Study setting and subjects
The data were drawn from inpatients’ and outpatients’ electronic medical records (EMR) and dual X-ray absorptiometry (DXA) database stored at The Second Hospital of Jilin University, which is a major diagnosis and treatment center for severe diseases (i.e., cancers and CVDs) in northeast China. The linked for EMR and DXA databases capture inpatients’ and outpatients’ demographics (i.e., age and sex), anthropometry data (e.g., body height and weight), clinical laboratory measurement records, prescription drug dispensation records, clinical diagnoses and BMD data.
We identified individuals with complete and valid data on femoral neck and total hip BMD, TC, LDL-C, HDL-C and TG in 2017 in this study. We excluded individuals: (1) with age < 50 years old at BMD test; (2) with early (< 40 years) menopause; (3) with a lipid-lowering therapy, synthyroid or hormone-replacement therapy; (4) with osteoporosis-related diseases such as cancer, thyroid disease, hypopituitarism, rheumatoid arthritis, chronic renal failure or renal dysfunction (creatinine > 442 µmol/L), chronic liver disease or liver dysfunction (aspartate aminotransferase or alanine aminotransferase > 80 U/L); (5) with a surgical history of bilateral salpingo-oophorectomy menopause or a history of bone surgery at the lumbar spine or hip. There are no data on osteoporosis medications in our study. Osteoporosis medications are prescribed after the patients are diagnosed with osteoporosis through DXA testing. To partially excluding the individuals taking anti-osteoporosis medication, we only include individuals at their first DXA tests in our study. This study was approved by The Second Hospital of Jilin University Life Science Ethics Committee (2017 Research Approval No.13).
BMD measurement
Femoral neck and total hip BMD were measured by a DXA fan-beam bone densitometer (Discovery Wi, Hologic, Bedford, MA, USA). The control spine phantom scan performed each day had a long-term (more than 5 years) coefficient of variation (CV) of < 0.5 %. A BMD T-score was calculated using the following formula:
$$\mathrm T=\left(\mathrm X-\mathrm\mu^{\mathrm{reference}}\right)/\mathrm{SD}^{\mathrm{reference}}$$
Where X = the observed BMD of the patients, µreference = BMD in young adults, and SDreference = standard deviation of BMD in young adults.
We defined osteoporosis as femoral neck or total hip T-score ≤ -2.5 [
26]; the reference for calculating T-score was based on BMD data from Whites 20–29 years old in the third National Health and Nutrition Examination Survey (NHANES III) [
27].
Blood draw and storage
Fasting blood samples (≥ 8 h) were drawn from individuals with spray-coated silica and a polymer gel evacuated sterile collection tubes (BD, Becton, Dickinson and Company, Franklin Lakes, New Jersey, USA) by nurses; these blood samples were centrifuged to obtain serum at clinical laboratory department.
Lipid biomarkers ascertainment
Within 2 h following blood draw, serum samples were used to directly measure TC, LDL-C, HDL-C and TG; their levels were determined by a biochemical analyzer (7600 model, Hitachi, Tokyo, Japan). The CVs for TC, LDL-C, HDL-C and TG measurements were 2.5 %, 3.1 %, 5.1 and 1.9 %, respectively.
Covariate ascertainment
The covariates were extracted from EMR and DXA databases. The covariates for this study included age, sex, body mass index (BMI), disease diagnoses (e.g., type 2 diabetes, hypertension, ischemic heart disease, ischemic stroke, osteoarthritis), diabetes complications (e.g., neuropathy, retinopathy, nephropathy), diabetes medications (e.g., insulin, biguanide, sulfonylureas, others), antihypertension medications (e.g., calcium channel blockers, angiotensin 2 receptor blockers, beta-blockers, diuretics, angiotensin converting enzyme inhibitors), liver function biomarkers, bone biochemical markers (e.g., alkaline phosphatase, calcium and phosphorus), thyroid function biomarkers and glucose metabolic biomarkers. The covariates included in the analysis could be potential confounders or effect modifiers that maybe associated with lipid biomarkers and femoral neck or total hip BMD [
16,
22,
24]. Height and weight were measured using a wall-mounted stadiometer (to the nearest 0.1 cm) and an electronic scale (to the nearest 0.1 kg). BMI was calculated as weight (kilograms) divided by the square of the height (meters). Any past use of diabetes medications and antihypertension medications were self-reported. Disease diagnoses were either self-reported and/or clinically diagnosed based on specific criteria. The biomarkers were measured using established methods in the clinical setting [
28]. To further examine whether menopausal age had an impact on the association between lipid biomarkers and osteoporosis, we also considered self-reported menopausal age in women.
Statistical analysis
Multivariable logistic regression models were used to test the associations of TC (per SD increase), LDL-C (per SD increase), HDL-C (per SD increase) and TG (per SD increase) with osteoporosis. Due to the high correlations between lipid biomarkers (Pearson r up to 0.70), we included each lipid biomarker in one model. Models were further adjusted for sex, age, BMI, type 2 diabetes, neuropathy, biguanide, calcium channel blockers, angiotensin converting enzyme inhibitors, alanine transaminase, albumin, total bilirubin and alkaline phosphatase; variables considered but not included in the adjusted model were hypertension, ischemic heart disease, ischemic stroke, osteoarthritis, retinopathy, nephropathy, insulin, sulfonylureas, others, angiotensin 2 receptor blockers, beta-blockers, diuretics, aspartate aminotransferase, total protein, gamma-glutamyltransferase, calcium, phosphorus, free thyroxine 4, thyroid-stimulating hormone, glycated hemoglobin A1c and fasting glucose, because they failed to meet P < 0.10 criteria under bivariate analyses with osteoporosis. Because total bilirubin and direct bilirubin are highly associated (r = 0.80), to avoid the effect of collinearity, direct bilirubin was not included in the multivariable logistic regression models.
Multivariable linear regression models were used to test the associations between lipid biomarkers and femoral neck or total hip BMD. Variables included in the model were the same as above.
TC, LDL-C, HDL-C and TG were classified as normal and abnormal groups according to their clinical reference values. In serum, TC ≥ 5.20 mmol/L and/or LDL-C ≥ 3.40 mmol/L and/or HDL-C < 1.00 mmol/L and/or TG ≥ 1.70 mmol/L were defined as abnormal groups [
29]. Multivariable logistic regression models were used to test the associations of each lipid biomarker group with osteoporosis. Again, adjusted covariates were the same as above.
Multivariable logistic regression models were used to test the associations of each lipid biomarker group with osteoporosis in women only and in individuals with any chronic disorders. In the subgroup analysis of women only, we further adjusted menopausal age in addition to the covariates as mentioned above.
All analyses were conducted with SPSS (version 24.0, IBM, Inc., New York, USA).