1595–1601www.elsevier.com/locate/bone

Bone 40 (2007)

Biphasic fracture risk in diabetes: A population-based study

William D. Leslie a,⁎, Lisa M. Lix b, Heather J. Prior b, Shelley Derksen b,Colleen Metge b, John O'Neil c

a Department of Medicine (C5121), University of Manitoba, 409 Tache Avenue, Winnipeg, Manitoba, Canada R2H 2A6b Manitoba Centre for Health Policy, University of Manitoba, Winnipeg, Canada R2H 2A6

c Community Health Sciences, University of Manitoba, Winnipeg, Canada R2H 2A6

Received 24 October 2006; revised 13 February 2007; accepted 21 February 2007Available online 1 March 2007

Abstract

Diabetes is associated with increased fracture rates but the effect size, time course and modifying factors are poorly understood. This study wasundertaken to assess the effect of diabetes on fracture rates and possible interactions with age, duration of diabetes and comorbidity. A retrospective,population-based matched cohort study (1984–2004) was performed using the Population Health Information System (POPULIS) for the Provinceof Manitoba, Canada. The study cohort consisted of 82,094 diabetic adults and 236,682 non-diabetic matched controls. Diabetes was subclassifiedas long term, short term, and newly diagnosed. Number of ambulatory diagnostic groups (ADGs) was an index of comorbidity. Poisson regressionwas used to study counts of combined hip, wrist and spine (osteoporotic) fractures (5691 with diabetes and 16,457 without diabetes) and hipfractures (1901 with diabetes and 5224 without diabetes). Independent effects of longer duration of diabetes (p-for-trend<0.0001) and number ofADGs (p-for-trend<0.0001) were observed on fracture rates. Newly diagnosed diabetes showed a reduction in osteoporotic fractures (rate ratio[RR] 0.91 [95% CI, 0.86–0.95]) and hip fractures (RR 0.83 [0.75–0.92]). Long-term diabetes showed an increase in osteoporotic fractures (RR 1.15[CI, 1.09–1.22]) and hip fractures (RR 1.40 [1.28–1.53]). We conclude that long-term diabetes is associated with increased fracture risk, whereasnewly diagnosed diabetes shows a reduction in fractures. It is hypothesized that the opposing effects of overweight/obesity and diabetes-relatedcomplications contribute to the observed biphasic fracture risk, though causality cannot be proven from this observational study.© 2007 Elsevier Inc. All rights reserved.

Keywords: Epidemiology; Population studies; Osteoporosis; Fractures; Diabetes

Introduction

Osteoporotic fractures are highly prevalent in developedcountries. About 40% of women experience an osteoporosis-related fracture in the course of their lifetime with men atapproximately one-third to one-half the risk of women [1,2].Low-trauma fractures, the clinical expression of osteoporosis,increase dramatically with age [3]. Since many chronic healthproblems become more prevalent with age, osteoporosis fre-

⁎ Corresponding author. Fax: +1 204 237 2007.E-mail addresses: bleslie@sbgh.mb.ca (W.D. Leslie),

lisa_lix@cpe.umanitoba.ca (L.M. Lix), heather_prior@cpe.umanitoba.ca(H.J. Prior), shelley_derksen@cpe.umanitoba.ca (S. Derksen),metge@ms.umanitoba.ca (C. Metge), oneilj@ms.umanitoba.ca (J. O'Neil).

8756-3282/$ - see front matter © 2007 Elsevier Inc. All rights reserved.doi:10.1016/j.bone.2007.02.021

quently occurs in conjunction with other disorders which mayinteract (positively or negatively) in terms of fracture risk.

Diabetes is much more frequent in the elderly [4]. Furt-hermore, there is a global increase in the prevalence of dia-betes and obesity [5], just as there is for osteoporosis [4].Initially it was felt that type 2 diabetes might be protectedagainst fractures due to the effect of overweight/obesity, a riskfactor for type 2 diabetes whereas underweight is a risk factorfor low bone density and fractures [6]. Subsequent studieshave shown that bone density is increased in type 2 diabetes,but fracture risk is paradoxically increased [7–9]. Diabeticcomplications leading to falls and/or compromised bonequality may underlie this effect. Duration of diabetes is there-fore likely to be an important modifier of this fracture risk,though not all studies have confirmed an effect of duration[10]. Age and chronic comorbidities could modify the fracture

1596 W.D. Leslie et al. / Bone 40 (2007) 1595–1601

risk associated with diabetes, but studies to date have not as-sessed these interactions.

Materials and methods

A retrospective, matched-cohort study was performed using the PopulationHealth Information System (POPULIS) data repository at the Manitoba Centrefor Health Policy (MCHP) [11]. Manitoba Health provides comprehensivehealth care coverage for residents of Manitoba, Canada, and maintains com-puterized databases of health services that include demographics, date and typeof service, and diagnoses from the International Classification of Disease-9-Clinical Modification (ICD-9-CM). An encrypted personal identifier allows forlinkage across datasets and creation of person-specific longitudinal records ofhealth service utilization. The study was approved by the Research Ethics Boardfor the University of Manitoba and Health Information Privacy Committee ofManitoba Health.

The study cohort included all diabetic adults (aged 20 years or older as ofDecember 31, 1993) who were continuously residents in the province from 1984to the end of 1993. Individuals who migrated into, or out of the province, or whodied during the pre-inception period, were excluded. The presence of diabetesprior to study inception (January 1, 1994) was based upon a validated definition:two physician office visits or a single hospitalization with a diagnosis of diabetes(ICD-9-CM code 250) in a 3-year period [4]. Records for preceding years wereused to subclassify diabetes as long term (diabetic more than 5 years prior toinception) and short term (diabetic less than 5 years prior to inception) [12].Individuals not meeting the diabetes definition at inception but who did meetthis definition during the follow-up study period were categorized as newlydiabetic. In total, 82,094 men and women met one of these mutually exclusivediabetes categories. Diabetes was classified as new onset in 42,874 subjects(52.2%), short term in 16,081 (19.6%) and long term in 23,139 (28.2%).

Up to three controls (non-diabetic at inception as well as throughout thefollow-up study period) were randomly matched to each diabetic subject. Con-trols were matched by birth year, gender, and area of residence within theprovince. First Nations Aboriginal ethnicity (also referred to as Natives or “In-dians”) identified from government files was an additional matching variablesince Canadian Aboriginals have higher rates of fracture and diabetes than non-Aboriginals [13,14]. No match could be found for 708 subjects and these wereexcluded from analysis. The final cohort therefore consisted of 82,094 diabeticadults and 236,682 non-diabetic controls.

Fractures that occurred from study inception (January 1, 1994) until the endof the follow-up period (March 31, 2004 unless the individual died or left theprovince) were identified. Vertebral fractures (ICD-9-CM code 805), wristfractures (ICD-9-CM code 813) and hip fractures (ICD-9-CM code 820–821)were taken to be indices of osteoporotic fractures [13]. Hip fractures had to beaccompanied by a physician claim for a site-specific fracture reduction orfixation. Hip fractures were studied as a specific subgroup due to the clinicalimportance of these fractures in terms of subsequent morbidity and mortality.

Area of residence in the province was defined at study inception and incomequintiles were defined using average household income from the 1996 StatisticsCanada Census [15]. The Johns Hopkins Ambulatory Care Group system wasused to develop an index of population comorbidity [16]. Ambulatory diagnosticgroups (ADGs) represent 32 comorbidity clusters of every ICD-9-CM diag-nostic code. The number of ADGs was categorized as none (reference category),1–2, 3–5 and 6 or more.

Crude (unadjusted) fracture rates were calculated per thousand person-years.Analysis of crude fracture rates could be confounded by the fact that diabetesprevalence, diabetes duration and multiple ADGs increase in older individuals.Therefore, multivariable models were constructed to assess the independenteffects of these variables as well as their possible interactions. Poisson regres-sion was used to model counts of the number of fractures in population strata asa function of the selected demographic, geographic, socioeconomic, and chronicdisease variables. The number of person-years at risk in each stratum was theoffset variable in the Poisson models. The base regression model included eth-nicity, age (10-year groups unless otherwise specified), gender, area of resi-dence, income quintile and the interactions of age*gender and residentialarea*income quintile. Matching variables were included in the models sincethey produced a statistically significant improvement in model fit. Full reg-

ression models included all of the variables in the base model in addition tochronic disease variables for diabetes diagnosis (Model 1), number of ADGs(Model 2), and their combination (Model 3). Diabetes diagnosis, number ofADGs, and age were included as main effects and two-way interactions [17].

Rate ratios (RR) and 95% confidence intervals (CI) are reported. Thedifference between the base and full models was evaluated using a likelihoodratio test [18]. To achieve satisfactory model fit, ages below 60 years werecombined. It was also necessary to combine income quintiles 1–2 (lowestincome quintiles) and 3–5 (highest income quintiles; reference category). Allregression analyses were performed using the GENMOD procedure of SASRelease 8.02 (SAS Institute Inc., Cary, NC).

Results

The composition and demographics of the cohorts are sum-marized in Table 1. The non-diabetic and diabetic groups werewell-matched in terms of age, gender, area of residence andincome level. Since duration of diabetes correlates with age,those with long-term diabetes tended to be slightly older thanshort-term or newly diabetic subjects. The distribution in ADGcategory was consistent with higher comorbidity in the diabeticcohort (p<0.0001) with a significant difference in those with 6or more ADGs (22.7% vs. 15.2%, p<0.0001).

The diabetic cohort provided 715,148 person-years of follow-up with 2,071,417 person-years of follow-up in the non-diabeticcontrols (Table 1). There were sufficient numbers of osteoporo-tic fractures (combined hip, wrist and spine n=22,148 [16,457in those without diabetes and 5691 in those with diabetes]) andisolated hip fractures (n=7125 [5224 in those without diabetesand 1901 in those with diabetes]) for analyses of main effectsand interactions.

The crude osteoporotic fracture rate in the non-diabetic con-trols was 8.2 per 1000 person-years (95% confidence interval[CI] 8.0–8.3) which was identical to the rate for those withdiabetes (8.2, 95% CI 7.9–8.4, p>0.2). Osteoporotic fracturerates increased with age in both cohorts (Table 2). Osteoporoticfracture rates were significantly greater in the diabetes cohortthan among non-diabetic controls prior to age 60 years, with anexcess in osteoporotic fractures after age 80 years in controlswithout diabetes compared to subjects with diabetes(p=0.0019). Overall hip fractures rates showed a borderlinedifference (diabetes 2.5 per 1000 person-years [95% CI 2.4–2.6] vs. controls 2.7 [2.5–2.8], p=0.06). Hip fracture rates weresignificantly greater in diabetic subjects than controls prior toage 80 years, but roughly equal in those age 80 years and older.Within the diabetic cohort there was a general pattern of higherage-specific fracture rates related to longer disease duration(Fig. 1) and this was statistically significant (osteoporotic frac-tures p-for-trend=0.026, hip fractures p-for-trend=0.025).

Crude fracture rates showed a strong relationship withcomorbidity as represented by the number of ADGs. Individualsin the lowest ADG category experienced 4.9 osteoporotic frac-tures per 1000 person-years (95% CI 4.7–5.1) as compared with6.5 (95% CI 6.3–6.6), 9.0 (8.8–9.2) and 13.5 (13.2–13.9) withprogressively higher ADG categories. Hip fractures showed thesame association, increasing from 1.3 per 1000 person-years(95% CI 1.2–1.4) in the lowest ADG category to 4.7 (95% CI4.5–4.9) in the highest ADG category. These patterns

Table 1Cohort demographics and number of fracture endpoints observed (percentages in parentheses)

ControlsN=236,682

Diabetes — allN=82,094

Diabetes — newN=42,874

Diabetes — short termN=16,081

Diabetes — long termN=23,139

Age (years), mean±SD 58.0±16.1 57.9±15.4 54.0±15.3 60.2±15.6 63.4±15.2Men 118,623 (50.1) 40,909 (49.8) 22,013 (51.3) 7813 (48.6) 11,083 (47.9)Aboriginal status 17,038 (7.2) 8776 (10.7) 4865 (11.3) 1702 (10.6) 2209 (9.5)Residence:

Rural north 13,769 (5.8) 5753 (7.0) 3230 (7.5) 1095 (6.8) 1428 (6.2)Rural south 91,753 (38.8) 31,949 (38.9) 16,665 (38.9) 6169 (38.4) 9115 (39.4)Urban 131,160 (55.4) 44,392 (54.1) 22,979 (53.6) 8817 (54.8) 12,596 (54.4)

Income quintile:Quintile 1 (lowest) 49,941 (21.1) 20,832 (25.4) 10,406 (24.3) 4201 (26.1) 6225 (26.9)Quintile 2 51,510 (21.8) 18,036 (22.0) 9316 (21.7) 3456 (21.5) 5264 (22.8)Quintile 3 49,814 (21.0) 16,762 (20.4) 8766 (20.5) 3374 (21.0) 4622 (20.0)Quintile 4 42,171 (17.8) 13,502 (16.4) 7452 (17.4) 2479 (15.4) 3571 (15.4)Quintile 5 (highest) 39,462 (16.7) 11,703 (14.3) 6601 (15.4) 2247 (14) 2855 (12.3)Not available 3784 (1.6) 1259 (1.5) 333 (0.8) 324 (2.0) 602 (2.6)

Number of ambulatory diagnostic groups (ADGs):None 39,733 (16.8) 7488 (9.1) 5893 (13.7) 584 (3.6) 1011 (4.4)1–2 81,308 (34.4) 24,787 (30.2) 14,224 (33.2) 4525 (28.1) 6038 (26.1)3–5 79,772 (33.7) 31,169 (38.0) 15,316 (35.7) 6558 (40.8) 9295 (40.2)6 or more 35,869 (15.2) 18,650 (22.7) 7441 (17.4) 4414 (27.5) 6795 (29.4)Follow-up (person-years) 2,071,417 715,148 410,703 133,140 171,304

Numbers of fractures:Hip 4570 (1.9) 1686 (2.1) 551 (1.3) 379 (2.4) 756 (3.3)Spine 3519 (1.5) 1287 (1.6) 599 (1.4) 242 (1.5) 446 (1.9)Wrist 8368 (3.5) 2718 (3.3) 1431 (3.3) 535 (3.3) 752 (3.2)

1597W.D. Leslie et al. / Bone 40 (2007) 1595–1601

wereevident within every age stratum (Fig. 2) and wereconfirmed in the linear trend analysis (osteoporotic and hipfractures p-for-trend<0.0001).

The regression analyses confirmed the importance of dia-betes duration on the occurrence of osteoporotic and hip frac-tures (Table 3). As compared with the non-diabetic referencegroup (Model 1), newly diagnosed diabetes showed asignificantly reduced risk of osteoporotic fractures (rate ratio[RR] 0.91 [95% CI, 0.86–0.95]) and hip fractures (RR 0.83[0.75–0.92]). Long-term diabetes showed an increase in bothosteoporotic fractures (RR 1.15 [CI, 1.09–1.22]) and hipfractures (RR 1.40 [1.28–1.53]). Short-term diabetes wasneutral in terms of fracture of fracture risk. The test for lineartrend across these three diabetes categories was significant forosteoporotic and hip fractures (p-for-trend<0.0001). Comor-bidity, as measured by the number of ADGs (Model 2), showed

Table 2Crude rates per 1000 person-years for osteoporotic fractures (spine, hip or wrist) anintervals are given in parentheses)

Age Osteoporotic fractures

Controls N=236,682 Diabetes N=82,094 p

<50 years 3.4 (3.3–3.5) 4.1 (3.9–4.4) <050–59 years 4.7 (4.5–4.9) 5.3 (4.9–5.6) 060–69 years 7.8 (7.5–8.0) 8.0 (7.6–8.5) >070–79 years 15.3 (14.9–15.8) 15.1 (14.4–15.9) >080 years and over 33.1 (32.0–34.2) 29.4 (27.6–31.3) 0Overall 8.2 (8.0–8.3) 8.2 (7.9–8.4) >0

a progressive increase in osteoporotic and hip risk with morediagnoses (p-for-trend<0.0001). No statistically significantinteraction was identified between diabetes category andcomorbidity (likelihood ratio test p>0.2). When both variableswere included in the regression (Model 3), there was noappreciable weakening in their contribution to fracture predic-tion when assessed separately.

Age showed a statistically significant interaction withdiabetes category and number of ADGs. Models that includedterms for age*diabetes and age*ADG showed a correspondingimprovement in model fit (likelihood ratio test p<0.0001).There was a gradient of increasing osteoporotic fracture riskwith longer duration of diabetes across the three diabetescategories which was statistically significant (p-for-trend<0.05) for all age groups except individuals age 80 and older(p-for-trend>0.2). Fig. 3 demonstrates the age interaction with

d hip fractures in non-diabetics controls and diabetic subjects (95% confidence

Hip fractures

Controls N=236,682 Diabetes N=82,094 p

.00001 0.1 (0.1–0.1) 0.3 (0.2–0.4) <0.00001

.01462 0.4 (0.4–0.5) 0.7 (0.6–0.9) 0.00003

.2 1.6 (1.5–1.7) 2.1 (1.9–2.3) 0.00005

.2 6.2 (6.0–6.5) 6.8 (6.3–7.3) 0.04595

.00189 19.2 (18.4–20.1) 17.9 (16.6–19.4) 0.12445

.2 2.5 (2.4–2.6) 2.7 (2.5–2.8) 0.05764

Fig. 1. Age-specific crude fracture rates (per 1000 person-years) in the diabeticcohort according to duration of diabetes. (A) Osteoporotic fractures (combinedspine, hip and wrist). (B) Hip fractures only. 95% Confidence intervals areshown.

Fig. 2. Age-specific crude fracture rates (per 1000 person-years) according tolevel of comorbidity (number of ambulatory diagnostic groups [ADGs]). (A)Osteoporotic fractures (combined spine, hip and wrist). (B) Hip fractures only.95% Confidence intervals are shown.

1598 W.D. Leslie et al. / Bone 40 (2007) 1595–1601

long-term diabetes less than age 50 showing increased fracturerisk (relative to controls) while newly diagnosed and short-termdiabetes showed fracture rates equal to controls. In contrast,newly diagnosed diabetes age 80 and older showed reducedfracture risk (relative to controls) while long-term term diabeteshad fracture rates equal to controls. A similar pattern wasevident for hip fractures (due to the small number of fracturessome age strata were combined). The test for linear trend acrossthe three diabetes categories was statistically significant forthose less than 60 years and 60–79 years age groups (p-for-trend<0.0001), and borderline in those age 80 and older (p-for-trend=0.08). The number of ADGs showed a consistentgradient of increasing osteoporotic and hip fracture risk (p-for-trend<0.0001 for all age groups).

Discussion

We found that longer diabetes duration correlated withhigher osteoporotic and hip fracture risk, consistent with thehypothesis that diabetes mediates its effect fractures through itsassociated complications. Newly diagnosed diabetic subjectswere actually at slightly reduced fracture risk. This is the first

time that a biphasic effect of diabetes on fracture rates has beenobserved. An index of comorbidity was also associated withincreased fracture risk, but this was independent of the effect ofdiabetes.

The effect of age on osteoporotic fracture rates is complex,since younger diabetics appeared have higher rates than controlswhereas older diabetics had reduced rates of fractures comparedwith controls. These opposing effects cancelled each other outso that the overall rates were the same. Although some studieshave found higher fracture rates in diabetic patients [7–9], apopulation-based study comparing 986 community diabetics(median age at diagnosis 61 years) with matched non-diabeticcontrols found no differences in fracture rates [19]. Theexplanation the age interaction that we observed is unclear.Shorter life expectancy in older diabetics could be one possibleexplanation, but Cox proportional models of time to firstfracture (which adjust for differential survival) gave similarresults to the Poisson regression models (data not shown).

It has been projected that the number of people with diabetesworldwide will double between 2000 and 2030 even if levels ofobesity remain constant [5]. Although most of this increase willbe in older adults, it is important to note that children and

Table 3Fracture rate ratios from Poisson regression models (95% confidence intervalsare given in parentheses)

Variables in model a Hip fractures Hip fractures

Model 1: diabetesDiabetes diagnosis

Controls Reference ReferenceDiabetes — new 0.91

(0.86–0.95)0.83(0.75–0.92)

Diabetes — short term (<5 years) 1.00(0.93–1.07)

1.13(1.00–1.28)

Diabetes — long term (>5 years) 1.15(1.09–1.22)

1.40(1.28–1.53)

Linear trend for diabetes categories p<0.0001 p<0.0001

Model 2: comorbidityNumber of ambulatory diagnostic groups (ADGs):

None Reference Reference1–2 1.13

(1.05–1.21)1.07(0.95–1.20)

3–5 1.33(1.24–1.42)

1.20(1.07–1.34)

6 or more 1.72(1.60–1.85)

1.53(1.36–1.71)

Linear trend for ADG categories p<0.0001 p<0.0001

Model 3: diabetes and comorbidityDiabetes diagnosis:

Controls Reference ReferenceDiabetes — new 0.89

(0.85–0.94)0.82(0.74–0.91)

Diabetes — short term (<5 years) 0.94(0.88–1.01)

1.10(0.97–1.24)

Diabetes — long term (>5 years) 1.09(1.05–1.15)

1.36(1.24–1.49)

Linear trend for diabetes categories p=0.0001 p<0.0001Number of ambulatory diagnosticgroups (ADGs):None Reference Reference1–2 1.13

(1.06–1.20)1.05(0.93–1.18)

3–5 1.32(1.24–1.40)

1.16(1.04–1.30)

6 or more 1.67(1.57–1.78)

1.48(1.32–1.66)

Linear trend for ADG categories p<0.0001 p<0.0001

a Adjusted for age, sex, income quintile, area of residence and ethnicity.

1599W.D. Leslie et al. / Bone 40 (2007) 1595–1601

adolescents are not spared [20,21]. Since young persons withdiabetes will have a lifetime during which to develop diabeticcomplications, they may be particularly vulnerable to theadverse skeletal effects. Therefore, our finding that youngerpersons with diabetes have the highest risk ratios (up to fourfoldfor hip fractures in long-term diabetes) may have importantclinical implications.

Our study provides a population-based perspective of howdiabetes of all types and durations affects fracture risk, but islimited by our inability to differentiate type 1 from type 2diabetes. The pathogenetic mechanism underlying osteoporosismay differ according to diabetes type [22], though the largeststudy to date found similar relative risk for fracture for type 1 andtype 2 diabetes. In a large case-control study fromDenmark, bothtypes of diabetes were associated with similarly increased risk

for fracture (OR 1.3, 95% CI 1.2–1.5 for type 1; OR 1.2, 95% CI1.1–1.3 for type 2) and hip fractures (OR 1.7, 95%CI 1.3–22 vs.OR 1.4, 95% CI 1.2–1.6) [23]. The population-based cohortfrom Tromso, Norway, noted increased non-vertebral and hipfracture risk regardless of diabetes type but no effect of diabetesduration [10]. Other studies report more hip fractures in type 1diabetes (RR 6.9, 95% CI 2.2–21.6) than in type 2 diabetes (RR1.8, 95% CI 1.1–2.9) [24]. In this latter study, type 2 diabetesshowed no increased risk of hip fractures if the duration ofdisease was less than 5 years. Relative risk for hip fracture in type1 diabetic men and women almost eightfold and 10-fold higherthan matched controls have been reported [25]. It is possible thatthe declining fraction of type 1 diabetes in older populationscontributes to the age interaction observed in our study.

The Rotterdam study assessed BMD and fractures in 792type 2 diabetes in relation to glucose tolerance tests [26,27].Diabetic subjects had higher bone mineral density with in-creased non-vertebral fracture risk (hazard ratio [HR] 1.33, 95%CI 1.00–1.77) after adjustment for age, gender, BMI, selectedclinical risk factors, and femoral bone density. The Health,Aging and Bone Composition (ABC) Study included 2979 menand women of mixed ethnicity age 70–79 years of age. Type 2diabetes was present in 566 (19%) and predicted higher bonedensity independent of body composition and fasting insulin[28]. Diabetic subjects had similar unadjusted fracture rates asnon-diabetic subjects (11.7 per 1000 person-years vs. 12.4 per1000 person-years). No significant effect of diabetes on fracturerisk was seen in regression models that adjusted for demo-graphic variables only (RR for diabetes 1.23, 95% CI 0.82–1.86) but those with diabetes were at higher fracture risk afteradjustment for higher femoral bone density (RR 1.71, 95% CI1.11–2.61) [9]. This study recently reported that thiazolidine-diones, which activate the peroxisome proliferator-activatedreceptor (PPAR)-gamma, may cause bone loss in older womenand suggests another potential mechanism contributing to higherfracture risk [29].

One previous study has examined women in whom diabetesdeveloped during follow-up [7], comparable to our newlydiagnosed diabetes category. Hip fracture risk (RR 1.60, 95% CI1.14–2.25) was increased which is contrary to our finding.Diabetes status, risk factors and hip fracture were ascertained bymailed questionnaire and it is not possible to exclude respon-dent bias.

The effect of overweight/obesity and fracture risk deservescomment. Insulin resistance is directly correlated with higherbone density in men and women, possibly mediated through theeffects of body weight rather than hyperinsulinemia [30], thoughthe mechanism through which weight affects bone turnover,bone density and fractures is not well defined and may involvecurrently unknown hormones or growth factors [31]. BMI25 kg/m2 confers a twofold reduction in hip fracture comparedto a BMI 20 kg/m2 with a 17% additional reduction in hipfracture risk with a BMI 30 kg/m2 [6]. Greater weight is a riskfactor for type 2 diabetes and could contribute to the apparentprotection from osteoporotic and hip fractures that we observedin newly diagnosed diabetes. Unfortunately, weight cannot beassessed in our data sources.

Fig. 3. Interaction between age and (A) diabetes category or (B) comorbidity (number of ambulatory diagnostic groups [ADGs]). Rate ratios are from Poissonregression models that included diabetes category (reference category: non-diabetic controls), number of ADGs (reference category: none), age⁎diabetes category andage⁎number of ADGs. 95% Confidence intervals are shown.

1600 W.D. Leslie et al. / Bone 40 (2007) 1595–1601

Our study of diabetes and fractures is the largest to datethanks to a comprehensive administrative health data repository.Fracture ascertainment from administrative health data is notwithout limitations, however. Fractures that produce few symp-toms may not lead to a physician interaction. We are also unableto determine whether increased fracture risk relates to reducedbone density, increased risk for falls, or other unidentified fac-tors. Risk estimates were not adjusted for weight, insulin ormedication use, diabetic control, or other clinical risk factors thataffect fractures [32]. Our diabetes definition has been validatedin terms of prevalence but is a relatively crude measure of di-sease duration [4]. The inclusion of individuals first diagnosedwith diabetes after inception in the diabetic cohort could bequestioned. Type 2 diabetes is characterized by a longasymptomatic period. Subjects in whom diabetes was diagnosedafter cohort inception probably had glucose dysregulation andrisk factors predisposing to diabetes for some time prior to diag-nosis. Therefore, it is more appropriate to include them with thediabetic group than with the non-diabetic controls.

In summary, long-term diabetes is associated with increasedfracture risk whereas newly diagnosed diabetes shows a reduc-tion in fracture risk, though this relationship is modified by aninteraction with age. Our findings are consistent with the viewthat disease-related complications are primarily responsible forthe excess fracture risk observed in diabetes. It is hypothesized

that the opposing effect of overweight/obesity in newly diag-nosed diabetes contributes to the observed biphasic fracture risk,though causality cannot be proven from this observational study.

Acknowledgments

Supported by a research grant from the Canadian Institutesfor Health Research. The authors are indebted to ManitobaHealth for providing the data used in this study, to the FirstNations and Inuit Health Branch and Indian and Northern Af-fairs Canada for permission to use the Status Verification Sys-tem, and to the Health Information Research Committee of theAssembly of Manitoba Chiefs for actively supporting this work.The results and conclusions are those of the authors, and noofficial endorsement by Manitoba Health is intended or shouldbe inferred. Sources of support: A research grant from the Ca-nadian Institutes for Health Research.

References

[1] Kanis JA, Johnell O, Oden A, Sembo I, Redlund-Johnell I, Dawson A, et al.Long-term risk of osteoporotic fracture in Malmo. Osteoporos Int 2000;11:669–74.

[2] Melton III LJ, Chrischilles EA, Cooper C, Lane AW, Riggs BL. Pers-pective. How many women have osteoporosis? J Bone Miner Res 1992;7:1005–10.

1601W.D. Leslie et al. / Bone 40 (2007) 1595–1601

[3] van Staa TP, Dennison EM, Leufkens HG. Epidemiology of fractures inEngland and Wales. Bone 2001;29:517–22.

[4] Blanchard JF, Ludwig S, Wajda A, Dean H, Anderson K, Kendall O.Incidence and prevalence of diabetes in Manitoba, 1986–1991. DiabetesCare 1996;19:807–11.

[5] Wild S, Roglic G, Green A, Sicree R, King H. Global prevalence ofdiabetes: estimates for the year 2000 and projections for 2030. DiabetesCare 2004;27:1047–53.

[6] De Laet C, Kanis JA, Oden A, Johanson H, Johnell O, Delmas P, et al. Bodymass index as a predictor of fracture risk: a meta-analysis. Osteoporos Int2005;16:1330–8.

[7] Nicodemus KK. Type 1 and type 2 diabetes and incident hip fractures inpostmenopausal women. Diabetes Care 2001;24:1192–7.

[8] Schwartz AV, Sellmeyer DE, Ensrud KE, Cauley JA, Tabor HK, SchreinerPJ, et al. Older women with diabetes have an increased risk of fracture: aprospective study. J Clin Endocrinol Metab 2001;86:32–8.

[9] Strotmeyer ES, Cauley JA, Schwartz AV, Nevitt MC, Resnick HE, BauerDC, et al. Nontraumatic fracture risk with diabetes mellitus and impairedfasting glucose in older white and black adults: the health, aging, and bodycomposition study. Arch Intern Med 2005;165:1612–7.

[10] AhmedLA, Joakimsen RM, Berntsen GK, FonneboV. Diabetesmellitus andthe risk of non-vertebral fractures: the Tromso study. Osteoporos Int 2005:1–6.

[11] Roos NP, Shapiro E. Revisiting the Manitoba Centre for Health Policy andEvaluation and its population-based health information system. Med Care1999;37:JS10–4.

[12] Forsen L, Meyer HE, Midthjell K, Edna TH. Diabetes mellitus and theincidence of hip fracture: results from the Nord–Trondelag Health Survey.Diabetologia 1999;42:920–5.

[13] Leslie WD, Derksen S, Metge C, Lix LM, Salamon EA, Wood SP, et al.Fracture risk among First Nations people: a retrospective matched cohortstudy. Can Med Assoc J 2004;171:869–73.

[14] MacMillan HL, MacMillan AB, Offord DR, Dingle JL. Aboriginal health.Can Med Assoc J 1996;155:1569–78.

[15] Income quintiles based on the 1996 census. URL: http://www.umanitoba.ca/academic/centres/mchp/concept/diet/income/income_quintile.html(Last accessed June 6, 2005).

[16] Smith NS, Weiner JP. Applying population-based case mix adjustment inmanaged care: the Johns Hopkins Ambulatory Care Group system. ManagCare Q 1994;2:21–34.

[17] McCulloch CE, Searle SR. Generalized, linear, and mixed models. NewYork: John Wiley and Sons; 2001.

[18] Fox J. Applied regression analysis, linear models, and related methods.CA: Sage, Thousand Oaks; 1997.

[19] Heath III H, Melton III LJ, Chu CP. Diabetes mellitus and risk of skeletalfracture. N Engl J Med 1980;303:567–70.

[20] Singh R, Shaw J, Zimmet P. Epidemiology of childhood type 2 diabetes inthe developing world. Pediatr Diabetes 2004;5:154–68.

[21] Wiegand S, Maikowski U, Blankenstein O, Biebermann H, Tarnow P,Gruters A. Type 2 diabetes and impaired glucose tolerance in Europeanchildren and adolescents with obesity—a problem that is no longerrestricted to minority groups. Eur J Endocrinol 2004;151:199–206.

[22] Inzerillo AM, Epstein S. Osteoporosis and diabetes mellitus. Rev EndocrMetab Disord 2004;5:261–8.

[23] Vestergaard P, Rejnmark L. Relative fracture risk in patients with diabetesmellitus, and the impact of insulin and oral antidiabetic medication onrelative fracture risk. Diabetologia 2005;48:1292–9.

[24] Forsen L, Meyer HE, Midthjell K, Edna TH. Diabetes mellitus and theincidence of hip fracture: results from the Nord–Trondelag Health Survey.Diabetologia 1999;42:920–5.

[25] Miao J, Brismar K, Nyren O, Ugarph-Morawski A, Ye W. Elevated hipfracture risk in type 1 diabetic patients: a population-based cohort study inSweden. Diabetes Care 2005;28:2850–5.

[26] van Daele PL, Stolk RP, Burger H, Algra D, Grobbee DE, Hofman A, et al.Bone density in non-insulin-dependent diabetes mellitus. The RotterdamStudy. Ann Intern Med 1995;122:409–14.

[27] Stolk RP, van Daele PL, Pols HA, Burger H, Hofman A, Birkenhager JC,et al. Hyperinsulinemia and bone mineral density in an elderly population:the Rotterdam Study. Bone 1996;18:545–9.

[28] Strotmeyer ES, Cauley JA, Schwartz AV, Nevitt MC, Resnick HE, ZmudaJM, et al. Diabetes is associated independently of body composition withBMDand bone volume in older white and blackmen andwomen: theHealth,Aging, and Body Composition Study. J Bone Miner Res 2004;19:1084–91.

[29] Schwartz AV, Sellmeyer DE, Vittinghoff E, Palermo L, Lecka-Czernik B,Feingold KR, et al. Thiazolidinedione use and bone loss in older diabeticadults. J Clin Endocrinol Metab 2006;9:3349–54.

[30] Dennison EM, Syddall HE, Aihie SA, Craighead S, Phillips DI. Type 2diabetes mellitus is associated with increased axial bone density in menand women from the Hertfordshire Cohort Study: evidence for an indirecteffect of insulin resistance? Diabetologia 2004;47:1963–8.

[31] Reid IR. Relationships among body mass, its components, and bone. Bone2002;31:547–55.

[32] Kanis JA, Borgstrom F, De Laet C, Johansson H, Johnell O, Jonsson B,et al. Assessment of fracture risk. Osteoporos Int 2005;16:581–9.