Osteoporos Int (2005) 16: 1330–1338 DOI 10.1007/s00198-005-1863-y

O R I GI N A L A R T IC L E

Body mass index as a predictor of fracture risk: A meta-analysis C. De Laet Æ J.A. Kanis Æ A. Ode´n Æ H. Johanson O. Johnell Æ P. Delmas Æ J.A. Eisman Æ H. Kroger S. Fujiwara Æ P. Garnero Æ E.V. McCloskey D. Mellstrom Æ L.J. Melton 3rd Æ P.J. Meunier H.A.P. Pols Æ J. Reeve Æ A. Silman Æ A. Tenenhouse

Received: 1 December 2004 / Accepted: 27 January 2005 / Published online: 1 June 2005 Ó International Osteoporosis Foundation and National Osteoporosis Foundation 2005

Abstract Low body mass index (BMI) is a well-documented risk factor for future fracture. The aim of this study was to quantify this effect and to explore the association of BMI with fracture risk in relation to age, gender and bone mineral density (BMD) from an international perspective using worldwide data. We studied individual participant data from almost 60,000 men and women from 12 prospective population-based cohorts comprising Rotterdam, EVOS/EPOS, CaMos, Rochester, Sheffield, Dubbo, EPIDOS, OFELY, Kuopio, Hiroshima, and two cohorts from Gothenburg, with a total follow-up of over 250,000 person years. The effects of BMI, BMD, age and gender on the risk of any fracture, any osteoporotic fracture, and hip fracture

alone was examined using a Poisson regression model in each cohort separately. The results of the different studies were then merged. Without information on BMD, the age-adjusted risk for any type of fracture increased significantly with lower BMI. Overall, the risk ratio (RR) per unit higher BMI was 0.98 (95% confidence interval [CI], 0.97–0.99) for any fracture, 0.97 (95% CI, 0.96–0.98) for osteoporotic fracture and 0.93 (95% CI, 0.91–0.94) for hip fracture (all p <0.001). The RR per unit change in BMI was very similar in men and women (p >0.30). After adjusting for BMD, these RR became 1 for any fracture or osteoporotic fracture and 0.98 for hip fracture (significant in women). The gradient of fracture risk without adjustment for BMD was

C. De Laet Scientific Institute of Public Health, Brussels, Belgium

S. Fujiwara Department of Clinical Studies, Radiation Effects Research Foundation, Hiroshima, Japan

J.A. Kanis (&) Æ E.V. McCloskey WHO Collaborating Centre for Metabolic Bone Diseases, University of Sheffield Medical School, Beech Hill Road, Sheffield, S10 2RX, UK E-mail: [email protected] Tel.: +44-114-2851109 Fax: +44-114-2851813

P. Garnero INSERM Unite´ 403, Lyon, France D. Mellstrom Department of Geriatric Medicine, Goteborg University, Gothenburg, Sweden

A. Ode´n Æ H. Johanson Consulting Statistician, Gothenburg, Sweden

L.J. Melton 3rd Division of Epidemiology, Mayo Clinic, Rochester, Minnesota, USA

O. Johnell Department of Orthopaedics, Malmo¨ General Hospital, Malmo¨, Sweden

P.J. Meunier INSERM Unit 403, Faculty R Laennec, Lyon, France

P. Delmas INSERM Unite´ 149, Villejuif, France

J. Reeve Strangeway’s Research Laboratory, Wort’s Causeway, Cambridge, UK

J.A. Eisman Bone and Mineral Research Program, Garvan Institute of Medical Research, St Vincent’s Hospital and University of New South Wales, Sydney, Australia

A. Silman ARC Epidemiology Unit, University of Manchester, Manchester, UK

H. Kroger Department of Surgery, Bone and Cartilage Research Unit, Kuopio University Hospital, Kuopio, Finland

A. Tenenhouse Division of Bone Metabolism, The Montreal General Hospital, Montreal, Canada

1331

not linearly distributed across values for BMI. Instead, the contribution to fracture risk was much more marked at low values of BMI than at values above the median. This nonlinear relation of risk with BMI was most evident for hip fracture risk. When compared with a BMI of 25 kg/m2, a BMI of 20 kg/m2 was associated with a nearly twofold increase in risk ratio (RR=1.95; 95% CI, 1.71–2.22) for hip fracture. In contrast, a BMI of 30 kg/ m2, when compared with a BMI of 25 kg/m2, was associated with only a 17% reduction in hip fracture risk (RR=0.83; 95% CI, 0.69–0.99). We conclude that low BMI confers a risk of substantial importance for all fractures that is largely independent of age and sex, but dependent on BMD. The significance of BMI as a risk factor varies according to the level of BMI. Its validation on an international basis permits the use of this risk factor in case-finding strategies. Keywords BMI Æ Fractures Æ Meta-analysis Æ Osteoporosis Æ Prospective studies Æ Risk

Introduction Low weight, or low body mass index (BMI), is a welldocumented risk factor for future fracture, whereas a high BMI appears to be protective [1–10]. The increasing prevalence of overweight and obesity in Western societies [11,12] might at first seem a promising development from the point of view of osteoporosis and fracture prevention. From a public health point of view, however, the story is more complicated. Obesity is associated with increased morbidity from diabetes, hypertension, and cardiovascular diseases, and is also associated with increased mortality. The same may also be true for overweight women. Recently, it was estimated that overweight (BMI>25 kg/m2) 41-year-old female nonsmokers lost on average 3.3 years of life, whereas, obese (BMI>30 kg/m2) female nonsmokers lost 7.1 life years [13]. It is important, therefore, to quantify the association between BMI and fracture risk and to explore its relationship to age, gender and bone mineral density (BMD) with the aim of being able to give balanced advice on lifestyle to patients. These relationships are also important when using BMI to assess fracture risk in case finding [2,14–16]. The aim of the present study was to explore the relationship of BMI with fracture risk (any fracture, any osteoporotic fracture and hip fracture alone) in men and women using data from 12 prospective population-based cohort studies in an international perspective. BMI was chosen rather than weight to explore this association, because of the wide variation in average weight and height between different countries, which is reduced by adjusting weight for height. Moreover, BMI is as good a predictor of fractures as weight in most studies of hip fracture outcomes [17,18].

Methods Participants We used baseline and follow-up data from 12 prospective population-based cohorts comprising the Rotterdam Study, The European Vertebral Osteoporosis Study (later the European Prospective Osteoporosis Study (EVOS/EPOS), The Canadian Multicentre Osteoporosis Study (CaMos), Rochester, Sheffield, Dubbo, Epide´miologie de l’Oste´oporose Study (EPIDOS), a cohort from France (OFELY), Kuopio, Hiroshima and two cohorts from Gothenburg. Details of each of the cohorts are published elsewhere, but are summarized briefly below and in Table 1. The Rotterdam Study, begun in 1990, is an ongoing prospective cohort study that aimed to examine and follow-up all residents aged 55 years and older living in Ommoord, a district of Rotterdam [19]. By 1993, 7,983 residents had been included (response rate 78%). Fracture follow-up was achieved through an automatic link with general practitioner computer systems and hospital admission data [20]. Fracture data were collected and validated by two independent research physicians. For this analysis, validated fracture follow-up was available for 6,851 participants (2,793 men) with an average follow-up time of 6 years. Femoral neck BMD was measured in 5,731 individuals (2,414 men) by dual X-ray absorptometry (DXA) (Lunar DPX-L). The European Vertebral Osteoporosis Study (EVOS) comprised age- and sex-stratified random samples from 36 centers in 19 European countries [21]. Equal numbers of men and women were drawn in each center within six 5-year age bands (50–74 and 75+ years). BMD was measured in 3,461 men and women from 13 centers by DXA at the femoral neck using pencil beam machines that were cross-calibrated using the European Spine Phantom. This sample provided the framework for the European Prospective Osteoporosis Study (EPOS), where repeated assessment was undertaken in 29 of the centers [22,23]. For this analysis, validated fracture follow-up was available for 13,490 participants (6,521 men) with an average follow-up time of 3 years. Femoral neck BMD was measured in 4,746 individuals (2,141 men). The Canadian Multicentre Osteoporosis study (CaMos) is an ongoing prospective age stratified cohort. The study is documenting the incidence of fractures and risk factors in a random sample of 9,424 men and women aged 25 years or more selected by telephone listings. The sampling frame is from nine study centers in seven provinces [24]. Characterization of individuals was by interview. BMD was measured by DXA at the femoral neck with Hologic QDR in seven centers and the Lunar DPX Alpha in two centers in 8,297 individuals (2,589 men). Machines were cross-calibrated using the same European Spine Phantom. For this analysis, validated fracture follow-up was available for 9,101 participants (2,801 men) with an average follow-up time of 3 years.

1332 Table 1 Details of cohorts studied (see Methods for cohort abbreviations, BMI body mass index) Cohort EVOS-EPOS

Mean Gender Sample Women Person- Any Hip Osteoporotic Mean Mean BMI Mean height (cm) weight (kg) size (%) years fracture fracture fracture age (years) (kg/m2)

Men Women Total CaMos Men Women Total Rochester Men Women Total Rotterdam Men Women Total DOES Men Women Total Gothenburg 1 Men Women Total Hiroshima Men Women Total OFELY Women Sheffield Women Kuopio Women Gothenburg 2 Women EPIDOS Women All men All women Overall

6,521 6,969 13,490 2,801 6,300 9,101 348 653 1,001 2,793 4,058 6,851 819 1,270 2,089 812 1,158 1,970 793 1,810 2,603 430 2,170 11,691 7,065 1,183 14,887 44,757 59,644

52 69 65 59 61 59 70 100 100 100 100 100 75

19,736 20,945 40,681 8,002 17,832 25,834 1,160 5,067 6,227 16,150 23,443 39,593 6,365 9,629 15,994 6,010 9,191 15,201 3,004 6,821 9,825 2,144 6,894 56,091 29,603 3,947 60,427 191,607 252,034

213 506 719 124 447 571 38 251 289 185 676 861 138 381 519 95 255 350 44 143 187 50 292 1,043 440 – 837 4,484 5,321

20 30 50 9 31 40 0 42 42 52 168 220 27 76 103 73 198 271 7 25 32 – 63 – 29 291 188 953 1,141

The Rochester cohort was recruited from two random population samples stratified by decade of age, one of women who were subsequently followed for up to 20 years [25], and another sample of women and men followed for 8 years [26]. BMD of the right femoral neck was measured by dual photon absorptiometry in the first cohort (cross-calibrated to DXA) and by DXA (Hologic QDR 2000) in the second group. Fractures were ascertained by periodic interview combined with review of the in-patient and out-patient medical records of all local care providers. For this analysis, validated fracture follow-up was available for 1,001 participants (348 men) with an average follow-up time of 6 years. Femoral neck BMD was measured in 993 individuals (345 men). The Sheffield cohort comprised women aged 75 years or more selected randomly from the population of Sheffield, UK, and surrounding districts between 1993 and 1999. Approximately 35,000 women, identified from general practitioner listings, were contacted by letter and invited to attend for the assessment of skeletal status; 5,873 women were willing to attend. Of these, 281 women were excluded and the remainder randomly allocated to treatment with placebo or a bisphosphonate, clodronate, to study its effects on fracture risk. The material for this study comprised 2,172 women allocated to treatment with placebo only [27]. All women had baseline assessment of BMD undertaken at the femoral neck using the Hologic QDR 4500. Outcomes were assessed by 6-monthly home visits. For this analysis, val-

213 506 719 59 248 307 25 219 244 130 516 646 110 297 407 95 255 350 12 78 90 – 243 – 312 – 644 2674 3318

64.2 63.5

27 27.1

171 159

79.1 68.4

59.9 62.9

27 26.9

174 160

81.4 68.6

55.4 57.6

27.3 25.5

175 161

84.3 66.3

68.4 69.9

25.7 26.7

175 161

78.3 69.4

70.1 70.7

26 25.4

173 160

78.2 64.8

77 78.6

25.4 25.3

173 159

75.7 63.9

63.2 65.9

22.7 23.1

163 150

60.7 52.3

64.1 80 52.3 58.9 82.4 66.4 62.2 63.2

24.2 26.7 26.2 24.6 25.4 26.2 25.9 26.0

158 156 161 165 153 172.6 160.4 163.3

60.5 64.8 68.2 67.2 59.5 77.9 66.9 69.5

idated fracture follow-up was available for 2,170 participants with an average follow-up time of 6 years. Femoral neck BMD was measured in 2,150 individuals. The Dubbo Osteoporosis Epidemiology Study (DOES) is a population-based study with multiple assessments of skeletal status in men and women aged 60 years or more from Dubbo, Australia [28]. Participation in the study was 56% of the population. Baseline measurements included BMD at the femoral neck assessed using DXA (GE-Lunar, DPX and Prodigy). Fractures are identified through radiologists’ reports from the two centers serving the region. For this analysis, validated fracture follow-up was available for 2,089 participants (819 men) with an average follow-up time of 8 years. Femoral neck BMD was measured in 2,060 individuals (801 men). The EPIDOS cohort is a prospective multicenter study of risk factors for hip fractures that included 7,575 French women age 75 years or older [29]. Participants were recruited through mailings using large populationbased listings such as voter-registration rolls. Baseline characteristics were obtained through a structured questionnaire as well as through clinical and functional examinations and BMD measurement (Lunar DPX). For this analysis, validated fracture follow-up was available for 1,183 participants comprising 291 hip fracture cases and age-matched controls with an average follow-up time of 3 years [30]. Femoral neck BMD was measured in 1,180 individuals.

1333

The OFELY cohort comprises a cohort of 1,039 women aged 31–89 years stratified by age randomly selected from the regional section of a large health insurance company (Mutuelle Ge´ne´rale d’Education Nationale, Lyon) [31]. Eighteen percent of women contacted participated in the study. Baseline characteristics were obtained using a standardized questionnaire, including the documentation of prior wrist, humeral, vertebral and hip fracture that occurred after the age of 45 years. Only low-trauma fractures (falls from a standing height or less) were recorded, but not the site of fracture. BMD was measured at the spine (L1–L4), at the proximal femur, distal radius and whole body by DXA using a QDR 2000 (Hologic). Women were reviewed annually and incident fractures registered. Peripheral fractures were confirmed by radiography. For this analysis, validated fracture follow-up was available for 430 participants with an average follow-up time of 5 years. Femoral neck BMD was measured in 427 individuals. The Kuopio osteoporosis risk factor and prevention (OSTPRE) study in Finland comprised a postal enquiry sent to all 14,220 women aged 47–56 who were residents of Kuopio province in 1989 [32]. Of 13,100 women responding to the enquiry, 938 were excluded for incomplete information. For this analysis, validated fracture follow-up was available for 11,691 participants with an average follow-up time of 5 years. The site of fracture was not recorded other than those at the forearm. Femoral neck BMD was measured in 1,743 individuals with the Lunar DPX [32]. The Gothenburg I study comprised four birth cohorts of 2,375 randomly sampled men and women aged 70 years or more followed for up to 20 years after a baseline BMD measurement [33]. The participants were drawn randomly from the population register in Gothenburg by the date of birth to provide cohorts aged 70, 76, 79 and 85 years at the time of investigation. Participation rate was 73%. Bone mineral density was measured at the right heel using dual photon absorptiometry. For this analysis, validated fracture follow-up was available for 1,970 participants (812 men) with an average follow-up time of 8 years. BMD was measured in 1,633 individuals (686 men). The Gothenburg II study comprised a randomly drawn population cohort of over 7,000 women aged 50– 70 years followed for 4 years [34]. The participation rate was 67%. Assessment included a standardized questionnaire that recorded information on risk factors for osteoporosis. Fractures were identified prospectively through the radiology departments serving the region. BMD was assessed at baseline at the distal forearm by using the Osteometer DTX-200. For this analysis, validated fracture follow-up was available for 7,065 participants with an average follow-up time of 4 years. BMD was measured in 7,056 individuals. The Adult Health Study (AHS) was established in 1958 to document the late health effects of radiation exposure among atomic bomb survivors in Hiroshima

and Nagasaki. The original AHS cohort consisted of about 15,000 atomic bomb survivors and 5,000 controls selected from residents in Hiroshima and Nagasaki using the 1950 national census supplementary schedules and the Atomic Bomb Survivors Survey. AHS subjects have been followed through biennial medical examinations since 1958. The participation rate has been around 80% throughout this period. BMD was measured at the proximal femur by DXA in 1994 (Hologic QDR-2000) in 2,588 individuals (791 men). Self-reported fractures were documented at 6-monthly intervals [35,36]. For this analysis, validated fracture follow up was available for 2,603 participants (793 men) with an average follow-up time of 4 years. Baseline and outcome variables Height and weight were measured using standard techniques in all cohorts. BMI was calculated as weight in kilograms divided by height squared in meters. For the purposes of this analysis, we utilized BMD assessed at the femoral neck by DXA, with the exception of the two Gothenburg cohorts where BMD was assessed by DXA at the distal forearm or by DPA at the right heel. We additionally analyzed the BMD data excluding those two cohorts. Fracture ascertainment was undertaken by self-report (Sheffield, EVOS/EPOS, Hiroshima, Kuopio, EPIDOS, OFELY) and/or verified from hospital or central databases (Gothenburg, CaMos, DOES, Kuopio, Sheffield, EVOS/EPOS, Rochester, Rotterdam). The EPOS and the Rotterdam study also included sequential systematic radiography to define incident vertebral deformities, but these were not used in this analysis. In the analysis, we used information on any clinical fracture and on fractures considered to be osteoporotic. In addition, hip fracture alone was considered separately. An osteoporotic fracture was one considered to be due to osteoporosis by the investigator in the EVOS/EPOS study and in CaMos. For the EVOS/EPOS study, osteoporotic fractures comprised hip, forearm, humeral or spine fractures. For the CaMos Study they comprised fractures of the spine, pelvis, ribs, distal forearm, forearm and hip. In the other cohorts, fractures at sites considered to be characteristic for osteoporosis were extracted [37]. Statistical methods The association of BMI with the risk of any fracture, osteoporotic fracture and hip fracture was examined using a Poisson regression model in each cohort separately. Covariates included current age and time since start of follow-up, and we performed analyses for both sexes separately, with and without taking BMD information into account. BMD was expressed as sex- and cohort-specific Z -scores. BMI was analyzed either

1334

continuously or using specific thresholds. The b-coefficients of each cohort and the two sexes were weighted according to the variance and merged to determine the weighted mean of the coefficient and its standard deviation. The risk ratios (RR) at different BMI levels are then given by e(weighted mean coefficient). We also analyzed the RR per unit difference of BMI for fracture at various levels of BMI, taking a BMI of 25 kg/m2 as the reference, since a BMI of 25 is the internationally accepted threshold between normal weight and overweight [11]. This was slightly below the average in most of the cohorts (see Table 1). To optimally describe the association of RR with level of BMI, we analyzed this relationship using spline functions. Heterogeneity between cohorts was tested by means of the I2 statistic [38]. Low heterogeneity was noted for hip fracture outcomes (I2=8%; p>0.30) and was moderate for osteoporotic fracture outcomes (I 2= 49%). When the interaction between BMI and current age (BMIÆcurrent age) was included, there was no significant heterogeneity between cohorts (p >0.30) for BMI (I 2 =0) nor for the interaction term (I 2 =0) and a fixed-effects rather than a random-effects model was used.

Results The total sample included almost 60,000 men and women from the 12 cohorts with a total follow-up of more than 250,000 person years. The distribution of BMI is shown in Table 2. Information on any fracture, on osteoporotic fracture and on hip fracture alone was available for about 58,000, 46,000 and 47,000 participants, respectively. During follow-up, there were 5,321 fractures registered including 3,318 osteoporotic fractures and 1,141 hip fractures. Bone mineral density measurements were available in 65% of individuals. Details by cohort and gender are shown in Table 1. Without information on BMD, low BMI in men and women combined was associated with a significantly increased age-specific risk of fracture, while at higher BMI values the risk of fracture was decreased. The risk ratio per unit increase in BMI (gradient of risk; GR) was for any fracture 0.98 (95% confidence interval [CI], Table 2 Distribution (%) of men and women categorized by intervals of BMI (BMI body mass index) BMI (kg/m2)

Men

Women

Total

<20 20–24 25–29 30–34 35–39 40+

7.5 30.9 47.2 12.4 1.7 0.2

8.9 38.5 35.8 12.9 3.1 0.8

8.5 36.5 38.8 12.8 2.7 0.7

0.97–0.99), for osteoporotic fracture 0.97 (95% CI, 0.96– 0.98) and for hip fracture 0.93 (95% CI, 0.91–0.94) (Fig. 1A). As the figure shows, the RRs per unit change of BMI in men and women were very similar and not significantly different (p >0.30). Since all osteoporotic fractures also included hip fractures, we analyzed osteoporotic fractures excluding hip fractures. As expected, the GR increased slightly (i.e., predictive value decreased) but remained significantly lower than 1 in men and women combined (data not shown). When information on BMD was included (Fig. 1B), the GRs changed markedly and remained significantly different from unity only for hip fractures in women. When the two studies from Gothenburg were excluded (because BMD was not measured at the femoral neck), the GR for hip fracture in women was not significantly different from unity. Since the GRs were similar in men and women both before and after correction for BMD and not significantly different, further results are presented for men and women combined, but all analyses were also carried out for men and women separately. The effect of age on gradient of risk is shown in Fig. 2. For any fracture and for any osteoporotic fracture, the GR per unit of BMI increased with advancing age (without adjustment for BMD). In contrast, for hip fractures the gradient of risk decreased with age, although this trend was not significant. When hip fractures were excluded from the osteoporotic fractures, a similar trend with age was observed as seen for all osteoporotic fractures. After correction for BMD, the risk gradients showed nonsignificant trends with age, and at most ages were not significantly different from 1. We also analyzed the risk ratio (RR) for fracture at various levels of BMI, taking a BMI of 25 kg/m2 as the reference. As expected, the RR increased with decreasing BMI. The magnitude of the effect was greater for hip fracture than for any osteoporotic fracture or any fracture. The RR of fracture risk with BMI was, however, nonlinear (Fig. 3 and Table 3). RR was markedly higher at the lower values of BMI, particularly with a BMI of 20 kg/m2 or less. By contrast, between a BMI of 25 kg/ m2 and 35 kg/m2 the differences in RR were small. For example, given the overall risk gradient for hip fracture (0.93/unit BMI) an increase of 5 BMI units from 25 kg/ m2 to 30 kg/m2 would be expected to correspond to a 32% reduction in hip fracture risk. The observed difference was, however, only 17% (1.00 vs 0.83, see Table 3). With a 10-unit BMI difference the expected risk reduction would be more than 50%, whereas, the difference between a BMI of 25 kg/m2 and one of 35 kg/m2 was only 25% (1.00 vs 0.75, see Table 3). At the low end of the BMI spectrum, on the other hand, a change of 5 units from a BMI of 25 kg/m2 to a BMI of 20 kg/m2, corresponded to a doubling of the hip fracture risk (1.95 vs 1.00, see Table 2). For any osteoporotic fracture there was a 27% difference in fracture risk comparing a BMI of 20 kg/m2 vs 25 kg/m2 and an 11% difference comparing a BMI of 30 kg/m2 v 25 kg/m2.

1335 Fig. 1 Risk Ratio (RR) for fracture per unit increase in BMI; (A) adjusted for current age and time since start of follow-up, and (B) additionally adjusted for BMD (BMI body mass index, BMD bone mineral density) [04Ca001]

Fig. 2 Relative fracture risk per unit increase in BMI by age for men and women combined; (A) adjusted for time since start of follow-up, and (B), additionally adjusted for BMD (BMI body mass index, BMD bone mineral density) [04Ca002]

Fig. 3 Relative fracture risk at various levels of BMI (kg/m2) for men and women combined. The reference is a BMI=25, (A) adjusted for current age and time since start of follow-up, and (B) additionally adjusted for BMD. The bold solid line describes hip

fracture, the solid line any osteoporotic fracture, and the dotted line any fracture (BMI body mass index, BMD bone mineral density) [04Ca003]

Re-analyzing the data using only those cohorts with a uniform acquisition of data on fractures (Rochester, Rotterdam, DOES, Hiroshima and Sheffield) did not alter the relationship of BMI and hip fracture risk. For osteoporotic fracture, a high BMI had a greater and significant protective effect in the absence of BMD (data not shown).

The data for Table 3 (unadjusted for BMD) were recalculated in the 65% of individuals who had also a BMD test. The findings did not differ from those utilizing the entire cohort. After adjustment for BMD, BMI was not predictive of fracture risk except for hip fracture at a BMI of

1336 Table 3 Risk Ratio (RR) for fracture at various levels of BMI (kg/m2) for men and women combined, adjusted for current age and time in study, without and with adjustment for BMD. The reference is a BMI of 25 kg/m2 (BMD bone mineral density, BMI body mass index, CI confidence interval) BMI

Any fracture

Osteoporotic fracture

Hip fracture

95% CI

RR

95% CI

RR

95% CI

Not adjusted for BMD 15 1.66 20 1.21 25 1.00 30 0.92 35 0.85

1.31–2.09 1.12–1.30 reference 0.85–1.00 0.74–0.98

1.79 1.27 1.00 0.89 0.74

1.35)2.37 1.16–1.38 reference 0.81–0.98 0.62–0.90

4.48 1.95 1.00 0.83 0.75

3.11–6.45 1.71–2.22 reference 0.69–0.99 0.50–1.11

Adjusted for BMD 15 1.00 20 0.98 25 1.00 30 1.01 35 0.99

0.75–1.33 0.9–1.08 reference 0.91–1.11 0.82–1.19

1.07 1.02 1.00 0.96 0.91

0.78–1.48 0.92–1.13 reference 0.86–1.08 0.73–1.13

2.16 1.42 1.00 1.00 1.18

1.42–3.28 1.23–1.65 reference 0.82–1.21 0.78–1.80

RR

20 kg/m2 or less (Table 3). There were no significant differences in these relationships between men and women. The results were not changed when cohorts with uniform characterization of fractures were analyzed. After adjustment for BMD, the risk of hip fracture was 33% higher comparing individuals with a BMI of 20 kg/ m2 with 25 kg/m2 (42% higher in all cohorts; see Table 3). For osteoporotic fractures, the I2 statistic was 49% (95% CI, 8–71), indicating that about half of the observed differences between studies was due to heterogeneity. This heterogeneity disappeared when account was taken of age (I 2 =0). For hip fractures there was also low heterogeneity (I 2 =8% (0–44)) indicating that the data about the relationship between BMI and fracture risk were homogenous between cohorts.

Discussion The principal finding of the present study, undertaken in large and internationally drawn cohorts, is the confirmation that low BMI is associated with a substantial increase in fracture risk of similar magnitude in men and women, whereas, a high BMI is protective. This risk associated with a low BMI was present at most ages and for all types of fracture studied, but was strongest for hip fracture. Overall, the RR for hip fracture decreased 0.93 per unit increase in BMI. This risk gradient was, however, not constant across the range of BMI. In the most common range of normal weight and overweight (the average in our populations was 26.0 kg/m2), the gradient of risk per unit BMI was relatively low (see Fig. 3A). In contrast, the gradients were steeper at low values of BMI, below 22 kg/m2. This threshold of 22 kg/m2 and lower corresponded overall in our cohorts to about 10% of men and 17% of women. A nonlinear relationship of BMI with fracture risk, with thresholds of BMI of approximately 22 kg/m2 and 26 kg/m2, has been described previously in case-control

studies of men and women [17,18]. In the present study, where the individual participant data were pooled from several prospective studies, we could derive the form of this relationship. The present study additionally demonstrates that this gradient of risk is markedly reduced when adjusted for BMD, suggesting that BMD is an important intermediary or confounder. These findings have important consequences for casefinding strategies based on clinical risk factors. Firstly, obesity should not be regarded as an important protective factor for hip fracture risk [39,40]. Rather, leanness should be regarded as a significant risk factor. Secondly, the use of low BMI as a risk factor will identify populations with a low BMD and hence a high risk of fracture. The finding that leanness is much more important as a risk factor for hip fracture—than obesity is a protective factor—means that advice concerning body weight and osteoporosis need not be inconsistent with the weight control advocated for the prevention of cardiovascular disease or diabetes. After adjustment of fracture risk by BMD, a low BMI was still a significant risk factor for hip fracture. Thus, low BMI can be used to enhance the predictive value of BMD in case finding. The mechanisms for this are conjectural but might include muscle weakness [41], perhaps associated with nutritional deficiencies of protein or vitamin D [41,42], decreased padding over the greater trochanter [43], or a greater liability to fall [10]. This BMD-independent risk was not observed, however, when only cohorts using femoral neck BMD were analysed. BMI and other risk factors have been studied mainly in relation to hip fracture risk. As expected, the risk gradient we found was higher for hip fractures than for all osteoporotic fractures. However, for any fracture or for all osteoporotic fractures, the impact of BMI was still significant, even when hip fractures were excluded from the analysis. For BMD, it is customary to express the relation with fracture risk as a gradient of risk per SD change. For hip fracture, for example, the most

1337

commonly cited number is that of an RR of 2.6 per SD decrease in BMD at the femoral neck [44]. In the populations that we studied, the average SD for BMI was around 4 kg/m2, corresponding to a RR for hip fracture of 1.4 per SD decrease in BMI, much lower than the estimate for BMD. The analogy is, however, not wholly appropriate in view of the nonlinearity of risk with BMI. We found no significant differences in the gradients of fracture risk between men and women; but, we found a significant increase of risk associated with low BMI with age, for any fracture and for osteoporotic fractures, but an opposite, though not significant, trend for hip fractures. In the case of any fracture or osteoporotic fracture, an increased exposure with age to relative gonadal deficiency might provide an explanation, at least in women [45]. A further possible explanation is that in young individuals, low BMI may be associated with physical fitness and a lower risk of fracture. In contrast, in the elderly, where hip fractures are more common, low BMI may more likely be related to frailty. The strength of the present study is that the estimate of risk is derived from several studies in an international setting from population-based cohorts, using the original individual participant data. The large sample size permitted us to examine the general relationship of BMI with fracture risk and also the detailed relationship with age and level of BMI. The study also has some limitations. The definition of what was considered an osteoporotic fracture was not the same in all cohorts. For hip fractures, the definition is similar in all cohorts and there is substantial homogeneity between studies; here we found higher gradients of risk across the range of BMI. For osteoporotic fractures our results were essentially similar when we analyzed only those cohorts with uniform documentation of fracture. The use of BMI, rather than weight, as a measure for body composition has the great advantage that there is less variability across countries and between sexes as can be seen from Table 1. A potential drawback is that BMI can be influenced by the height loss associated with vertebral deformities. Therefore, in individuals with important height loss, the risk conferred through BMI on fracture risk could be underestimated [46]. The use of maximal attained height rather than current height might be a solution in clinical practice, if it could be shown that risk prediction could thereby be improved. We conclude that low BMI confers a risk of fracture of substantial importance that is largely independent of sex. The significance of BMI as a risk factor varies according to the level of BMI and to a lesser extent on age. Its validation on an international basis permits the use of this risk factor, at least in the absence of a BMD measurement, in case-finding strategies. Even with BMD, a low BMI may remain an independent risk factor for hip fracture in those with a BMI of less than 20 kg/m2. These data also show that there should be no conflict between advice for weight control with reasonable target values, such as for the prevention of diabetes

or cardiovascular disease, and the prevention of osteoporotic and hip fractures. Acknowledgements We are grateful to Drs. T.V. Nguyen and J.R. Center for their help with the DOES Study. We would like to thank the Alliance for Better Bone Health, Hologic, IGEA, Lilly, Lunar, Novartis, Pfizer, Roche and Wyeth for their unrestricted support of this work. We are also grateful to the EU (FP3/5), the International Osteoporosis Foundation, the International Society for Clinical Densitometry and the National Osteoporosis Foundation for supporting this study.

References 1. Cummings SR, Nevitt MC, Browner WS, Stone K, Fox KM, Ensrud KE, Cauley J, Black D, Vogt TM (1995) Risk factors for hip fracture in white women. Study of Osteoporotic Fractures Research Group. N Engl J Med 332:767–773 2. Eddy DM, Johnston CC, Cummings SR, Dawson-Hughers B, Lindsay R, Melton LJ et al (1998) Osteoporosis: Review of the evidence for prevention, diagnosis and treatment and costeffectiveness analysis. Osteoporos Int 8: [Suppl 4] 3. Honkanen RJ, Honkanen K, Kroger H, Alhava E, Tuppurainen M, Saarikoski S (2000) Risk factors for perimenopausal distal forearm fractures. Osteoporos Int 11:265–270 4. Galvard H, Elmstahl S, Elmstahl B, Samuelsson SM, Robertsson E (1996) Differences in body composition between female geriatric hip fracture patients and healthy controls: body fat is more important as explanatory factors for the fracture than body weight and lean body mass. Aging (Milano) 8:282–286 5. Joakimsen RM, Fonnebo V, Magnus JH, Tollan A, Sogaard AJ (1998) The Tromso study: body height, body mass index and fractures. Osteoporos Int 8:436–442 6. Turner LW, Faile PA, Tomlinson R (1999) Osteoporosis diagnosis and fracture. Orthop Nurs 18:21–27 7. Van der Voort DJ, Geusens PP, Dinant CJ (2001) Risk factors for osteoporosis related to their outcome: fractures. Osteoporos Int 12:630–638 8. Young Y, Myers AH, Provenzano G (2001) Factors associated with time to first hip fracture. J Aging Health 13:511–526 9. Roy DK, O’Neill TW, Finn JD, Lunt M, Silman AJ, Felsenberg D, Ambrecht G, Banzer D, Benevolenskaya LI, Bhalla A, Bruges Armas J, Cannata JB, Cooper C, Dequeker J, Eastell R, Yershova O, Felsch B, Gowin W, Havelka S, Hoszowski K, Jajic I, Janot J, Johnell O, Kanis JA, Krag G, Lopez Vaz A, Lorenc R, Lyritis G, Masaryk P, Matthis C, Miazgowski T, Gennari C, Pols HAP, Poor G, Raspe HH, Reid DM, Reisinger W, Scheidt-Nave C, Stepan J, Todd C, Weber K, Woolf AD, Reeve J (2003) Determinants of incident vertebral fracture in men and women: results from the European Prospective Osteoporosis Study (EPOS). Osteoporos Int 14; 19–26 10. Willig R, Luukinen H, Jalovaara P (2003) Factors related to occurrence of hip fractures during a fall on the hip. Public Health 117:25–30 11. World Health Organization (1998) Obesity. Preventing and managing the global epidemic. World Health Organization, Geneva 12. Mokdad AH, Bowman BA, Ford ES, Vinicor F, Marks JS, Koplan JP (2001) The continuing epidemics of obesity and diabetes in the United States. JAMA 286:1195–200 13. Peeters A, Barendregt JJ, Willekens F, Mackenbach JP, Al Mamun AA, Bonneux L; NEDCOM, the Netherlands Epidemiology and Demographic Compression of Morbidity Research Group (2003) Obesity in adulthood and its consequences for life expectancy: a life-table analysis. Ann Intern Med 138:24–32 14. Kanis JA, Delmas P, Burckhardt P, Cooper C, Torgerson D, on behalf of the European Foundation for Osteoporosis and Bone Disease (1997) Guidelines for diagnosis and management of osteoporosis. Osteoporos Int 7:390–406

1338 15. Royal College of Physicians (1999) Clinical Guidelines for strategies to prevent and treat osteoporosis. Royal College of Physicians, London 16. Kwaliteitsinstituut voor de Gezondheidszorg CBP (2002) Osteoporose: tweede herziene richtlijn. Utrecht, the Netherlands. ISBN 90–76906–23–8 17. Johnell O, Gullberg B, Kanis JA, Allander E, Elffors L, Dequeker J, Dilsen G, Gennari C, Vaz AL, Lyritis G, Mazzuoli G, Miravet L, Passeri M, Cano RP, Rapado A, Ribot C (1995) Risk factors for hip fracture in European women—the Medos Study. J Bone Miner Res 10:1802–1815 18. Kanis J, Johnell O, Gullberg B, Allander E, Elffors L, Ranstam J, Dequeker J, Dilsen G, Gennari C, Vaz A, Lyritis G, Mazzuoli G, Miravet L, Passeri M, Perez CR, Rapado A, Ribot C (1999) Risk factors for hip fracture in men from southern Europe: the MEDOS study. Mediterranean Osteoporosis Study. Osteoporos Int 9:45–54 19. Hofman A, Grobbee DE, de Jong PT, van den Ouweland FA (1991) Determinants of disease and disability in the elderly: The Rotterdam Elderly Study. Eur J Epidemiol 7:403–422 20. De Laet CEDH, Van Hout BA, Burger H, Hofman A, Weel AEAM, Pols HAP (1998) Hip fracture prediction in elderly men and women: validation of the Rotterdam Study. J Bone Miner Res 13:1587–1593 21. O’Neill TW, Felsenberg D, Varlow J, Cooper C, Kanis JA, Silman AJ (1996) The prevalence of vertebral deformity in European men and women: the European Vertebral Osteoporosis Study. J Bone Miner Res 11:1010–1017 22. Felsenberg D, Silman AJ, Lunt M, Ambrecht G, Ismail AA, Finn JD, Cockerill WC, Banzer D, Benevolenskaya LI, Bhalla A, Bruges Armas J, Cannata JB, Cooper C, Dequeker J, Eastell R, Ershova O, Felsch B, Gowin W, Havelka S, Hoszowski K, Jajic I, Janot J, Johnell O, Kanis JA, Kragl G, Lopez Vaz A, Lorenc R, Lyritis G, Masaryk P, Matthis C, Miazgowski T, Parisi G, Pols HAP, Poor G, Raspe HH, Reid DM, Reisinger W, Scheidt-Nave C, Stepan J, Todd C, Weber K, Woolf AD, Reeve J, O’Neill TW (2002) Incidence of vertebral fracture in Europe: results from the European Prospective Osteoporosis Study, Epos. J Bone Miner Res 17:716–724 23. Ismail AA, Pye SR, Cockerill WC, Lunt M, Silman AJ, Reeve J, Banzer D, Benevolenskaya LI, Bhalla A, Bruges Armas J, Cannata JB, Cooper C, Delmas PD, Dequeker J, Dilsen G, Falch JA, Felsch B, Felsenberg D, Finn JD, Gennari C, Hoszowski K, Jajic I, Janott J, Johnell O, Kanis JA, Kragl G, Lopez Vaz A, Lorenc R, Lyritis G, Marchand F, Masaryk P, Matthis C, Miazgowski T, Naves-Diaz M, Pols HAP, Poor G, Rapado A, Raspe HH, Reid DM, Reisinger W, Scheidt-Nave C, Stepan J, Todd C, Weber K, Woolf AD, O’Neill TW (2002) Incidence of limb fracture across Europe: results from the European prospective osteoporosis study, EPOS. Osteoporos Int 13:565–571 24. Kreiger N, Tenenhouse A, Joseph L et al (1999) The Canadian Multicentre Osteoporosis Study CaMos: background, rationale, methods. Can J Aging 18:376–387 25. Melton LJ III, Crowson CS, O’Fallon WM, Wahner HW, Riggs BL (2003) Relative contributions of bone density, bone turnover and clinical risk factors to long-term fracture prediction. J Bone Miner Res 18:312–318 26. Melton LJ III, Atkinson EJ, O’Connor MK, O’Fallon WM, Riggs BL (1998) Bone density and fracture risk in men. J Bone Miner Res 13:1915–1923 27. Johansson H, Oden A, Johnell O, Jonsson B, De Laet C, Oglesby A, McCloskey EV, Kayan K, Jalava T, Kanis JA (2004) Optimization of BMD measurements to identify high risk groups for treatment—a test analysis. J Bone Miner Res 19(6):906-913

28. Jones G, Nguyen TV, Sambrook PN, Kelly PJ, Gilbert C, Eisman JA (1994) Symptomatic fracture incidence in elderly men and women. The Dubbo Osteoporosis Epidemiology Study, DOES. Osteoporos Int 4:277–282 29. Dargant-Molina P, Favier F, Grandjean H et al (1996) Fallrelated factors and risk of hip fracture: the EPIDOS prospective study. Lancet 348:145–149 30. Chapurlat R, Garnero P, Breart G, Meunier P, Delmas PD (2000) Serum estradiol and sex hormone-binding globulin and the risk of hip fracture in elderly women: The EPIDOS study. J Bone Miner Res 15:1835–1841 31. Garnero P, Sornay-Rendu E, Claustrat B, Delmas PD (2000) Biochemical markers of bone turnover, endogenous hormones and the risk of fracture in postmenopausal women: the OFELY study. J Bone Miner Res 15:1526–1536 32. Honkanen R, Tuppurainen M, Kroger H, Alhava E, Saarikoski S (1998) Relationships between risk factors and fractures differ by type of fracture; a population-based study of 12,192 perimenopausal women. Osteoporos Int 8:25–31 33. Svanborg A (1977) Seventy-year-old people in Gothenburg. A population study in an industrialised Swedish city ii. Journal Presentation of Social and Medical Conditions. Acta Medica Scand 611 [Suppl]:5 34. Johansson C, Black D, Johnell O, Oden A, Mellstrom D (1998) Bone mineral density is a predictor of survival. Calcif Tissue Int 63:190–196 35. Fujiwara S, Kasagi F, Yamada M, Kodama K (1997) Risk factors for hip fracture in Japanese cohort. J Bone Miner Res 12:998–1004 36. Fujiwara S, Kagasi F, Masunari N, Naito K, Suzuki G, Fukunage M (2003) Fracture prediction from bone mineral density in Japanese men and women. J Bone Miner Res 18:1547–1553 37. Kanis JA, Oden A, Johnell O, Jonsson B, De Laet C, Dawson A (2002) The burden of osteoporotic fractures: a method for setting intervention thresholds. Osteoporos Int 12:417–424 38. Higgins JPT, Thompson SG, Deeks JJ, Altman DG. (2003) Measuring inconsistency in meta-analyses. BMJ 327:557–60 39. Kiel DP, Felson DT, Anderson JJ, Wilson PWF (1987) Hip fracture and the use of estrogens in postmenopausal women. The Framingham Study. N Eng J Med 317:1169–1174 40. Malmivaara A, Heliovaara M, Knekt P, Reunanen A, Aromaa A (1993) Risk factors for injurious falls leading to hospitalization or death in a cohort of 19,500 adults. Am J Epidemiol 138:384–394 41. Bischoff HA, Stahelim HB, Dick W, Akos R, Kuecht M, Salis C, Nebiker M, Thieler R, Pfeifer M, Bergerow B, Lew RA, Conzelmann M (2003) Effects of vitamin D and calcium supplementation on falls: a randomized controlled trial. J Bone Miner Res 18:343–351 42. Wootton R, Brereton PJ, Clark MB, Hesp R, Hodkinson HM, Klenerman L, Reeve J, Slavin G, Tellez M (1979) Fractured neck of femur in the elderly: an attempt to identify patients at risk. Clin Sci 57:93–101 43. Nilsson BE (1970) Spinal osteoporosis and femoral neck fracture. Clin Orthop Relat Res 68:93–95 44. Marshall D, Johnell O, Wedel H (1996) Meta-analysis of how well measures of bone mineral density predict occurrence of osteoporotic fractures. BMJ 312:1254–1259 45. Sirola J, Kroger H, Honkanen R, Sandini L, Tuppurainen M, Jurvelin JS, Saarikoski S (2003) Risk factors associated with peri- and postmenopausal bone loss: does HRT prevent weight loss-related bone loss? Osteoporos Int 14:27–33 46. Davison KS, Siminoski K, Chik C, Jen H, Warshawski R, Lee K (2003) Impact of height loss due to vertebral fractures on body mass index. J Bone Miner Res [Suppl]:S243