Fracture Risk Stratification by REMS: A 5 Years Prospective Longitudinal Multicentre Study

Abstract

Osteoporosis is a skeletal disorder characterized by low bone mass and deterioration of bone microarchitecture. Current predictive tools, mainly based on DXA-BMD, do not full capture all the factors contributing to bone deterioration. By combining information on bone quantity (T-score) and skeletal fragility (Fragility Score, FS), the Radiofrequency Echographic Multi-Spectrometry provides an integrated measure of fracture susceptibility known as fracture risk stratification. We aimed To evaluate the ability of REMS fracture risk stratification through a 5 years prospective longitudinal multicentre study. A non-parametric Mann–Whitney test was used to assess differences of REMS T-score and FS between fractured and non-fractured patients. Additionally, Chi-squared tests, Cochran-Armitage trend tests, class-specific unadjusted logistic regressions and standardized residual analyses were performed to evaluate the proposed stratification model. A total of 4443 female patients was recruited and n = 4017 underwent REMS scans. The Chi-squared and the Cochran-Armitage trend tests showed that the distribution of fractured patients differed across the 7 risk classes with a strong positive monotonic trend at lumbar spine (Z = 47.83; p < 0.0001) and femoral neck (Z = 75.41; p < 0.0001). From a clinical perspective, both Class-specific binary logistic regression and standardized residuals clearly identified two opposite risk profiles: patients belonging to the low-risk classes exhibited a significantly reduced probability of fracture, whereas those in the high-risk classes showed a markedly increased fracture risk. This 5-year longitudinal study supports the clinical consistency of a REMS composite approach for fracture risk stratification based on the integration of REMS BMD T-score and Fragility Score.

Introduction

Osteoporosis is a skeletal disorder that is chronic and progressive in nature. It is characterized by low mineral density and microarchitectural deterioration of bone tissue, leading to an increased bone fragility and susceptibility to fractures [1]. Globally, it represents a major public health issue, particularly in elderly population [2–5]. Due to its silent progression accompanied by an absent screening program and a poor awareness of the problem, people generally receive a diagnosis of osteoporosis when a fracture has already occurred. Thus, the high social and economic impact of osteoporosis makes the early identification of individuals at high-risk of utmost importance.

Currently, the standard technique used for the diagnosis of osteoporosis is the dual-energy X-ray absorptiometry (DXA). According to the World Health Organization (WHO) diagnostic criteria, osteoporosis is defined by a T-score ≤ − 2.5 at lumbar spine, femoral neck or total hip [6]. Despite its widespread use and clinical utility, DXA has several well-recognized limitations. First, it provides a 2D areal measurement of BMD and not a true 3D volumetric assessment, which may not fully reflect bone strength: areal BMD, in fact, is susceptible to bone size and, thus, will overestimate fracture risk in individuals with a small skeletal size, who will have lower areal BMD than normal-sized individuals, and vice versa for individuals with a large skeletal size [7]. Second, DXA does not capture information about bone quality, including factors such as microarchitecture, turnover, mineralization and micro-damage, all of which significantly contributing to overall bone strength and fracture risk [8]. Finally, DXA involves exposure to ionizing radiation. Although DXA involves low radiation exposure, it may be less suitable for frequent monitoring or for the examinations of fragile population, in addition to high-costs and non-portable equipment that limit the accessibility in some healthcare settings [9].

In response to these limitations, recent advances in medical imaging and signal processing have led to the development of innovative diagnostic techniques able to provide a more detailed and clinically relevant assessment about bone status. Among these, Radiofrequency Echographic Multi Spectrometry (REMS) is a non-invasive, radiation-free and portable technology based on the analysis of native radiofrequency ultrasound signals. The technology provides an estimate of BMD and derived parameters (T-score and Z-score) reflecting the structural and compositional properties of the bone. Several studies have demonstrated that REMS shows a high diagnostic agreement with DXA [10, 11] and a strong ability to avoid clinically relevant artefacts that may lead to BMD overestimation in DXA acquisitions [12, 13] and its efficacy for short-term monitoring of bone health [14, 15]. In addition, REMS has introduced also the Fragility Score (FS), a proprietary parameter designed to enhance the prediction of fracture risk by covering microstructural information not captured by traditional BMD measures. Previous published studies have demonstrated that the FS is more strongly associated with a history of fractures than the DXA and REMS T-score alone, suggesting its superior discriminative capability in identifying individuals at high-risk of fracture [16–18]. All these characteristics support the use of the REMS technology in osteoporosis management as a reliable and accessible alternative to DXA [9, 19].

In this context, an accurate and accessible method for the assessment of fracture risk is essential as many fragility fractures occur in individuals with T-score values above the osteoporotic threshold, thus highlighting the importance of adopting more comprehensive methods for assessing fracture risk [4, 20, 21]. Currently, several risk assessment tools combining different risk factors (including DXA-BMD or not) have been developed, with different levels of complexity, but only few of them (QFracture, Fracture Risk Assessment (FRAX) tool, Garvan, Derived Fracture Risk Assessment (DeFRA) among others) have been validated in a population-based settings according to a proper methodological approach [22]. However, none of these perform better than the others [23] and none of these is able to meet different specific needs [24]. Considering that both BMD and bone quality contribute to define bone strength and fracture risk, the combination of these two sources of information could potentially enhance diagnostic accuracy.

To this end, a novel composite parameter that integrates data from both REMS T-score and FS could generate an integrated measure of fracture susceptibility. The underlying hypothesis is that this approach could provide a more comprehensive assessment of skeletal fragility, thereby improving the stratification of patients based on fracture risk and supporting more effective clinical decision-making procedures.

Therefore, the primary aim of the present 5-year longitudinal multicentre study was to evaluate the ability of the REMS fracture risk stratification approach, based on the combined use of REMS BMD T-score and FS, to classify women into predefined classes of increasing fracture risk. Specifically, lumbar spine classes were evaluated against major osteoporotic fractures (MOF), whereas femoral neck classes were evaluated against incident hip fractures. By integrating information related to bone quantity and bone quality with a single non-ionizing assessment, this approach was intended to explore whether REMS-derived parameters may support fracture risk stratification beyond conventional densitometric assessment alone. In this context, the present study was designed to assess the clinical behaviour and internal consistency of predefined REMS risk classes over a 5-year follow-up period.

Materials and Methods

This multicentre prospective observational study was conducted across the participating clinical centres. The patient recruitment started on January 2016 and lasted until July 2019. Patient follow-up lasted until July 2024. The study protocol was approved by the Ethics Review Board of the participating institutions.

Patients and Measurements

The clinical data were collected from a study involving patients who fulfilled the following inclusion criteria: Caucasian ethnicity, female gender, age between 30 and 90 years, medical prescription for spinal and/or femoral densitometric examinations and absence of significant ambulation difficulties. All the patients voluntarily entered the study after giving written informed consent. According to the study protocol, patients underwent an ultrasound REMS scan of lumbar spine and femoral neck. REMS acquisitions were performed at the anatomical site(s) specified in the medical prescription (femoral neck and/or lumbar spine). During the follow-up period, patients were regularly (every 6 months) seen in-person or contacted by phone to monitor for any occurrence of incident fractures, which were confirmed through medical reports based on imaging investigations (such as radiographic imaging, vertebral morphometry, etc., when available) acquired during the follow-up period lasting up to 5 years. Investigators responsible for fracture ascertainment were blinded to the participants’ baseline REMS results and assigned risk classes. For the purposes of the study, fractures resulting from trauma were excluded in order to focus exclusively on fractures attributed to bone fragility.

Examinations

REMS scans were carried out using an EchoStation device (Echolight S.p.a., Lecce, Italy) equipped with a convex transducer operating at a nominal frequency of 3.5 MHz, following the manufacturer's instructions. The scanning procedure has been thoroughly described in previously published works [25, 26]. In brief, lumbar scans were conducted by positioning the echographic transducer trans-abdominally under the sternum to initially visualize the L1 vertebra, then moving it down to L4, following the on-screen and audible instructions from the EchoStudio software (Echolight S.p.a., Lecce, Italy). Femoral neck scans were performed by placing the echographic transducer parallel to the head-neck axis of the femur to visualize the typical proximal femur profile, including the interfaces of femoral head, neck and trochanter. Once the target interfaces were identified, the operator held the image for 40 s according to the indications provided by the device.

All acquisitions were performed by operators who had received the specific training according to the guidelines of the device manufacturer and had at least 3 months of previous experience in REMS acquisitions.

Parameters

BMD (g/cm²) and T-score values, the latter derived from the standard reference database for Caucasians, were included in the subsequent statistical analysis.

REMS software also produces a proprietary parameter, named the Fragility Score, used as an indicator of skeletal fragility. For the calculation of the FS, ultrasound radiofrequency signals reflected by the bone target are processed through spectral and statistical analyses [27, 28]. An algorithm embedded in EchoStudio automatically identifies the bone interface and the region of interest (ROI). In detail, the analysis of FS involved the following steps: (1) automatic segmentation of vertebral or femoral interfaces, (2) automatic selection of a ROI for each segmented interface, (3) classification of each radiofrequency segment as “frail” or “not frail”, (4) for each considered ROI, calculation of the FS values, defined as the percentage of the analysed RF portions that were previously classified as “frail”, (5) calculation of the FS value as average of all the ROI values. A graphical abstract of the FS is shown in Fig. 1.

Fig. 1.

Diagram representing the workflow of Fragility Score calculation

Then, the spectral profile of the patient ROI is classified as frail or not-frail by comparing it with reference spectral models previously built from fractured and non-fractured patients [27, 28]. From this comparison, the system automatically provides a quantitative estimation through a dimensionless parameter ranging from 0 (maximum similarity with the healthy model) to 100 (maximum similarity with the fractured model). Based on the manufacturer’s instructions, the bone quality of a patient can be classified as normal when the FS value is included within the range indicated by the green code, as reduced when the FS value is included within the range indicated by the yellow code, and as low when the FS value is included within the range indicated by the red code.

By combining the information obtained from the FS and the T-score values, a 3 × 5 matrix was generated. To correctly account for “borderline” cases, a tolerance of ± 0.3 to the T-score thresholds was applied [10]. Similar risk ranges were subsequently merged, generating 7 final risk classes. When FS and T-score values were measured through lumbar REMS scanning, patients were stratified into 7 risk classes (from R1 to R7), with a related probability range of MOF. Analogously, when the REMS scan was performed at the femoral neck, the relationship between the corresponding parameters generated the 7 risk classes along with the fracture probability range for hip fractures.

The combination of the REMS T-score and FS parameters to generate the 7 risk classes is illustrated in Table 1.

Table 1.

Combination of T-score (normal, osteopenic and osteoporotic) and fragility score (normal, reduced and decreased) classes and the generated specific risk classes (R1-R7)

		T-score classification
		Normal		Osteopenic			Osteoporotic
		≥ − 0.7	]− 1.0;− 0.7[	]− 1.3; − 1.0[	[− 2.2; − 1.3[	[− 2.5; − 2.2[	[− 2.8; − 2.5[	< − 2.8
Fragility Score classification	Normal	R1	R2	R2	R3	R4	R4	R5
	Reduced	R2	R3	R3	R4	R5	R5	R6
	Decreased	R3	R4	R4	R5	R6	R6	R7

Statistical Analysis

The analysis was conducted separately at lumbar spine and femoral neck. Parameters derived from lumbar spine scans were matched with data on MOF, while parameters derived from femoral neck acquisitions were matched exclusively with data on hip fractures.

Continuous variables, including patients' age, height, weight, body mass index (BMI), REMS T-scores and FS are presented as mean ± SD. All continuous variables were tested to assess the normality by Shapiro–Wilk test. A non-parametric Mann–Whitney test was performed to assess statistically significant differences between patient groups with or without incident fragility fractures. A DeLong test was performed to compare the areas under the receiver operating characteristic curves (AUCs) [29].Discrete variables (7 risk classes, number of patients enrolled, number of fractured and non-fractured patients) were reported as absolute frequencies. In order to assess the distribution of fractured patients across the 7 risk classes and to evaluate the presence of an ordered trend, a Chi-squared test followed by a Cochran-Armitage trend test were performed. Subsequently, to compare the likelihood of experiencing a fragility fracture for individuals classified in a specific risk class vs. those classified in all the remaining classes, class-specific unadjusted logistic regression models were used. For each risk class, a separate unadjusted logistic regression model was fitted by comparing patients assigned to that specific class with patients assigned to all the remaining classes combined. Therefore, class-specific odds ratios (ORs) should be interpreted as the relative odds of fracture associated with belonging to a given class compared with belonging to any other class. Finally, to evaluate the contribution of each risk class to the overall frequency distribution between fractured and non-fractured individuals, standardized residuals from the Chi-squared test were examined.

In all the statistical tests listed, a p-value < 0.05 was considered statistically significant.

All statistical analyses were conducted using MedCalc® (version 19, MedCalc Software, Mariakerke, Belgium), and MATLAB® (MathWorks, Natick, MA, USA) software.

Results

Patient Characteristics T-Score and Fragility Score in Fracture Occurrence Prediction

A total of 4443 female patients was recruited. During the follow-up period, 426 participants (9.6%) had incomplete follow-up data or were lost to follow-up, leaving 4017 women who completed the study. At total of 5458 scans was obtained from this cohort of patients and was used for the subsequent analysis. Among these, 2744 were scans of the lumbar spine, whereas 2714 were scans of the femoral neck. During the follow-up period, 418 scans performed at the lumbar spine (15.2%) and 275 scans performed at the femoral neck (10.1%) were associated to patients who experienced MOF or hip fragility fractures, respectively. Specifically, n = 1441 underwent REMS examinations of both anatomical sites, n = 1303 underwent REMS examination at lumbar spine only, and n = 1273 underwent REMS examination at femoral neck only.

Table 2 summarizes the general baseline characteristics of the participants and the bone-related statistics of examinations grouped by anatomical site of the scan (lumbar spine or femoral neck). As expected, the mean REMS T-score values were significantly lower, and the mean FS measured were significantly higher in scans from patients who experienced a fragility fracture during the follow-up compared to those who did not, regardless the anatomical site of the scan.

Table 2.

General baseline characteristics of the participants and scan characteristics of the patients assessed based on anatomical location (lumbar spine or femoral neck). The reported values are expressed as mean ± standard deviation (SD). An independent non-parametric Mann–Whitney test was used to compare the groups of patients without and with fragility fractures occurred during the follow-up period. MOF: major osteoporotic fractures

	Parameter
General population	Age (years)	65.5 ± 10.1
	Weight (kg)	63.7 ± 10.3
	Height (cm)	158.4 ± 6.9
	BMI (kg/m2)	25.4 ± 4.0

Anatomical site	Parameter	Patients without fragility fracture	Patients with fragility fracture (MOF/hip)	p value
Lumbar spine	N° of scans	2326	418
	REMS T-score	– 2.1 ± 1.1	– 2.5 ± 1.1	< 0.0001
	Fragility score	42.5 ± 16.9	54.0 ± 17.3	< 0.0001
Femoral neck	N° of scans	2439	275
	REMS T-score	– 2.0 ± 1.0	– 2.4 ± 0.9	< 0.0001
	Fragility score	36.6 ± 16.9	49.2 ± 15.9	< 0.0001

ROC Curve Analysis

A De Long test was performed to compare the differences in the AUC between a model including T-score + Fragility Score and a model including T-score alone at both anatomical sites (lumbar spine and femoral neck). The results showed a statistically significant difference (p < 0.001) indicating the Fragility Score provides incremental discriminative information beyond the REMS-derived T-score.

Fracture Risk Analysis

In Fig. 2, for women analysed at lumbar spine and femoral neck, the percentage of scans associated to patients who experienced an incident fracture during follow-up is highlighted as a proportion of the total number of scans in each risk class (R1-R7).

Fig. 2.

Frequency charts depicting the percentage of patients without and with an incident fracture during follow-up (in orange and blue, respectively) for each risk class across the two patient cohorts are presented as follows: women analysed for lumbar spine (a), women analysed for femoral neck (b)

The Chi-squared test showed a statistically significant p value (p < 0.0001) for both lumbar spine and femoral neck scans, indicating that the distribution of the fractured patients differed across the 7 risk classes. Likewise, the Cochran-Armitage trend test showed high values at lumbar spine (Z = 47.83; p < 0.0001) and femoral neck (Z = 75.41; p < 0.0001), demonstrating a strong positive monotonic trend in both anatomical sites. Furthermore, Class-specific binary logistic regression analyses confirmed this pattern, revealing progressively increasing odds ratios from R1 to R7 for both lumbar spine (Table 3) and femoral neck (Table 3). At lumbar spine, the analysis of class-specific OR revealed a clear and progressive gradient of fracture risk across the seven risk classes. The low-risk classes (R1–R2), in fact, showed a significantly reduced risk compared with the remaining population collocated in the other risk classes, with OR ranging from 0.35 to 0.58. Conversely, the high-risk classes (R5–R7) exhibited a progressively increased risk, with OR moving from 1.04 (R5) to 1.91 (R7). When examining the analysis at femoral neck, similar results were found: while low-risk classes (R1–R3) were associated to a significantly reduced risk (OR ranging from 0.20 to 0.57), the high-risk classes (R5–R7) were associated to a progressively increased odds of fracture, with OR ranging from 1.41 (R5) to 2.26 (R7).

Table 3.

For lumbar spine and femoral neck scans the odds ratio (OR) with the corresponding 95% confidence interval (CI) and p value were reported within each risk class (from R1 to R7)

Lumbar spine scans				Femoral neck scans
Risk class	Odds ratio	95% CI	p value	Odds ratio	95% CI	p value
R1	0.35	0.19 to 0.63	< 0.001	0.20	0.074 to 0.545	< 0.01
R2	0.58	0.40 to 0.82	< 0.01	0.32	0.20 to 0.52	< 0.0001
R3	0.73	0.51 to 1.04	> 0.05	0.57	0.39 to 0.843	< 0.01
R4	0.85	0.63 to 1.14	> 0.05	1.01	0.73 to 1.4	> 0.05
R5	1.04	0.81 to 1.32	> 0.05	1.41	1.03 to 1.921	< 0.05
R6	1.64	1.27 to 2.10	< 0.0001	1.89	1.4 to 2.5	< 0.0001
R7	1.91	1.42 to 2.55	< 0.0001	2.26	1.6 to 3.2	< 0.0001

Standardized residuals were used to identify which risk classes contributed most to the overall Chi-squared association. Negative residuals below − 2 indicate fewer fractured patients than expected under the hypothesis of no association between risk class and fracture occurrence, whereas positive residuals above +2 indicate more fractured patients than expected. At the lumbar spine (Table 4), low-risk classes R1 and R2 showed less fractured patients than expected, as indicated by strongly negative residuals (− 3.23 and − 2.63), whereas high-risk classes R6 and R7 showed an excess of fractures (with respect to the assumption that no differences exist among the seven risk classes). A similar pattern was observed at the femoral neck (Table 4), with low-risk classes (R1-R3) showing less fractured patients (residuals ranging from − 3.20 to − 2.46) than expected and high-risk classes (R6-R7) showing more fractured patients than expected. These patterns support the ability of the REMS-derived stratification to separate lower- and higher-risk groups.

Table 4.

Standardized residuals from the Chi-squared frequency distribution at lumbar spine and femoral neck. Residual values greater than + 2 indicate more fractured or non-fractured patients than expected (i.e., under the assumption that no differences exist among the seven risk classes), whereas values lower than − 2 indicate fewer than expected (i.e., under the assumption that no differences exist among the seven risk classes). Residuals between − 2 and + 2 fall within the range of expected frequencies (i.e., under the assumption that no differences exist among the seven risk classes)

Lumbar spine			Femoral neck
Risk class	Standardized residual (fractured patients)	Standardized residuals (non-fractured patients)	Standardized residual (fractured patients)	Standardized residuals (non-fractured patients)
R1	− 3.23	1.37	− 3.20	1.08
R2	− 2.63	1.11	− 4.14	1.39
R3	− 1.50	0.64	− 2.46	0.83
R4	− 0.93	0.39	0.05	− 0.02
R5	0.24	− 0.10	1.87	− 0.63
R6	3.25	− 1.38	3.78	− 1.27
R7	3.83	− 1.62	4.11	− 1.38

Discussion

Currently, the most used clinical technologies are not able to provide a comprehensive diagnosis of osteoporosis that might predict a reliable occurrence of fragility fractures in a clinically sustainable framework [9, 30]. Bone mineral density, in fact, still represents the main driver of treatment decision, and it is generally used to identify people with osteoporosis or prone to be osteoporotic. Nevertheless, several studies have demonstrated that BMD alone does not reliably predict whether a patient will develop a fragility fracture [31]. As a direct consequence, a high percentage of patients remain underdiagnosed, receiving a treatment only when the clinical outcome is already compromised [32]. In order to encompass the multifactorial nature of osteoporosis, different predictive tools have been developed, including, but not limited to the FRAX questionnaire [33], Qfracture algorithm [34], Garvan Fracture-Risk calculator (Garvan) [35] and Derived Fracture Risk Assessment (DeFRA) [36]. This strategy has certainly improved the ability to diagnose osteoporosis and to identify individuals who are at higher risk of fragility fracture, but an evident gap still exists [30, 37]. A significative amount of fragility fractures, in fact, still occurs in individuals who are not classified at high-risk; furthermore, none of these tools include information about bone quality and skeletal fragility, which play a crucial role in fracture risk. For this, the interest for a new technology that can provide an accurate diagnosis without further complications has increasingly grown over the past few years. Therefore, the present study aimed to introduce a novel composite stratification risk analysis comprising quantitative information from both bone mass (REMS T-score) and bone quality (REMS FS) to generate an integrated and reliable measure of fracture susceptibility. To date, this is the first study evaluating the REMS fracture risk stratification as a parameter for prediction of fragility fractures.

This prospective observational multicentre analysis included data from female patients who underwent REMS scans at lumbar spine and/or femoral neck and who were monitored over a 5-year period for the occurrence of fragility fractures. At the time of recruitment, patients were examined, the REMS parameter were annotated (T-score, FS and risk stratification derived by the combination of the two REMS parameters) and close monitoring (every 6 months in-person or by telephonic interview) for the occurrence of any fragility fracture was conducted. A recent study performed by Pisani et al. [16] demonstrated the higher performance of single parameters (REMS FS and T-score) in predicting fragility fractured compared to DXA T-score, which is currently recognized as a clinically available indicator to predict fracture risk. In addition, FS consistently demonstrated superior predictive capacity compared with both REMS and DXA T-score values, with an ability to identify above the cut-off a nine-fold and a six-fold higher risk to develop a MOF or hip fracture, respectively [17]. These findings further support the hypothesis that BMD alone has a limited fracture prediction ability, emphasizing the need for reliable parameters that estimate fracture risk independently from or in addition to BMD [38, 39].

The stronger association of FS with fracture outcomes should be interpreted as a confirmation that BMD and FS provide complementary information on different domains of skeletal fragility: the former reflects bone mass, whereas the latter reflects bone-quality-related features. The rationale of the present stratification approach is therefore to integrate both of them into a clinically interpretable classification of fracture risk.

Based on this rationale, the combination of REMS T-score and FS was used to define 7 risk classes (here named from R1 to R7), each associated to a progressively increasing probability of fracture. Therefore, based on this assumption, at the time of scan examination, patients were stratified twice across 7 different risk classes (once for lumbar spine and once for femoral neck examination) according to the measured values of REMS T-score and FS. Then, according to the information reported about MOF or hip fractures during the 5 years follow-up period, the percentages of fractured patients across each risk class were annotated. The statistical analysis confirmed the clinical relevance of the proposed risk classification at both anatomical sites. The Chi-squared test demonstrated that the distribution of fractured patients differed significantly across the seven risk classes, indicating a non-random association between class assignment and fracture occurrence. The Cochran–Armitage trend test further showed a highly significant and progressive increase in fracture prevalence from the lowest (R1) to the highest (R7) risk class for both the considered anatomical sites.

From a clinical perspective, the class-specific odds ratios clearly identified two opposite risk profiles: patients belonging to the low-risk classes (R1–R2/R3) exhibited a significantly reduced odds of fracture, whereas those in the high-risk classes (R6–R7) showed a markedly increased fracture risk. This finding supports the ability of the classification system to effectively stratify patients according to clinically meaningful levels of fracture susceptibility. For example, patients classified as R1 (a low-risk class) showed an OR = 0.35 at lumbar spine and an OR = 0.20 at femoral neck, which correspond to a 65% and 80% reduced odds of developing a fragility fracture compared with patients in all other risk classes. On the other hand, patients classified as R6 (a high-risk class) showed an OR = 1.64 at lumbar spine and OR = 1.89 at femoral neck, which correspond to a 64% and 89% increased odds of developing a fragility fracture compared with patients in all other risk classes. The analysis of standardized residuals provided additional insight into the contribution of each class to this overall pattern. Low-risk classes consistently showed fewer fractures than expected under the hypothesis of no association between risk class and fracture occurrence, whereas high-risk classes were characterized by an excess of fractured patients. Intermediate classes are expected to show less marked separation because fracture risk is biologically continuous. This is consistent with the absence of a clear-cut threshold between normal and pathological skeletal conditions [40, 41].

Taken together, these findings support REMS 5 years fracture risk classification. The strong agreement between global association tests, trend analysis, odds ratio estimates and residual diagnostics highlights the internal consistency of this stratification and supports its potential clinical relevance for identifying individuals at progressively increasing risk of fracture. Compared with questionnaire-based fracture risk calculators (e.g., FRAX, QFracture, Garvan, DeFRA, etc.), the REMS approach provides an automatic stratification based on skeletal parameters acquired during a single echographic examination. Importantly, it focuses on a shorter time horizon (5 years with respect to the 10 years of the current standards), which may be clinically relevant for near-term prediction of fracture risk and treatment prioritization.

The present study has some limitations. First, the cohort was composed of women referred for densitometric assessment in clinical settings, and therefore may not be representative of the general population. This may partly explain the relatively high fracture rates observed during the follow-up. Second, only female patients were included in the analysis. Although women (especially post-menopausal women) represent the group at highest risk for osteoporotic fragility fractures, the incidence of fragility fractures in men is increasing as well, with mortality and disability rates even higher than in women [42], therefore deserving inclusion in studies dedicated to the evaluation of fracture risk prediction tools. Third, REMS examinations were not paired with contemporaneous DXA scans at the same anatomical sites, preventing a direct head-to-head comparison with DXA-based fracture risk assessment. Fourth, the absence of measurements at both sites in all participants prevents a complete within-patient comparison between lumbar and femoral risk classification. Fifth, detailed information on anti-osteoporotic treatment at baseline and during follow-up was not collected. Finally, the study did not include a formal comparative analysis against established fracture risk calculators such as FRAX, DeFRA, QFracture or Garvan.

Since osteoporosis is becoming prevalent also in middle-income countries, future studies will be specifically dedicated to populations of those areas [43].

Conclusions

Osteoporosis and related fragility fractures show a multifactorial nature that cannot be fully captured by current predictive tools. This 5 years prospective longitudinal multicentre study supports the clinical consistency of a REMS composite approach for fracture risk stratification based on the integration of REMS BMD T-score and Fragility Score. The seven predefined classes R1-R7 showed a progressive association with incident fractures, with lumbar spine-derived classes evaluated against major osteoporotic fractures and femoral neck-derived classes evaluated against hip fractures.

These findings support the role of REMS in 5 years estimation of fracture risk by combining information related to bone quantity and bone quality within a single echographic examination.

Author Contributions

Conceptualization: CC, MLB, SG; Methodology: CC, MLB, SG; Formal analysis and investigation: CC, MLB, GG, AG, MM, SG; Writing—original draft and preparation: CC, MLB, SG; Writing—review and editing: CC, MLB, GG, AG, MM, SG; Supervision: CC, MLB, SG.

Funding

No funding was received for conducting this study.

Data Availability

All relevant data will be available upon reasonable request and under a dedicated agreement to the corresponding author.

Declarations

Conflict of interest

Nothing to declare.

Ethical Approval

The study protocol was approved by the Ethics Review Board of the participating institutions, and it was conducted in accordance with the ethical standards of the Declaration of Helsinki (1964).

Human and Animal Rights and Informed Consent

No animals were used in the study. Informed consent for participation and publication has been obtained from all the participants included in the study.