Association of hemoglobin levels with bone mineral density for adults over 18 years of age: a cross-sectional study

The overall objective of this study was to determine the association between hemoglobin (HGB) and bone mineral density (BMD) in the lumbar and thoracic spine of adults aged ≥ 18 years. This cross-sectional study utilized the non-institutionalized US population from the National Health and Nutrition Survey (NHANES) as the sample source. A multiple linear regression model was used to assess the relationship between HGB and BMD in the lumbar and thoracic spine, with analysis of subgroups conducted according to sex and race. Smooth curve fitting was performed to explore the potential nonlinear relationship. When nonlinearity was found, we further constructed a weighted two-piecewise linear regression model and used a recursive algorithm to calculate the inflection point. After accounting for relevant confounding variables, HGB was found to be negatively associated with lumbar spine BMD in multiple regression models. However, in the subgroup analyses stratified by sex and race, the relationship between HGB and thoracic spine BMD and lumbar spine BMD was only found in women and other races and races that were not recorded. In Non-Hispanic Asian subjects, the relationship between HGB and BMD in the lumbar spine and thoracic spine showed a U-shaped curve. In addition, the relationship between HGB and BMD in the lumbar spine formed an inverted U-shaped curve among participants in other races and those whose race was not reported. Our study shows that HGB has a non-linear relationship with lumbar and thoracic BMD. Further studies are required to elucidate the mechanisms underlying this association.


Materials and methods
Study population. Data for the analysis were obtained from the NHANES (2013-2018), a large, comprehensive, and regularly updated probability sample of the non-institutionalized U.S. population [20][21][22] .
The study population consisted of participants aged ≥ 18 years who had complete data on HGB and lumbar and thoracic spine BMD. The NCHS Ethics Review Committee approved the implementation of the NHANES, and informed consent was obtained from all participants. The study adhered to the relevant guidelines and regulations. The encryption procedure was uniform to make it possible to link complaints belonging to the same patient in the NHANES. Detailed documentation of the ethics application and written informed consent are available on the NHANES website 23-25 . Data collection. The following information was collected by two researchers (XZH and SQL): Demographic data [gender, age, race, annual household income and education level]. Examination data [BMD of the lumbar spine and the thoracic spine (mg/cm 2 ), systolic blood pressure (SBP) (mmHg), diastolic blood pressure (DBP) (mmHg), body mass index (BMI) (kg/m 2 )].
Questionnaire data [alcohol status (average alcohol drinks/day in the past 12 months), smoking status, diabetes (Has a doctor told you that you have diabetes?), rheumatoid arthritis (RA) (Which type of arthritis was it?), chronic obstructive pulmonary disease (COPD) (Ever told you had copd?), cancer or malignancy (Ever told you had cancer or malignancy?)].
Weight value [Depending on the rules for selecting weight values offered on the NHANES website, "Full Sample Two-Year Mobile Examination Center Exam Weight (WTMEC2YR)" was Selected as representative weighting value].
For the selection of confounding variables, we followed the following principles: (1) based on clinical significance, we selected indicators related to HGB and BMD; (2) from the perspective of statistical methodology, we chose to include factors that have an impact greater than 10% on the results; (3) based on the above two points, confounding factors were selected according to the biological point of view.  (2) the missing values of a categorical variable were separated into a "not recorded" group.

Evaluation of variables. HGB was measured at the NHANES Mobile Examination Centers (MECs). A
Beckman Coulter DxH 800 instrument was used to obtain a complete blood count (CBC) on blood specimens and provide a distribution of blood cells for all participants. Transmittance of light at 525 nm through a lysed WBC solution in the HGB cuvette was compared to the transmittance of the same light through a reagent blank. The system converted this ratio to an HGB value using a calibration factor. The weight (mass) of HGB was determined by the degree of absorbance recorded from the photocurrent transmittance expressed in g/dl. A detailed description of the measurement of HGB is mentioned in the Laboratory Method Files section on the NHANES website 26 .
The low-level X-rays of the Hologic Discovery model A densitometer (Hologic, Inc., Bedford, Massachusetts) with Apex 3.2 software was used to scan the patient's body to estimate BMD under standard operating conditions by radiographers who participated in the DXA examination; the entrance dose of the examinee for a whole-body scan was less than 1 mR 1 (a standard X-ray is approximately 35 R). More information regarding the measurement of the DXA examination protocol can be obtained from the NHANES website 27 .
Information about age, gender, race, education level, annual household income, alcohol status, smoking status, diabetes, RA, and COPD was provided to the participants by trained interviewers using a computer-assisted personal interview (CAPI) system. Data for serum albumin, blood urea nitrogen, uric acid, phosphorus, ALP, ALT, iron, CPK, glucose, LDH, potassium, total bilirubin, total protein, creatinine and calcium were obtained from standard biochemical profile analysis using a Beckman Synchron LX20. Total cholesterol were analyzed using a Roche Modular P chemistry analyzer (enzymatic method). Metals determination for blood cadmium and blood lead was performed by inductively coupled plasma mass spectrometry on whole blood samples from the Laboratory Sciences Department of the National Center for Environmental Health. Triglycerides (TG), low density lipoprotein cholesterol (LDL-C) and high density lipoprotein cholesterol (HDL-C).The details of the variables mentioned above are available on the NHANES website 28 and Appendix 1.

Statistical analysis.
We used the weighted chi-square test to calculate the differences between the classification variables and a weighted linear regression model for continuous variables. Finally, the means (continuous variables) or proportions (categorical variables) were used to describe the baseline characteristics of all participants included in the study. Weighted multiple linear regression analysis was used to calculate the relationship between HGB and BMD. Following the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines 29 , we calculated the unadjusted, minimally adjusted, and fully adjusted results. To further investigate the relationship between HGB and BMD of the thoracic and lumbar spine in different populations, a subgroup analysis was performed by separating the participants by age and sex. Additionally, smooth curve fitting was performed to explore the nonlinear relationship. Accordingly, we further constructed a weighted twopiecewise linear regression model and used a recursive algorithm to calculate the inflection point.

Consent for publication.
All participating authors give their consent for this work to be published.

Results
Participant selection and general characteristics. Initially, the information of 29,400 participants was extracted from the NHANES database for 2013-2018. After excluding individuals without HGB data (n = 5211) and BMD data (n = 10,942), under 18 years of age (n = 4133), and with RA, COPD, cancer, or any other malignancy (n = 623), a total of 8491 participants were included in this study (Fig. 1).
The weighted socio-demographic and medical characteristics of the participants are presented in Table 1. The mean age of these participants was 37.8 ± 12.1 years, of which 52.5% were male and 47.2% were female, with non-Hispanic whites being the most numerous. There were significant differences in sex, age, race, annual household income, education level, SBP, DBP, BMI, red blood cell count, albumin, ALP, ALT, blood urea nitrogen, total calcium, iron, CPK, creatinine, glucose, LDH, phosphorus, total bilirubin, total protein, uric acid, albumin creatinine ratio, blood cadmium, blood lead, phosphorus, potassium, triglyceride, LDL-C, HDL-C, total cholesterol, alcohol status, smoking status and diabetes across the different HGB groups (quartiles, Q1 to Q4).

Relationship between HGB and BMD.
To explore the relationship between HGB and BMD, we constructed three weighted univariate and multivariate linear regression models: Model 1, unadjusted; Model 2, adjusted for sex, age, and race; and Model 3, adjusted for the covariates in Table 1.
After fully adjusting for confounders, HGB was negatively correlated with lumbar spine BMD (β = − 0.0416, P = 0.031873) ( Table 2, Fig. 2). However, the negative correlation between HGB and thoracic spine BMD was not statistically significant (β = − 0.0127, P = 0.362458) ( Table 2, Fig. 3). In fully adjusted subgroup analyses stratified by sex and race, the negative association between HGB and thoracic spine BMD and lumbar spine BMD was only present in females and other races and races that were not recorded (Tables 3 and 4).
In addition, fully adjusted smoothened plots showed a non-linear relationship between HGB and BMD in the lumbar and thoracic spine after stratification by sex or race (Figs. 4 and 5). Among participants of Non-Hispanic Asian, BMD in the thoracic spine decreased with HGB until the turning point (turning point: HGB 14.9 g/dL). Similarly, there was a turning point between HGB and BMD in the lumbar spine (turning point: HGB 15.6 g/ dL). There is also a turning point in other races and races that were not recorded (turning point: HGB 13.4 g/ dL) (Table 5).
Overall, the relationship between HGB and BMD in the lumbar spine and BMD in the thoracic spine in Non-Hispanic Asian showed a U-shaped curve. In other races and races that were not recorded, there was an inverted U-shaped curve between HGB and lumbar BMD.

Discussion
The main goal of this study was to clarify whether HGB is correlated with BMD in the lumbar and thoracic spine. In our comprehensive adjusted model multiple linear regression analysis, we found that HGB was negatively correlated with lumbar BMD, and the completely adjusted smoothed curve fits showed a non-linear correlation. When stratified by sex and race, the relationship between HGB and thoracic spine BMD and lumbar spine BMD was only found in women and other races and races that were not recorded. In Non-Hispanic Asian subjects, the relationship between HGB and BMD in the lumbar spine and thoracic spine showed a U-shaped curve. In addition, the relationship between HGB and BMD in the lumbar spine formed an inverted U-shaped curve among participants in other races and those whose race was not reported.
Similarly, a cross-sectional study involving 3626 Korean participants found that HGB levels were inversely associated with low BMD in the lumbar spine among non-anemic adults 30 . Furthermore, in a recently published study, BMD was negatively correlated with HGB levels in younger and older women 31 . imilar results were obtained by Cesari et al. in a survey of 950 older adults 32 . Nevertheless, a study examining the association between serum HGB levels, BMD, and fracture risk using estimated scores from the Fracture Risk Assessment Tool (FRAX) in 662 male patients concluded that HGB was positively associated with BMD but negatively associated with the risk of hip fracture and major osteoporotic fracture 33 . It is possible that age and ethnic differences in the study population, or the limited sample size, may have influenced the results. Based on this, the cross-sectional study has a broad and large sample size, targeting both males and females aged 18 years and older, and provides subgroup analysis across gender and race. The mechanism of the link between HGB and BMD is not yet clear; however, based on the theory that both osteoblasts and cells of the hematopoietic microenvironment that are responsible for maintaining hematopoietic tissue have a common progenitor, namely mesenchymal stem cells (MSC) 34 , Gurevitch et al. proposed a hypothesis that the differentiation pathways of osteogenesis and the hematopoietic microenvironment compete with each other, with osteogenic stimulation predominating during the growth phase of the organism. Nonetheless, after maturation, there is a gradual increase in differentiation of MSCs towards cells of the hematopoietic microenvironment and a decrease in intraosseous differentiation. This subsequently leads to a reduction in bone mass and enlargement of the bone marrow cavity in hematopoietically active cancellous bones 35 . They speculated that continuous overproduction of blood cells leads to excessive depletion of the hematopoietic system and is a non-negligible component in the etiology of osteoporosis. Blood loss promotes the proliferation of hematopoietic progenitor cells, leading to an increase in the number of hematopoietic cells including osteoblasts, which enhances bone tissue resorption. In addition, a reduction in blood volume stimulates bone development and increases the number of osteoblasts, thereby promoting new bone formation 36 . It has also been shown that acute bleeding stimulates the secretion of bone morphogenetic protein 2 and BMP6 by hematopoietic stem cells, thereby driving MSC differentiation along the osteogenic pathway 37 .  www.nature.com/scientificreports/ This cross-sectional study not only confirmed the association between HGB and BMD, but also provided substantial clinical significance. The U-shaped curve in the relationship between HGB and BMD in the Non-Hispanic Asian population implies that BMD in the thoracic and lumbar spine may be quietly decreasing as HGB levels approach 14.9 (g/dL) or 15.6 (g/dL) levels. More notably, however, the inverted U-shaped curve in the relationship between HGB and BMD of the lumbar and thoracic spine in people of other races and unreported races demonstrates that for these patients, clinicians should be aware of low levels of HGB while being alert to the risk of reduced bone mass and the need for close monitoring of BMD and early intervention.
This study analyzed NHANES 2013-2018 data conducted by the U.S. Centers for Disease Control and Prevention. Owing to the rigorous design of the NHANES database, the accuracy of the data requires the completion The area between the two blue dashed lines is represented as a 95% CI. The area between the lines is expressed as 95% CI. Each point represents the size of the hemoglobin and is connected into a continuous line. Age, sex, race, annual household income, education level, SBP, DBP, BMI, red blood cell count, albumin, ALP, ALT, blood urea nitrogen, total calcium, iron, CPK, creatinine, glucose, LDH, phosphorus, total bilirubin, total protein, uric acid, albumin creatinine ratio, blood cadmium, blood lead, phosphorus, potassium, triglyceride, LDL-C, HDL-C, total cholesterol, alcohol status, smoking status and diabetes were adjusted.  The area between the two blue dashed lines is represented as a 95% CI. The area between the lines is expressed as 95% CI. Each point represents the size of the hemoglobin and is connected into a continuous line. Age,sex, race, annual household income, education level, SBP, DBP, BMI, red blood cell count, albumin, ALP, ALT, blood urea nitrogen, total calcium, iron, CPK, creatinine, glucose, LDH, phosphorus, total bilirubin, total protein, uric acid, albumin creatinine ratio, blood cadmium, blood lead, phosphorus, potassium, triglyceride, LDL-C, HDL-C, total cholesterol, alcohol status, smoking status and diabetes were adjusted. www.nature.com/scientificreports/ of a large sample size and stratified analysis for this study. It is regrettable that for some confounding factors that may affect the results, such as chronic kidney disease, chronic inflammatory disease, long-term infection, and the use of certain drugs, due to the lack of relevant information in the 2013-2018 NHANES database, this study cannot describe the current cases in the study. Nonetheless, this study relied on a cross-sectional design; therefore, it was not possible to ascertain a causal relationship between BMD and HGB.

Conclusions
Our study shows that HGB has a non-linear relationship with lumbar and thoracic BMD. Further studies are required to elucidate the mechanisms underlying this association. Table 3. Association between HGB and BMD stratified by gender. Expose variable HGB (g/dL), ending variable lumbar spine BMD, the results in the table are expressed as β (95% CI). Model 1, no covariates were adjusted. Model 2, age, race were adjusted. Model 3, age, race, annual household income, education level, SBP, DBP, BMI, red blood cell count, albumin, ALP, ALT, blood urea nitrogen, total calcium, iron, CPK, creatinine, glucose, LDH, phosphorus, total bilirubin, total protein, uric acid, albumin creatinine ratio, blood cadmium, blood lead, phosphorus, potassium, triglyceride, LDL-C, HDL-C, total cholesterol, alcohol status, smoking status and diabetes were adjusted. In the subgroup analysis stratified by gender, the model is not adjusted for the stratification variable itself.  Table 4. Association between HGB and BMD stratified by race. Expose variable HGB (g/dL), ending variable lumbar spine BMD, the results in the table are expressed as β (95% CI). Model 1, no covariates were adjusted. Model 2, age, sex were adjusted. Model 3, age, sex, annual household income, education level, SBP, DBP, BMI, red blood cell count, albumin, ALP, ALT, blood urea nitrogen, total calcium, iron, CPK, creatinine, glucose, LDH, phosphorus, total bilirubin, total protein, uric acid, albumin creatinine ratio, blood cadmium, blood lead, phosphorus, potassium, triglyceride, LDL-C, HDL-C, total cholesterol, alcohol status, smoking status and diabetes were adjusted. In the subgroup analysis stratified by race, the model is not adjusted for the stratification variable itself.    Hemoglobin (g/dl) Lumbar bone mineral density (mg/cm 2 ) (a) (b) Figure 5. The hemoglobin and bone mineral density relationship, stratified by race. Age, sex, annual household income, education level, SBP, DBP, BMI, red blood cell count, albumin, ALP, ALT, blood urea nitrogen, total calcium, iron, CPK, creatinine, glucose, LDH, phosphorus, total bilirubin, total protein, uric acid, albumin creatinine ratio, blood cadmium, blood lead, phosphorus, potassium, triglyceride, LDL-C, HDL-C, total cholesterol, alcohol status, smoking status and diabetes were adjusted.  Table 5. Threshold effect analysis of HGB on Thoracic Spine BMD and Lumbar Spine BMD using twopiecewise linear regression. Age, sex, annual household income, education level, SBP, DBP, BMI, red blood cell count, albumin, ALP, ALT, blood urea nitrogen, total calcium, iron, CPK, creatinine, glucose, LDH, phosphorus, total bilirubin, total protein, uric acid, albumin creatinine ratio, blood cadmium, blood lead, phosphorus, potassium, triglyceride, LDL-C, HDL-C, total cholesterol, alcohol status, smoking status and diabetes were adjusted.