Introduction

It has been estimated that 1.31 million new hip fractures occurred in 1990. In the same year, the prevalence of hip fractures associated with disability was 4.48 million; 740,000 deaths have been imputed to hip fracture [1]. A threefold increase in incident hip fractures is anticipated within 2030 [2]. In addition, a threefold increase in total direct costs has been documented among patients with hip fracture in the year following the event, as compared with other age-matched subjects [3]. However, apart from hormonal factors, knowledge about the determinants of osteoporosis in older populations is still poor.

Anemia represents a common condition in older populations, which prevalence has been found in about 13% of subjects older than 70 [4]. Noticeably, hypoxemia has been proven to be a risk factor for osteoporosis [5–7]. Several studies have reported on the concurrence of osteoporosis and anemia in selected conditions, such as sickle-cell anemia [8], chronic inflammatory conditions [9] or renal failure [10]. However, available information on the association between haemoglobin levels and osteoporosis in general older populations is scanty [5].

We tested whether haemoglobin levels are associated with bone mineral density in older community-dwelling populations, independently of underlying conditions that might induce both anemia and osteoporosis. Also, we searched for haemoglobin cutoff levels, if any, that might best predict osteoporosis in older subjects.

Patients and methods

Participants

The study involved all subjects, without exclusion criteria, aged 75 or older living in Tuscania (Italy) on January 1, 2004. This town had been selected for a study of the genetic determinants of health status in older populations, along with five other Italian towns. Of the 387 participants enrolled in the site, we excluded 23 subjects with missing data for the study variables and six participants treated with biphosphonates. None of the participating women were on hormonal treatment during the survey. Data were recorded using dedicated software. All participants underwent ambulatory or home visits by the study physicians, who performed detailed physical and anamnestical examination, standard electrocardiography, Doppler echocardiography and bone densitometry, and collected blood samples for serum chemistry and genomic analyses. Also, the study researchers completed a questionnaire which included data on socioeconomic status, lifestyle habits and physical activity, among others. Smoking was considered as total lifetime pack-years for current and former smokers. Alcohol consumption was considered when it consisted of at least two drinks per week. Education was expressed as years of school attendance. Nutritional parameters were collected using a validated questionnaire already adopted for testing large Italian populations [11]. Physical activity was measured using the Physical Activity Scale for the Elderly questionnaire [12]. Participants’ diagnoses and treatments were obtained by their general practitioners before the study visits and further confirmed by the study physicians, who received specific training and whose concordance had been tested using dummy cases. Drugs were coded according to the anatomical, therapeutic and chemical codes [13]. Diagnoses were coded according to the International Classification of Diseases, ninth edition, Clinical Modification codes [14]. Blood samples were obtained in the morning after an overnight fast; after processing, the specimens were aliquoted into cryovials, frozen at −70°C and shipped to the Department of Experimental Pathology, University of Bologna. Measures for interleukin-6 (IL-6) and high-sensitivity C reactive protein (CRP) were obtained from frozen stored plasma and measured in duplicate. IL-6 and CRP plasma levels were measured by enzyme-linked immunosorbent assay kits.

Diagnosis of anemia

Anemia was defined according to the World Health Organization (WHO) criteria: haemoglobin <130 g/L in male and <120 g/L in female participants [15].

Measurement of bone mineral density

As the study enrolled the whole older population of the site, including homebound subjects, a portable instrument was used to assess bone densitometry (Achilles Express, GE Medical Systems, Madison, WI, USA). This validated instrument measures bone density and structure using ultrasound bone densitometry [16]. Ultrasound densitometry enables measurement of the physical properties of bone, specifically bone mineral density. The ultrasound measurement is based on two criteria: the velocity [speed of sound (s), SOS] and frequency attenuation [broadband ultrasound attenuation (dB/MHz), BUA] of sound wave as it travels through the bone [17]. The stiffness is an index combining SOS and BUA, which is calculated by the spread speed of supersonic waves. The formula is (BUA − 50) × 0.671 + (SOS − 1,380) × 0.28. This charts the SOS and BUA into biologically relevant ranges.

According to current WHO recommendations, the ultrasound-derived T score was automatically calculated by the built-in software as the difference between the subject bone mineral density (represented by the stiffness index) and the average density in a reference healthy 30-year-old woman of the same ethnicity, divided by the standard deviation of the reference population [18]. The reference population was derived by the NHANES III study population [19]. According to WHO guidelines, osteoporosis is defined as a T score value of −2.5 or lower, meaning a bone density that is two and a half standard deviations below the mean of a 30-year-old woman. The ultrasound-derived Z score was automatically calculated by normalising the T score for age, as recommended by WHO. The intra-observer and inter-observer correlation coefficients for ultrasound bone densitometry that were analysed during the researchers’ training were 0.73 and 0.68, respectively.

Data analyses

Data of continuous variables are presented as mean values ± SD. Statistical analyses were performed using the SPSS for Windows 13.0 software (Chicago, IL, USA); differences were considered significant at the p < 0.050 level. Analysis of variance (ANOVA) for normally distributed variables in relation to diagnosis of anemia was performed by ANOVA comparisons; otherwise, the nonparametric Kruskal–Wallis H test was adopted. Chi-square analysis was used for dichotomous variables. Haemoglobin was considered as a continuous variable. High-sensitivity CRP and IL-6 were analysed after log transformation.

Linear regression analysis was used to estimate the association of variables of interest, including haemoglobin levels, with the ultrasound-derived T score, Z score, and the stiffness index. Linearity of variables was assessed using the ANOVA test of linearity; the linearity assumption was assumed to be satisfied at a p < 0.050 level. To assess independent correlates of bone mineral density, which might confound the association between the ultrasound-derived T score, Z score and the stiffness index with haemoglobin levels, groups of variables (demographics, co-morbid conditions, medications, and objective tests, as shown in Table 1) were first examined in separate age- and sex-adjusted regression models with the simultaneous introduction of covariates (Table 2). Those variables, significant at the p < 0.050 level in these initial models, were simultaneously entered into a summary age- and sex-adjusted regression model (Table 2). In addition, the association between bone mineral density parameters and haemoglobin levels was assessed (Table 3), adjusting for those variables which differed significantly according to diagnosis of anemia (Table 1).

Table 1 Characteristics of participants by diagnosis of anemia (according to the World Health Organization criteria: <130 g/L in men and <120 g/L in women)
Full size table
Table 2 Association (β coefficients and 95% confidence intervals) between haemoglobin levels and the stiffness index, T score and Z score according to the age- and sex-adjusted and the fully adjusted regression models
Full size table
Table 3 Association (β coefficients and 95% confidence intervals) between haemoglobin levels and the stiffness index, T score and Z score, adjusted for variables that differed significantly according to the diagnosis of anemia
Full size table

Eventually, linear discriminant analysis was used in separate models to identify haemoglobin levels which best predicted ultrasound-derived T score levels diagnostic of osteoporosis (i.e., <−2.5) in men and women. The Bayesian characteristics of dichotomised haemoglobin levels for identifying osteoporosis were also calculated.

Results

Anemia was detected in 43/358 (12%) participants; osteoporosis was found in 153/358 (43%) participants. The main characteristics of participants, according to diagnosis of anemia, are shown in Table 1. In the three initial linear regression models, age, sex, protein consumption, physical activity scale for the elderly, body mass index and haemoglobin were all associated with indices of bone mineral density at a p < 0.050 level (Table 2). In the final regression models, haemoglobin levels were still associated with the ultrasound-derived T score (β = 0.13; 95% confidence intervals = 0.01–0.25; p = 0.030), Z score (β = 0.11; 95% confidence intervals = 0.01–0.22; p = 0.045) and stiffness index (β = 1.87; 95% confidence intervals = 0.51–3.21; p = 0.007) after simultaneously adjusting for all these potential confounders (Table 2). Also, in the regression model (Table 3) including those variables which were found to differ significantly according to diagnosis of anemia (Table 1), haemoglobin levels were still associated with ultrasound-derived T score (β = 0.23; 95% confidence intervals = 0.09–0.36; p = 0.001), Z score (β = 0.15; 95% confidence intervals = 0.01–0.30; p = 0.039) and stiffness index (β = 2.45; 95% confidence intervals = 0.71–4.20; p = 0.006).

Eventually, the linear discriminant analysis indicated in separate models that haemoglobin levels below 140 g/L in men and 130 g/L in women best predicted (p < 0.0001) participants having ultrasound-derived T score levels <−2.5. Using such cutoff levels, low haemoglobin had 35% sensitivity, 77% specificity, 53% positive predictive value and 61% negative predictive value.

Discussion

The results of the present study indicate that haemoglobin levels are independently associated with all the ultrasonographic bone mineral parameters in unselected, community-dwelling older populations. Also, haemoglobin levels <140 g/L in men and 130 g/L in women might best identify older subjects having ultrasound-derived T score levels <−2.5.

The association of haemoglobin levels with bone mineral density does not represent a mere academic oddity. In fact, several studies have reported that the prevalence and incidence rates of osteoporotic fractures are constantly increasing in Western countries, paralleling the aging of populations [2, 20, 21]. Accordingly, several reports have documented that the mortality rates and the financial burden associated with fractures are currently boosting [3, 22, 23]. In this setting, a threefold increase in total direct costs has been documented among patients with hip fracture in the year following the event, as compared with age-matched subjects without fractures [3].

On the other hand, epidemiological data indicate that anemia is a prevalent condition in older populations, affecting about 13% of subjects aged 70+ [21]. In addition, available data demonstrate that anemia is a powerful risk factor for increased mortality and disability in the elderly [24–26]. This might be partially due to the increased risk of falls [27] because reduced haemoglobin levels have been associated both with decreased muscular strength and poorer physical performance [19, 24–26, 28]. According to these findings, anemia might further increase the risk of fractures in the elderly.

Several conditions characterised by low haemoglobin levels or anemia might also be associated with increased prevalence of bone loss or osteoporosis without any clear cause–effect relationship [9, 10, 29]. However, a study has suggested that anemia might be associated with osteoporosis independently of underlying conditions [30]. Our results support the hypothesis of independent association between low haemoglobin levels and bone mineral density. The role of confounding in this setting is crucial. In fact, several variables, including demographics, lifestyle habits, co-morbid conditions, drug treatment and objective parameters have been associated with bone mineral density in older subjects [31]. To minimise the effect of potential confounders, we built regression models adjusting both for those factors associated with bone mineral density (Table 2) and for variables associated with anemia in the study population (Table 3).

The present study does not allow us to explain the pathophysiology of reduced bone mineral density in participants with low haemoglobin levels. It might be hypothesised that inflammation, frequently found in older subjects, might affect both bone density parameters and haemoglobin levels. In fact, several studies have focused on the role of proinflammatory cytokines in the development of anemia and osteoporosis in older populations [29, 32, 33]. Specifically, healthy elderly present higher circulating levels of IL-6 as compared with younger subjects [34]; this proinflammatory cytokine yields an inhibitory control on haematopoiesis [35].

Indeed, the study previously carried out in an elderly population has suggested that the association between low haemoglobin levels and osteoporosis might be independent of circulating proinflammatory cytokines [30]. In our study, the association between haemoglobin levels and bone mineral density parameters persisted even after adjusting for circulating levels of IL-6 and CRP.

Experimental and clinical studies have rather suggested that hypoxemia might cause osteoporosis [5, 28]. Accordingly, in patients with chronic obstructive pulmonary disease, an independent association between severity of the illness and bone mineral density parameters has been demonstrated [6]. More recently, a study has found an association between left ventricular function and bone mineral density parameters in older women [7]. Thus, hypoxemia might have contributed to the association between low haemoglobin levels and bone mineral density parameters in the present study.

In our study, the association of haemoglobin levels with bone mineral density parameters persisted (Table 3) after adjusting for those variables that differed according to diagnosis of anemia (Table 1). This might be relevant to clinical practice, as it might indicate a potential therapeutic approach to osteoporosis through the correction of low haemoglobin levels. Even independently of such a possible therapeutic approach, the finding of low haemoglobin levels might add to currently acknowledged risk factors for the identification of higher-risk subpopulations for screening purposes. In fact, currently adopted clinical tools for the assessment of increased probability of osteoporosis yield low specificity and positive predictive values, as compared with those associated with low haemoglobin levels in this study [36].

A possible limitation of the study is represented by its cross-sectional design that did not allow us to ascertain any cause–effect relationships. Another limitation is the lack of measurement of serum parathormone and vitamin D levels. Information on sunlight exposure is included in the physical activity scale for the elderly, which mainly explores outdoor activities. The T score cutoff of −2.5 refers to dual-emission X-ray densitometric assay. Indeed, several threshold levels have been proposed for ultrasound measurements. The adoption of the −2.5 cutoff in this study allows lower sensitivity, but higher specificity (> 90%) as compared with other proposed thresholds [37]; this represents a conservative bias that further supports the validity of our results. Noticeably, ultrasound-derived indices of bone mineral density have been found to predict the risk of nonspinal fractures [38].

Independently of its pathophysiological determinants, the finding of increased probability of osteoporosis in older participants, whose haemoglobin levels were below 140 g/L in men and 130 g/L in women, allows us to select subjects with higher pretest probability of osteoporosis in whom screening might yield higher effectiveness.