoporosis in Women (GLOW) collects risk factor and fracture information on more than 50,000 women from 10 countries, and thus provides the opportunity to assess the relationship between previous fractures at 10 skeletal locations and incident fractures.

Methods

GLOW is an observational cohort study that is being conducted in physician practices in 17 sites in 10 countries (Australia, Belgium, Canada, France, Germany, Italy, Netherlands, Spain, UK, and USA). Details of the study design and methods have been described previously.⁽⁶⁾ In brief, study sites were selected on the basis of geographic distribution and the presence of lead investigators with expertise in osteoporosis and access to a clinical research team capable of managing a large cohort of subjects. These lead investigators identified primary care practices in their region that were members of local research or administrative networks and were able to supply names and addresses of their patients electronically. The composition of groups varied by region and included health-system owned practices, managed practices, independent practice associations, and health maintenance organizations. Participating physicians were defined as those who spent the majority of their time providing primary healthcare to patients, and included internists, family practitioners, and general practitioners.

Each practice provided a list of the names and addresses of women aged 55 years or older who had been attended by their physician in the past 24 months. All eligible women aged 65 years or older and a random sample of half that number less than 65 years were recruited from each practice by mail. Patients were not included if they were unable to complete the study survey due to cognitive impairment, language barriers, institutionalization, or illness. The median participation rate across the 17 study sites was 62%.

The questionnaires were designed to be self administered and covered domains that included: patient characteristics and risk factors; perception about fracture risk and osteoporosis; medication use (currently taking or ever taken); selected medical diagnoses; healthcare access and use; physical activity; and physical and emotional health status. Where possible, items from published validated instruments were used, including the National Health and Nutrition Examination Survey (NHANES),⁽⁷⁾ EuroQol EQ-5D,^(8,9) and the physical function component of SF-36.^(10–12)

All information was self-reported. For the baseline survey, subjects were asked to identify any fractures they had experienced since the age of 45 by checking boxes for any of 10 specified locations (clavicle, upper arm, wrist, spine [clinical vertebral], rib, hip, pelvis, upper leg, lower leg, and ankle). When more than one fracture was indicated, each fracture was counted when obtaining the total number of fractures by site for the study population. Prior fractures were not confirmed.

In the follow-up questionnaires, completed 12 and 24 months after baseline, subjects were asked how many times they had sustained a fracture in the previous 12 months (none, once, twice or more), the date the fracture occurred and, if a single instance was reported, to identify which bone or bones were broken from the list of 10, with the option to write in fractures occurring at other locations. Data collected also included the circumstances of the fracture (including where and how it occurred) and resulting medical care obtained. Because of the complexity of obtaining this latter information for subjects reporting more than one episode of incident fracture, identification of fracture locations was not attempted by questionnaire when multiple fracture episodes were indicated, but was obtained later by telephone interview. Incident fractures were not confirmed.

Incident fractures included the 10 locations specified in the checklist as well as locations identified from the “other” responses; these were fractures of the shoulder, knee, elbow, hand, and foot (shoulder and upper arm were combined for the analysis). These were grouped into five fracture categories: any bone, hip, spine, other weight-bearing bone (pelvis, upper leg, knee, lower leg, ankle, and foot), and non-weight-bearing bone (rib, clavicle, upper arm/shoulder, elbow, wrist, and hand). Neither of the latter two categories included hip or spine. When more than one incident fracture was reported, each fracture was included, so the total number of fractures may exceed the number of subjects with fracture.

Statistical analysis

Data from 51,762 women with complete information on fracture history who provided information at 1- and/or 2-year follow-up were analyzed. The number of days from baseline survey date to date of first incident fracture was calculated for each woman with an incident fracture and complete fracture information. The prevalence of prior fracture is reported in percent. Kaplan-Meier estimates of 2-year incident fracture rates by prior fracture number and location are reported as percentages. We assessed associations between prior and incident fractures using Cox regression models controlling for age and physician practice site. Age was used as a linear continuous variable in the models. No adjustments were made for osteoporosis medication use, because the results from including it in the model did not materially change the hazard ratios (HRs). HRs are reported for incident fractures for each particular type of prior fracture, controlling for other types of prior fracture. Potential interactions were assessed between age and prior fracture location, based on prior research Kanis #1, vanStaa #14). To help control for type I error, we used alpha=0.005 for their statistical significance level. Any interactions found to be statistically significant in the adjusted Cox model (Table 3) are reported by age group in Table 4 as 2-year relative fracture risks (using the Kaplan-Meier method) between women with and without a prior fracture. All analyses were conducted using SAS software package (SAS Institute, Cary, NC, version 9.2).

Table 3.

Hazard Ratios (95% Confidence Intervals) for Groups of Incident Fractures by Location and Number of Prior Fractures (Adjusted for Age, Physician Practice Site, and Multiple Previous Fractures) (n=51,296)

Previous fracture site	Any bone (n=3149)	Hip (n=237)	Spine (n=340)	Other weight-bearing bonea (n=1199)	Non-weight-bearing boneb (n=1528)
Any prior fracture	2.19 (2.03, 2.35)	2.02 (1.55, 2.63)	2.93 (2.34, 3.66)	2.21 (1.96, 2.49)	2.15 (1.93, 2.39)
Number of prior fractures
1 (n=9096)	1.81 (1.66, 1.97)	1.60 (1.17, 2.18)	2.16 (1.67, 2.80)	1.94 (1.69, 2.22)	1.75 (1.55, 1.98)
2 (n=2085)	2.98 (2.63, 3.38)	2.95 (1.98, 4.40)	3.97 (2.80, 5.63)	2.64 (2.12, 3.28)	3.07 (2.58, 3.65)
3 or more (n=828)	4.80 (4.11, 5.6)	3.66 (2.19, 6.15)	9.05 (6.28, 13.05)	4.44 (3.40, 5.79)	4.50 (3.6, 5.63)
Prior fracture sites
Rib (n=2161)	2.03 (1.79, 2.31)c	1.29 (0.79, 2.10)	2.28 (1.64, 3.17)	1.63 (1.30, 2.05)	2.37 (2.00, 2.81)c
Hip (n=863)	1.65 (1.36, 2.00)	3.50 (2.30, 5.32)c	1.60 (0.95, 2.67)	1.27 (0.88, 1.84)	1.51 (1.13, 2.00)
Wrist (n=4411)	1.59 (1.44, 1.76)	1.04 (0.71, 1.51)	1.37 (1.01, 1.85)	1.50 (1.26, 1.8)	1.81 (1.58, 2.09)
Spine (n=1138)	1.93 (1.64, 2.27)	1.22 (0.66, 2.26)	7.34 (5.42, 9.92)	1.37 (1.00, 1.88)	1.66 (1.31, 2.11)
Upper arm (n=1479)	1.42 (1.20, 1.67)	1.49 (0.89, 2.49)	1.27 (0.80, 2.02)c	1.34 (1.01, 1.78)	1.47 (1.17, 1.85)
Ankle (n=3201)	1.40 (1.24, 1.58)c	1.45 (0.95, 2.20)	1.24 (0.86, 1.80)	1.83 (1.52, 2.19)	1.15 (0.95, 1.39)
Lower leg (n=1286)	1.18 (0.98, 1.43)c	0.94 (0.48, 1.86)	1.03 (0.60, 1.77)	1.44 (1.09, 1.91)	1.07 (0.81, 1.41)c
Upper leg (n=479)	1.36 (1.04, 1.77)	2.15 (1.12, 4.14)	0.86 (0.37, 2.00)	1.94 (1.31, 2.87)	1.00 (0.65, 1.53)
Clavicle (n=686)	1.05 (0.82, 1.35)	0.35 (0.11, 1.15)	1.48 (0.82, 2.67)	0.90 (0.57, 1.42)	1.29 (0.93, 1.79)
Pelvis (n=527)	1.03 (0.79, 1.34)	2.62 (1.44, 4.77)	0.85 (0.43, 1.68)	1.17 (0.75, 1.82)	0.95 (0.65, 1.40)

^aOther weight-bearing bone: Includes pelvis, upper leg, lower leg, ankle, foot and knee (excludes hip and spine).

^bNon-weight-bearing bone: Includes rib, wrist, upper arm, clavicle, hand, elbow and shoulder (excludes hip and spine).

^cMain effect of prior fracture is reported here in italics, but a statistically significant interaction between age and prior fracture was found and is presented in detail in Table 4 (p-value ≤ 0.005). P-values for the interactions are respectively: 0.001, 0.0002, <0.0001, 0.0002, 0.003, 0.003, 0.004.

Table 4.

Statistically Significant (p-value ≤ 0.005) Age by Prior Fracture Interactions: Estimated 2-Year Incident Fracture Rates by Prior Fracture and Age Group (in parentheses are numbers of 2 year outcomes, number of women in each prior fracture group)

Baseline fx	2 Year Outcome : Any bone			Hip			Spine			Non-weight-bearing
	Prior fx	No prior fx	RR	Prior fx	No prior fx	RR	Prior fx	No prior fx	RR	Prior fx	No prior fx	RR
Rib:	(334, 2161)	(2805, 49515)								(n=187, 2143)	(n=1336, 49322)
< 65 yrs	13.2	4.9	2.7							7.9	2.3	3.4
65–74 yrs	17.4	5.7	3.1							9.6	2.8	3.5
75–84 yrs	17.8	8.3	2.2							11.1	4.0	2.7
85+ yrs	17.9	11.0	1.6							7.6	5.2	1.5
Ankle:	(334, 3201)	(2804, 48477)
< 65 yrs	11.7	4.9	2.4
65–74 yrs	10.1	5.9	1.7
75–84 yrs	12.7	8.5	1.5
85+ yrs	11.4	11.5	1.0
Lower leg:	(n=147, 1286)	(n=2988, 50349)								(n=62, 1275)	(n=1459, 50150)
< 65 yrs	15.1	5.0	3.1							5.9	2.4	2.4
65–74 yrs	9.9	6.1	1.6							5.6	3.0	1.9
75–84 yrs	12.5	8.7	1.4							4.2	4.5	0.9
85+ yrs	13.3	11.5	1.2							4.7	5.4	0.9
Hip:				(n=34, 856)	(n=201, 50476)
< 65 yrs				4.6	0.1	34.4
65–74 yrs				3.2	0.3	11.9
75–84 yrs				4.7	1.0	4.7
85+ yrs				5.1	2.2	2.4
Upper arm:							(n=23, 1470)	(n=316, 50172)
< 65 yrs							2.0	0.3	5.8
65–74 yrs							1.0	0.6	1.6
75–84 yrs							2.8	1.3	2.2
85+ yrs							(a)	1.6	-

Fx= fracture, RR=relative risk(a)- no 2-year spine fractures in this group

Results

The characteristics of women in GLOW have been described in detail in a previous publication.⁽⁶⁾A summary of risk factors and numbers of participants by country is shown in Table 1. The mean age of study subjects was 68 years. The prevalence of fracture-related risk factors ranged from 23.2% for history of previous fracture to 0.5% for consumption of three or more alcoholic drinks per day. Almost half the subjects were from the United States with the other nine countries contributing approximately 2000–4000 participants each.

Table 1.

Characteristics of GLOW Women (n=51,762)

Continuous variables	Mean (SD), range
Age	68 (8.6), 53
Body mass index	27 (5.9), 123
Discrete variables	n (%)
Current smoker	4506 (8.8)
Currently using anti-osteoporosis medicationa	9758 (19.5)
Alcohol ≥ 3 drinks/day	255 (0.5)
Secondary osteoporosis	10,359 (20.7)
Prior fracture	12,009 (23.2)
Parental hip fracture	8019 (17.3)
Cortisone use	1500 (3.0)
Rheumatoid arthritis	384 (0.8)
Country
Australia	2634 (5.1)
Belgium	3151 (6.1)
Canada	3607 (7.0)
France	4421 (8.5)
Germany	2764 (5.3)
Italy	2717 (5.3)
Netherlands	2655 (5.1)
Spain	2185 (4.2)
UK	3485 (6.7)
USA	24,143 (46.6)

^aAlendronate, calcitonin, etidronate, ibandronate, pamidronate, risedronate, raloxifene, strontium ranelate, teriparatide, tibolone, zoledronic acid.

Prior fractures

Of the 60,393 women enrolled in the study, follow-up fracture data were available for 51,762 (85.7%). Of these, 12,009 (23.2%) reported having sustained a prior fracture at any of the 10 designated locations since the age of 45 years. Of these, 9096 (17.6%) women indicated a single bone was fractured, 2085 (4.0%) had two bones fractured, and 828 (1.6%) said three or more bones had been broken. This resulted in a total of 16,231 fractures, most commonly in the wrist (27.2% of fractures) and ankle (19.7% of fractures) (Table 2).

Table 2.

Number, Percent of Prior Fractures, and Estimates^a of 2-year Incidence of Any Fracture by Fracture History Among 51,762 Women

Prior fracture history	Prior fractures		2-year estimated incidence
	n	%	%
No prior fracture	39,753		5.0
Any prior fracture	12,009		11.6
Single fracture	9,096	75.7	9.5
Two fractures	2,085	17.4	15.8
Three or more fractures	828	6.9	24.5
Prior fracture sites
Total	16,231	100.0
Wrist	4411	27.2	12.5
Ankle	3201	19.7	11.3
Rib	2161	13.3	16.5
Upper arm	1479	9.1	13.7
Lower leg	1286	7.9	12.3
Spine	1138	7.0	19.2
Hip	863	5.3	17.8
Clavicle	686	4.2	12.0
Pelvis	527	3.2	16.6
Upper leg	479	3.0	16.6

^aKaplan-Meier estimates of 2-year incidence of any fracture.

Incident fractures

Among the 12,009 women with prior fracture, 11.6% were estimated to have a fracture by 2 years after baseline (Table 2). Of these, 9.5% of women with a single prior fracture, 15.8% with two prior fractures, and 24.5% of those with three or more prior fractures were estimated to have an incident fracture by 2 years. The prior fractures most likely to lead to an incident fracture were spine (19.2%), hip (17.8%), and pelvis and upper leg (both 16.6%).

Predictors of incident fractures

Table 3 shows the HRs and 95% confidence intervals (CIs) for groupings of incident fractures by number and location of prior fracture. Having one or more prior fracture added significantly to the likelihood of any incident fracture in the multivariable modeling (Cox regression analyses that took into account age, study site, and multiple prior fractures). HRs increased from 1.81 for a single prior fracture to 2.98 and 4.80 for two and three or more prior fractures, respectively. The increased risk associated with multiple prior fractures was most dramatic for incident fractures of the spine. Women with a history of a single prior fracture had an HR of about 2 but those with a history of three or more carried a 9-fold increased likelihood of sustaining a new vertebral fracture. When combinations of multiple prior fractures were examined, 321 variations were observed with no particular patterns predominating.

When incident fractures of “any bone” were considered, seven of the 10 prior fracture locations were significantly associated with any incident fracture. Of these, the greatest HRs were for rib and spine (2.03 and 1.93, respectively) followed by hip and wrist (1.65 and 1.59 respectively) (Table 3). The strongest predictors of incident hip and spine fractures were prior fracture of the hip and spine, respectively. Women with a history of hip fracture had an HR of 3.50 for future hip fracture and those with previous spine fractures were 7.3 times as likely to suffer another vertebral fracture as those without such a history. Prior rib fractures were associated with a 2.3-fold risk of subsequent vertebral fracture but carried no increased risk for incident fractures of the hip. A history of upper leg fracture predicted a 2.2-fold increased risk of hip fracture but did not predict incident spine fracture. The HR for prior fractures of the pelvis was significant only for future hip fracture (HR 2.6). Previous clavicle fractures were not significantly linked with any type of subsequent fracture

The incidence of fractures of other weight-bearing bones was significantly associated with seven of the 10 prior fracture locations, with previous upper leg and ankle fractures being the most predictive (Table 3). Non-weight-bearing bone incident fractures were significantly associated with five of 10 previous fracture sites, with prior fractures of the rib, wrist, and spine having the highest HRs.

Statistically significant (p ≤.005) interactions were found between age and prior fracture in 3 of the 10 sites (rib, ankle, and lower leg), for the outcome of any incident fracture by 2 years and in 3 other sites for other fracture outcomes (prior hip for hip, upper arm for spine, rib and lower leg for non-weight-bearing bones). These are displayed in Tables 3 and 4. The general pattern in these interactions (Table 4) is one of decreasing relative risk (RR) for prior fracture, as a woman ages.

Discussion

These data from a large, contemporary international registry of postmenopausal women suggest that a broader range of previous fracture sites than is commonly considered is associated with future fractures, and that occurrence of multiple past fractures substantially increases subsequent risk. We found that nine of 10 fracture sites (hip, wrist, spine, rib, upper arm/shoulder, ankle, upper and lower leg, and pelvis) were significantly related to the occurrence of future fractures. Fractures of the rib, upper arm, pelvis, upper and lower leg, and ankle made up 56% of total previous fractures. Adding these to the traditional sites of hip, spine, and wrist (40%) more than doubled the number of prior fractures that should be considered as placing women at increased risk of future fracture.

Our results amplify findings from prior meta-analyses and cohort studies.^{(1,2,13,–15)} Investigators from the Study of Osteoporotic Fractures (SOF) noted that, among over 3200 women aged 65 years or older who self-reported a non-hip, non-spine fracture since the age of 50, there was a 23% increased incidence of hip fracture during 10 years of follow-up.⁽¹³⁾ Two cohort studies examined the effect of prior fracture at specific sites on future fractures. Similar to our findings, Van Staa et al.⁽¹⁴⁾ using data from the General Practice Research Database, found that among 119,000 women who had sustained a previous fracture of the femur/hip, wrist, spine, humerus, rib, tibia/fibula, and ankle, there was a 2- to 3-fold higher risk of subsequent fracture at any of a variety of sites.

Center et al.⁽¹⁵⁾ found that, with the exception of the ankle, all initial fractures observed using data from the Dubbo study, including those of hip, vertebra, pelvis, distal femur, proximal tibia, multiple rib, proximal humerus, and other “minor” locations, conferred a 1.6–2.4-fold increased risk for subsequent fracture.

Expanding the range of prior fracture sites that are associated with future fracture is supported by studies that assess the relationship between bone mineral density (BMD) and fracture.^(16,17) Stone et al.⁽¹⁶⁾ reported on fractures that occurred during more than 8 years of follow-up among women in the SOF study. Of 18 fracture sites assessed, all but two (ankle and face) showed a significant association with central BMD; ankle fractures were significantly associated with BMD measured peripherally. Similarly, data from the Rotterdam study⁽¹⁷⁾ found a relationship between reduced BMD and six non-vertebral fracture sites, including upper humerus, wrist, hand, hip, foot, and lower leg, among a smaller cohort of both men and women. These results demonstrate that – with the exception of fractures of the face (and possibly the ankle) – many fractures occur more frequently as BMD declines and thus low BMD is a mechanism shared by a variety of fracture types.

Specific locations of prior fracture had varying implications for different incident fracture sites Incident hip fractures were most likely among women with previous hip, pelvis, or upper leg fractures and subsequent vertebral fractures occurred most frequently among women with a history of spine fractures. These findings are consistent with previous research.^(1,2) Ankle fractures were the second most common type of prior fracture (after wrist fractures), contributing almost 20% of prior fractures and were associated with an increased risk of subsequent fractures of “any” and “other weight-bearing” bones. Data from previous cohort studies on the predictive role of ankle fractures are inconsistent, however. The General Practice Research Database (GPRD) cohort study grouped ankle with tibia and fibula and found significant increased risks for all six incident fracture groups analyzed.⁽¹⁴⁾ In the Australian cohort, however, an association between subsequent fracture and ankle fracture was found only in men, and not in women.⁽¹⁵⁾

Rib fractures were the third most frequently occurring previous fracture, and carried the highest HR for future fractures at any site as well as more than 2-fold risks for future fractures of the spine and non-weight-bearing bones. Data from the European Prospective Osteoporosis Study⁽¹⁸⁾ also indicate that women with a history of rib fracture are twice as likely to sustain an incident fracture of “any” limb as women without, although these investigators found a large increase in the HR of subsequent hip fracture (HR 7.7) that was not seen in our data. Neither previous ankle nor rib fractures have been generally regarded as important predictors of future fracture. However, our current results coupled with the findings from SOF⁽¹⁶⁾ and GPRD⁽¹⁴⁾ should raise awareness of their potential importance. Previous fractures of the pelvis and lower leg were not significantly associated with incident fractures of “any” bone but carried significant HRs for future fractures of hip and other weight-bearing bones, respectively. Such specific associations may not be unanticipated given the varying determinants of fracture (age, low BMD, falls, and frailty) that relate to different bones⁽¹⁹⁾. Increasing age and falls are associated with both hip and pelvis fractures and fractures of the lower leg likely share precipitants in common with fractures of other weight-bearing bones.

The added risk conferred by more than one prior fracture has been sparsely documented. Several reports have demonstrated a dramatic increase in fractures among women with multiple previous vertebral fractures^(20–22); and case-control studies that detailed risk factors for fractures of the pelvis⁽²³⁾ and proximal humerus⁽²⁴³⁾ have indicated an OR of about 1.4 for each (unspecified) previous fracture since the age of 45. Our results greatly amplify these findings.

We found that, while prior fracture increased the likelihood of incident fracture, the relationship was modified by age. RRs for the youngest group of women were higher than for the oldest women across all 10 bone sites (though differences were statistically significant for only three of these). Several investigators have reported that the risk of subsequent fracture varies with age.^(1,14) The meta-analysis by Kanis et al⁽¹⁾ showed a non-significant trend towards lower relative risk of an “osteoporotic” fracture among men and women with a prior history of fracture with advancing age and a marked decline in risk ratio for hip fracture at older ages. Van Staa and colleagues⁽¹⁴⁾ reported that standardized incidence ratios of subsequent fractures for a variety of bones were higher for subjects aged 65–74 years than for those aged 85 and above and postulated that the effect modification “could reflect varying proportions of high-risk patients over age” without offering more specific details. Other explanations for the age interaction may be that prior fractures among younger women are likely to be more recent and have a greater influence on subsequent fractures,^(13,25) or that increased mortality among prior fracture patients, particularly those with hip fracture, removes susceptible subjects from the prior fracture group at older ages.

As in most studies of human populations, our sample is a selected group – in this case of primary care practices and women who elected to participate. As a result, we may have included practices and subjects with a greater interest in bone health and fracture than is typical of broader populations. However, comparison of demographic characteristics and clinical risk factors of our subjects with those of a representative sample of US women suggested that, although women in our study were better educated and enjoyed better self-reported health status, the prevalence of risk factors related to fracture was similar in the two groups.⁽⁶⁾

While our survey approach enabled the sampling of a large number of subjects from a broad geographic range, the self-reported nature of the data pose some limitations. We do not have validating information on reports of fracture. Over- or under-reporting of fractures is possible, but would likely be non-differential. A substantial literature on the subject demonstrates some variability in the accuracy of self-reporting. In general, reports of fractures of the hip and wrist can be confirmed 90% of the time, while reports of fractures at other sites are less reliable. Still, data from several large cohort studies suggest an overall accuracy of fracture self-reports in the range of 80%.^(26–28)

A further limitation to our fracture data comes from the method of collecting information on prior fractures. These data were acquired from a single question that asked women to identify any of 10 bones they had fractured since the age of 45. If multiple bones were identified, we cannot distinguish whether they were broken at different times or in the same episode. Nor were we able to capture multiple fractures of a single type of bone. It is likely that there are women who sustained more than one prior vertebral fracture, for example, who are characterized as having experienced only one. Thus, our data likely underestimate the number of women with multiple past fractures. Reported vertebral fractures are likely to represent only those that came to medical attention and morphometric fractures would not be included. Because we lacked information on the timing of prior fractures, calculating the interval to incident fracture occurrence was not possible. Several reports have indicated that the risk of subsequent fracture declines with time from the prior fracture.^{(13, 25)} We grouped incident fractures as a single outcome (“any fracture”) and in categories such as “weight-bearing” and “non-weight-bearing” to provide adequate statistical power as well as economy of presentation. Grouping sites by their “weight-bearing” function carries clinical plausibility, while attempting to avoid the loss of specific fracture risk information identified by Kelsey and Samuelson⁽¹⁹⁾ as a problem of combining sites with disparate determinants of risk.

To conclude, in this large, international cohort of women, nine of 10 prior fracture sites were associated with an increased risk of incident fractures in the 2 years that followed. Prior fractures of the rib and ankle made up one third of reported prior fractures, and were significant predictors of future fracture risk. For those with a history of one or more previous fracture since the age of 45, the risk was multiplied by a factor of 2- to 5-fold. Appreciation of this expanded role of prior fractures is important to both those in clinical practice and those developing fracture risk models.