Best Performance Parameters of HR-pQCT to Predict Fragility Fracture: Systematic Review and Meta-Analysis.

Osteoporosis is a systemic skeletal disease characterized by low bone mass and bone structural deterioration that may result in fragility fractures. Use of bone imaging modalities to accurately predict fragility fractures is always an important issue, yet the current gold standard of dual-energy X-ray absorptiometry (DXA) for diagnosis of osteoporosis cannot fully satisfy this purpose. The latest high-resolution peripheral quantitative computed tomography (HR-pQCT) is a three-dimensional (3D) imaging device to measure not only volumetric bone density, but also the bone microarchitecture in a noninvasive manner that may provide a better fracture prediction power. This systematic review and meta-analysis was designed to investigate which HR-pQCT parameters at the distal radius and/or distal tibia could best predict fragility fractures. A systematic literature search was conducted in Embase, PubMed, and Web of Science with relevant keywords by two independent reviewers. Original clinical studies using HR-pQCT to predict fragility fractures with available full text in English were included. Information was extracted from the included studies for further review. In total, 25 articles were included for the systematic review, and 16 articles for meta-analysis. HR-pQCT was shown to significantly predict incident fractures and/or major osteoporotic fractures (MOFs). Of all the HR-pQCT parameters, our meta-analysis revealed that cortical volumetric bone mineral density (Ct.vBMD), trabecular thickness (Tb.Th), and stiffness were better predictors. Meanwhile, HR-pQCT parameters indicated better performance in predicting MOFs than incident fractures. Between the two standard measurement sites of HR-pQCT, the non-weight-bearing distal radius was a more preferable site than distal tibia for fracture prediction. Furthermore, most of the included studies were white-based, whereas very few studies were from Asia or South America. These regions should build up their densitometric databases and conduct related prediction studies. It is expected that HR-pQCT can be used widely for the diagnosis of osteoporosis and prediction of future fragility fractures.

Introduction

With the aging population, age‐associated diseases have become a major concern worldwide. Osteoporosis is a skeletal disease characterized by low bone mass and bone structural deterioration leading to increased risk of fracture.( ) Osteoporosis has no symptoms until an individual encounters a fracture. Currently, the dual energy X‐ray absorptiometry (DXA) is the gold‐standard imaging tool in assessing bone mineral density (BMD) and fracture risk.( ) DXA can measure an integral BMD (including cortical and trabecular BMD) and provide spatial distribution of the bone mass. With the World Health Organization (WHO) definition, a T‐score derived from DXA is clinically useful in identifying patients with osteoporosis (T‐score of −2.5 or below).( ) Previous prospective studies showed that the risk of fracture is inversely related to the areal BMD (aBMD) measured by DXA.( , , ) Despite this, many fragility fractures are found in osteopenic patients who have relatively high aBMD.( ) aBMD values overlap substantially between patients with and without fractures. The fracture prediction power of DXA is therefore suboptimal.

According to the National Institutes of Health (NIH), major parameters for defining osteoporosis is based on both bone mass and structure that are the key determinants of bone strength.( ) Recent advances in bone imaging permit the assessment of bone microstructure in vivo using high‐resolution peripheral quantitative computed tomography (HR‐pQCT). HR‐pQCT is a noninvasive three‐dimensional (3D) imaging technique that quantitatively measures volumetric BMD (vBMD) at the peripheral skeletal sites (distal radius and distal tibia). It provides accurate evaluation of the volumetric bone density in a 3D manner within cortical and trabecular compartments, independent of the bone size. vBMD is of intriguing potential to be used for early detection of osteoporosis and incorporation into the algorithm of fracture prediction. Therefore, HR‐pQCT is of good potential for clinical detection of osteoporosis and an updated guideline of standardizing HR‐pQCT imaging techniques has recently been published.( )

HR‐pQCT has shown a good ability in predicting fracture risk, as reported in a number of studies using HR‐pQCT.( , , , , ) However, these studies reported successful predictions with different and inconsistent HR‐pQCT parameters, such as total vBMD (Tt.BMD), trabecular vBMD (Tb.BMD), cortical thickness (Ct.Th), and trabecular thickness (Tb.Th) in the Os des Femmes de Lyon (OFELY) study( ); Tt.BMD, Tb.BMD, trabecular number (Tb.N,) and trabecular separation (Tb.Sp) in the Canadian Multicentre Osteoporosis Study (CaMos) study( ); Tb.N, Tb.Sp, and connectivity density (Conn.D) in Structure of Aging Men's Bones (STRAMBO) study.( ) HR‐pQCT can generate up to 18 geometric, microstructural, and biomechanical (derived by finite element analysis) parameters. These parameters carry different meanings and implications in various applications, yet they may be difficult for users to understand their underlying impact. It is therefore useful to understand these parameters and compare their strength in fracture risk prediction. The objective of this systematic review and meta‐analysis is to investigate which HR‐pQCT bone quality parameters at distal radius and/or distal tibia can best predict fragility fractures in various ethnic populations.

Materials and Methods

Search strategy

A literature search was performed in Embase, PubMed, and Web of Science databases. The keywords “HR‐pQCT OR high resolution peripheral quantitative computed tomography OR XtremeCT” AND “fracture*” were used to search in all fields. Last access to these databases was on January 31, 2021. The Preferred Reporting Items for Systematic reviews and Meta‐Analyses (PRISMA) guidelines were followed.

Search criteria

Inclusion criteria were as follows: (i) clinical studies, (ii) subjects aged ≥50 years or postmenopausal, (iii) fragility fractures, ie, those resulting from falling from standing height.( ) Exclusion criteria were as follows: (i) non‐English articles, (ii) review, conference abstract, case report, protocol article, (iii) involving diseases affecting bone metabolism, (iv) nonstandard HR‐pQCT parameters or evaluation, (v) precision or validation or machine learning studies, and (vi) fractures of skull, face, fingers, and toes.

Selection of studies

Study selection was conducted by two independent reviewers. Duplicates were removed. Irrelevant papers were screened out through the titles and abstracts based on inclusion and exclusion criteria. Full text of relevant articles was retrieved and reviewed for the eligibility. Disagreements were resolved by discussion and consensus.

Data extraction

The following information was extracted by reviewers: study design, ethnicity, sex, age, sample size, trauma degree, fragility fracture site, fracture prediction results (odds ratio [OR], hazard ratio [HR]), model of HR‐pQCT, site of measurement, and data of HR‐pQCT–related parameters.

Assessment of quality of included studies

Two authors independently conducted quality assessment of the selected studies. Disagreements were settled by discussion. The methodological quality was assessed using the Newcastle‐Ottawa Scale validated for observational studies, for which separate tools are developed for cohort and case control studies.( ) The form assigns a maximum of four points for selection, two points for comparability, and three points for exposure or outcomes.

Data analysis

A meta‐analysis on peripheral bone microarchitecture parameters and estimated bone strength in patients with fragility fracture was performed by computing risk ratios (RRs) and 95% confidence intervals (CIs) using a fixed‐effects model. Quantitative analyses were performed on a time‐to‐event basis and were confined to data derived from the period of follow‐up. Data analysis was carried out based on (i) bone microarchitecture parameters and estimated bone strength and (ii) study design (by cohort studies or case–control studies), followed by an overall effect by combining the outcomes from both study types. Data heterogeneity was represented using chi‐square test and heterogeneity (I 2) estimate. Test for overall effect was performed using Z‐statistics. Subgroup analysis was tested for the study types (ie, cohort versus case control studies). Publication bias was analyzed using Egger's test. Publication bias was indicated when p < 0.05. The Egger's test was carried out by Meta‐Essentials (Erasmus Research Institute of Management [ERIM], Rotterdam, Netherlands) and the corresponding funnel plots were prepared by RevMan software (version 5.4; The Cochrane Collaboration, London, UK). RevMan was used for all analyses and production of plots. The inverse variance (IV) method was used to weight the study effect size. Test of contrast comparing the effect sizes of the three best performance HR‐pQCT parameters were carried out using Cohen's d, which was derived through the effect size calculations for t test in IBM SPSS version 27 (IBM Corp, Armonk, NY, USA). In brief, Cohen's d point estimates of ~0.2 are regarded as small effects, values ~0.5 as medium‐sized effects, and those ≥0.8 as large effects.( )

Results

Search results

In the search, 1941 papers were identified from various databases. After removing duplicates, 695 papers remained, from which 544 papers were excluded based on the selection criteria after screening. The full texts of 151 papers were retrieved for further assessment of eligibility; 126 papers were further excluded, because these studies did not involve fracture discrimination or prediction. Hence, 25 papers were finally included in this systematic review including 18 case–control studies( , , , , , , , , , , , , , , , , , ) and seven cohort studies.( , , , , , , ) Out of 25 papers, only 16 papers were included for meta‐analysis, in which there were 11 case–control studies( , , , , , , , , , , ) and five cohort studies.( , , , , ) Figure 1 summarizes the selection process of the included papers.

Fig 1

Flowchart showing the selection process of literature search.

Characteristics of the papers

Among 25 included papers, the sample size ranged from 48 to 1794 participants, whereas one study involved up to 7254 participants because it grouped the participants from eight cohorts for analysis.( ) Some studies used the participants from the same cohort for different analyses; eg, five studies from OFELY (France),( , , , , ) four studies from MrOS (The Osteoporotic Fractures in Men Study, USA or Sweden),( , , , ) two studies from STRAMBO (France),( , ) two studies from CaMOS (Canada),( , ) one study from Hertfordshire Cohort Study (HCS, UK),( ) one study from Geneva Retirees Cohort (GERICO, Switzerland), ) and one study from Global Longitudinal Study of Osteoporosis in Women (GLOW, UK).( ) One study assessed eight cohorts (Framingham, Mayo Clinic, Qualité Osseuse Lyon Orléans [QUALYOR, France], STRAMBO, OFELY, GERICO, CaMos, and MrOS).( )

The mean age of the participants in the 25 studies was 65 to 85 years. Nine studies analyzed the prediction for incident fractures,( , , , , , , , , ) eight studies for both incident and major osteoporotic fractures,( , , , , , , , ) five studies for vertebral fractures only,( , , , , ) two studies for hip fractures only,( , ) and one for wrist fracture only.( ) Table 1 summarizes the included 25 studies.

Table 1

Summary of the Included Studies in the Systematic Review

Study design	Included studies	Cohort	Ethnicity	Follow‐up	Sex/age	Sample size	Trauma degree	Fracture site	Fracture prediction
Cohort a	Biver and colleagues( 36 )	GERICO study	Swiss	5.0 ± 1.8 years	Women Age: 65.0 ± 1.5 years	Women: 740	Low‐trauma only (n = 85); 47% had prior low trauma fracture after age 20 years	Incident fractures – exclude fingers, toes, skull, face Subgroup analysis in major osteoporotic fracture	HR
Cohort a	Burt and colleagues( 11 )	CaMOS Cohort	Canadian	5.0 ± 0.5 years	Women Age: 68.4 ± 7.0 years	Women: 149	Fragility fracture (n = 22); 45% had previous fracture	Incident fractures ‐ exclude fingers, toes, nose	OR
Cohort	Ohlsson and colleagues( 37 )	MrOS Sweden Cohort	Swedish	5.3 ± 2.0 years	Men Age: 80.2 ± 3.5 years	Men: 456	Fragility fracture (n = 71)	Incident fractures exclude skull, face Subgroup analysis in major osteoporotic fracture	HR; no crude data
Cohort a	Langsetmo and colleagues( 12 )	MrOS US, at year 14 visit	American	Mean: 1.7 years (IQR 1.4–2.3, max. 2.9)	Men Age: 84.4 ± 4.2 years	Men: 1794	Regardless of site and trauma (n = 127); 33% had prior fracture after age 50 years	All incident fractures Subgroup analysis in major osteoporotic fracture	HR
Cohort	Samelson and colleagues( 14 )	8 Cohorts	Multiple (American, Canadian, European)	4.63 ± 2.41 years (range 0.01–11.04)	Men Age: 67 ± 9 years Women Age: 72 ± 9 years	Men: 2486 Women: 4768	Regardless of site and trauma (n = 765); 23% had previous fracture	All incident fractures	HR (repeated cohorts)
Cohort a	Sornay‐Rendu and colleagues( 35 )	OFELY study at Year 14 visit	French	9.4 years (IQ:1.0)	Women Age: 68 ± 9.0 years	Women: 589	Low‐trauma only (n = 135); 32% had prior low trauma fracture after age 40 years	Incident fractures exclude head, toes, fingers Subgroup analysis in major osteoporotic fracture	HR
Cohort a	Szulc and colleagues( 13 )	STRAMBO study	French	8 years (IQR: 6.0–8.0)	Men Age: 72.5 years	Men: 819	Fragility fracture (n = 105); 23% had previous fracture	Incident fractures – exclude skull, face, hand, fingers, toes Subgroup analysis in major osteoporotic fracture	HR
Case control a	Boutroy and colleagues( 18 )	OFELY cohort at Year 13 visit	French	N/A	Women Age: 73.4 ± 6.0 years	Case: 33 Control: 33	Low‐trauma only	Wrist fracture only	OR
Case control	Boutroy and colleagues( 30 )	Multicenter analysis	Multiple (American, European, Asian, Argentines)	N/A	Women Age: 66 ± 8 years	Case: 470 Control: 909	Low‐trauma only	All incident fractures Subgroup analysis in major osteoporotic fracture	OR; no crude data
Case control a	Edwards and colleagues( 25 )	HCS cohort‐UK at follow‐up	British	N/A	Men Age: 75.8 ± 2.4 years Women Age: 76.4 ± 2.5 years	Case: 44(M); 48(F) Control: 133(M); 111(F)	No information	All incident fractures	OR
Case control	Fink and colleagues( 32 )	MrOS US, at Year 14 visit	American	N/A	Men Age: 84.4 ± 4.2 years	Case: 344 Control: 1450	Regardless of site and trauma	All incident fractures	OR; no crude data
Case control	Johansson and colleagues( 31 )	Gothenburg Swedish Women Cohort	Swedish	N/A	Women Age: 77 ± 1.5 years	Case: 277 Control: 750	By VFA	Vertebral fracture only	OR; no crude data
Case control	Litwic and colleagues( 34 )	GLOW study at Year 5	British	N/A	Women Age: 70.6 ± 5.4 years	Case: 63 Control: 258	No information	All incident fractures	Cluster analysis; no crude data
Case control	Melton and colleagues( 28 )	Mayo Clinic	American	N/A	Women Age: 74.8 ± 7.9 years	Case:40 Control:40	Minimal or moderate trauma	Vertebral fracture only	OR; no crude data
Case control	Melton and colleagues( 29 )	Mayo Clinic	American	N/A	Women Age: 69.6 ± 9.9 years	Case:193 Control: 90	Minimal or moderate trauma	Vertebral fracture only	OR; no crude data
Case control a	Sornay‐Rendu and colleagues( 10 )	OFELY study at Year 13 visit	French	N/A	Women Age: 73.7 ± 8.0 years	Case: 101 Control: 101	Low‐trauma only	Incident fracture exclude head, toes, fingers Subgroup analysis in wrist fracture	OR
Case control a	Sornay‐Rendu and colleagues( 19 )	Case: 30% from OFELY; 70% new admitted Control: OFELY study at Year 14 visit	French	N/A	Women Age: 74 ± 9 years	Case: 100 Control: 362	Low‐trauma only	Vertebral fracture only	OR
Case control a	Stein and colleagues( 21 )		American	N/A	Women Age: 68 ± 7 years	Case: 68 Control:101	Low‐trauma only	All incident fractures	OR
Case control a	Sundh and colleagues( 24 )	MrOS‐Sweden Gothenburg cohort at Year 5	Swedish	N/A	Men Age: 80.2 ± 3.5 years	Case: 87 Control: 369	No information	Incident fracture exclude hand, finger, foot, toe, skull Subgroup analysis in peripheral fracture, osteoporotic fracture, multiple fractures	OR
Case control a	Sundh and colleagues( 27 )	Gothenburg Swedish Women Cohort	Swedish	N/A	Women Age: 77.7 ± 1.6 years	Case: 46 Control: 361	No information	Hip fracture only	OR
Case control	Torres and colleagues( 33 )		Brazilian	N/A	Men Age: 79.0 ± 4.1 years Women Age: 79.6 ± 4.1 years	Case: 28(M); 75(F) Control: 72(M); 101(F)	By VFA	Vertebral fracture only	OR; no crude data
Case control a	Vilayphiou and colleagues( 20 )	OFELY cohort at Year 13 visit	French	N/A	Women Age: 73.7 ± 8.1 years	Case: 101 Control: 101	Low‐trauma only	Incident fracture‐ exclude head, toes, fingers	OR
Case control a	Vilayphiou and colleagues( 22 )	STRABMO cohort	French	N/A	Men Age: 71 ± 10 years	Case:185 Control: 185	Low‐trauma only	Incident fracture‐ exclude head, toes, fingers	OR
Case control a	Wong and colleagues( 23 )	CaMOS cohort	Canadian	N/A	Women Age: 75 ± 9 years	Case: 40 Control: 57	Low‐trauma only	Incident fracture exclude skull, fingers, toes	OR
Case control a	Zhu and colleagues( 26 )		Asian	N/A	Women Age: 79.2 ± 5.5 years	Case: 24 Control: 24	Low‐trauma only	Hip fracture only	OR

HR = hazard ratio; OR = odds ratio; VFA = vertebral fracture assessment.

Paper included in meta‐analysis.

Quality of selected studies

Tables 2 and 3 summarizes the quality of the 25 studies using the Newcastle‐Ottawa Quality Scale. All studies were of good quality and suitable for quantitative analysis.

Table 2

Quality Assessment Using Newcastle‐Ottawa Scale for the Case Control Studies

	Selection				Comparability	Exposure
Cross‐sectional studies	Adequate definition of cases	Representativeness of cases	Selection of controls	Definition of controls	Control for important factor or additional factor	Ascertainment of exposure	Same method of ascertainment for cases and controls	Nonresponse rate	Total score
Boutroy and colleagues( 18 )	*	*	*		**	*	*		7
Edwards and colleagues( 25 )			*	*	**	*	*		6
Sornay‐Rendu and colleagues( 10 )	*	*		*	**	*	*	*	8
Sornay‐Rendu and colleagues( 19 )	*		*	*	*	*	*		6
Stein and colleagues( 21 )	*	*	*	*	**	*	*		8
Sundh and colleagues( 24 )		*	*	*	**	*	*		7
Sundh and colleagues( 27 )	*		*	*	**	*	*		7
Vilayphiou and colleagues( 20 )	*	*	*		**	*	*		7
Vilayphiou and colleagues( 22 )	*		*	*	**	*	*	*	8
Wong and colleagues( 23 )		*	*	*	*	*	*		6
Zhu and colleagues( 26 )	*	*		*	**	*	*	*	8

* Each star represents 1 score.

Table 3

Quality Assessment Using Newcastle‐Ottawa Scale for the Cohort Studies

	Selection				Comparability	Exposure
Cohort studies	Representativeness of the exposed cohort	Selection of the non‐exposed cohort	Ascertainment of exposure	Demonstration that outcome of interest was not present at start of study	Control for important factor or additional factor	Ascertainment of outcome	Was follow‐up long enough for outcome to occur	Adequacy of follow‐up of cohorts	Total score
Biver and colleagues( 36 )	*	*	*	*	**	*	*	*	9
Burt and colleagues( 11 )	*	*	*	*	**	*	*	*	9
Langsetmo and colleagues( 12 )		*	*	*	**	*	*	*	8
Sornay‐Rendu and colleagues( 35 )		*	*	*	**	*	*	*	8
Szulc and colleagues( 13 )	*	*		*	**	*	*	*	8

* Each star represents 1 score.

*

Study design

Among 25 included studies, 18 papers were case–control studies( , , , , , , , , , , , , , , , , , ) and seven were cohort studies.( , , , , , , ) In the case–control studies, the fracture participants mostly had low or moderate trauma, whereas four studies did not specify the trauma degree of participants.( , , , ) In the cohort studies, the fracture participants were also of low trauma and 23% to 47% of participants had previous fractures (Table 1).

Gender and ethnicity

Sixteen of 25 studies included women only,( , , , , , , , , , , , , , , , ) six studies men only,( , , , , , ) and three studies both genders.( , , ) For ethnicity, the 25 studies included participants as follows: seven French,( , , , , , , ) five American,( , , , , ) four Swedish,( , , , ) two Canadian,( , ) two British,( , ) two multiple,( , ) one Swiss,( ) one Asian,( ) and one Brazilian.( ) In summary, seven were from North America, one from South America, 14 from Europe, one from Asia, and two from multiple countries (Table 1).

Follow‐up period

In the seven cohort studies included, the mean follow‐up period ranged from 1.7 to 9.4 years. There were two studies with mean follow‐up time of less than 5 years( , ) and three studies with up to around 5 years.( , , ) Two studies followed the patients up to a mean period of 8 years( ) and 9.4 years,( ) respectively.

HR‐pQCT machine

Almost all (22/25) studies used Scanco HR‐pQCT (XtremeCT version I, XT I; SCANCO Medical AG, Brüttisellen, Switzerland) machines. Two studies used Scanco XtremeCT version II (XT II) model( , ); one study used a Scanco prototype machine.( ) Among the 16 studies in meta‐analysis, all of them used XT I, except a cohort study by Langsetmo and colleagues( ) using XT II (Table 4). According to Scanco information (http://www.scanco.ch/), XT I and XT II are compatible but their major differences include resolution (82 μm in XT I versus 61 μm in XT II), scanning time (3 minutes in XT I versus 2 minutes in XT II). For the prototype machine, the spatial resolution was at 165 μm.( ) In Melton and colleagues' study( ) using a prototype machine, only microstructural parameters were reported, whereas volumetric density and related finite element analysis were measured by a single‐energy QCT and a third‐party software (ON Diagnostics, Berkeley, CA, USA), respectively.

Table 4

Summary of HR‐pQCT Details of the Included Studies

Study design	Included studies	Model of HR‐pQCT	Site(s)	Volumetric density parameter (mg HA/cm3)	Bone geometry (mm2)	Trabecular microarchitecture	Cortical structural parameters	μFEA	μFEA inputs
Cohort a	Biver and colleagues( 36 )	XtremeCT I	Tibia, radius	Tt.BMD Tb.BMD Ct.BMD	Tt.Ar Tb.Ar Ct.Ar	Tb.N (mm−1) Tb.Th (mm) Tb.Sp (mm) Tb.Sp.SD (mm)	Ct.Th (mm) Ct.Po (%)	Stiffness (N/mm) Est. failure load (N) Modulus (N/mm2) [Scanco IPL]	Elastic modulus of 10 GPa; Uniaxial compression; Yield criterion 0.7% critical strain, 2% critical volume
Cohort a	Burt and colleagues( 11 )	XtremeCT I	Tibia, radius	Tt.BMD Tb.BMD Ct.BMD	Tt.Ar Tb.Ar Ct.Ar	Tb.N (mm−1) Tb.Th (mm) Tb.Sp (mm)	Ct.Th (mm) Ct.Po (%)	Est. failure load (N) [Numerics88]	Elastic modulus of 6.829 GPa; Uniaxial compression; Yield criterion 0.7% critical strain, 2% critical volume
Cohort	Ohlsson and colleagues( 37 )	XtremeCT I	Tibia	Ct.BMD	Ct.Ar	BV/TV(%) Tb.N (mm−1) Tb.Th (mm) Tb.Sp (mm)	Ct.Th (mm) Ct Pm (mm) Ct.Po (%)	Stiffness (N/mm) Est. failure load (N) [Scanco IPL]	Elastic modulus of 10 GPa; Uniaxial compression; Yield criterion 0.7% critical strain, 2% critical volume
Cohort a	Langsetmo and colleagues( 12 )	XtremeCT II	Tibia, diaphyseal tibia, radius	Tt.BMD Tb.BMD Ct.BMD	Tb.Ar Ct.Ar	Tb.N (mm−1) Tb.Th (mm)	Ct.Th (mm) Ct.Po (%)	Est. failure load (N) [Scanco IPL]	Elastic modulus of 10 GPa; Uniaxial compression; Yield criterion 0.7% critical strain, 7.5% critical volume
Cohort	Samelson and colleagues( 14 )	XtremeCT I	Tibia, radius	Tt.BMD Tb.BMD Ct.BMD	Ct.Ar/Tt.Ar	Tb.N (mm−1) Tb.Th (mm) Tb.Sp (mm)	Ct.Th (mm) Ct.Po (%)	Est. failure load (N) [Scanco IPL]	Elastic modulus of 6.829 GPa; Axial compression; Yield criterion 0.7% critical strain, 2% critical volume
Cohort a	Sornay‐Rendu and colleagues( 35 )	XtremeCT I	Tibia, radius	Tt.BMD Tb.BMD Ct.BMD	Tt.Ar Tb.Ar Ct.Ar	Tb.N (mm−1) Tb.Sp.SD (mm) Conn.D SMI	Ct.Th (mm) Ct.Po (%)	Stiffness (N/mm) Est. failure load (N) [Scanco IPL]	Elastic modulus of 20 GPa (cortical bone) and 17.5 GPa (trabecular bone); Axial compression; Yield criterion 0.35% critical strain, 2% critical volume
Cohort a	Szulc and colleagues( 13 )	XtremeCT I	Tibia, radius	Tt.BMD Tb.BMD Ct.BMD	—	Tb.N (mm−1) Tb.Th (mm) Tb.Sp (mm) Tb.Sp.SD (mm) Conn.D		Stiffness (N/mm) Est. failure load (N) [Scanco IPL]	Elastic modulus of 20GPa (cortical bone) and 17.5GPa (trabecular bone); Axial compression; Yield criterion 0.35% critical strain, 2% critical volume
Case control a	Boutroy and colleagues( 18 )	XtremeCT I	Radius	Tt.BMD Tb.BMD Ct.BMD	Tt.Ar	Tb.N (mm−1) Tb.Th (mm) Tb.Sp (mm) Tb.Sp.SD (mm)	Ct.Th (mm)	Stiffness (N/mm) Est. failure load (N) % load trab distal % load trab proximal Tb Von Mises stress Ct Von Mises stress [Scanco IPL]	Elastic modulus of 20 GPa (cortical bone) and 17.5 GPa (trabecular bone); Axial compression; Yield criterion 0.7% critical strain, 2% critical volume
Case control	Boutroy and colleagues( 30 )	XtremeCT I	Tibia, radius	Tb.BMD Ct.BMD	Tt.Ar Tb.Ar Ct.Ar	Tb.N (mm−1) Tb.Th (mm) Tb.Sp (mm) Tb.Sp.SD (mm)	Ct.Th (mm)	—	—
Case control a	Edwards and colleagues( 25 )	XtremeCT I	Radius	Tb.BMD Ct.BMD	Tt.Ar Tb.Ar Ct.Ar	Tb.N (mm−1) Tb.Th (mm) Tb.Sp (mm)	Ct.Th (mm) Ct.Po (%)	—	—
Case control	Fink and colleagues( 32 )	XtremeCT II	Tibia, diaphyseal tibia, radius	Tt.BMD Tb.BMD Ct.BMD	Tt.Ar Tb.Ar Ct.Ar	Tb.N (mm−1) Tb.Th (mm)	Ct.Th (mm) Ct.Po (%)	Est. failure load (N) [Scanco IPL]	Elastic modulus of 10 GPa; Uniaxial compression; Yield criterion 0.7% critical strain, 1% critical volume
Case control	Johansson and colleagues( 31 )	XtremeCT I	Tibia, 14% tibia, radius, 14% radius	Tt.BMD Ct.BMD	Ct.Ar	BV/TV(%) Tb.N (mm−1) Tb.Th (mm) Tb.Sp (mm)	Ct.Th (mm) Ct.Po (%)	—	—
Case Control	Litwic and colleagues( 34 )	XtremeCT I	Tibia, radius	Tb.BMD Ct.BMD	Tt.Ar Tb.Ar Ct.Ar	Tb.N (mm−1) Tb.Th (mm) Tb.Sp (mm)	Ct.Th (mm) Ct.Po (%)	—	—
Case control	Melton and colleagues( 28 )	XtremeCT prototype	Radius	—	—	BV/TV(%) Tb.N (mm−1) Tb.Th (mm) Tb.Sp (mm) Tb.Sp.SD (mm)	—	—	—
Case control	Melton and colleagues( 29 )	XtremeCT I	Radius	Tb.BMD Ct.BMD	Ct.Ar	Tb.N (mm−1) Tb.Th (mm) Tb.Sp (mm) Tb.Sp.SD (mm) Conn.D	Ct.Th (mm)	—	—
Case control a	Sornay‐Rendu and colleagues( 10 )	XtremeCT I	Tibia, radius	Tt.BMD Tb.BMD Ct.BMD	—	BV/TV(%) Tb.N (mm−1) Tb.Th (mm) Tb.Sp (mm) Tb.Sp.SD (mm)	Ct.Th (mm)	—	—
Case control a	Sornay‐Rendu and colleagues( 19 )	XtremeCT I	Tibia, radius	Tt.BMD Tb.BMD Ct.BMD	—	BV/TV(%) Tb.N (mm−1) Tb.Th (mm) Tb.Sp (mm) Tb.Sp.SD (mm)	Ct.Th (mm)	—	—
Case control a	Stein and colleagues( 21 )	XtremeCT I	Tibia, radius	Tt.BMD Tb.BMD Ct.BMD	Tt.Ar	Tb.N (mm−1) Tb.Th (mm) Tb.Sp (mm) Tb.Sp.SD (mm)	Ct.Th (mm)	Stiffness (N/mm) [Scanco IPL]	Elastic modulus of 15 GPa; Uniaxial compression; Yield criterion 0.7% critical strain, 2% critical volume
Case control a	Sundh and colleagues( 24 )	XtremeCT I	Tibia	Ct.BMD	Ct.Ar	BV/TV(%) Tb.N (mm−1) Tb.Th (mm) Tb.Sp (mm)	Ct.Th (mm) Ct.Po (%) Ct.Po Dm (mm)	—	—
Case control a	Sundh and colleagues( 27 )	XtremeCT I	Tibia, 14% tibia	Ct.BMD	Ct.Ar	BV/TV(%) Tb.N (mm−1) Tb.Th (mm)	Ct.Th (mm) Ct.Po (%)	—	—
Case control	Torres and colleagues( 33 )	XtremeCT I	Tibia, radius	Tt.BMD Tb.BMD Ct.BMD	—	Tb.N (mm−1) Tb.Th (mm) Tb.Sp (mm)	Ct.Th (mm) Ct.Po (%) Ct.Po Dm (mm)	Stiffness (N/mm) Est. failure load (N) [Scanco IPL]	Elastic modulus of 10 GPa; Uniaxial compression; Yield criterion 0.7% critical strain, 2% critical volume
Case control a	Vilayphiou and colleagues( 20 )	XtremeCT I	Tibia, radius	Tt.BMD Tb.BMD	Tt.Ar	Tb.N (mm−1) Tb.Sp.SD (mm)	Ct.Th (mm)	Stiffness (N/mm) Est. failure load (N) % load trab distal % load trab proximal Tb Von Mises stress Ct Von Mises stress [Scanco IPL]	Elastic modulus of 20 GPa (cortical bone) and 17.5 GPa (trabecular bone); Axial compression; Yield criterion 0.35% critical strain, 2% critical volume
Case control a	Vilayphiou and colleagues( 22 )	XtremeCT I	Tibia, radius	Tt.BMD Tb.BMD Ct.BMD	Tt.Ar	Tb.N (mm−1) Tb.Sp (mm) Tb.Sp.SD (mm)	Ct.Th (mm)	Stiffness (N/mm) Est. failure load (N) % load trab distal % load trab proximal Tb Von Mises stress Ct Von Mises stress [Scanco IPL]	Elastic modulus of 20 GPa (cortical bone) and 17.5 GPa (trabecular bone); Axial compression; Yield criterion 0.35% critical strain, 2% critical volume
Case control a	Wong and colleagues( 23 )	XtremeCT I	Tibia, radius	Tt.BMD Tb.BMD Ct.BMD	—	BV/TV(%) Tb.N (mm−1) Tb.Th (mm) Tb.Sp (mm)	Ct.Th (mm)	—	—
Case control a	Zhu and colleagues( 26 )	XtremeCT I	Tibia, radius	Tt.BMD Tb.BMD Ct.BMD	Tt.Ar Ct.Ar	Tb.N (mm−1) Tb.Th (mm) Tb.Sp (mm) Tb.Sp.SD (mm)	Ct.Th (mm) Ct.Po (%)	Stiffness (N/mm) Est. failure load (N) [Scanco IPL]	Elastic modulus of 10 GPa; Uniaxial compression; Yield criterion 0.7% critical strain, 2% critical volume

Paper included in meta‐analysis.

Due to the differences in spatial resolution, the image processing method is also different between XTI and XTII machines. Higher spatial resolution in XTII allows direct measurement of trabecular microarchitecture (bone volume/total volume [BV/TV], Tb.Th, Tb.Sp) and cortical thickness (Ct.Th), whereas a derived (indirect) morphological method of evaluation was used in XTI. The image analysis methods for measuring bone area, vBMD and Tb.N are the same in both generations of HR‐pQCT machine.( , , ) Fully automated threshold‐based approach for cortical porosity (Ct.Po) evaluation, as one of the standard parameters, was used in XTII. In XTI, extended cortical analysis by semiautomated contouring process for Ct.Po evaluation (threshold‐based approach) was required.( )

Scanning position

Most of the studies (18/25) measured both distal tibia and distal radius using HR‐pQCT; three measured distal tibia only( , , ) and four measured distal radius only.( , , , ) All of them used standard protocol (fixed offset scanning) to measure the extremities, whereas four studies also measured diaphyseal region in addition to the standard sites( , , , ) (Table 4). Regarding the standard scanning protocol, the total scanning region was 9.02, 10.2, and 10 mm (in length) in XT I, XT II, and prototype machine, respectively. For XT I, the first CT slice was 9.5 mm (for radius) and 22.5 mm (for tibia) proximal to the reference line (endplate). For XT II, the first CT slice was 9.0 mm (for radius) and 22.0 mm (for tibia) proximal to the reference line (endplate).( ) For the prototype HR‐pQCT machine, the first CT slice was 6.0 mm proximal to the reference line (endplate) for distal radius only.( )

Finite element analysis

In the 25 included studies, all of them measured volumetric density, geometric and/or structural parameters, whereas only 14 studies reported estimated bone strength generated by micro‐finite element analysis (μFEA).( , , , , , , , , , , , , , ) Of the 14 studies, almost all of them used Scanco default Image Processing Language (IPL) for μFEA analysis, except one using a third‐party finite element software (FAIM, version 6; Numerics88 Solutions, Calgary, AB, Canada).( ) μFEA model inputs are summarized in Table 4.

Meta‐analysis

In the meta‐analysis, only studies with crude OR or HR results available or provided by the authors are included for analysis. Nine of 25 studies were excluded from the analysis, because no response was received for our request for crude data acquisition( , , , , , , , ) or they involved repeated cohorts.( ) Although a few studies adopted the patients from the same cohort, the 16 included studies should have no overlapping of database, because they involved different fracture types, patient groups, or ethnicities. In the meta‐analysis, the included studies have two types of study design: cohort and case–control. A total of 56 analyses were performed in the meta‐analysis to analyze 14 HR‐pQCT parameters, covering vBMD, cortical and trabecular microarchitecture, and FEA parameters, involving two scanning sites (distal radius and distal tibia) for predicting both incident fracture and major osteoporotic fracture (MOF).

There were 16 studies included in the meta‐analysis (five cohort studies and 11 case control studies). The total number of subjects for 16 studies was 6904 (female, n = 3288; male, n = 3616). The total number of incident fracture was 1392, in which 709 was MOF. Publication bias was not present in all parameters (all p > 0.05), except Ct.Th at both distal radius (p = 0.02), tibia (p = 0.01), and BV/TV at distal tibia (p = 0.04) (Supplementary Table S1). Therefore, the risk of bias from the selected publications was minimal.

For selection of the best performance HR‐pQCT parameters in predicting incident fracture and MOF, the meta‐analysis results should meet the following criteria: (i) risk ratio or effect size >1.0 (except cortical porosity and trabecular separation <1.0), (ii) I 2 estimate <65%, (iii) generalizability in predicting both incident fractures and MOFs, and (iv) their successful prediction at both distal radius and tibia.

The following highlights the meta‐analysis results with effect size over 1.0 (except cortical porosity and trabecular separation <1.0) with low‐to‐moderate heterogeneity (I 2 < 65%) in predicting fractures. The parameters not meeting these criteria are summarized in Supplementary Table S2.

Meta‐analysis of incident fracture prediction

vBMD

Cortical vBMD measured at both radius and tibia could significantly predict incident fracture (radius: RR = 1.50, 95% CI = 1.39–1.61, I 2 = 18%, p < 0.01; tibia: RR = 1.64, 95% CI = 1.52–1.76, I 2 = 43%, p < 0.01) (Fig. 2). Trabecular vBMD measured at radius only showed significant prediction (RR = 1.81, 95% CI = 1.68–1.96, I 2 = 60%, p < 0.01) (Supplementary Fig. S1).

Fig 2

Forest plot of Ct.vBMD measured at distal radius and distal tibia for predicting incident fracture and MOF, (A) distal radius; incident fracture, (B) distal tibia; incident fracture, (C) distal radius; MOF, and (D) distal tibia; MOF. Ct.vBMD = cortical volumetric bone mineral density; MOF = major osteoporotic fracture.

Cortical parameters

Ct.Th measured at radius only could significantly predict incident fracture (RR = 1.64, 95% CI = 1.51–1.78, I 2 = 42%, p < 0.01) (Supplementary Fig. S2).

Trabecular bone microarchitecture parameters

Tb.N measured at both radius and tibia could predict incident fracture significantly (radius: RR = 1.68, 95% CI = 1.56–1.82, I 2 = 62%, p < 0.01; tibia: RR = 1.39, 95% CI = 1.29–1.49, I 2 = 61%, p < 0.01) (Supplementary Fig. S3). Tb.Th measured at both radius and tibia could predict incident fracture significantly (radius: RR = 1.37, 95% CI = 1.25–1.49, I 2 = 0%, p < 0.01; tibia: RR = 1.30, 95% CI = 1.19–1.42, I 2 = 15%, p < 0.01) (Fig. 3). Tb.Sp measured at radius only could predict incident fracture significantly (RR = 0.63, 95% CI = 0.59–0.68, I 2 = 60%, p < 0.01) (Supplementary Fig. S4).

Fig 3

Forest plot of Tb.Th measured at distal radius and distal tibia for predicting incident fracture and MOF, (A) distal radius; incident fracture, (B) distal tibia; incident fracture, (C) distal radius; MOF, and (D) distal tibia; MOF. MOF = major osteoporotic fracture; Tb.Th = trabecular thickness.

Fig 4

Forest plot of stiffness measured at distal radius and distal tibia for predicting incident fracture and MOF, (A) distal radius; incident fracture, (B) distal tibia; incident fracture, (C) distal radius; MOF, (D) distal tibia; MOF. MOF = major osteoporotic fracture.

Finite element analysis parameters

Stiffness measured at both radius and tibia could predict incident fracture significantly (radius: RR = 1.67, 95% CI = 1.53–1.83, I 2 = 30%, p < 0.01; tibia: RR = 1.71, 95% CI = 1.54–1.89, I 2 = 50%, p < 0.01) (Fig. 4). Estimated failure load measured at both radius and tibia could predict incident fracture significantly (radius: RR = 1.72, 95% CI = 1.57–1.88, I 2 = 44%, p < 0.01; tibia: RR = 1.76, 95% CI = 1.60–1.93, I 2 = 0%, p < 0.01) (Supplementary Fig. Supplementary Fig. S5 A,B).

Meta‐analysis of MOF prediction

Cortical vBMD measured at radius and tibia could significantly predict MOF (radius: RR = 1.65, 95% CI = 1.46–1.86, I 2 = 0%, p < 0.01; tibia: RR = 1.80, 95% CI = 1.63–2.00, I 2 = 34%, p < 0.01) (Fig. 2). Trabecular vBMD measured at radius only showed significant prediction (RR = 2.24, 95% CI = 1.95–2.57, I 2 = 42%, p < 0.01) (Supplementary Fig. S1 C).

Ct.Th measured at radius only could significantly predict MOF (RR = 1.77, 95% CI = 1.55–2.00, I 2 = 62%, p < 0.01) (Supplementary Fig. S2 C). Only cortical porosity measured at tibia could predict MOF significantly (RR = 0.59, 95% CI = 0.51–0.68, I 2 = 57%, p < 0.01) (Supplementary Fig. Supplementary Fig. S6).

Tb.N measured at radius only could predict MOF significantly (RR = 2.10, 95% CI = 1.86–2.37, I 2 = 50%, p < 0.01) (Supplementary Fig. S3). Tb.Th measured at both radius and tibia could predict MOF significantly (radius: RR = 1.31, 95% CI = 1.06–1.62, I 2 = 0%, p < 0.01; tibia: RR = 1.31, 95% CI = 1.12–1.53, I 2 = 0%, p = 0.01) (Fig. 3).

Stiffness measured at both radius and tibia could predict MOF significantly (radius: RR = 1.85, 95% CI = 1.60–2.15, I 2 = 0%, p < 0.01; tibia: RR = 1.90, 95% CI = 1.62–2.23, I 2 = 0%, p < 0.01) (Fig. 4). Estimated failure load measured at tibia could predict MOF significantly (RR = 2.07, 95% CI = 1.80–2.38, I 2 = 31%, p < 0.01) (Supplementary Fig. Supplementary Fig. S5).

Meta‐analysis in predicting incident fracture and MOF by DXA

Among the 16 studies included in meta‐analysis, there were six studies (one cohort and five case control) involving both HR‐pQCT parameters and DXA areal BMD (aBMD) in predicting incident fracture or MOF. Due to limited number of studies, we grouped the aBMD data of all DXA measuring sites (lumbar spine, proximal femur and distal radius) together to perform the meta‐analysis. Results showed that aBMD could predict incident fracture and MOF significantly (incident fracture: RR = 1.30, 95% CI = 1.28–1.32, p < 0.01; MOF: RR = 1.53, 95% CI = 1.41–1.65, p < 0.01). However, there were significant heterogeneity from the dataset (incident fracture: I 2 = 89%, p < 0.01; MOF: I 2 = 93%, p < 0.01) (Supplementary Table S3).

Effect sizes of the three best performance HR‐pQCT parameters

Cohen's d was used to compare the effect size among cortical vBMD, Tb.Th, and stiffness. All d values were equal to or larger than 0.8, which means that large effects (not overlapping) were observed (Supplementary Table S4).

Discussion

This review aimed to investigate which HR‐pQCT parameters at the distal radius and/or distal tibia can best predict fragility fractures. All the included papers showed positive prediction of HR‐pQCT for incident fractures and/or MOFs, in which almost all of the HR‐pQCT parameters showed significant prediction. Among all the parameters, the meta‐analysis showed that cortical vBMD, Tb.Th, and stiffness were best prediction parameters of HR‐pQCT, in terms of their effect size, heterogeneity, and their generalizability in predicting incident fractures and MOFs at both measurement sites. The three factors measured at distal radius and distal tibia could successfully predict both incident fractures and MOFs with the effect size up to 1.90, after screening out the analyses with high heterogeneity (I 2 > 65%).

Of the three outstanding predictors, stiffness is a HR‐pQCT parameter of estimated bone strength. This indicates that estimated bone strength can predict fragility fractures well, whereas vBMD or microarchitecture parameters are other important ones. On the other hand, geometric parameters, eg, total area, cortical area, trabecular area, show less predictive power. Almost all the included studies with FEA consistently used the default Image Processing Language (IPL) of HR‐pQCT for analysis, except one using third‐party FEA software.( ) The possible reason why estimated bone strength shows a good prediction power is that μFEA analysis has included geometric, microarchitecture, and vBMD data altogether to estimate mechanical parameters: stiffness and estimated failure load.( ) Therefore, estimated bone strength is relatively more all‐round and advanced than other individual geometric, structural or vBMD parameter.

The other two outstanding predictors are cortical vBMD and Tb.Th. Because trabecular bone is metabolically more active than cortical bone, trabecular bone loss usually occurs early in osteoporosis. The decrease in trabecular bone is caused by thinning of the trabeculae and especially in early postmenopausal women,( ) by disruption of the trabecular microstructure and loss of trabecular elements. This will eventually reduce the trabecular number and increase the trabecular separation in later phase. Hence, trabecular thinning is an early phenomenon of osteoporosis, thus Tb.Th is a sensitive parameter. During trabecular bone loss, cortical bone will share more loading. With the progression of osteoporosis, cortical bone will deteriorate to be more porous and cortical trabecularization will be resulted,( ) leading to a great loss of cortical bone mass (yet Ct.vBMD may increase, because the zone formally included in the cortical compartment will be included in the trabecular compartment instead), whereas fracture risk will also be sharply increased. This may explain why cortical vBMD is a sensitive HR‐pQCT parameter for fragility fracture prediction.

Between incident fracture and MOF prediction, most of the HR‐pQCT parameters performed better in MOF. A few parameters showed the effect size >2.0 in MOF prediction, such as Tb.vBMD, Tb.N measured at distal radius, and estimated failure load measured at distal tibia, whereas none of them demonstrated the effect size >2.0 in incident fracture prediction. These results revealed that HR‐pQCT parameters might reflect the mechanical competence or bone quality of three MOF sites (wrist, spine, and hip) better than other regions. A study by Liu and colleagues( ) substantiated that the prediction of stiffness of proximal femur and vertebrae by HR‐pQCT parameters were comparable with direct measurements of proximal femur and lumbar spine by DXA and central QCT, respectively. Hence, HR‐pQCT‐based images and FEA of distal radius and tibia were good indicators of mechanical properties of the lumbar spine and proximal femur.( )

HR‐pQCT has two standard measurement sites: distal radius and distal tibia. Our review showed that more parameters measured at the distal radius could significantly predict both incident fractures and MOFs than those at the distal tibia. The distal radius is therefore a more preferable measurement site for fracture prediction. The distal radius has been reported to be an ideal site for screening primary osteoporotic distal radius fracture.( ) Miyamura and colleagues( ) showed that fracture group had significantly lower BMD and T‐score of ultradistal, mid‐distal, and one‐third distal forearm than control group; DXA measurements exhibited high correlation with vBMD measured by computed tomography (r = 0.83–0.92). Because distal radius of nondominant side is a non‐weight‐bearing region, osteoporosis developed in this region cannot be offset by most osteogenic weight‐bearing activities, thus providing a sensitive site for osteoporosis detection. On the other hand, distal tibia, as a weight‐bearing site, will be influenced by most daily activities and cannot predict fragility fractures as good as the distal radius.

Cortical porosity (Ct.Po) is a unique parameter provided by HR‐pQCT that no other imaging system can measure. This is defined as the average fraction of void volumes within the cortical bone volume.( ) Ct.Po has been reported to have prominent increase at the fifth decade and reach a plateau before the sixth decade,( ) which may be an important determinant of fracture risk. In our meta‐analysis, the results revealed that Ct.Po measured at distal tibia could predict MOF significantly. Hung and colleagues( ) also substantiated that Ct.Po at the distal tibia increased from around 2.5% to 7.5% from the age of 50 to 60 years, which should weaken the cortical strength substantially, thus supporting the MOF prediction power of Ct.Po. However, more studies should be conducted to validate its ability of prediction, because many analyses on Ct.Po in our review showed high data heterogeneity, leading to difficult interpretation.

DXA is currently the gold standard tool in assessing BMD and predicting fracture risk. Our findings support that bone quality measured by HR‐pQCT could predict both incident fracture and major osteoporotic fracture. However, based on the results of this meta‐analysis, we cannot conclude whether HR‐pQCT parameters predict fractures better than DXA. The major reason is that significant heterogeneity (I 2 = 89–93) was found in the analysis of DXA in predicting fractures, which is not suitable for direct comparison with HR‐pQCT. A large cohort prospective consortium study by Samelson and colleagues( ) have shown that HR‐pQCT parameters could predict incident fractures independently of femoral neck aBMD and Fracture Risk Assessment Tool (FRAX) score. Moreover, the predictive model (area under the receiver operating characteristic curve) involving HR‐pQCT parameters plus aBMD or FRAX score was significantly improved, as compared with aBMD or FRAX score alone. Further review is suggested to compare the predictive ability between HR‐pQCT and DXA in predicting incident fractures.

This systematic review indicated that most of the studies (21/25) included were from North America and Europe, whereas very few were from South America, Asia, or multi‐countries. This means the fragility fracture prediction by HR‐pQCT is mainly white‐based, yet these analyses may not be directly applicable to Asians. Ethnic differences in BMD has been well reported; eg, total hip and spine BMD were 4% to 5% lower among Chinese women than US whites.( ) Vertebral fracture rate of Asians was higher than that of whites but lower hip fracture rate, resulting in a high vertebral‐to‐hip fracture ratio.( ) Also, a report demonstrated that Chinese women had cortical and trabecular microstructural differences from white women.( ) This is therefore highly recommended to establish their countries' bone databases of HR‐pQCT and conduct related fracture prediction studies. Also, there is a lack of robust data from Africa and among Africa descent, which will be an area of future research.

There are some limitations in this study. In the included cohort studies, many of them involved participants with previous fractures, while those in case–control studies were mostly first‐time fracture patients. Also, case–control studies were retrospective in nature. These might cause substantial data heterogeneity among the studies.

In conclusion, our study showed that HR‐pQCT could predict fragility fractures well and many parameters were of good prediction power to predict fragility fractures or incident fractures. Of all the HR‐pQCT parameters, stiffness, cortical vBMD, and Tb.Th were the outstanding predictor parameters. Also, of the two standard measurement sites by the HR‐pQCT of distal radius and tibia, distal radius was shown to be a more preferable measurement site for fracture prediction. To date, most of the studies were white‐based; Asian and African regions should build up their own densitometric databases and conduct related prediction studies, and Asians have been reported to exhibit different bone structure from whites. It is expected that HR‐pQCT can ultimately predict and prevent fragility fractures in the future, if applied widely in the community.

Disclosures

All authors state that they have no conflicts of interest.

PEER REVIEW

The peer review history for this article is available at https://publons.com/publon/10.1002/jbmr.4449.

Supplementary Fig. S1. Forest plot of trabecular vBMD measured at distal radius and distal tibia for predicting incident fracture and MOF, (A) distal radius; incident fracture, (B) distal tibia; incident fracture, (C) distal radius; MOF, (D) distal tibia; MOF.

Click here for additional data file.

Supplementary Fig. S2. Forest plot of cortical thickness (Ct.Th) measured at distal radius and distal tibia for predicting incident fracture and MOF, (A) distal radius; incident fracture, (B) distal tibia; incident fracture, (C) distal radius; MOF, (D) distal tibia; MOF.

Click here for additional data file.

Supplementary Fig. S3. Forest plot of trabecular number (Tb.N) measured at distal radius and distal tibia for predicting incident fracture and MOF, (A) distal radius; incident fracture, (B) distal tibia; incident fracture, (C) distal radius; MOF, (D) distal tibia; MOF.

Click here for additional data file.

Supplementary Fig. S4. Forest plot of trabecular separation (Tb.Sp) measured at distal radius and distal tibia for predicting incident fracture and MOF, (A) distal radius; incident fracture, (B) distal tibia; incident fracture, (C) distal radius; MOF, (D) distal tibia; MOF.

Click here for additional data file.

Supplementary Fig. S5. Forest plot of estimated failure load measured at distal radius and distal tibia for predicting incident fracture and MOF, (A) distal radius; incident fracture, (B) distal tibia; incident fracture, (C) distal radius; MOF, (D) distal tibia; MOF.

Click here for additional data file.

Supplementary Fig. S6. Forest plot of cortical porosity (Ct.Po) measured at distal radius and distal tibia for predicting incident fracture and MOF, (A) distal radius; incident fracture, (B) distal tibia; incident fracture, (C) distal radius; MOF, (D) distal tibia; MOF.

Click here for additional data file.

Supplementary Table S1. Egger's test results of publication bias

Supplementary Table S2. Summary table of all meta‐analysis data and heterogeneity test results for HR‐pQCT parameters

Supplementary Table S3. Meta‐analysis of DXA in predicting incident fracture and MOF

Supplementary Table S4. Determining the effect size overlapping among cortical vBMD, Tb.Th and stiffness in different fracture sites using Cohen's d

Click here for additional data file.