Author + information
- Roman Pfister, MD∗∗ (, )
- Guido Michels, MD∗,
- Stephen J. Sharp, MSc†,
- Robert Luben, BSc‡,
- Nick J. Wareham, MBBS, PhD† and
- Kay-Tee Khaw, MBBChir, PhD‡
- ∗Department III of Internal Medicine, Heart Centre of the University of Cologne, Cologne, Germany
- †Medical Research Council Epidemiology Unit, Institute of Metabolic Science, University of Cambridge, Cambridge, United Kingdom
- ‡Department of Public Health and Primary Care, Institute of Public Health, University of Cambridge, Cambridge, United Kingdom
- ↵∗Reprint requests and correspondence:
Dr. Roman Pfister, Department III of Internal Medicine, Heart Centre of the University of Cologne, Kerpenerstrasse 62, 50937 Cologne, Germany.
Objectives It is unknown whether bone mineral density as a measure of osteoporosis is associated with development of heart failure.
Background Recent evidence suggests shared risk factors between heart failure and osteoporosis. Additionally, patients with osteoporosis are at increased risk for cardiovascular disease.
Methods We examined the prospective association of bone mineral density measured as broadband ultrasound attenuation by quantitative ultrasound of the heel with incident heart failure events in 13,666 apparently healthy persons 42 to 82 years of age participating in the EPIC (European Prospective Investigation into Cancer and Nutrition) study in Norfolk, United Kingdom.
Results During a mean follow-up of 9.3 years, 380 incident cases of heart failure occurred. The risk of heart failure decreased with increasing bone mineral density. The hazard ratios comparing each quartile with the lowest were 0.40 (95% confidence intervals [CI]: 0.27 to 0.59), 0.54 (95% CI: 0.37 to 0.79), and 0.46 (95% CI: 0.32 to 0.68) in analysis adjusting for age, sex, smoking, alcohol consumption, physical activity, occupational social class, educational level, systolic blood pressure, diabetes, cholesterol concentration, and body mass index (p for trend = 0.002), with a 23% risk decrease associated with every increase in 1 standard deviation of bone mineral density (hazard ratio [HR]: 0.77; 95% CI: 0.66 to 0.89). The association was stronger with heart failure without (HR: 0.75; 95% CI: 0.63 to 0.89) than with antecedent myocardial infarction (HR: 0.82; 95% CI: 0.62 to 1.09).
Conclusions We observed an inverse association between bone mineral density and the risk of heart failure in apparently healthy individuals. Our findings give support for cardiac assessment in people with reduced bone mineral density and warrant further exploration of underlying biological mechanisms linking osteoporosis and heart failure.
Heart failure remains a major public health issue. Approximately 1% to 2% of the adult population have heart failure, and the prevalence is rising to more than 10% in people older than 70 years. Despite the best current treatment, patients with heart failure cause enormous health care costs as the result of frequent hospitalizations and have adverse outcomes, with a 1-year mortality rate of 6% to 24% (1,2). Hence, strong emphasis is put on the identification of new risk factors for heart failure to better understand the pathophysiology of the disease and to early detect people at risk and implement preventive interventions (3).
Osteoporosis is a multifactorial skeletal disease characterized by low bone mass and micro-architectural deterioration of bone tissue, which is diagnosed by quantitative assessment of bone mineral density (4,5). Currently, approximately 21% of women and 6% of men 50 years of age or older were shown to have osteoporosis (6). Because the disease shows strong relation to age, the prevalence is projected to further increase as the result of the demographic development.
Osteoporosis and heart failure are generally considered two distinct diseases, but recent evidence suggests a link between both diseases. Several epidemiological studies demonstrated an association of heart failure and risk of future osteoporosis and related fractures (7–9). This has been partly attributed to catabolic bone remodeling associated with systemic processes of advanced chronic heart failure (10), but there is also evidence suggesting common pathophysiological mechanisms during development of both heart failure and osteoporosis. A twin study showed that common genetic predisposition and early environmental sharing largely explained the association between prevalent heart failure and increased risk of hip fracture (7). On the basis of these observations, we hypothesized that low bone mineral density as a measure of osteoporosis might be a risk marker for development of heart failure. Thus far, this has not been examined.
The aim of the present study was to investigate the prospective association of bone mineral density as a quantitative measure of osteoporosis with the risk of developing heart failure in a population of apparently healthy middle-aged men and women participating in the EPIC (European Prospective Investigation into Cancer and Nutrition) study in Norfolk.
The EPIC-Norfolk study is a prospective population study of 25,639 men and women between 39 and 79 years of age and recruited between 1993 and 1997, resident in Norfolk, United Kingdom. Details of the recruitment process, study design, and population characteristics have been published earlier (11). The EPIC-Norfolk population is broadly similar to the U.K. population in terms of the distribution of anthropometric, smoking, and cardiovascular risk factors. The EPIC-Norfolk study was approved by the Norfolk Local Research Ethics Committee, and participants gave signed informed consent at each contact.
Three years after the baseline survey between January 1998 and October 2000, participants were invited for a follow-up assessment, which constitutes the baseline for the current analyses. A total of 15,786 participants who were then 42 to 82 years of age attended and completed a detailed health and lifestyle questionnaire, including questions on history of hypertension, diabetes, heart attack, stroke, cancer, osteoporosis, fractures of the wrist, hip or spine, smoking, drug treatment, educational level (less than A-level, A-level and higher), and occupational-social class (manual, non-manual). Diet over the previous 12 months was assessed by means of a food frequency questionnaire as reported earlier (12). Daily intake of vitamin D and total energy intake were calculated by means of a standardized food composition data base (13). At this visit, quantitative ultrasound measurements of the calcaneum were obtained (14). Broadband ultrasound attenuation (BUA) (db/MHz) was measured at least twice on each calcaneum with a CUBA sonometer (McCue Ultrasonics, Winchester, United Kingdom). We used the mean of left and right ultrasound measures for analysis. The coefficient of variation was 3.5%; 15,667 of the 15,786 participants who attended the health examination and completed the health and lifestyle questionnaire had measures of BUA available. We excluded 1,898 participants who reported a history of heart attack, stroke, or any cancer at the clinic visit and 103 participants with baseline medical heart failure treatment.
Total physical activity energy expenditure (metabolic equivalent h/week) was calculated on the basis of the EPIC physical activity questionnaire, with the past year used as a reference frame. Comparison against minute-by-minute heart rate monitoring and maximal aerobic capacity (VO2max) suggested that the questionnaire is reasonably valid for ranking individuals, and we grouped participants by quartiles of physical activity energy expenditure (15). Trained nurses examined individuals at a clinic visit. Height and weight were measured, and body mass index (BMI) was estimated as weight (kg) divided by height (m2). Blood pressure was measured with the use of an Accutorr non-invasive blood pressure monitor (Datascope Medical, Huntington, United Kingdom) after the participant had been seated for 5 min. We used the mean of two measurements for analysis. Non-fasting blood samples were taken by venepuncture into plain and citrate bottles. Blood samples for assay were stored at 4°C and assayed at the Department of Clinical Biochemistry, University of Cambridge, Cambridge, United Kingdom. We measured serum total cholesterol with the RA 1000 (Bayer Diagnostics, Basingstoke, United Kingdom).
We defined prevalent heart failure by self-reported intake of drugs that were recommended for treatment of heart failure in clinical practice at the time of the survey (loop diuretics in combination with digitalis or angiotensin-converting enzyme inhibitors) (16). Ninety percent of patients in U.K. general practices received any of these treatment regimens within 6 months after diagnosis of heart failure in 1996 (17). Our definition of prevalent heart failure showed high specificity in an earlier validation study (18). Incident heart failure cases were ascertained by means of death certificate data and hospital record linkage with virtually complete follow-up, which showed high specificity for the diagnosis of heart failure according to criteria recommended by the European Society of Cardiology (19). All participants are flagged for death at the U.K. Office of National Statistics. Death certificates are coded by trained nosologists with the use of International Classification of Disease (ICD)-revision 10. Participants are also linked to National Health Service hospital information systems so that admissions anywhere in the United Kingdom are reported to EPIC-Norfolk through routine annual record linkage. Heart failure death was defined as ICD-10 I50 anywhere on the death certificate. Incident heart failure was defined as heart failure death or hospital discharge code ICD-10 I50 at any position. For secondary analyses, ischemic heart disease and myocardial infarction were defined by hospital discharge code ICD-10 I20–25 and I21–22, respectively. The current study is based on follow-up through March 2009.
We examined the association between quartiles of BUA and baseline characteristics through the use of linear and logistic regression analyses; we present results adjusted for age and sex because both variables are strong determinants of bone mineral density and may mask or pretend associations of BUA with other risk factors. We calculated age-sex–adjusted hazard ratios (HRs) and 95% confidence intervals (CIs) for combined fatal and nonfatal heart failure by means of categories of BUA with the use of Cox proportional hazard models. Multivariable Cox regression was used to determine the independent contribution of BUA for incident heart failure. Covariates previously shown to be significantly associated with heart failure in this cohort were included in the multivariable model (18). Because approximately 20% of participants had missing values of one or more covariates, we performed a secondary analysis, imputing missing data on covariates through the use of a chained equations approach (20,21). Moreover, to address the issue of unbalanced risk factors across quartiles of BUA, we performed propensity score analyses with the use of the same set of covariates as included in the fully adjusted Cox model. We used logistic regression to calculate the probability of having high BUA (being in quartiles 2 to 4) compared with the reference of quartile 1 because propensity score analysis requires a binary exposure variable. We estimated HRs by means of Cox regression, including the propensity score as a covariate. We also used single-nearest-neighbor matching with no replacement and a caliper (0.20 standard deviation [SD] of the propensity score) to match patients in quartile 1 to those in quartiles 2 to 4 (with the use of Stata command psmatch2) (22). We dropped patients from this analysis without overlapping propensity scores between the two groups and estimated hazard ratios by use of Cox regression on the exposure of being in quartiles 2 to 4 compared with quartile 1 within the matched groups. Online Table 1 shows that differences in baseline characteristics across quartiles of BUA were considerably smaller after propensity score matching. We performed additional secondary analyses, excluding heart failure events occurring during the first 2 years of follow-up and participants with history of osteoporosis or fractures and diuretic drug treatment to address the issue of reverse causality. To assess whether the association of BUA and heart failure was mediated by preceding coronary heart disease or vitamin D intake, we added hospitalization for ischemic heart disease as a time-dependent variable and dietary vitamin D intake adjusted for total energy intake to the model. Because the use of non-steroidal anti-inflammatory drugs (NSAIDs), steroids, calcium, and vitamin D supplements might influence bone mineral density and potentially might affect heart failure risk, we also excluded participants who used these drugs in secondary analyses. We performed pre-specified, stratified analysis categorized by age, sex, BMI, and smoking. Finally, we examined the association between BUA and heart failure with and without antecedent myocardial infarction and with heart failure with systolic dysfunction. Echocardiographic data on systolic function were available from an earlier validation study performed on a random subset of incident heart failure events (19).
All analyses were undertaken with the use of Stata statistical software, version 12.1 (Stata Corporation, College Station, Texas).
The mean age of the study population was 61.5 ± 9.0 years. BUA of the heel was 80.2 ± 19.1 db/MHz. Table 1 shows characteristics of the participants according to quartiles of BUA. Increasing BUA was strongly associated with male sex and decreasing age. When adjusting for age and sex, increasing BUA was associated with increasing BMI and rate of diuretic and steroid/NSAID use and was inversely associated with cholesterol concentration, smoking, manual social class, low educational level, and history of osteoporosis or fractures.
During a mean follow-up of 9.3 years, 380 (18 fatal and 362 nonfatal) incident cases of heart failure were identified (incidence, 3.0 per 1,000 person-years), with a time to heart failure event of 6.0 ± 2.8 years. Table 2 shows HRs for heart failure, comparing each quartile of BUA with the lowest quartile. The risk of heart failure decreased with increasing quartiles of BUA in age and sex–adjusted analysis (p for trend = 0.05). The association was enhanced in multivariate analysis adjusting for age, sex, smoking, alcohol consumption, physical activity, occupational-social class, educational level, prevalent diabetes, systolic blood pressure, cholesterol concentration, and BMI (p for trend = 0.002). Every increase of 1 SD in BUA was associated with a 23% risk reduction for heart failure. Replacing BMI by weight and height did not change the results (data not shown).
When comparing combined quartiles 2 to 4 with quartile 1, the decreased risk of heart failure was consistent in analyses of conventional multivariable Cox regression, propensity score adjustment, and propensity score matching (Table 3).
Secondary analyses showed similar results when imputing missing data of covariates (HR: 0.82; 95% CI: 0.73 to 0.93 per increase in 1 SD of BUA), after exclusion of participants with incident heart failure events occurring during the first 2 years of follow-up, participants with prevalent osteoporosis or history of fractures, and participants who used diuretics, steroids/NSAIDs, or calcium/vitamin D supplements (Table 4). Results were also virtually unchanged after adjustment for ischemic heart disease as a time-dependent variable (HR: 0.77; 95% CI: 0.67 to 0.89 per increase in 1 SD of BUA) and after adjustment for dietary vitamin D intake (HR: 0.79; 95% CI: 0.67 to 0.94 per increase in 1 SD of BUA). There was a consistent trend for reduced heart failure risk with increasing BUA in analyses stratified by sex, age, BMI, and smoking.
The association of BUA was stronger with heart failure without antecedent myocardial infarction than with heart failure with antecedent myocardial infarction (Table 5). Data on systolic function were available in 146 (38.4%) of the 380 incident heart failure events, with systolic dysfunction present in 111 (76%) cases. The association between BUA and heart failure with systolic dysfunction was similar as for overall heart failure.
In a prospective, population-based study, we demonstrated an inverse association between bone mineral density assessed by BUA of the heel and risk for incident heart failure in apparently healthy, middle-aged individuals. Every increase of 1 SD in ultrasound attenuation was associated with a 23% relative decrease in risk for heart failure after adjustment for age, sex, BMI, systolic blood pressure, diabetes, cholesterol level, smoking, alcohol consumption, physical activity, occupational-social class, and educational level. The association was stronger with non-ischemic than with ischemic heart failure cases. To our knowledge, this is the first study to establish an independent association between bone mineral density and the subsequent development of heart failure.
What could account for the association between low bone mineral density and increased heart failure risk? One possible explanation is that both disorders share common risk factors such as age, physical inactivity, smoking, and diabetes (3,23). We observed a significant association of bone mineral density with some of these risk factors in our cohort. After adjusting analyses for these factors and other established risk factors for heart failure, though, we still observed a significant association between bone mineral density and risk of heart failure; thus, traditional risk factors do not appear to mediate the observed association.
Heart failure is an established predictor of future osteoporosis (7–9). A second explanation for our findings is reverse causality, for example, latent cases of heart failure at baseline that could be associated with low bone density might only manifest as clinical disease during follow-up. We excluded participants with medically treated heart failure and history of myocardial infarction, which is the most frequent cause of heart failure, thus reducing the likelihood of prevalent heart failure. Nonetheless, cases of latent heart failure might be included because we did not have cardiac imaging available to exclude cardiac dysfunction. However, when excluding events occurring during the first 2 years of follow-up, which probably reflect pre-existing heart failure cases (24), our results virtually did not change. Furthermore, excluding participants with baseline diuretic drug treatment, which might be an indicator of heart failure and itself was shown to be associated with osteoporosis risk (25), also did not change our results. Thus, our findings are less likely to be attributable to reverse causality, particularly when considering the long time period of 6 years between assessment of bone density and manifestation of heart failure.
A third possible explanation is that coronary heart disease mediates the association between bone mineral density and heart failure. Bone formation and vascular calcification share underlying biological mechanisms (26), and a body of experimental and epidemiological evidence suggests that low bone mineral density increases the risk of developing cardiovascular disease (27–30). However, when we adjusted our analysis for ischemic heart disease occurring during follow-up, our results did not change substantially. In addition, the association was stronger with heart failure cases without antecedent myocardial infarction than with heart failure cases with antecedent myocardial infarction, which suggests that the association was independent of interim coronary heart disease and actually was driven by non-ischemic causes of heart failure.
A fourth explanation for our findings is that common underlying biological processes might contribute to low bone density and development of heart failure. In support of this, epidemiological studies showed that disorders leading to bone catabolism such as vitamin D deficiency and increased parathyroid hormone levels are associated with development of heart failure (31,32). Additionally, experimental studies showed that vitamin D is a negative regulator of the renin-angiotensin-aldosterone system, which plays a key role in the pathophysiology of heart failure, and genetic knock-out of vitamin D receptors leads to myocardial hypertrophy, which is a common cause of heart failure (33,34). We did not observe an attenuation of the association between bone mineral density and heart failure risk when adjusting for dietary vitamin D intake, but this does not take into account endogenous synthesis of vitamin D, which substantially contributes to the overall vitamin D status. Furthermore, adrenergic agonists, which are key mediators of heart failure–related remodeling, were demonstrated to stimulate bone “resorption stimulating receptor activator of NF-κB ligand” (RANKL) in vitro (35), and RANKL was also shown to be elevated in bone marrow plasma of patients with heart failure (10). Conversely, RANKL increased total matrix metalloproteinase activity in human fibroblasts, thus contributing to left ventricular remodeling processes in experimental studies (36). Thus, common pathophysiology is a plausible explanation for the association between bone mineral density and heart failure risk.
Our findings might have important implications. Low bone mineral density, and, in consequence, osteoporosis, which is diagnosed through low bone mineral density, are new markers that indicate increased risk for heart failure, independent of established risk factors. The 23% risk reduction observed for every SD change in ultrasound attenuation is comparable to the effect of a 12–mm Hg decrement in systolic blood pressure or a 3-U decrement in BMI (37,38). This is of major clinical interest because osteoporosis and low bone density are common, particularly in the elderly, affecting approximately 52 million persons in the United States (23), and screening for osteoporosis is recommended by the U.S. Preventive Service Task Force for all women 65 years of age or older and all younger women with a similar disease risk (39). Hence, information on the presence of this risk factor will be available in a large part of the elderly population and might be used to select people for further diagnostics of cardiac dysfunction. Identification of early stages of heart failure is crucial because effective treatments are readily available for prevention of systolic heart failure, such as angiotensin-converting enzyme inhibitors (40), and we demonstrated that bone mineral density was associated with heart failure with systolic dysfunction.
Our findings also contribute to the pathophysiological understanding of the link between osteoporosis and heart failure in showing that not only heart failure is associated with development of osteoporosis but that also low bone mineral density with or without manifest osteoporosis is associated with development of heart failure and precedes the clinical manifestation of heart failure for many years. This adds to existing evidence suggesting that both diseases share common mechanisms in pathogenesis, and further research is warranted to explore underlying biological processes to identify new therapeutic targets for preventive interventions. If, indeed, reverse causality is underlying our findings and a detectable reduction of bone mineral density is caused by heart failure processes many years before clinical manifestation, this might also have important clinical impact in suggesting that patients with cardiac dysfunction might benefit from early osteoporosis screening and initiation of anti-osteoporotic treatment. However, before that, further study is needed to examine whether there is an association between bone density and asymptomatic cardiac dysfunction.
Our approach of ascertaining cases through hospital records will tend to result in the detection of more severe cases and thus is a specific approach to finding heart failure but is likely to be relatively insensitive. This limits the generalizability of our conclusions to less severe heart failure. Additionally, we used quantitative ultrasound of the heel as a measure of osteoporosis, yet dual-energy X-ray absorptiometry (DEXA) is the standard method recommended to diagnose osteoporosis in current guidelines (5,39). However, although results of both methods show only moderate correlation (correlation factor, 0.47 in our cohort ), a recent meta-analysis as well as an analysis within our cohort demonstrated that quantitative ultrasound of the heel performed at least equally well compared with DEXA in predicting different fracture outcomes (41–43) and is a generally accepted measure of bone mineral density (5). Furthermore, we have no detailed information on the etiology of the incident heart failure cases. However, we showed that the association was driven by non-ischemic heart failure cases, and the association was apparent in cases with systolic dysfunction, although the latter findings were of borderline significance because of the low number of cases with available data on systolic function. Nonetheless, further study is needed to characterize the cardiac phenotype that is associated with low bone mineral density. Finally, observational studies are susceptible to residual confounding. This is a particular problem when underlying biology is poorly understood and hence potential mediators of the association are unknown. However, we performed several analyses to account for confounding such as adjustment of an extensive set of covariates, propensity score adjustment, and propensity score matching, which all showed consistent results.
We observed an inverse association between bone mineral density and the risk of heart failure in apparently healthy individuals. Our findings give support for cardiac assessment in people with reduced bone mineral density and warrant further exploration of underlying biological mechanisms.
The authors thank all of the participants in this study and the EPIC-Norfolk study staff at the University of Cambridge, Department of Public Health and Primary Care.
For a supplemental table, please see the online version of this article.
This work was supported by the Medical Research Council U.K. (G0401527, G1000143) and Cancer Research U.K.http://dx.doi.org/10.13039/501100000289 (C864/A8257). The authors have reported that they have no relationships relevant to the contents of this paper to disclose. Drs. Pfister and Michels have contributed equally to this paper.
- Abbreviations and Acronyms
- body mass index
- broadband ultrasound attenuation
- confidence interval
- hazard ratio
- International Classification of Disease
- non-steroidal anti-inflammatory drug
- Received February 20, 2014.
- Accepted March 7, 2014.
- American College of Cardiology Foundation
- McMurray J.J.,
- Adamopoulos S.,
- Anker S.D.,
- et al.
- Tavazzi L.,
- Senni M.,
- Metra M.,
- et al.
- Schocken D.D.,
- Benjamin E.J.,
- Fonarow G.C.,
- et al.
- Strom O.,
- Borgstrom F.,
- Kanis J.A.,
- et al.
- van D.S.,
- Majumdar S.R.,
- Bakal J.A.,
- et al.
- Carbone L.,
- Buzkova P.,
- Fink H.A.,
- et al.
- Leistner D.M.,
- Seeger F.H.,
- Fischer A.,
- et al.
- Wareham N.J.,
- Jakes R.W.,
- Rennie K.L.,
- et al.
- Dargie H.J.,
- McMurray J.J.
- ↵Leuven E, Sianesi B. PSMATCH2: Stata module to perform full Mahalanobis and propensity score matching, common support graphing, and covariate imbalance testing. Available at: http://ideas.repec.org/c/boc/bocode/s432001.html. Version 4.0.6, May 17, 2012. Accessed September 2013.
- Ueland T.,
- Yndestad A.,
- Oie E.,
- et al.
- Loehr L.R.,
- Rosamond W.D.,
- Poole C.,
- et al.