Association between total cholesterol and total bone mineral density in US adults: National Health and Nutrition Examination Survey (NHANES), 2011–2018

The study was approved by the National Center for Health Statistics' Ethics Review Board (https://​www.​cdc.​gov/​nchs/​nhanes/​irba98.​htm), and informed consent was obtained from all NHANES participants for their data to be adopted in further study.

The NHANES is an ongoing representative epidemiological study in the USA that intends to survey multifarious evidence about the nutrition and health statistics utilizing a multistage, complex, probability sampling design. Our analysis was based on data from 2011 to 2018, which represents four cycles of the NHANES datasets. All data of subjects was collected from the NHANES database, which is part of the U.S. Centers for Disease Control and Prevention. The inclusion criteria were: (i) individuals who had accessible total BMD data; (ii) individuals who had accessible total cholesterol data; and (iii) individuals aged 20 to 59 years old. Moreover, the subjects with cancer were excluded from the study. Of a total 39,156 participants, we excluded 21,282 subjects with missing total BMD data, 1,358 participants with missing total cholesterol data, 387 subjects with cancer diagnoses, and 6,090 subjects younger than 20 years or older than 59 years. Finally, 10,039 participants were used in the research (Fig. 1).

Total cholesterol as the exposure variable was measured via a serum sample by Roche Cobas 6000, which is an enzymatic method where esterified cholesterol is converted to cholesterol by cholesterol esterase and then acted upon by cholesterol oxidase to produce cholest-4-en-3-one and hydrogen peroxide. The hydrogen peroxide then reacts known as the Trinder reaction with 4-aminophenazone in the presence of peroxidase to produce a colored product that is measured at 505 nm (secondary wavelength = 700 nm). Total BMD testing was performed by Dual-energy X-ray absorptiometry (DEXA) by Hologic QDR 4500A bone densitometers and Apex software (version: 3.2) throughout by certified radiology technologists. For categorical variables, the study included the following covariates: gender, race/ethnicity, educational level, moderate activity, smoking, marital status, high blood pressure status, and diabetes status. The study included ratio of family income-to-poverty, waist circumference, BMI, calcium intake, serum calcium, alcohol consumption, ALT, AST, blood urea nitrogen, serum phosphorus, γ-glutamyl transferase, lactate dehydrogenase, serum uric acid and triglycerides as continuous covariates. The categories "other Hispanic" and "other multiracial race" from the original NHANES were assigned to the variable "other race/ethnicity" to lower the bias of sample quantity. The NHANES program (https://​www.​cdc.​gov/​nchs/​nhanes/​) describes more detailed information about how the variables were measured (Fig. 2).

We used weighted and variance estimation analysis to explain the significant differences in our data sets. Weighted multivariate logistic regression models were applied aiming to evaluate the link between total cholesterol and total BMD. We applied weighted multivariate linear regression models and weighted χ2 tests to assess the distinction among each group for the categorical and continuous variables. To subgroup analysis, stratified multivariate regression analysis was also made. Moreover, nonlinear associations between total cholesterol and total BMD were also discussed via smooth fitting curves and generalized additive models. For nonlinear models, if the nonlinearity was detected, then the two-piecewise linear regression models was performed on both parts of the inflection point using a recursive algorithm. All analyses were performed using EmpowerStats (version: 2.0, X&Y Solutions, Inc., Boston, MA. http://​www.​empowerstats.​com) and R software (version: 4.0.3, http://​www.​R-project.​org), with a P value < 0.05 considered significant, and the frequency histogram of total cholesterol was completed via ggplot2 package in R.

The weighted characteristics of participants were separated into quartiles based on total cholesterol concentration (Q1: 1.50–4.20 mmol/L; Q2: 4.21–4.80 mmol/L; Q3: 4.81–5.51 mmol/L; and Q4: 5.52–21.02 mmol/L), as displayed in Table 1. A total of 10,039 participants aged 20–59 years old were covered in our study. Our study showed significant differences in baseline characteristics between the total cholesterol concentration quartiles (Table 1). Participants from the Q4 total cholesterol quartile were intended to be males, non-Hispanic whites, with greater values of BMI (29.39 ± 5.89 kg/m²), waist circumference, alcohol consumption, blood urea nitrogen, ALT, AST, serum phosphorus, serum calcium, γ-glutamyl transferase, lactate dehydrogenase, serum uric acid, triglycerides, and ratio of family income-to-poverty and lower total BMD, moderate activity, and low calcium intake compared to the other level groups.

Overall, the mean age of participants was 39.15 ± 11.62 years, the male (50.57%) and female (49.43%) were almost equal, and non-Hispanic whites account for 60.64%. Most of participates had an above high marriage rate (62.37%), high school education level (65.15%) and the ratio of family income-to-poverty was 2.89 ± 1.63. The participants had an educational level over high school, high blood pressure, and diabetes, which accounted for 65.15%, 21.94%, and 5.69%, respectively. Moreover, the distribution of total cholesterol categorized by gender and race is presented in Fig. 3. The specific results and other baseline characteristics are displayed in Table 1.

The multivariate regression analysis results are presented in Table 2. When no covariates are adjusted, total cholesterol was negatively associated with total BMD in Model 1 (β = − 0.0070, 95%CI − 0.0090 to − 0.0051, P < 0.001). After covariates were adjusted, this negative relationship was still present in Model 2 (β = − 0.0038, 95%CI − 0.0058 to − 0.0019, P < 0.001) and Model 3 (β = − 0.0036, 95%CI − 0.0057 to − 0.0015, P < 0.001). Participants of the highest total cholesterol quartile (Q4) had a 0.0109 g/cm² lower total BMD than those in the base total cholesterol quartile after total cholesterol had been transformed from a continuing variable to a quartile’s variable.

Subgroup analyses, stratified by gender, age, and, race/ethnicity, are presented in Table 2. The adverse connection of total cholesterol with total BMD remained in both females (β = − 0.0037, 95%CI − 0.0065 to − 0.0008, P = 0.011) and males (β = − 0.0037, 95%CI − 0.0067 to -0.0006, P = 0.020), as well as in non-Hispanic white (β = − 0.0040, 95%CI − 0.0075 to − 0.0004, P = 0.028) and other races (β = − 0.0047, 95%CI − 0.0086 to − 0.0008, P = 0.017), but not in Mexican Americans and non-Hispanic blacks. Moreover, the significant negative correlation still existed in participants aged 50–59 years (β = − 0.0048, 95%CI − 0.0069 to − 0.0028, P < 0.001).

Figures 4, 5, 6, and 7 illustrate fitted smooth curve and generalized additive model applied to describe the linear or nonlinear relationship between total cholesterol and total BMD. In other races (including Hispanic and Multi-Racial), the relationship between total cholesterol and total BMD was a reverse-L-curve, and the inflection point was 6.7 mmol/L (Table 3). Among participants aged 40–49 years, the association between total cholesterol and total BMD was a U-shaped curve (Table 4). For a total cholesterol < 5.84 mmol/L, every 1 mmol/L increase in total cholesterol was associated with a 0.0063 g/cm decrease in total BMD (95%CI − 0.0117 to − 0.0009); by comparison, for individuals with a total cholesterol > 5.84 mmol/L, a 1 mmol/L increase in total cholesterol was associated with a 0.0092 g/cm increase in total BMD (95%CI 0.0017 to 0.0167). Among participants at borderline diabetes status (Table 5), a U-shaped curve (inflection point: 4.65 mmol/L) was presented. For total cholesterol < 4.65 mmol/L, every 1 mmol/L growth was related to a 0.0504 g/cm² lower total BMD (95%CI − 0.0941 to − 0.0068); on contrast, for participants with a total cholesterol > 4.65 mmol/L, a 1 mmol/L increase in total cholesterol was linked with a 0.0293 g/cm² increase in total BMD (95%CI 0.0093 to 0.0493).