High-Sensitivity C-Reactive Protein and Risk of Nontraumatic Fractures in the Bruneck Study FREE

C-reactive protein (CRP) is a 23-kDa protein named according to its ability to bind proteoglycan C of pneumococci.¹ It is a member of the pentraxin family and part of a phylogenetically old defense system against microbes. C-reactive protein is predominantly produced by the liver and increases up to 100-fold during inflammatory stimuli. Testing for CRP is widely used to detect immediate-phase responses in the context of neoplasms, tissue injuries, and infectious and autoimmune diseases. In the late 1990s, high-sensitivity (hs) CRP assays became available, allowing assessment of serum concentrations at the lower end of its distribution. In the subsequent years, hs-CRP levels far beneath the cutoff value, indicative of an immediate-phase response, predicted the risk of cardiovascular disease, diabetes mellitus, and cognitive impairment.^2- 5 The term low-grade systemic inflammation was coined and categorizations of low, average, and high vascular risk proposed.⁶

Chronic inflammatory diseases, such as rheumatoid arthritis, ankylosing spondylitis, systemic lupus erythematosus, and Crohn disease, are major triggers for bone loss.^7- 8 The detrimental effects are mediated through cytokines that alter the balance between bone formation and bone resorption.⁹ Despite the compelling evidence that prominent inflammation in the course of the previously mentioned diseases leads to an increased fracture risk, it is unknown whether similar effects extend to low-grade systemic inflammation of healthy individuals or whether there is a threshold of inflammatory activity required for bone damage to emerge. Our prospective population-based study was designed to investigate the potential association between hs-CRP level and risk of nontraumatic fractures in the general community.

The Bruneck Study is a prospective population-based survey of the epidemiology and pathogenesis of atherosclerosis and disorders of the brain and bone.^10- 13 The study protocol was reviewed and approved by the appropriate ethics committees. All study subjects gave their written informed consent. At the 1990 baseline, the study population was recruited as an age- and sex-stratified random sample of all inhabitants of Bruneck (125 women and 125 men in each of the fifth to eighth decades of age). A total of 93.6% participated, with data assessment completed in 919 subjects. Thirteen subjects with nontraumatic fractures before baseline were excluded, leaving 906 men and women for the current analysis. Follow-up examinations were performed every 5 years (1995, 2000, and 2005),¹⁰ and information on fractures was complete in more than 98% of study subjects.

Lifetime fractures were carefully recorded for all study participants, thereby using 3 sources of information: the subject's self-report, a review of medical records from the Bruneck Hospital and general practitioners, and a standardized evaluation of all radiographs ever taken of study subjects. The situation in Bruneck is unique in that (1) the only x-ray filming facility in the whole district is located at the hospital and all radiographs ever taken of study subjects were available for review, (2) radiographs were obtained in virtually all cases of injury and in moderate-to-severe or long-lasting back pain, and (3) population mobility in the survey area was low during the past 15 years. For all radiologically confirmed fractures, localization, date of occurrence, and associated circumstances were recorded. Fractures were classified as nontraumatic if resulting from a fall from standing height or less or manifesting without any trauma.¹⁴ Vertebral fractures were diagnosed according to established criteria.¹⁵

Lifestyle variables were assessed in 1990, 1995, 2000, and 2005. Body mass index was calculated as weight in kilograms divided by the square of height in meters. Smoking status and alcohol consumption were recorded as detailed previously.¹⁰ The activity score was composed of the scores for work (3 categories) and sports/leisure activities (0, ≤2, or >2 h/wk).⁹ Socioeconomic status was defined with a 3-category scale (low, medium, and high) based on information about occupational status and educational level of the person with the highest annual income in the household.⁹ Diabetes mellitus was diagnosed according to American Diabetes Association criteria, with a cutoff level of fasting plasma glucose of 126 mg/dL or more (≥7.0 mmol/L).¹⁶ Autoimmune diseases were identified by an extensive 2-phase screening procedure detailed previously.¹³ A diagnosis of malignancy required confirmation by histological/cytological examinations. Dementia was coded present if prediagnosed by a neurologist and/or in case of a Mini-Mental State Examination score of less than 24. Cardiovascular disease subsumes fatal and nonfatal strokes (National Survey of Stroke criteria), fatal and nonfatal definite myocardial infarctions (World Health Organization definition), and definite peripheral artery disease.¹² Bone ultrasonographic data (broadband ultrasonic attenuation and speed of sound) were assessed in 2000 and 2005 at the heel bones using quantitative ultrasonographic equipment (SAHARA; Hologic, Inc, Bedford, Mass).

Blood samples were drawn after an overnight fast and 12 hours of abstinence from smoking, and were available from 919 subjects in 1990, 826 subjects in 1995, 700 subjects in 2000, and 572 subjects in 2005. Samples taken in 1990 and 1995 were analyzed in 2005 after 10 and 15 years of storage at −70°C (without any cycle of thawing-freezing), and 2000 and 2005 samples were immediately processed. The hs-CRP level was analyzed with an immunoturbimetric latex CRP assay (OLYMPUS OSR6199; Olympus Diagnostics, Melville, NY) with an IFCC/BCR/CAP CRM 470 CRP standard. Serum levels of osteocalcin, parathyroid hormone, and β-crosslaps were measured by electrochemiluminescence immunoassay (Roche Diagnostics, Mannheim, Germany), 25-hydroxyvitamin D by automated chemiluminescence assay (Nichols Advantage; Nichols Institute Diagnostics, San Juan Capistrano, Calif), and soluble receptor activator of nuclear factor κB ligand and osteoprotegerin by a commercial sandwich enzyme-linked immunosorbent assay (Biomedica, Vienna).^11- 12

Person-years of follow-up for each participant were accrued from the 1990 baseline until diagnosis of nontraumatic fracture, death, or October 1, 2005, whichever came first. Subjects experienced nontraumatic vertebral or hip fractures. Participants were divided into 3 approximately equal groups according to tertiles of hs-CRP. Relative risks were estimated with rate ratios comparing the incidence of nontraumatic fractures in each tertile with that of the lowest (referent) tertile. Risk factor–adjusted relative risks were estimated from pooled logistic regression analyses.¹⁷ This technique treated each observation period (1990-1995, 1995-2000, and 2000-2005) as a mini follow-up study in which updated risk factor measurements were used to predict fracture risk (ie, risk factor levels assessed in 1990 were used to predict the risk of fractures occurring in 1990-1995, risk factor levels assessed in 1995 were used to predict the risk of fractures occurring in 1995-2000, and risk factor levels assessed in 2000 were used to predict the risk of fractures occurring in 2000-2005). Updates were performed for all variables in the analyses and not only for hs-CRP level. In case of a missing hs-CRP level, follow-up measurement during the subsequent 5-year period was not considered in the analysis. Observations over all 3 periods were pooled into a single sample, as usual in pooled logistic regression models. The pooled logistic regression approach has been shown to be asymptotically equivalent to the Cox proportional hazards regression model, with time-dependent covariates given short intervals between reevaluations and low rates of events.¹⁷ Multivariate models were adjusted for age (in years), sex (men, premenopausal women, and postmenopausal women), follow-up period (1990-1995, 1995-2000, or 2000-2005), income (low, medium, or high), smoking (number of cigarettes smoked daily), alcohol consumption (measured in grams per day), physical activity (score), diabetes mellitus (no or yes), body mass index, creatinine levels (measured in milligrams per deciliter), sex hormone therapy (no or yes), and facultative variables of bone metabolism. We performed tests for linear trend by treating the medians in each category of CRP level as a continuous variable. All reported P values are 2-sided, and all analyses were prespecified except for the subgroup analyses in subjects with prominent inflammation arbitrarily defined by an hs-CRP cutoff of 7.5 mg/L.

Levels of hs-CRP in the Bruneck cohort were lower than those recently reported from a standard US population (Table 1).¹⁸ The Bruneck participants are quite lean and that might explain the relatively low CRP level. Geometric means of hs-CRP level in the 1990, 1995, 2000, and 2005 assessments were highly consistent when accounting for the continuous aging of study subjects and minor shifts in sex ratios (1.7, 1.8, 2.0, and 2.2 mg/L), which suggests considerable stability of CRP level in frozen samples.¹⁹ Individual levels of hs-CRP were stable during a 5-year period, as indicated by Spearman rank correlation coefficients of 0.65, 0.64, and 0.64 for assessments 1990 vs 1995, 1995 vs 2000, and 2000 vs 2005, respectively. The intraclass correlation calculated for all assessments and the entire 15-year follow-up was high, at 0.52. Classification of hs-CRP level into tertile groups showed a considerable agreement during the three 5-year intervals (61.2%, 58.0%, and 59.2% of subjects remained in the same tertile group; unweighted κ coefficients of 0.42, 0.37, and 0.39, respectively). Overall, long-term consistency was at least as high as that recorded in previous studies (113-473 subjects) with shorter sampling intervals.^2,20- 21

Levels of clinical and laboratory characteristics according to tertile groups for hs-CRP are depicted in Table 2. The hs-CRP level was significantly and inversely associated with variables of bone turnover, such as β-crosslaps and osteocalcin concentration (P<.001), whereas it emerged as unrelated to bone ultrasonographic data of the heel. These findings did not apply to the small subsegment of subjects with hs-CRP levels exceeding 7.5 mg/L in whom hs-CRP level showed a positive correlation with β-crosslaps (eg, 1990 baseline: r = 0.28, P=.03) and an inverse correlation with ultrasonographic measures of bone density (eg, 2005 follow-up examination: bone ultrasonographic data, r = −0.33 [P=.007]; and speed of sound, r = −0.32 [P=.009]).

Between 1990 and 2005, 69 men and women experienced nontraumatic hip and vertebral fractures. The incidence of nontraumatic fractures varied from 1.3 per 1000 person-years in the lowest tertile to 13.9 per 1000 person-years in the highest tertile group for hs-CRP (Table 3). In pooled logistic regression analysis (n = 2379) adjusted for age, sex/menopausal status, and period of follow-up, the 5-year risk of nontraumatic fracture continuously increased across CRP tertile groups (Table 4). In multivariate analysis, after controlling for a variety of demographic, lifestyle, and risk variables and variables of bone metabolism, results were similar (Table 4). In this model, apart from hs-CRP level, sex (odds ratio [OR], 2.1; 95% confidence interval [CI], 0.2-18.5), menopausal status (OR, 4.4; 95% CI, 2.1-9.1), and serum receptor activator of nuclear factor κB ligand (OR, 2.3 [95% CI, 1.0-5.2] and 4.3 [95% CI, 2.1-8.8] for subjects in the middle and highest, respectively, vs lowest tertile group for receptor activator of nuclear factor κB ligand) emerged as highly significant predictors of fracture risk (P<.001 each). Further weaker associations were obtained for age (OR, 1.4 [95% CI, 1.0-1.9] per 10-year increase; P=.06), alcohol consumption (OR, 1.5 [95% CI, 1.0-2.2] per 50-g/d increase; P=.03), and physical activity (OR, 0.8 [95% CI, 0.7-1.0] per 1-score point increase; P=.07), whereas all other variables were not statistically significant.

There was no evidence of a differential association between hs-CRP level and nontraumatic fracture in men (n = 19; incidence, 3.4 per 1000 person-years; multivariate OR, 13.2 [95% CI, 1.3-126.7]) for subjects in the highest vs lowest tertile group for hs-CRP level (P=.004 for trend) and in women (n = 50; incidence, 8.8 per 1000 person-years; multivariate OR, 9.1 [95% CI, 3.1-27.2]) for subjects in the highest vs lowest tertile group for hs-CRP level (P<.001 for trend) or in different age groups, even though absolute levels of fracture risk differed substantially in these groups. Moreover, associations were significant in all three 5-year intervals and in analyses separately focusing on nontraumatic hip fractures (n = 11; incidence, 1.0 per 1000 person-years; multivariate OR, 8.0 [95% CI, 1.0-69.6] for subjects in the highest vs lowest tertile group for hs-CRP level; P=.02 for trend) and nontraumatic vertebral fractures (n = 58; incidence, 5.2 per 1000 person-years; multivariate OR, 9.9 [95% CI, 3.4-28.7] for subjects in the highest vs lowest tertile group for hs-CRP level; P<.001 for trend). The exclusion of subjects with malignancies, autoimmune disease, cardiovascular disease, and dementia and of those with an hs-CRP level greater than 7.5 mg/L had little effects on the results obtained (Table 4). Results were also confirmed when log_e-transformed hs-CRP level was treated as a continuous variable (multivariate OR, 1.8 [95% CI, 1.4-2.3] per 1-U increase; P<.001) or hs-CRP categorizations other than the predefined tertile groups were applied (multivariate OR, 3.5 [95% CI, 1.3-9.2] and 7.5 [95% CI, 2.8-19.8] for subjects with an hs-CRP level of 1-3 and >3 vs <1 mg/L; P<.001) (Table 1); the multivariate OR values were 1.0 (95% CI, 0.3-4.3), 2.8 (95% CI, 0.8-9.0), 5.1 (95% CI, 1.7-15.6), and 7.1 (95% CI, 2.4-21.3) for subjects in the second to fifth vs first quintile group, respectively, for hs-CRP level (P<.001).

When restricting the analysis to the 2000 to 2005 period and allowing for ultrasonographic measures of bone density (2000), this variable showed the expected inverse relation to fracture risk (OR, 4.5 [95% CI, 1.7-12.2] and 2.5 [95% CI, 0.9-7.2] in the lowest and middle vs highest tertile group for bone ultrasonographic data; P=.002) and the association between fracture risk and hs-CRP tertile groups remained virtually unchanged.

In contrast to nontraumatic vertebral and hip fractures, other (traumatic) fractures (n = 312) were not related to hs-CRP level in multivariate pooled logistic regression analysis (OR, 1.0 [95% CI, 0.7-1.4] and 1.1 [95% CI, 0.8-1.6] in the middle and highest vs lowest tertile group for hs-CRP level; P=.65).

Chronic inflammatory diseases cause skeletal breakdown and enhance the risk of nontraumatic fractures.^{8- 9,22- 23} This prospective population-based study demonstrates that hs-CRP level is a strong risk predictor of nontraumatic fractures in the general community, and that its predictive significance extends to levels commonly regarded as low-grade inflammation. Subjects in the bottom and top tertile groups for hs-CRP level differed in their fracture risk by a factor of 7.8 (lower limit of the 95% CI, 3.1). This finding emerged as largely independent of age, sex, menopausal status, lifestyle characteristics, and laboratory variables of bone metabolism. The conjecture is similar to that observed in cardiovascular disease. There is solid evidence of an enhanced burden of vascular disease in patients with chronic inflammatory diseases, like rheumatoid arthritis, systemic lupus erythematosus, chronic infections, and common allergic disorders,^13,24- 26 and hs-CRP level is a well-established predictor of vascular risk in healthy individuals.^2- 3

Several observations suggest that the key finding of our study is not based on chance: (1) The association obtained was strong, specific for nontraumatic fractures and of a dose-response type and remained robust after correction for potential confounders and in several sensitivity analyses. (2) P values for the main findings were in a comfortable range well below .001. (3) Data were highly consistent in all three 5-year intervals and various subgroups according to age, sex, and menopausal state. Furthermore, we considered the possibility that subjects in the highest tertile group for hs-CRP level have confounding diseases featured by elevated hs-CRP levels and an increased fracture risk. Autoimmune diseases and malignancies are such conditions, as are cardiovascular disease and dementia. The latter conditions confer a higher risk of falls and may affect fracture risk.²⁷ Actually, exclusion of subjects with any of these diseases and/or those with an hs-CRP level greater than 7.5 mg/L of unknown origin had little effect on the results obtained (Table 3). The possibility of unidentified confounding, however, cannot be excluded.

More important, hs-CRP level was not related to bone density and both measures predicted the risk of nontraumatic fractures independently of each other. These findings, however, should be interpreted in view that ultrasonographic data were assessed at the heel and standard radiological measurements of bone density at predilection sites of osteoporosis were not available. Nevertheless, fracture risk is not exclusively determined by bone density but is also influenced by other factors, like age and genetic variables,²⁸ that also affect bone quality aside from bone mass. More important, levels of osteocalcin and β-crosslaps decreased with increasing hs-CRP concentration, suggesting a state of low bone turnover and reduced bone remodeling. Speculatively, this could affect bone microarchitecture and quality and account for the elevated fracture risk.²⁹ An inhibitory effect of inflammation on osteoblast differentiation is known from experimental models and may indirectly also affect osteoclast maturation via coupling mechanisms.³⁰ In contrast, the conjecture in subjects with autoimmune disease^31- 32 and in the small segment of the general community with more prominent inflammation (hs-CRP level, >7.5 mg/L) seems to be characterized by a preferential up-regulation of bone resorption.

The strength of this population study is its representative nature for the general community, which is ascertained by (1) near complete participation and follow-up and (2) a broad age range between 40 and 94 years and a high degree of accuracy in assessing and classifying fractures. Furthermore, assessments of risk estimates for other variables were consistent, revealing established variables, such as sex, menopausal state, age, alcohol consumption, and physical activity, and newly identified variables, such as receptor activator of nuclear factor κB ligand level, as relevant for fracture risk.¹¹ One potential limitation is the fact that clinically unapparent vertebral fractures were not assessed. Radiographic measures detect fractures that are clinically unapparent and the prevalence of these events is higher than that of clinically symptomatic fractures. Estimates suggest that every third vertebral fracture is clinically apparent.^33- 34 However, the relevance of such events is small from a clinical viewpoint, and nonassessment of minor disease phenotypes may be expected to weaken evident relations rather than to create spurious ones. Another limitation is that bone density has been assessed by heel ultrasonography rather than dual x-ray absorptiometry of the spine or hip. It can, thus, not be fully excluded that the association between CRP level and fracture risk is independent from bone mineral density, because bone ultrasonographic data, although well correlated with dual x-ray absorptiometry measures, do not formerly measure bone mineral density.^35- 36 However, most studies^37- 42 show that the prognostic validity of bone ultrasonography for vertebral and hip fractures is at least as good as that of hip and vertebral dual x-ray absorptiometry, suggesting no major effect on the strength of association between CRP level and fracture risk.

In summary, this study demonstrates that hs-CRP level predicts the risk of nontraumatic fractures at levels previously shown to be linked with cardiovascular disease and diabetes mellitus. This finding is intriguing from a pathophysiological viewpoint, suggesting a tight interplay between the immune system, inflammation, and bone quality, and may well find clinical application if confirmed in independent population samples.

Correspondence: Georg Schett, MD, Department of Internal Medicine III and Institute for Clinical Immunology, University of Erlangen-Nuremberg, Erlangen, Germany (Georg.Schett@med3.imed.uni-erlangen.de).

Accepted for Publication: August 17, 2006.

Author Contributions: Drs Schett and Kiechl contributed equally to this work. Study concept and design: Schett, Kiechl, Redlich, and Willeit. Acquisition of data: Schett, Kiechl, Weger, Pederiva, Mayr, Petrangeli, Oberhollenzer, Lorenzini, Redlich, Axmann, Zwerina, and Willeit. Analysis and interpretation of data: Schett and Kiechl. Drafting of the manuscript: Schett and Kiechl. Critical revision of the manuscript for important intellectual content: Schett, Pederiva, Mayr, Petrangeli, Lorenzini, and Willeit. Statistical analysis: Schett and Kiechl. Obtained funding: Schett, Oberhollenzer, and Willeit. Administrative, technical, and material support: Schett, Pederiva, Mayr, Oberhollenzer, and Axmann. Study supervision: Schett and Willeit.

Financial Disclosure: None reported.

Funding/Support: This study was supported by the START prize from the Austrian Ministry of Sciences; the Pustertaler Verein zur Prävention von Herz- und Hirngefässerkrankungen; the Sanitätseinheit Ost; and the Assessorat für Gesundheit.

Role of the Sponsor: The funding bodies had no role in data extraction and analyses, in the writing of the manuscript, or in the decision to submit the manuscript for publication.