Age-, Site-, and Sex-Specific Normative Centile Curves for HR-pQCT-Derived Microarchitectural and Bone Strength Parameters in a Chinese Mainland Population

Abstract

High-resolution peripheral quantitative computed tomography (HR-pQCT) is an advanced 3D imaging technology that has the potential to contribute to fracture risk assessment and early diagnosis of osteoporosis. However, to date no studies have sought to establish normative reference ranges for HR-pQCT measures among individuals from the Chinese mainland, significantly restricting its use. In this study, we collected HR-pQCT scans from 863 healthy Chinese men and women aged 20 to 80 years using the latest-generation scanner (Scanco XtremeCT II, Scanco Medical AG, Brüttisellen, Switzerland). Parameters including volumetric bone mineral density, bone geometry, bone microarchitecture, and bone strength were evaluated. Age-, site-, and sex-specific centile curves were established using generalized additive models for location, scale, and shape with age as the only explanatory variable. Based on established models, age-related variations for different parameters were also quantified. For clinical purposes, the expected values of HR-pQCT parameters for a defined age and a defined percentile or Z-score were provided. We found that the majority of trabecular and bone strength parameters reached their peak at 20 years of age, regardless of sex and site, then declined steadily thereafter. However, most of the cortical bone loss was observed after the age of 50 years. Among the measures, cortical porosity changed most dramatically, and overall, changes were more notable at the radius than the tibia and among women compared with men. Establishing such normative HR-pQCT reference data will provide an important basis for clinical and research applications in mainland China aimed at elucidating microstructural bone damage driven by different disease states or nutritional status.

Introduction

Osteoporosis is a systemic metabolic disorder characterized by low bone mass, deterioration in bone microarchitecture, and increased risk of fragility fractures.⁽¹⁾ It is estimated that approximately 40% of Chinese women and 8.7% of Chinese men over the age of 50 years will sustain one or more osteoporotic fractures in their remaining lifetime.⁽²⁾ Fractures lead to significantly increased morbidity, mortality, and medical costs for patients.^(3,4) Therefore, screening for osteoporosis among high-risk populations coupled with early diagnosis and management are key to improving clinical outcomes.

Currently, bone mineral density (BMD) is measured at the lumbar spine and hip using dual-energy X-ray absorptiometry (DXA), the gold standard tool for diagnosing osteoporosis. However, osteoporosis diagnosis solely based on BMD has its limitations. First, DXA measures two-dimensional areal BMD (aBMD) with planar projection technology. As a result, it is subject to technical errors in a number of clinical situations, such as hyperostosis, fracture, vascular calcification, as well as inappropriate positioning during acquisition.⁽⁵⁾ Second, under most clinical circumstances, DXA measures the femoral neck and lumbar spine only and therefore does not provide information about appendicular sites. It has been reported that only 44% of women and 21% of men meet the DXA-based diagnostic criteria when non-vertebral fractures occur.⁽⁶⁾

The advent of high-resolution peripheral quantitative computed tomography (HR-pQCT) introduced a low-radiation technique to measure three-dimensional volumetric BMD (vBMD) as well as cortical and trabecular microarchitecture at peripheral sites (specifically, the distal radius and tibia), thus compensating for some of the key limitations of DXA mentioned above. Indeed, HR-pQCT-based studies in recent years have shown that age-related skeletal deterioration is driven not only by BMD decline but also by alteration of the bone microarchitecture.^(5,7–17) Before HR-pQCT can be applied in clinical practice to screen for osteoporosis or detect bone damage, a population-based normative reference database and reference intervals must be established. However, because HR-pQCT parameters are highly dependent on age, the reference intervals should be replaced by reference centile curves.⁽¹⁸⁾ Such curves have been published for HR-pQCT in the UK, US, Canada, Denmark, Brazil, and Hong Kong.^{(5,7–10,12–17)} Given the variability in BMD, bone microarchitecture, and fracture incidence among different races and even potentially those of the same race from different geographic regions, data from these prior studies are not necessarily directly applicable to the Chinese mainland population.

The present study aims to establish the age-, site-, and sex-specific centile curves for the Chinese mainland population, and in doing so, to quantify age-related patterns of change in various HR-pQCT parameters. Establishing such normative HR-pQCT reference data will not only help to clarify bone microstructural features of mainland Chinese but also provide an important foundation for HR-pQCT-based clinical and research applications in mainland China aimed at elucidating structural bone damage driven by different disease states or nutritional status.

Materials and Methods

Participants

Participants in this cross-sectional study were drawn from two groups: (i) Healthy participants (n = 666) recruited from Peking Union Medical College Hospital (PUMCH) from June 2015 to February 2017, consisting of medical staff and persons presenting to the PUMCH Outpatient Clinic for routine physical examination. Posters and flyers that included a brief introduction of HR-pQCT, general inclusion criteria, and possible benefits one might derive from the study were disseminated at PUMCH Outpatient Clinic Physical Examination Center to attract potential participants. The study was also advertised on the PUMCH intranet for staff/students. Each potential participant was screened via telephone to determine eligibility. Those that were eligible were invited for a face-to-face encounter at the PUMCH Outpatient Endocrinology Clinic, where written informed consent was obtained, followed by collection of HR-pQCT and anthropometric measures. (ii) Healthy postmenopausal women recruited from the community in Dongcheng District of Beijing, China, from January 2017 to May 2017 as part of the Chinese Vertebral Osteoporosis Study (ChiVOS). In brief, ChiVOS was a nationwide multicenter epidemiologic study investigating the prevalence of vertebral fracture among Chinese community-dwelling postmenopausal women. More than 2700 women were recruited from 10 sites across five geographical regions across China. Women were eligible for inclusion in ChiVOS if they were ≥50 years at the time of enrollment, had lived in an urban community in China for more than 6 months, and were postmenopausal (amenorrhea lasting for more than 12 months with an elevated serum follicle-stimulating hormone [FSH] over 40 mIU/mL, or over 6 months after bilateral oophorectomy with or without hysterectomy). Study procedures included a face-to-face encounter, coordinator-administered study questionnaire, DXA measurement, thoracolumbar X-ray, and biospecimen (blood and urine) sample collection. As part of the original protocol, all patients who were recruited for the ChiVOS study in Beijing (n = 274) were enrolled at PUMCH, a large tertiary-care hospital, and also obtained an additional HR-pQCT test. ChiVOS was reviewed and approved by the institutional review board of PUMCH, and written informed consent was obtained from all participants before engagement in any study activities.

For the present study, participants with the following characteristics were excluded from the final analysis: (i) history of inherited or acquired metabolic bone or endocrine disorders such as rickets, osteomalacia, osteogenesis imperfecta, hyper/hypoparathyroidism, and hyper/hypothyroidism, and other conditions associated with secondary osteoporosis or fracture, including rheumatic disease, chronic liver or kidney disease, type 1 diabetes mellitus, and malignancy; (ii) use of medications known to affect bone metabolism, such as hormone replacement therapy (HRT), glucocorticoids, anti-osteoporotic drugs (with the exception of vitamin D and calcium); (iii) history of bilateral oophorectomy before menopause; (iv) history of prior fragility fractures. Participants were further divided into seven age groups by decade.

Anthropometric data acquisition

A digital stadiometer (Hampton, Seritex, East Rutherford, NJ, USA) and the RGZ-120 digital electronic weight scale (Xiheng, China) were used to measure standing height and body weight to the nearest 0.1 cm and 0.1 kg, respectively. Body mass index (BMI) was calculated as weight (kg)/height (m)².

HR-pQCT scans

For each participant, the nondominant forearm and left ankle were scanned by HR-pQCT (Xtreme CT II, Scanco Medical AG, Brüttisellen, Switzerland). If the participant had a history of fracture in the nondominant arm or left leg, the contralateral side was scanned instead. Images were obtained with an isotropic resolution of 61 μm. Subjects were asked to keep their wrists and ankles immobilized in a carbon-fiber cast from the manufacturer without talking for at least 2 minutes. The reference line was placed at the distal end plate of the non-dominant radius and tibia. A stack of 168 CT slices was acquired 9.0 and 22.0 mm proximal to the reference line for the distal radius and tibia, respectively. At the time of acquisition, the images obtained were manually scored for motion on a scale of 1 (no motion) to 5 (significant blurring of periosteal surface, discontinuities in the cortical shell, or streaking in the soft tissue), and up to three repeated scans at each site were permitted in cases with significant motion artifacts. After reevaluating, images with grade 4 or greater motion artifact were deemed to be of insufficient quality and were excluded.^(19,20)

All image analyses were performed according to standard in vivo acquisition protocols provided by the manufacturers, with the scanner set at standard mode (voltage 68 kVp, current 1.47 mA, integration time 43 ms, matrix size 2304 × 2304). The periosteal surface of the bone was identified automatically. For identification of the endosteal surface and segmentation of the cortical and trabecular compartments, an Image Processing Language (v5.42, Scanco Medical) algorithm was applied.^(21,22) Segmentation failures were detected automatically by measuring slice-wise variation in total cross-sectional area. Cases with an absolute slice-wise difference of 2 mm² at the diaphyseal tibia, and 4 mm² at the distal sites, were visually reviewed and manually corrected, as needed. Total, trabecular, and cortical volumetric bone mineral density and cross-sectional area (Ar) were directly measured after successful compartment segmentation. Cortical porosity (Ct.Po) and thickness (Ct.Th) and bone volume fraction (Tb.BV/TV), trabecular thickness (Tb.Th), separation (Tb.Sp), and number (Tb.N) were calculated directly.^(23,24)

All micro finite element (μFE) analyses were performed with the Scanco Finite Element (FE) software (vision 1.13, Scanco Medical). After the binary image was turned into a mesh of isotropic brick elements, a uniaxial compression test with a 1000 N load applied was performed with 1% apparent strain. All elements were assigned a homogenous elastic modulus of 10 GPa and a Poisson’s ratio of 0.3 was specified. Failure load was assumed to occur when 2% of the elements exceed a local effective strain of 0.7%.⁽²⁵⁾

Statistical analysis

Sex-specific reference centile curves for each HR-pQCT parameter (the response variables) with age as the only explanatory variable were constructed using the lms function of Generalized Additive Models for Location, Scale, and Shape (GAMLSS) package implemented in R, version 3.5.1.^(26,27) The complex nature of the relationships between age and HR-pQCT parameters indicates that traditional linear or quadratic functions are insufficient to model the centile curves. GAMLSS is a more flexible semiparametric statistical modeling technique, suited to modeling a response variable that does not follow exponential family distributions (eg, leptokurtic or platykurtic and/or skewed data).^(26,28,29) The exponential family distribution assumption for the response variable is replaced instead by a general distribution family, which is capable of modeling both skewness and kurtosis in GAMLSS. It offers general linear predictors for four distribution parameters, μ (location, median), σ (scale, variability), ν (shape, skewness), and τ (shape, kurtosis), as well as a choice of error distributions. The original LMS method, developed in 1992, allowed modeling of the first three distribution parameters (μ, σ, and ν) and was later modified to incorporate the fourth parameter (τ) into the analysis.^(26,28,29) Default distribution families including the Box-Cox Cole and Green (BCCG), Box-Cox power exponential (BCPE), and the Box-Cox t (BCT) distributions and normal distribution (NO) are tested, then compared using the Generalized Akaike Information Criterion as a measure of the fit of the model, and ultimately selected automatically by the lms function of the GAMLSS package. For all models, penalized B-spline was used for curve smoothing. A power transformation (λ) to age was applied as appropriate to stretch or compress the age scale. Goodness of fit was assessed generally and locally in these selected models by Q-Q plot of the normalized quantile residuals,⁽²⁶⁾ Q statistics,⁽³⁰⁾ and worm plots.⁽³¹⁾ If inadequacies were detected by diagnostic plots, the effective degrees of freedom (df) of specific distribution parameters (μ, σ, ν, and τ) were increased to improve the fit of the model.⁽²⁶⁾ All automatically selected models were deemed adequate except for tibial trabecular area (Tb.Ar) in men and radial cortical porosity (Ct.Po) in women. For the former, inadequate local fitness of τ was identified and was addressed by increasing the effective df for the τ curve. For Ct.Po at the radius in women, a combination of data transformation [lg(1.001 + Ct.Po)] and increase in effective df for μ were needed to adequately fit the data. Revised models were subsequently rechecked using Q statistics and worm plots and deemed adequate.

Predicted values of HR-pQCT parameters for a defined age and percentile or Z-score were estimated from the model. A Z-score is a standard score indicating how many standard deviations a measurement is from the age-specific mean. To investigate the differences in parameters that may be based upon age, the maximum and minimum absolute values with corresponding age for each HR-pQCT parameter were identified from the centile curves. The maximum values were deemed “peak values” except for six parameters—total area (Tt.Ar), Tb.Ar, Ct.Po, cortical perimeter (Ct. Pm), trabecular spacing (Tb.Sp), and Tb.Inhomogeneity—for which a higher value indicates a poorer trait. Age-related variations for different parameters were calculated using the equation (maximum–minimum/maximum) for overall change and further investigated separately for specific periods (eg, from peaking age to 50, 50 to 65, 65 to 80 years).

Models and graphs were developed using the package gamlss (http://www.gamlss.org/)⁽²⁷⁾ and ggplot2⁽³²⁾ implemented in R 3.5.1 (http://cran.us.r-project.org/).

Results

Participants’ characteristics

A total of 940 participants were recruited, among which 253 men and 610 women were eligible for inclusion after careful exclusion (Supplemental Fig. S1). After excluding low-quality and missing images, 849 HR-pQCT scans of the radius (from 250 men and 599 women) and 840 HR-pQCT scans of the tibia (from 249 men and 591 women) were included in the analysis. Distribution of participants by age group and their anthropometric measures are shown in Table 1. μFE analyses were not available for patients recruited through the ChiVOS study; therefore, μFE analysis data were available from 619 participants (250 men and 369 women) at the radius and 603 participants (235 men and 368 women) at the tibia. Given the ChiVOS study purposefully recruited postmenopausal women, the overall sample for the present study was not equally balanced across age groups, and there were more women (70.7%) than men (29.3%). Women had an average age of 54.8 ± 16.2 years, whereas the average age for men was 46.4 ± 15.2 years.

Table 1.

Anthropometric Parameters by HR-pQCT Site and Age Group

Age (years)	20–29	30–39	40–49	50–59	60–69	70–79	≥80	Total
No. of participants
Men	40	65	52	42	34	15	5	253
Women	56	76	91	134	135	92	26	610
Total	96	141	143	176	169	107	31	863
Radius study
Men	40	64	51	41	34	15	5	250
Women	55	76	90	133	131	88	26	599
Total	95	140	141	174	165	103	31	849
Tibia study
Men	38	64	51	42	34	15	5	249
Women	54	74	89	131	132	86	25	591
Total	92	138	140	173	166	101	30	840
Anthropometrya
Age (years)
Men	27.1 ± 1.9	35.0 ± 2.8	44.5 ± 2.8	55.3 ± 3.1	64.7 ± 2.6	75.3 ± 3.2	81.0 ± 1.0	46.4 ± 15.2
Women	26.3 ± 2.3	34.8 ± 2.6	45.2 ± 2.7	55.2 ± 3.1	65.1 ± 2.9	74.8 ± 3.0	82.6 ± 1.9	54.8 ± 16.2
Weight (kg)
Men	72.1 ± 10.0	75.9 ± 12.3	73.7 ± 12.5	71.7 ± 11.1	71.7 ± 7.2	68.3 ± 8.4	69.9 ± 16.3	73.0 ± 11.2
Women	56.9 ± 9.3	59.1 ± 8.3	61.7 ± 10.3	61.2 ± 9.8	63.2 ± 11.6	60.3 ± 9.4	59.6 ± 9.3	60.9 ± 10.1
Height (cm)
Men	173.4 ± 4.7	171.3 ± 6.5	171.0 ± 5.9	170.2 ± 5.6	170.7 ± 5.1	169.8 ± 6.7	161.4 ± 9.7	171.0 ± 6.1
Women	162.1 ± 5.0	160.6 ± 5.0	160.1 ± 5.3	159.3 ± 5.0	156.7 ± 6.2	155.1 ± 6.0	152.6 ± 5.2	158.4 ± 6.0
Body mass index (kg/m2)
Men	24.0 ± 3.2	25.8 ± 3.2	25.1 ± 3.5	24.7 ± 3.3	24.6 ± 2.3	23.8 ± 3.3	26.5 ± 3.8	24.9 ± 3.2
Women	21.7 ± 3.5	22.9 ± 3.3	24.1 ± 3.8	24.1 ± 3.8	25.6 ± 4.3	24.6 ± 3.3	25.8 ± 4.3	24.2 ± 3.9

^aThe anthropometric data were calculated for 253 men and 610 women included in the study.

Fitting the GAMLSS models for HR-pQCT parameters

Table 2 shows the transformation power (λ), response variable distribution, and effective df for smoothing the specific distribution parameter (μ, σ, ν, and τ) in the GAMLSS model fitted for each HR-pQCT parameter. Effective df(μ) in most models were greater than 2 regardless of sex or site, indicating that the median did not have a linear relationship with age for most bone parameters. For both men and women, models for cortical parameters including Ct.Po, cortical volumetric bone mineral density (Ct.vBMD), and cortical area (Ct.Ar) almost always showed the highest df(μ) compared with models fitted for other parameters, indicating that the medians of these three HR-pQCT parameters might change with age in more sophisticated ways. The majority of models required modeling of skewness (ν). Effective df(ν) equaled 2 in the majority of models, indicating that skewness either had a positive or negative linear relationship with age.

Table 2.

Peak (or Nadir) Value, Age at Peak Value, and Maximum Percent Difference Between Age Groups for Each HR-pQCT Parameter

	Men						Women
HR-pQCT parameters	Agepeak	μpeaka	SDpeak	Agenadir	μnadira	Percent difference (%)b	Agepeak	μpeaka	SDpeak	Agenadir	μnadira	Percent difference (%)b
Radius
Tt.vBMD (mgHA/cm3)	31	356.7	53.1	≥80	274.5	−23.06	33	363.0	59.9	≥80	206.3	−43.18
Ct.vBMD (mgHA/cm3)	46	919.2	39.3	≥80	835.1	−9.15	40	969.2	38.9	≥80	839.4	−13.39
Tb.vBMD (mgHA/cm3)	≤20	195.0	35.6	≥80	132.8	−31.89	31	143.4	35.6	≥80	79.3	−44.71
Tt.Ar (mm2)	≤20	321.1	71.8	≥80	355.9	+10.84	≤20	223.3	34.9	≥80	252.4	+13.03
Ct.Ar (mm2)	47	82.8	12.4	≥80	63.1	−23.79	35	62.6	7.1	≥80	43.7	−30.19
Tb.Ar (mm2)	≤20	242.2	72.4	≥80	282.0	+30.59	≤20	161.7	39.9	≥80	211.0	+30.49
Ct.Th (mm)	48	1.233	0.200	≥80	0.965	−21.72	33	1.167	0.164	≥80	0.772	−33.86
Ct.Po (%)	≤20	0.346	0.277	≥80	1.437	+315.03	≤20	0.136	0.030	≥80	0.745	+447.79
Ct.Pm (mm)	≤20	74.4	8.3	≥80	82.0	+10.22	≤20	62.1	4.650	≥80	66.6	+7.25
Tb.BV/TV	29	0.283	0.049	63	0.204	−27.91	32	0.210	0.055	≥80	0.123	−41.47
Tb.N (mm−1)	≤20	1.481	0.232	≥80	1.322	−10.78	≤20	1.359	0.191	≥80	1.037	−23.70
Tb.Th (mm)	≤20	0.249	0.016	≥80	0.233	−6.45	36	0.225	0.017	≥80	0.216	−4.20
Tb.Sp (mm)	≤20	0.608	0.113	≥80	0.729	+19.74	≤20	0.676	0.128	≥80	0.979	+44.82
Tb.1/N.SD (mm)	≤20	0.235	0.048	≥80	0.286	+21.28	≤20	0.243	0.064	≥80	0.413	+69.55
Stiffness (kN/mm)	≤20	93.8	17.0	≥80	60.8	−35.14	34	63.1	11.2	≥80	36.6	−42.01
Failure Load (N)	≤20	5092.7	887.0	≥80	3242.9	−36.32	33	3479.0	612.7	≥80	1920.2	−44.81
Tibia
Tt.vBMD (mgHA/cm3)	≤20	378.2	73.6	≥80	240.2	−36.49	≤20	328.7	45.6	≥80	192.9	−41.31
Ct.vBMD (mgHA/cm3)	≤20	945.0	24.5	≥80	812.2	−14.05	34	975.4	32.3	≥80	790.3	−18.98
Tb.vBMD (mgHA/cm3)	≤20	223.9	35.6	≥80	137.2	−38.74	≤20	158.9	35.7	≥80	106.8	−32.76
Tt.Ar (mm2)	≤20	779.4	129.0	≥80	848.1	+8.81	≤20	591.6	76.9	≥80	659.8	+11.53
Ct.Ar (mm2)	≤20	158.3	22.4	≥80	130.0	−17.85	34	121.6	15.4	≥80	84.3	−30.67
Tb.Ar (mm2)	≤20	631.9	144.1	≥80	722.7	+14.37	30	482.4	85.3	75	576.9	+19.59
Ct.Th (mm)	≤20	1.661	0.296	≥80	1.368	−17.62	33	1.449	0.239	≥80	1.048	−27.71
Ct.Po (%)	≤20	0.619	0.416	≥80	4.591	+641.68	≤20	0.827	0.721	≥80	3.865	+367.35
Ct.Pm (mm)	≤20	109.5	9.0	≥80	115.5	+5.48	≤20	94.4	5.6	≥80	100.3	+6.25
Tb.BV/TV	≤20	0.326	0.109	67	0.217	−33.25	≤20	0.235	0.052	≥80	0.177	−24.65
Tb.N (mm−1)	≤20	1.431	0.277	≥80	1.185	−17.17	≤20	1.234	0.210	≥80	1.069	−13.40
Tb.Th (mm)	≤20	0.286	0.026	45	0.252	−11.77	≤20	0.244	0.021	≥80	0.241	−1.32
Tb.Sp (mm)	≤20	0.659	0.147	≥80	0.833	+26.40	≤20	0.777	0.158	≥80	0.938	+20.72
Tb.1/N.SD (mm)	≤20	0.293	0.069	≥80	0.351	+20.14	≤20	0.305	0.081	≥80	0.390	+27.87
Stiffness (kN/mm)	≤20	274.2	52.1	≥80	180.8	−34.07	≤20	180.3	28.6	≥80	106.6	−40.87
Failure Load (N)	≤20	14577.1	2600.7	≥80	9847.7	−32.44	≤20	9883.7	1585.6	≥80	5901.1	−40.29

vBMD = volumetric bone mineral density; HA = hydroxyapatite; Ct. = cortical; Tb. = trabecular; Ar = area; Th = thickness; Po = porosity; Pm = perimeter; BV/TV = bone volume fraction; N = number; Sp = separation; 1/N.SD = inhomogeneity.

^aPeak refers to the highest value, whereas worst refers to the lowest value, except in six parameters: total area, trabecular area, cortical porosity, cortical perimeter, trabecular separation, and trabecular inhomogeneity. In these parameters, a higher value indicates poorer structural trait and, therefore, μ_peak was identified as the nadir.

^bPercent difference was calculated by [(μ_peak − μ_nadir)/μ_peak] × 100.

Centile curves and patterns of change in HR-pQCT parameters

Site- and sex-specific centile curves (1st, 3rd, 15th, 25th, 75th, 85th, 95th, 97th, and 99th) for some key HR-pQCT parameters are presented in Figs. 1 and 2. Figures for all parameters are available in the supplemental material (Supplemental Fig. S2). Peak value, age at peak value, and percentage change from peak to nadir for each parameter are presented in Table 2.

Fig 1.

Reference centile curves built with GAMLSS for some key parameters at the distal radius in males (left column) and females (right column). From top to bottom, curves represent the 97th, 90th, 75th, 50th, 25th, 10th, and 3rd percentiles. Cortical volumetric bone mineral density (Ct.vBMD; A, B), trabecular volumetric bone mineral density (Tb.vBMD; C, D), cortical thickness (Ct.Th; E, F), cortical porosity (Ct.Po; G, H), trabecular number (Tb.N; I, J), trabecular thickness (Tb.Th; K, L), and trabecular separation (Tb.Sp; M, N).

Fig 2.

Reference centile curves built with GAMLSS for some key parameters at the distal tibia in males (left column) and females (right column). From top to bottom, curves represent the 97th, 90th, 75th, 50th, 25th, 10th, and 3rd percentiles. Cortical volumetric bone mineral density (Ct.vBMD; A, B), trabecular volumetric bone mineral density (Tb.vBMD; C, D), cortical thickness (Ct.Th; E, F), cortical porosity (Ct.Po; G, H), trabecular number (Tb.N; I, J), trabecular thickness (Tb.Th; K, L), and trabecular separation (Tb.Sp; M, N).

For cortical parameters at the radius and tibia, an accelerated loss in Ct.vBMD and Ct.Th, along with a rapid increase in Ct.Po was observed after 50 years of age in women. In men, Ct.vBMD and Ct.Th shared similar but more gradual trends. Most cortical bone parameters at the radius peaked before the age of 50 years in men, whereas this peak was observed before the age of 40 years in women. Cortical bone parameters at the tibia peaked much earlier, mostly before the third decade in both men and women. However, almost all trabecular compartment parameters peaked before 20 years old and demonstrated a steady decline thereafter regardless of scanning site and sex. Notable exceptions included Tb.Th, Tb.BV/TV, and Tb.vBMD in women’s radius, whose peak ages ranged from 31 to 36 years. Generally, parameters representing tibial bone structures peaked earlier than their radial counterparts, whereas trabecular bone peaked earlier than cortical bone.

The age-related differences for each HR-pQCT parameter were estimated from fitted models and shown in Fig. 3. Ct.Po varied most dramatically with age regardless of measurement site and sex. In addition to Ct.Po, volumetric bone mineral density parameters also displayed obvious variation across age groups. Generally, based on the fitted models, the age-related percent differences were more pronounced at the radius than tibia (Fig. 4A). Cortical bone showed greater structural decline with increasing age, but the trend was attenuated for trabecular bone. The tibia retained its bone strength more consistently across increasing age strata than the radius in both sexes. Fig. 4B further displays the proportion of the overall difference attributable to specific age strata estimated from the models. Compared with the tibia, the radius suffered more of its overall age-related differences after 50 years of age, whereas compared with males, females generally displayed more structural variations after 50 years of age.

Fig 3.

The maximum percent difference between age groups for each HR-pQCT parameter.

Fig 4.

(A) Average of annual percent differences in HR-pQCT parameters for specific age strata (before age 50, 50 to 65, 65 to 80 years), estimated from the models. (B) Proportion of the overall difference attributable to specific age strata, estimated from the models.

For clinical application and research purposes, the reference centile values (1st, 3rd, 5th, 15th, 25th, 50th, 75th, 85th, 95th, 97th, 99th) and Z-scores (−3.0, −2.5, −2.0, −1.5, −1.0, 0, 1.0, 1.5, 2.0, 2.5, 3.0) for the study population, by 5-year intervals, are provided in Supplemental Tables S3 and S4.

Discussion

This is the first study to establish normative centile curves for HR-pQCT parameters in the healthy Han population in mainland China. Using established statistical models and second-generation HR-pQCT technology, we identified patterns of age-related differences in HR-pQCT parameters between sexes and bone compartments. We believe that this study not only sheds light on the skeletal microstructural features of this population but also provides a foundation for future work focused on elucidating the abnormalities in bone structure that arise under a variety of metabolic disorders or malnutritional conditions.

Similar to prior studies, we found that peak trabecular microarchitecture was acquired before the age of 20 years regardless of measurement site and sex, followed by a steady structural decline.^(9,10,16,17) Across progressively older age groups, we observed increases in Ct.Po, Ct.Pm, Tt.Ar, Tb.Ar, Tb.Sp, Tb.Inhomogeneity and decreases in volumetric BMD, Ct.Ar, Ct.Th, Tb. BV/TV, Tb.N, and Tb.Th in both sexes, indicating gradual “trabecularization” of the cortex.^(33,34) In addition to the endocortical bone loss and compensatory periosteal expansion, the secular difference in overall body/bone size and associated change in measurement site could also partly explain the trends observed for certain geometric HR-pQCT parameters, including Tt.Ar, Tb. Ar, and Ct.Pm. It should be noted that frank cortical bone loss based upon HR-pQCT was not evident until mid-adulthood. This pattern of early loss of trabeculae starting in young adulthood followed by increasing cortical loss after 50 years has also been confirmed in other populations.^{(15,17,35–40)} This may be explained by the accessibility of the trabecular bone compartment to remodeling during young adulthood due to its larger surface area; however, this process becomes self-limited with age as more and more trabeculae are disrupted. Conversely, intracortical porosity increases with age (due to chronic slight negative imbalances at each remodeling unit), gradually providing more bone surface for intracortical bone remodeling to occur.^(34,36,41) Understanding the prevailing drivers of structural changes during different periods of life can help guide the adoption of targeted management strategies for patients at each stage.

Contrary to several previous studies, we did not observe an increase in Tb.N with decreases in Tb.Th and Tb.Sp in young men. For example, Shanbhogue and colleagues and Khosla and colleagues reported that in young men, the radius showed an increase in trabecular number with age together with a decrease in trabecular thickness and separation with no significant changes in trabecular vBMD.^(15,37,42) One hypothesis that has been advanced is that thicker trabeculae are converted into more numerous thinner trabeculae; however, the reasons and mechanisms for this finding ultimately remain unclear. From a technical perspective, the difference in our findings may be attributable to the more accurate measurement of Tb.Th by the second-generation scanner used in our study. The aforementioned studies all employed the first-generation HR-pQCT, and partially resorbed cortical remnants could have been detected as new, thinner trabeculae due to the limited resolution of first-generation HR-pQCT technology. Indeed, trabecular number and trabecular thickness are two measures that demonstrate poor agreement between first- and second-generation HR-pQCT.⁽⁴³⁾ The lack of agreement may also reflect the sensitivity of Tb.Th to spatial resolution and partial volume effects.^(24,44) Another hypothesis is that this finding could also be a result of racial differences. A study from HK using the first-generation HR-pQCT found results similar to ours, suggesting that race may actually play a more prominent role compared with technical aspects of the machine itself.⁽¹⁷⁾

In general, we found that HR-pQCT parameters differed by sex, age period (before versus after 50 years of age), site (tibia versus radius), and bone compartment (trabecular versus cortical bone). After the age of 50 years, greater age-related differences were observed in radial parameters compared with tibial parameters among both men and women. This site-specific discrepancy has also been observed in several longitudinal studies from other countries.^(33,42) Khosla and colleagues attributed this to a possible “threshold effect,” meaning that weight-bearing bones might have a lower sensitivity to bioavailable sex hormones compared with non-weight-bearing bones and thus have a higher threshold to withstand sex-hormone deficiency.⁽¹⁵⁾ We speculate that weight-bearing bones (eg, the tibia) may also be more subject to the effects of direct impact loading and muscle-derived myokines and that the muscle-bone interaction may also help to preserve bone structure. Compared with the site-specific response to sex hormones, the concept of a “threshold effect” has been better established for compartment-specific sex hormone responses. According to Khosla’s theory, the threshold for estrogen deficiency in cortical bone appears to be lower than that in trabecular bone for both women and men.^(45,46) The theory was supported by the finding that estrogen receptor (ER)-α and ER-β, which demonstrate different sensitivities to estrogen exposure, are expressed differently in trabecular versus cortical bone.⁽⁴⁷⁾ The theory also helps to explain the late bone loss observed in the cortical compartment when estrogen levels fall below the postulated threshold for estrogen deficiency in cortical bone. But the hypothesis does not explain the early trabecular bone loss observed among young individuals when levels of sex hormones are still replete.

Although the peak bone microarchitecture, peak age, and age-related differences were derived from established models, the reliability of these results must be evaluated carefully. First, it should be noted that some parameters, such as Tb.Th and Ct. Pm, generally displayed quite flat curves and did not vary significantly with age change. As a result, the peak/nadir levels could be susceptible to the influence of extreme values and outliers. Second, it has been established that genetics accounts for 60% to 80% of adult peak bone mass, with another 20% to 40% explained by lifestyle choices. Calcium and vitamin D intake, micronutrient supplements, alcohol or cigarettes, dietary patterns, as well as physical activity could have significant impact on one’s potential for bone accretion, maintenance, and loss.^(48–50) Unfortunately, because data on nutritional intake was only collected as part of the ChiVOS study, we did not have comprehensive data for the entire sample regarding this variable and therefore could not incorporate it as a covariate in our analyses. Participants in our study were born from the 1930s to 1990s in Chinese mainland, and lifestyle habits in this population have changed tremendously over the past 60 years. Some elderly participants may have suffered from poverty and famine during their youth and thus may have failed to achieve their full potential for bone accrual and development, whereas younger participants may have grown up in period of comparative food security with access to vitamin D and calcium supplements from an early age. Therefore, as with other similar studies adopting a cross-sectional population-based design, the magnitude of “age-related differences” derived from our models could be somewhat overestimated compared with what might be found using a longitudinal cohort design where the same individuals are followed over time. On the other hand, in recent decades, with increasing urbanization and technological development, people may be less likely to engage in manual labor occupations compared with those born when agriculture was still the pillar industry, offsetting some of impact from nutritional status on skeletal development.

This study had several strengths. First, GAMLSS, which is one of the most widely recommended statistical modeling methods for centile curve construction, was adopted in our study, and the model properly addressed the problems of kurtosis and skewness of original data.^(27,29,51) Second, our study is the first to establish a reference database of HR-pQCT parameters for the Chinese mainland population. A total of 610 women were included in the analysis, which is, to date, the largest reference database for women population. Third, we used the second-generation HR-pQCT scanner, which provides markedly improved resolution, enabling us to detect more delicate changes in some resolution-dependent bone parameters such as Tb.Th and Tb.Sp.⁽⁴⁴⁾ The ability to measure these parameters directly, rather than relying on indirect calculations based on algorithms for the first-generation HR-pQCT, enables us to measure trabecular bone structures with greater confidence.^(24,52)

Our study also had some limitations. First, the analysis is based upon cross-sectional data and therefore cannot represent the natural history of bone structural changes over the course of a lifetime. Second, only 250 men were included, and even fewer were older than 65 years. This might subject the models to the influence of extreme values, outliers, and boundary effect. Even for women, our models lack statistical power to generate normative data for those over the age of 80 years because of the limited sample size for this group. Third, the GAMLSS model allowed for only one explanatory variable, so the possible effect of other variables such as height and weight were not included. Finally, because the fixed offset scan position was adopted in our study, difference in limb length and/or bone size could have impacted HR-pQCT measures. And similarly, this could bias the comparisons between male and female groups who have significant differences in height. Whether the height differences might reflect differences in the secular bone size or limb length, thus accounting for the age-related variation in HR-pQCT parameters, or just age-related decline remains unknown. Lastly, this study focused on the Han population in mainland China, and therefore may lack generalizability to other ethnic groups or populations.

In summary, we present the first age-, site-, and sex-specific HR-pQCT normative centile curves for the healthy Han population in mainland China. We demonstrate that in this population, trabecular bone structure peaks during the second decade of life and declines steadily thereafter. However, in the cortical compartment, accelerated bone loss is not observed until about mid-adulthood. This acceleration is concurrent with onset of menopause in women but is also evident in men, suggesting mechanisms beyond simple sex hormone deficiency. Finally, greater age-related differences were observed in radial parameters compared with tibial parameters among both men and women. The data generated from this study provide a critical point of reference for future studies aimed at assessing microstructural alteration of the bone under diverse pathological conditions. Although our study has revealed important cross-sectional associations, additional longitudinal studies are needed to understand how these parameters change over time and temporal associations with key clinical characteristics.