Public Notices

Biphasic fracture risk in diabetes: A population-based study

Description
Biphasic fracture risk in diabetes: A population-based study
Categories
Published
of 7
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
Share
Transcript
  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 2A6   b  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 of ADGs (  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 at approximately 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 health  problems become more prevalent with age, osteoporosis fre-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 risk factor for type 2 diabetes whereas underweight is a risk factor for 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 Bone 40 (2007) 1595 – 1601www.elsevier.com/locate/bone ⁎  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  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 for linkage across datasets and creation of person-specific longitudinal records of health service utilization. The study was approved by the Research Ethics Boardfor the University of Manitoba and Health Information Privacy Committee of Manitoba Health.The study cohort included all diabetic adults (aged 20 years or older as of December 31, 1993) who were continuouslyresidents 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 diabetes prior to study inception (January 1, 1994) was based upon a validated definition:two physicianoffice visits or a singlehospitalization with a diagnosis of diabetes(ICD-9-CM code 250) in a 3-year period [4]. Records for preceding years were used 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 meet this 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 the province. First Nations Absrcinal ethnicity (also referred to as Natives or   “ In-dians ” ) identified from government files was an additional matching variablesince Canadian Absrcinals have higher rates of fracture and diabetes than non-Absrcinals [13,14]. No match could be found for 708 subjects and these were excluded 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 the province) were identified. Vertebral fractures (ICD-9-CM code 805), wrist fractures (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 be accompanied by a physician claim for a site-specific fracture reduction or fixation. 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 wasusedtodevelopan indexofpopulationcomorbidity[16].Ambulatorydiagnostic groups (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 diabetes prevalence, diabetes duration and multiple ADGs increase in older individuals.Therefore, multivariable models were constructed to assess the independent effects 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 of ADGs, 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 were combined. It was also necessary to combine income quintiles 1 – 2 (lowest income 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).Thediabeticcohortprovided715,148person-yearsoffollow-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 cohort than 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 higher age-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) with progressively 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 1596  W.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 linear trend across these three diabetes categories was significant for osteoporotic and hip fractures (  p -for-trend < 0.0001). Comor- bidity, as measured by the number of ADGs (Model 2), showeda progressive increase in osteoporotic and hip risk with morediagnoses (  p -for-trend < 0.0001). No statistically significant interaction 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 risk with 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 Table 1Cohort demographics and number of fracture endpoints observed (percentages in parentheses)Controls  N  =236,682Diabetes  —  all  N  =82,094Diabetes  —  new  N  =42,874Diabetes  —  short term  N  =16,081Diabetes  —  long term  N  =23,139Age (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)Table 2Crude rates per 1000 person-years for osteoporotic fractures (spine, hip or wrist) and hip fractures in non-diabetics controls and diabetic subjects (95% confidenceintervals are given in parentheses)Age Osteoporotic fractures Hip fracturesControls  N  =236,682 Diabetes  N  =82,094  p  Controls  N  =236,682 Diabetes  N  =82,094  p < 50 years 3.4 (3.3 – 3.5) 4.1 (3.9 – 4.4)  < 0.00001 0.1 (0.1 – 0.1) 0.3 (0.2 – 0.4)  < 0.0000150 – 59 years 4.7 (4.5 – 4.9) 5.3 (4.9 – 5.6) 0.01462 0.4 (0.4 – 0.5) 0.7 (0.6 – 0.9) 0.0000360 – 69 years 7.8 (7.5 – 8.0) 8.0 (7.6 – 8.5)  > 0.2 1.6 (1.5 – 1.7) 2.1 (1.9 – 2.3) 0.0000570 – 79 years 15.3 (14.9 – 15.8) 15.1 (14.4 – 15.9)  > 0.2 6.2 (6.0 – 6.5) 6.8 (6.3 – 7.3) 0.0459580 years and over 33.1 (32.0 – 34.2) 29.4 (27.6 – 31.3) 0.00189 19.2 (18.4 – 20.1) 17.9 (16.6 – 19.4) 0.12445Overall 8.2 (8.0 – 8.3) 8.2 (7.9 – 8.4)  > 0.2 2.5 (2.4 – 2.6) 2.7 (2.5 – 2.8) 0.057641597 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 for those 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 consistent gradient 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 of diabetes.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 out so that the overall rates were the same. Although some studieshave found higher fracture rates in diabetic patients [7 – 9], a population-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 first fracture (which adjust for differential survival) gave similar results 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 of obesity remain constant  [5]. Although most of this increase will be in older adults, it is important to note that children and 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  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 younger  persons with diabetes have the highest risk ratios (up to fourfoldfor hip fractures in long-term diabetes) may have important clinical 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 largest studytodatefoundsimilarrelativeriskforfracturefortype1andtype2diabetes.Inalargecase-controlstudyfromDenmark,bothtypes of diabetes were associated with similarly increased risk forfracture(OR 1.3,95%CI 1.2 – 1.5 fortype1;OR 1.2,95%CI1.1 – 1.3fortype2)andhipfractures(OR1.7,95%CI1.3 – 22vs.OR 1.4, 95% CI 1.2 – 1.6) [23]. The population-based cohort from 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 (RR 1.8, 95% CI 1.1 – 2.9) [24]. In this latter study, type 2 diabetesshowed no increased risk of hip fractures if the duration of diseasewaslessthan5years.Relativeriskforhipfractureintype1 diabetic men and women almost eightfold and 10-fold higher than matched controls have been reported [25]. It is possible that the 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 per 1000 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 after adjustment 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 womenandsuggestsanotherpotentialmechanismcontributingtohigher fracture 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 higher  bone density in men and women, possibly mediated through theeffectsofbodyweightratherthanhyperinsulinemia[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/m 2 confers a twofold reduction in hip fracture comparedto a BMI 20 kg/m 2 with a 17% additional reduction in hipfracture risk with a BMI 30 kg/m 2 [6]. Greater weight is a risk factor for type 2 diabetes and could contribute to the apparent  protection from osteoporotic and hip fractures that we observedin newly diagnosed diabetes. Unfortunately, weight cannot beassessed in our data sources. Table 3Fracture rate ratios from Poisson regression models (95% confidence intervalsare given in parentheses)Variables in model a  Hip fractures Hip fractures  Model 1: diabetes Diabetes diagnosisControls 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: comorbidity  Number 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 comorbidity Diabetes 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 < 0.0001 Number 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.1599 W.D. Leslie et al. / Bone 40 (2007) 1595  –  1601
Search
Similar documents
View more...
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks
SAVE OUR EARTH

We need your sign to support Project to invent "SMART AND CONTROLLABLE REFLECTIVE BALLOONS" to cover the Sun and Save Our Earth.

More details...

Sign Now!

We are very appreciated for your Prompt Action!

x