Development and validation of osteoporosis risk assessment score in patients with heart failure: comparison with the osteoporosis self-assessment tool for Asians

Abstract

Aims

Osteoporosis is prevalent and is associated with poor prognosis in patients with heart failure (HF). However, bone mineral density measurement by a dual-energy X-ray absorptiometry (DEXA) scan is not always available in a daily clinical setting or large-scale population-based studies.

Methods and results

A single-centre, cross-sectional observational study was conducted with 387 patients [median age: 77 years (interquartile range: 68–83 years); 37% women]. Bone mineral densities were measured by DEXA scans, and osteoporosis was diagnosed as ≤−2.5 standard deviation of the bone mineral densities in healthy young adults. Osteoporosis risk assessment score (ORAS) was developed using significant predictors from a logistic regression model for osteoporosis and was subsequently validated. Osteoporosis was found in 103 (27%) of the 387 HF patients. Multivariate logistic regression analyses yielded the ORAS based on sex, body mass index, handgrip strength, and anti-coagulant therapy utilization. The C-index of ORAS in the developmental set (0.796, 95% confidence interval: 0.747–0.845) was similar to the bootstrap validation of the prediction model (0.784) and tended to be higher than that of the osteoporosis self-assessment tool for Asians (OSTA). A nomogram of ORAS, established on the basis of the final logistic regression model, demonstrated 100% sensitivity at the lowest score (35 points), with an optimal cut-off point of 127 points, yielding 85% sensitivity and 62% specificity.

Conclusion

Osteoporosis risk assessment score exhibits superior predictive performance to OSTA in predicting osteoporosis in HF patients, establishing itself as a valuable tool for early detection in both daily clinical practice and large-scale population-based studies.

Graphical Abstract

BMI, body mass index; CI, confidence interval; HF, heart failure; ORAS, osteoporosis risk assessment score; OSTA, osteoporosis self-assessment tool for Asians.

Open in new tabDownload slide

Introduction

Heart failure (HF), a critical public health concern, affects ∼26 million individuals worldwide, incurring large medical costs.^1,2 The incidence of HF is strongly correlated with age, with over 80% of diagnosed patients being more than 65 years of age, resulting in a rapidly increasing prevalence in many countries.¹ Despite advancements in therapeutic interventions and preventive strategies, HF patients, particularly older individuals, continue to experience high readmission rates.³ This leads to a progressive decline in overall functionality,⁴ causing substantial functional dependence, diminished quality of life, and elevated mortality and morbidity rates,⁵ Intriguingly, factors beyond cardiac dysfunction contribute to functional dependence and decreased quality of life in older HF patients. Those factors include chronic kidney disease (CKD), anaemia, malnutrition, sarcopenia, frailty, and orthopaedic disorders such as osteoporosis.^6–11

Osteoporosis, a systemic skeletal disorder, is characterized by reduced bone mineral densities (BMDs) and microarchitectural deterioration of bone tissue, significantly affecting daily activity and healthy life expectancy, primarily through increased susceptibility to fractures.^12,13 Osteoporosis has been recognized as a critical concern for HF patients for several reasons. First, research findings, including results of our study, showed that whole-body BMDs are lower in HF patients than in non-HF patients, resulting in a high prevalence of osteoporosis in HF patients. Second, large-scale epidemiological studies have shown a close association between HF and an elevated hip fracture risk,^14–18 a serious clinical consequence of osteoporosis.¹⁹ Finally, the presence of osteoporosis in HF patients correlates with a severity-dependent deterioration in clinical outcomes.¹⁸ Thus, earlier identification of osteoporotic patients or those at increased risk could potentially enhance the quality of life and prognosis for HF patients. However, the gold standard for measuring BMDs, dual-energy X-ray absorptiometry (DEXA),²⁰ has limitations in terms of accessibility and cost, particularly in clinical settings or large-scale population-based research. This necessitates the development of a simpler and more economical tool for osteoporosis risk screening that maintains high sensitivity and specificity. For this reason, several screening tools for predicting the presence of osteoporosis were developed, one of which is the osteoporosis self-assessment tool for Asians (OSTA). However, OSTA is determined by only two factors, i.e. age and body weight, which may lead to inaccurate prediction of osteoporosis in HF patients, since several factors other than age and sex were selected as powerful predictors of osteoporosis in our previous study.¹⁸ These notions led us to attempt the development of HF-specific screening tools for osteoporosis.

Therefore, the aim of this study was to develop a simple screening tool, osteoporosis risk assessment score (ORAS), designed specifically to identify HF patients at risk of osteoporosis and to validate its predictive ability in a cross-sectional study. Moreover, we compared the predictive performance of ORAS with that of OSTA, highlighting the potential advantages of using a more tailored approach to osteoporosis risk assessment in HF patients.

Methods

Study design and study subjects

An ambispective, single-centre, and cross-sectional design was used in this study to assess consecutive HF patients admitted to our institute over a 5-year span (1 April 2017 to 31 March 2022). This design encompassed both a retrospective phase, in which we examined previously collected data (1 April 2017 to 10 April 2019), and a prospective phase, in which we enrolled and evaluated new HF patients (11 April 2019 to 31 March 2022). These periods coincided with routine DEXA scans at our institute, enriching our analysis with comprehensive osteoporosis risk data. Heart failure was diagnosed according to the Japanese Circulation Society/Japanese Heart Failure Society Guidelines.⁴ The inclusion criteria were consistent across both phases, targeting HF patients aged 50 years or older,²¹ excluding those with rheumatoid arthritis, hyperthyroidism, or hyperparathyroidism, those on steroids or anti-osteoporotic drugs, and those with missing data (Figure 1). By adopting this ambispective design, we leveraged existing data for robust retrospective analysis, while substantiating our findings with prospectively collected data. Different cohorts were chosen in respective phases to capture a broad range of HF patient data during routine DEXA scans, thereby enhancing our study’s comprehensiveness and generalizability.

This study adhered to the Declaration of Helsinki and was approved by the Clinical Investigation Ethics Committee of Sapporo Medical University Hospital (Number 302-243).

Body composition analysis

Body composition was analysed using a DEXA scan (Horizon A DXA System; HOLOGIC, Waltham, MA, USA), as previously reported.²² Bone mineral densities were measured at the hip with the femoral neck, total femoral bones, and the lumbar spine (L2–L4) and expressed as g/cm². Osteoporosis was diagnosed as ≤−2.5 standard deviation (SD) of the BMDs (T-score ≤−2.5 SD) in healthy young adults at any site according to the WHO criteria.²³ Appendicular skeletal muscle mass (ASM) was calculated as the sum of bone-free lean masses in the arms and legs. Appendicular skeletal muscle mass index was defined as ASM (kg)/height (m)².

Collection of data for clinical parameters

Demographic information, medications, laboratory results, echocardiographic data, and muscle strength were obtained from the patients’ medical records.

Laboratory data for N-terminal pro-brain natriuretic peptide (NT-proBNP), serum albumin, haemoglobin, creatinine, estimated glomerular filtration rate (eGFR), and corrected calcium and phosphorus concentrations were obtained within 7 days of DEXA measurements. The creatinine-based eGFR (eGFRcre) was calculated by using equations developed for Japanese subjects.²⁴ Chronic kidney disease was defined as eGFRcre < 60 mL/min/1.73 m². Anaemia was defined as haemoglobin <14.0 g/dL for men and <12.0 g/dL for women.

Transthoracic echocardiography was performed according to the standard protocol within 14 days of DEXA measurements, and the left ventricular ejection fraction (LVEF) was measured by using the modified Simpson method. Heart failure with reduced ejection fraction (HFrEF) was defined as LVEF <40%.

Handgrip strength, an index of muscle strength, was measured using a digital handgrip strength dynamometer (TKK-5401; Takei Scientific Instruments, Tokyo, Japan) within 7 days of DEXA measurements. Patients alternated between left and right hands for maximum effort, with the absolute value of the maximum reading calculated to 0.1 kg.

Calculation of osteoporosis self-assessment tool for Asians score

Osteoporosis self-assessment tool for Asians score was calculated as follows: OSTA score = 0.2 [weight (kg) − age (years)].^25,26 Classification of risk groups based on the OSTA score is as follows: low risk, OSTA score >−1; medium risk, −4 to ≤−1; high risk, <−4.²⁵

Sample size calculation

We calculated the minimum sample size for ORAS development using the ‘pmsampsize’ package for R, considering the prevalence of osteoporosis at 40%,¹⁸ four predictor parameters in the logistic regression model, and a C-index of 0.8. The required minimum sample size was determined to be 369 cases in this study.

Statistical analysis

Data are presented as means ± SD or medians [interquartile range (IQR): 25th–75th percentiles] depending on the results of the Shapiro–Wilk test. Categorical data are expressed as numbers with percentages.

Baseline characteristics were compared between patients with and without osteoporosis by using the Welch test, Mann–Whitney U-test, or χ² test, as appropriate. A multivariate logistic regression model was used to predict osteoporosis probability and develop ORAS. In addition to the traditional or previously reported risk factors,^18,19,27 significant variables in the analyses of osteoporosis prevalence in the subgroups were incorporated into the multivariate logistic regression model. Then, the non-significant predictors were removed through a backward-forward stepwise process based on the minimum Akaike’s Information Criterion.²⁸ Only predictors with a statistical significance of <0.05 based on the Wald test were included in the final model, and overall model performance was evaluated using the Nagelkerke pseudo R². The discrimination ability was assessed with the C-index.

Internal validation of the model was performed by correcting measures of predictive performance for ‘optimism’ or overfit using the bootstrap method with 1000 bootstrap samples. The bootstrap-corrected C-index was also generated. Model calibration (correspondence between model-predicted probabilities and observed probabilities) was assessed through the Hosmer–Lemeshow goodness-of-fit test and evaluating the slope of the calibration line (plotting the precited probabilities vs. the observed probabilities) that deviates from the ideal of 1.0.

A prediction scoring table and nomogram of ORAS were created. The optimal cut-off value of ORAS for predicting osteoporosis was calculated by using the Youden index. Harrell’s C-indexes were compared between ORAS and OSTA using the DeLong test.

Analyses were performed using JMP version 15.2.1 (SAS Institute Inc., Cary, NC, USA) and R version 4.2.1 (R Foundation for Statistical Computing, Vienna, Austria). In all analyses, a two-tailed P < 0.05 indicated statistical significance.

Results

Six hundred and eighty-two patients were initially screened, and 289 patients were excluded by using the exclusion criteria. Accordingly, 387 patients (154 from the retrospective phase and 233 from the prospective phase) were included in the analyses, as shown in Figure 1.

Baseline characteristics

The median age of the patients was 77 years (IQR, 68–83 years), and 37% were female patients. The median body mass index (BMI) of the patients was 22.5 kg/m² (IQR, 20.4–24.9 kg/m²). At the time of the DEXA scan, 9, 64, and 27% of the patients had New York Heart Association Classes I, II, and III, respectively. The median LVEF was 48% (IQR, 34–62%), and 36% of the patients were classified as patients with HFrEF. The most frequent aetiology of HF was valvular heart disease (37%), followed by cardiomyopathy (29%) and ischaemic cardiomyopathy (18%). Hypertension, diabetes mellitus (DM), and CKD were present in 65, 41, and 72% of the patients, respectively, and 11% of the patients had experienced a fracture event. Osteoporosis was present in 27% of the patients (103 patients: 45 from the retrospective phase and 58 from the prospective phase), and the mean T-scores at the femoral neck, total femoral bones, and lumbar spine were −1.4 ± 1.2, −1.0 ± 1.3, and −0.1 ± 2.1, respectively.

Comparison between patients with and those without osteoporosis

Osteoporotic patients were significantly older than non-osteoporotic patients, and the former comprised of higher percentages of women, patients with CKD, and patients with anaemia, had lower BMI, lower haemoglobin concentration, and lower serum creatinine levels, and used warfarin more frequently. In addition, osteoporotic patients tended to have a higher level of NT-proBNP and higher prevalence of DM, tended to use direct-oral anti-coagulants less frequently, and experienced a fracture event more frequently than non-osteoporotic patients (Table 1).

Development and validation of osteoporosis risk assessment score

Table 2 shows the results of multivariate logistic regression analysis for predicting osteoporosis. Among well-known predictors and variables having statistical significance in Table 1, sex (women), BMI, handgrip strength, and type of anti-coagulant therapy were selected as significant predictors for osteoporosis by the backward stepwise method (Table 2). Hence, the following final logistic regression model for the probability of osteoporosis was developed (Nagelkerke pseudo R² = 0.299):

where p is the probability of osteoporosis. The C-index of the final logistic regression model was 0.796 [95% confidence interval (CI): 0.747–0.845], and the bootstrap-corrected C-index was 0.784. Based on the calibration slope and plot, the prediction model seemed to be well calibrated (Figure 2).

A nomogram of ORAS was developed on the basis of the final logistic regression model, ranging from 0 to 207 total points (Figure 3). Sensitivity and specificity were 100 and 0%, respectively, for the lowest ORAS (35 points) and they were 2 and 100%, respectively, for the highest ORAS (207 points). The optimal cut-off point of ORAS for predicting osteoporosis was 127 points according to the Youden index of the receiver operating characteristic curve, with sensitivity and specificity of 85 and 62%, respectively. When applying one of the established and most widely used screening scores for osteoporosis, OSTA,²⁶ to the present study subjects, the C-index of ORAS for osteoporosis tended to be higher than that of OSTA (P = 0.057, Figure 4). We also conducted further analyses to assess the impact of adding OSTA to a model developed solely by ORAS and vice versa. While the addition of OSTA to a model developed solely by ORAS did not yield significant improvement in the C-index for osteoporosis [0.796 (95% CI: 0.747–0.845) vs. 0.802 (95% CI: 0.753–0.852); P = 0.31], the inclusion of ORAS in an OSTA-based model resulted in a significant improvement in the C-index [0.751 (95% CI: 0.693–0.808) vs. 0.802 (95% CI: 0.753–0.852); P < 0.01], indicating a complementary relationship between the two tools. The optimal cut-off point of OSTA for osteoporosis was ≤−4.72 in the present study subjects, with sensitivity and specificity of 67 and 73%, respectively.

Furthermore, post hoc analyses were performed to demonstrate the predictive performance of ORAS for osteoporosis among subgroups of patients. As shown in Figure 5, the C-index values were similar in the subgroups including subgroups for sex, age, severity of HF, HF phenotype (i.e. HFrEF vs. non-HFrEF), comorbidities (i.e. DM, CKD, and atrial fibrillation), functional status, and medication use.

Further exploration of the predictive capabilities of ORAS revealed correlations between ORASs and T-scores at various sites: femoral neck (r = −0.540), total femur (r = 0.533), and lumbar spine (r = −0.448, Supplementary material online, Figure S1). This indicates that an increase in the ORAS could suggest a higher severity of osteoporosis, extending its clinical applicability.

Discussion

In the present study, we developed and validated a simple scoring system, ORAS, for assessing the probability of osteoporosis in HF patients using four factors: sex, BMI, handgrip strength, and type of anti-coagulant therapy. Analyses of the predictive ability of ORAS yielded high sensitivity and specificity. The overall predictive ability of the C-index in ORAS was better than that in a previously developed screening score, and predictive abilities of ORAS were consistent across subgroups of patients, including subgroups for severity of HF and comorbidities. The results of the present study suggest that further assessment of BMDs by a DEXA scan needs to be considered in HF patients with ORAS of ≥127 points. To our knowledge, ORAS is the first risk screening score for predicting osteoporosis in HF patients.

Several established risk factors contribute to osteoporosis, including non-modifiable factors such as advanced age, female gender, postmenopausal state, and genetic/environmental factors.²⁹ Modifiable risk factors include malnutrition, limited physical activity, low BMI, cigarette smoking, and alcohol consumption. Osteoporosis can also result from specific diseases such as rheumatoid arthritis, hyperparathyroidism, hyperthyroidism, Cushing syndrome, CKD, and DM or chronic medication use such as the use of glucocorticoids.³⁰ To develop an HF-specific screening tool for osteoporosis, our study excluded patients under 50 years of age,²¹ those receiving anti-osteoporotic medications and glucocorticoid therapy, and those with chronic diseases causing osteoporosis. In the selected HF population, female gender was a powerful predictor for HF patients aged ≥50 years. Additionally, low BMI and reduced handgrip strength independently explained osteoporosis, while older age, CKD, and DM were not selected as predictors of osteoporosis in HF patients. This suggests that reduced mechanical stress such as weight bearing or muscle forces on the bone rather than an imbalance of bone resorption and formation due to ageing and metabolic disorders could be a primary mechanism of osteoporosis in HF patients.

On the other hand, unintentional weight loss and a decline in muscle strength are hallmarks of cachexia, a systemic syndrome characterized by low-grade inflammation and malnutrition, leading to disability and death.^31,32 Co-existence of osteoporosis with low BMI and reduced handgrip strength may be a surrogate marker of cachexia in HF patients.³³ Interestingly, the possible involvement of receptor activator of nuclear factor kappa B (NF-κB) ligand (RANKL), a cytokine promoting osteoporosis through the NF-κB pathway, has been implicated in cachexia pathogenesis. Receptor activator of NF-κB ligand blockade, a clinically available therapy for osteoporosis, attenuated cachexic conditions in mice with ovarian cancer.³⁴ Further analyses are needed to determine whether this applies to HF-mediated cachexia, as serum RANKL levels were up-regulated in HF patients.

In this study, 17 and 29% of the HF patients received warfarin and direct-oral anti-coagulants (DOACs), respectively. Direct-oral anti-coagulants have shown superiority in safety and even effectiveness in patients with non-valvular atrial fibrillation and venous thrombosis.^35,36 This seems to be the case in patients with risks of osteoporosis: the incidence of fragility fracture has been shown to decrease in non-valvular atrial fibrillation patients taking DOACs than in those taking warfarin,^37–39 which may be attributable to the property of warfarin as a Vitamin K antagonist that suppresses the carboxylation of glutamic acid to γ-carboxyglutamyl residues of osteocalcin. However, the use of DOACs is limited in patients with severe renal dysfunction and low BMI, i.e. patients at risk of osteoporosis, which was also the case in HF patients in the present study: patients who took warfarin were older, included a larger percentage of women and had a lower eGFRcre than those who took DOACs (see Supplementary material online, Table S1). In addition to the property of warfarin as a Vitamin K antagonist, the prescription pattern is likely to contribute to a close association between the use of anti-coagulants and osteoporosis. Therefore, ORAS may be applicable to non-HF patients with atrial fibrillation and mechanical valve implantation and to cancer-bearing patients with venous thrombosis, although its utility should be analysed in those populations in future studies.

We have developed ORAS, a novel osteoporosis risk assessment tool specifically tailored for HF patients, and we compared its performance with that of OSTA, a screening tool widely used in practice.^25,26,40 Osteoporosis risk assessment score, which incorporates the variables of sex, BMI, handgrip strength, and anti-coagulant use, demonstrated a trend of superior predictive ability over OSTA (P = 0.057, Figure 4). The utility of ORAS remained consistent across different patient subgroups (Figure 5). Interestingly, incorporating ORAS into an OSTA-based model significantly improved the C-index for predicting osteoporosis. This enhancement suggests a complementary relationship between ORAS and OSTA, wherein their combined application might yield better predictive performance than that with the individual application of each tool. This finding highlights the unique contribution of ORAS, with its inclusion of a distinct set of variables compared with OSTA, in assessing osteoporosis risk. Therefore, while both tools have their respective strengths, their combined use could offer a more comprehensive and precise assessment of osteoporosis risk in HF patients. Based on our findings, we suggest that Japanese HF patients with ORAS ≥127 points and OSTA ≤−4.72 should be recommended for DEXA measurement for potential osteoporosis diagnosis. However, the utility of ORAS and OSTA requires further investigation in different data sets of Japanese HF patients. Overall, ORAS represents a promising alternative to OSTA for estimating osteoporosis risk in HF patients, with the potential to improve early detection and intervention.

Beyond its predictive accuracy, our additional analyses revealed that ORAS has the potential to quantify the severity of osteoporosis. We discovered correlations between ORASs and T-scores at various anatomical sites: femoral neck, total femur, and lumbar spines (see Supplementary material online, Figure S1). These correlations indicate that a higher ORAS may denote a higher severity of osteoporosis, providing a graded measure of risk that could assist in clinical decision-making. This nuanced understanding of risk can help shape more personalized, targeted interventions for patients with HF who face varying degrees of osteoporosis risk. These findings extend the utility of ORAS beyond a binary predictive tool, demonstrating its potential as a measure of risk severity.

This study has several strengths. First, our study was designed according to the calculation of the minimum sample size required for prediction model development. This enabled us to minimize overfitting when developing a prediction model and led to more robust models being developed. Second, we used a resampling method (i.e. bootstrapping with 1000 iterations) for internal validation, rather than a data splitting method (i.e. into model developing and testing samples). Accordingly, our results showed that ORAS could effectively predict the probability of osteoporosis in HF patients. Third, the variables used in ORAS are easily available to clinicians and physical therapists, making their application in a clinical setting more convenient. Finally, this is the first study showing the utility of OSTA in HF patients.

However, our study has several limitations. First, the single-centre, retrospective cross-sectional design with a small number of Japanese patients has the potential to introduce selection bias and ethnic specificity. To confirm the accuracy of ORAS, it should be tested in outpatients with stable conditions and in diverse racial and ethnic groups. Second, due to the retrospective study design, we could not obtain information on two important factors related to women’s health, menstruation and menopause, which might impact the results. Although patients under 50 years of age were excluded in our study, considering the average age of menopause among Japanese women,^41–43 menstruation status could still be influential. Third, established risk factors for osteoporosis such as cigarette smoking, alcohol consumption, and family history were not considered in our study. Finally, the impact of osteoporosis and fractures on activity limitation was found to be relatively small in a large Japanese population study.⁴⁴ Therefore, the prevalence and significance of other musculoskeletal disorders in HF patients should also be analysed in future research.

Conclusions

We developed and validated an ORAS that incorporates sex, BMI, handgrip strength, and anti-coagulant therapy in HF patients. Osteoporosis risk assessment score shows reasonable accuracy in the prediction of osteoporosis and exhibits superior predictive performance compared with that of OSTA in this patient group. This makes ORAS a valuable tool in daily clinical settings and large-scale population-based studies. Osteoporosis risk assessment score offers valuable information for general clinicians and healthcare professionals to decide whether to refer patients for a detailed osteoporosis examination.

Author contributions

R.N.: conceptualization, data curation, funding acquisition, formal analysis, investigation, methodology, visualization, and writing—original draft; S.K.: conceptualization, data curation, funding acquisition, formal analysis, investigation, methodology, project administration, supervision, visualization, writing—original draft, review, and editing; T.Y.: conceptualization, formal analysis, methodology, project administration, resources, supervision, writing—original draft, review, and editing; M.K.: conceptualization, formal analysis; R.N.: data curation, funding acquisition, and investigation; Y.F.: data curation and investigation; K.Y.: data curation and investigation; S.H.: data curation, formal analysis, investigation, and methodology; K.O.: data curation, formal analysis, investigation, and methodology; H.K.: data curation, formal analysis, investigation, and methodology; M.K.: supervision, writing—review and editing; M.F.: resources, supervision, writing—review and editing; K.T.: resources, writing—review and editing; A.H.: conceptualization, project administration, resources, supervision, and writing—review and editing.

Supplementary material

Supplementary material is available at European Journal of Cardiovascular Nursing online.

Acknowledgements

The authors are highly grateful to the participants in the study and to the staff at Sapporo Medical University Hospital.

Funding

This study was supported by Sapporo Medical University grants for Programs Promoting Academic Advancements, Sapporo, Japan, a grant from Yuasa Memorial Foundation, and JSPS KAKENHI (grant nos JP18K17677, JP20K19313, and JP22K11288).

Data availability

The deidentified participant data will not be shared.

References

Author notes