In this cross-sectional study, 89 men and women with diabetes mellitus of ages ranging from 49 to 81 were recruited between November 2022 to September 2023, as outlined in Figure 1 . Patients at the endocrinological and rheumatological departments of the Charité Universitätsmedizin Berlin and a local doctor’s office specializing in diabetic patients (Endokrinologikum), were eligible for participation. Screened patients were excluded from further analysis if any of the following exclusion criteria were met: (1) BMI > 35 kg/m², (2) edema of the lower legs or metal implants in the lower legs, (3) inability to understand the nature of the study or inability to consent, (4) lack of indication for DXA scan or DXA scan older than one year.

During the measurement appointment, participants’ height and weight were measured, and a comprehensive medical history questionnaire was recorded. The assessment included details on diabetes type, disease duration, microvascular and macrovascular complications of diabetes, diabetes-specific medication, fracture history, lifestyle, and use of anti-osteoporotic drugs. Additionally, information on other medical conditions and medications was collected. Self-reported fractures from the past ten years, excluding fractures in fingers and toes, were considered eligible and verified with medical records when possible (40).Metabolic control in patients was assessed by measuring hemoglobin A1C (HbA1c) values at the time of the study visit or within a few days of the visit for some participants. Bone turnover parameters were assessed according to the German guidelines (40).

An age- and sex-matched control group of 76 non-diabetic patients, with and without fragility fractures, was formed from participants enrolled in a different study (“Identification of increased fracture risk through non-invasive determination of cortical micro- and macro-structural properties using quantitative bone ultrasound imaging” German Clinical Trial Register number: DRKS00025849). After recruitment, 26 patients were excluded from the analysis for the following reasons: CortBS measurement did not meet quality criteria (n = 14) or a bone surface was only detected in the anterior region (n = 3), incomplete DXA data, i.e., measurements could not be performed or analyzed at femur and spine (n = 7), CortBS quality criteria met and incomplete DXA data (n = 2), and undefined diabetes type other than T1DM or T2DM (n = 2). Diabetic participants were then stratified into four groups based on fracture history and diabetes type: (1) T2DM without fracture (T2DM^noFx), (2) T2DM with fracture (T2DM^Fx), (3) T1DM without fracture (T1DM^noFx), and (4) T1DM with fracture (T1DM^Fx).

All participants gave written consent before study participation. The study was registered in the German Clinical Trial Register (DRKS00029331) and approved by the local ethics committee of the Charité–University Hospital Berlin (EA4/140/22).

DXA scans were performed using a medical narrow-angle fan beam densitometer, GE Lunar Prodigy, for the diabetes group and part of the control group, and GE Lunar iDXA for the remaining control patients, in accordance with the vendor’s manual. Seven participants from the diabetes group and two participants from the control group provided externally taken DXA scans (GE Lunar Prodigy device for diabetes patients) conducted less than one year prior to recruitment. We did not conduct a cross-calibration of the devices but consider the potential variations to have minimal impact on the results. For all other participants, measurements were taken both at the femora and the lumbar spine (L1-L4). Each lumbar spine scan was checked for artifacts to ensure that inadequate scans were excluded from the analysis. At least one femur scan and two valid vertebrae in the lumbar spine were necessary for a participant’s inclusion in the analysis. aBMD values and corresponding T-scores were assessed at the lumbar spine, femoral neck, and total proximal femur area. If measurements from both femoral sides were available, the lowest BMD and T-score values of the total area from either side were used for further analysis. If only one side was measured, the BMD and T-score values from that side were used.

Ultrasound scans were performed with a medical ultrasound scanner, SonixTOUCH, equipped with a SonixDAQ single-channel data acquisition system and a 4DL14-5/38 3-D linear array transducer (Ultrasonix, Richmond, Canada). The system was operated via a custom-developed user interface to perform a Cortical Backscatter (CortBS) measurement, as described in (39). Briefly, the measurement consists of 1) an image-guided 3-D multidirectional scan of a slightly focused ultrasound beam across a selected cortical bone volume of interest, 2) the reconstruction of a 3D volume from the acquired channel data, 3) the automatic detection of the periosteal bone surface and the sound beam inclination for all acquisitions, 4) the calculation of an inclination-corrected reference spectrum from signals reflected from the periosteal bone interface, 5) the calculation of a normalized depth-dependent spectrogram, 6) the estimation of the frequency-dependent attenuation and backscatter coefficients α(f) and BSC(f), respectively, and 7) the estimation of the intracortical pore diameter distribution Ct.PoDm.D. Subsequently, microstructural and viscoelastic parameters, the intracortical pore diameter index Ct.Po.Dm.I (see Figure 2 ) were derived. CortBS risk scores were calculated using multivariate logistic regression models based on ultrasound variables associated with prevalent fractures. Different scores were derived for vertebral, non-vertebral, and any type of fractures. For comparison with DXA T-scores, the means and standard deviations of the predictor variable were used. All variables obtained from the CortBS measurement are summarized in Table 1 .

All ultrasound measurements were taken at the central anteromedial tibia midshaft on the leg with the lowest femoral DXA T-Score. The tibia length was measured from the medial knee joint cleft to the medial malleolus, with both landmarks manually palpated and the 50% intersection marked with a skin marker pencil. Ultrasound gel and an ultrasound coupling pad (Aquaflex, Parker Laboratories, Inc., Fairfield, NJ, USA) were used to attach the transducer to the skin at the marked position. Conventional B-mode images facilitated the positioning of the probe, i.e. to center a cross-sectional image of the periosteal tibia bone interface. The probe was manually tilted until the bone surface was nearly perpendicular to the sound beam. Focus position was aligned with the periosteal bone interface (note that the software was programmed such that the focus was set 1 mm below to ensure that all beams are focused inside the cortical bone tissue), and gain was adjusted to obtain approximately 100% signal level at the periosteal bone interface. The multidirectional scan was performed automatically by means of 1) electronic beam steering and 2) sweeping the transducer array using the built-in motor in the direction perpendicular to the image plane.

Two measurements were taken for each participant to guarantee that at least one would meet the requisite quality standards. The data acquisition was performed within 3 seconds per scan. The signal quality of the captured channel data was inspected directly after completion of the scan. If the signal level was too low, the signal gain was readjusted, and the scan was repeated.

Continuous variables are presented as means and standard deviations (SD) or standard error (SE). Data were tested for normal distribution by a Lilliefors test. Differences with respect to sex, prevalent fractures, diabetes type, co-morbidities, and medications were analyzed using N-way ANOVA. If data were not normally distributed, observed effects were confirmed using non-parametric Wilcoxon rank sum tests. Associations between continuous variables were performed using stepwise multivariate regression analysis the calculation of Spearman’s rank sum correlation coefficient ρ.

The fragility fracture discrimination performance of CortBS and DXA was assessed by multivariate partial least squares (PLS) linear discrimination analysis with Leave-One-Out Cross-Validation (PLS-LDA LOOCV) using the libPLS library (41). For variable selection, a sub-window permutation analysis (SPA) using 2,000 Monte Carlo samplings was repeated until a stable set of significant model variables was found. The number of PLS components was restricted to one-tenth of the observation number. Different discrimination models were developed to predict vertebral, non-vertebral, and any fragility fractures from DXA-based BMD values and T-scores, CortBS parameters, and for combinations of DXA or CortBS parameters with each participant’s anthropometric data (weight, height, BMI) and age. The mean and SE of the area under the curve (AUC) of the receiver operation characteristics (ROC), accuracy, and sensitivity were calculated. Differences between the AUC values were evaluated using the Hanley & McNeil test using MedCalc 22.032 (MedCalc Software Ltd, Ostend, Belgium).

To evaluate the correlation of multiple QUS parameters with DXA BMD, PLS regression with threefold cross-validation was used. Spearman’s rank sum correlation coefficient ρ and root mean square error (RMSE) between the predicted QUS-based parameter and those measured by DXA were computed.

Except for the PLS-LOOC and SPA analyses, all statistical tests were performed using the Statistics Toolbox of Matlab R2023b (MathWorks, Natick, MA, USA). Statistical results were considered significant for p values < 0.05.

During the recruitment phase, the diabetic group initially included 39 T1DM, 48 T2DM patients, and 2 patients with undefined diabetes type. However, due to the exclusion criteria, data from 31 T1DM and 35 T2DM patients were finally used in the statistical analysis. Among the T1DM patients, eight had at least one fracture. Of those, one had a vertebral fracture, seven had non-vertebral fractures, and none had both types of fractures. Among the T2DM patients, 17 patients had at least one fracture. Of those, four had vertebral fractures, 9 had a non-vertebral fracture, and four had both. Three patients with T2DM and a previous fracture received antiresorptive treatment. Vitamin D intake was observed in 42 patients of the diabetes cohort.

In the analyzed diabetes population, 32 subjects were females, all of whom were postmenopausal. The average duration of diabetes was 19 ± 15 years. Forty-five subjects were on insulin, and 29 were on oral antidiabetic drug treatment, with metformin being the most common (n = 24). No patients were on thiazolidines, sulfonylureas or glucosidase inhibitors at the time of the study. The average HbA1c value of diabetic patients was 7.2 ± 1.2. Eleven patients in the T1DM group were diagnosed with late-onset autoimmune diabetes in adults (LADA).

Further patient characteristics in juxtaposition to controls are summarized in Table 2 . The patient’s characteristics were similar in the control and diabetes groups, except for the expected differences in anthropometric data and slightly higher mean age. No significant differences were observed between fractured and non-fractured groups in the control, T1DM, and T2DM groups regarding anthropomorphic data and medical history.

Complete spine and femur measurements were missing for 5 and 4 patients, respectively, leading to their exclusion from subsequent analysis. A strong positive association with weight or BMI was observed for all DXA parameters ( Supplementary Table A1 ). Height had a minor negative association with the spine and total femur aBMD. All aBMD values and T-scores were significantly reduced in patients who were treated with antiresorptive or anabolic drugs, with the reduction more pronounced in femur measurements compared to spine measurements. Minor associations with other drug treatments were also observed. Differences with respect to fractures and DM type are summarized in Table 3 and Supplementary Table A1 . DXA aBMD values were significantly reduced in patients with fractures, particularly in total femur T scores (F = 11.3) and to a lesser extent in the spine T scores (F = 4.7). After adjustment for the impact of fractures, all T scores and aBMD values were significantly higher in patients with T2DM compared to controls (4.7 ≤ F ≤ 15.1). The T scores and aBMD values in patients with T1DM fell between those observed in controls and patients with T2DM, with some significant differences compared to controls or T2DM. It should be noted that the relative impact of DM on the DXA aBMD values in the spine or total femur was much stronger than that of prevalent fractures. Moreover, the mean T scores for all groups with fractures were higher than -2.5, thus not meeting the criteria for an osteoporotic T-Score.

The measurements of 3 patients were excluded, because only the anterior, but not the antero-medial region of the tibia was visible in the image. Moreover, the measurements of 16 patients (of which 2 had incomplete DXA), had to be discarded due to low quality.

Most anthropometric parameters were not associated with CortBS parameters. The mean attenuation and backscatter amplitudes (α_6MHz and BSC_mean) were significantly reduced in participants with higher weight and patients with T2DM ( Supplementary Table A1 ). Moreover, height had a minor negative association with α₀. Several parameters, particularly α₀ (F = 19.9), were affected by sex. Antiresorptive and anabolic drug treatments had distinct effects on different CortBS parameters. While pore diameter distributions did not significantly differ in patients treated with antiresorptive drugs, they were increased in those treated with anabolic drugs. Conversely, backscatter and attenuation coefficients changed in patients on antiresorptive drugs. Porosity was increased in patients receiving either treatment. Differences related to fractures and DM are summarized in Table 4 . All CortBS parameters, except for α_6MHz and BSC_mean, were significantly altered in patients with fractures. The most pronounced association was observed in the CortBS risk score (F = 15.7) and, to the least extent, in cortical porosity (F = 5.1). Notably, only the parameters that were not associated with fractures (α_6MHz and BSC_mean) were significantly reduced in patients with T2DM compared to controls (4.2 ≤ F ≤ 7.3).

The mean backscatter and attenuation coefficients and pore diameter distributions in control and diabetes groups are shown in Figure 3 . In the control group, individuals with fractures exhibited a shift in BSC peak towards smaller frequencies ( Figure 3A ), which was associated with an altered pore diameter distribution ( Figure 3G ). In particular, the number of small pores was reduced, the peak position was shifted towards larger values, and the number of large pores was increased. This was reflected in significant increases in pore size distribution parameters, e.g., peak, 10% and 90% quantiles, FWHM, FWHM_max, pore diameter distribution index, and cortical porosity ( Table 4 , Supplementary Table A1 ). Moreover, the frequency-dependent attenuation was significantly reduced ( Table 4 , Figure 3B ). Similar differences were observed between individuals with or without fractures in the T2DM group ( Table 4 , Figures 3C, F, I ). However, backscatter and attenuation amplitudes were lower compared to the control group. This difference was significant, as quantified by BSC_mean and α_6MHz ( Supplementary Table A1 ), indicating that the scatterer density (i.e., pore density) was lower in T2DM patients compared to controls. Similar to the controls, prevalent fractures in T2DM patients were associated with the accumulation of large pores. In contrast, the T1DM group exhibited distinct differences compared to both the control and T2DM groups ( Figures 3B, E, H ). BSC and attenuation in patients with diabetes without fractures had a similar shape but were lower in amplitude compared to controls. This difference was reflected in significantly reduced α_6MHz (F = 7.3) and BSC_mean (F = 4.2) values ( Table 4 , Supplementary Table A1 ). Notably, the pore diameter distribution in T1DM patients with fractures was similar to that of controls without fractures. In these patients, backscatter and attenuation coefficients matched the shape and amplitude of those in controls. These results suggest that T1DM patients without fractures had a lower pore density compared to healthy controls. In contrast, T1DM subjects with fractures exhibited a pore diameter distribution comparable to that of controls without fractures, indicating that pore density and diameter distribution in T1DM patients with fractures reach the level of controls without fractures.

The distinct differences in pore diameter distribution with respect to fractures and diabetes type compared to controls without fractures are shown in Figure 4 . The T2DM group exhibited the strongest reduction in normal-sized pores and the highest increase in large pores. In contrast, T1DM patients with fractures exhibited similar pore morphology to controls without fractures, with the exception of a non-significant reduction in the number of small pores.

Figure 5 shows the results of the discrimination analyses based on prevalent fractures. The AUC values obtained from the CortBS discrimination models (0.66 ≤ AUC ≤ 0.68) were significantly higher than those from DXA (0.59 ≤ AUC ≤ 0.63, best models from either BMD values or T-scores, see Table 5 ) for each type of fractures, reaching a significance level of p < 0.0001 for all fracture types. The combination of age and anthropometric information in the models resulted in a marginal improvement in the CortBS discrimination model for vertebral fractures (from AUC = 0.66 to AUC = 0.69) and in the DXA discrimination for other fractures (from AUC = 0.59 to AUC = 0.61). In other models, the additional degrees of freedom introduced by a higher number of variables rather led to a reduction in the discrimination performance.

Among the evaluated DXA parameters, the total femur T-score provided the best discrimination performance for all fracture types ( Table 5 ). Distinct combinations of CortBS parameters were selected to discriminate different fracture types.

The total femur T-scores were significantly reduced in patients with fractures, but they were also affected by diabetes and diabetes type ( Figures 6A–C ). Notably, the impact of diabetes (12.0 ≤ F ≤ 13.9) was considerably greater than that of prevalent fractures (4.3 ≤ F ≤ 11.3). The CortBS risk scores were significantly reduced in patients with fractures across all fracture types ( Figures 6D–F ) and notably, were not significantly affected by diabetes.

The prediction of aBMD from CortBS parameters using multivariate PLS models is shown in Figure 7 . Moderate correlations were observed for all prediction models (0.53 ≤ ρ ≤ 0.68).

The application of QUS in DM patients, mostly T2DM patients, has yielded mixed results. To our knowledge, this is the first time a cortical pulse-echo device was used to study diabetic patients. Most devices have been applied in anatomic locations other than the tibia. Tao et al. (33) used an axial transmission device in postmenopausal women with T2DM at the tibia in addition to the phalanges and the radius. Lower SOS values and elevated aBMD values were found at all locations in the diabetic cohort. At the radius, they were also associated with disease duration. Lower QUS parameters have also been found by other QUS devices in diabetic patients. More recently, the application of REMS, which analyses the ultrasound backscatter of the trabecular tissue at the femoral neck and spine, revealed lower REMS-BMD and elevated DXA T-Scores in postmenopausal women with T2DM (37). Similar results were found in a similar cohort with a device that employs a cortical transverse transmission method at the phalanges. The longitudinal study found no difference between diabetic and control patients at baseline. Three years later, the Amplitude-dependent Speed Of Sound (Ad.SOS) and Bone Transmission Time (BTT) decreased. The authors attribute this finding to an increased cortical porosity, in line with our results, as well as to an increased trabecular density (47). The application of the same device in premenopausal women with T1DM yielded reduced QUS parameters (Ad.SOS, Ultrasound Bone Profile Index (USBPI), T-Score, Z-Score), which could be associated with poor glycemic control (48). Another longitudinal study has used this device on young women with T1DM. Both at baseline and 10 years later, the QUS parameters were lower compared to controls, but no significant decline was found when the baseline was compared to the 10-year follow-up (49). Another study in a cohort of T2DM and non-diabetic men (50) focused on investigating the impact of lifestyle factors on bone health. In both groups, a decrease in heel bone stiffness was observed with age, although the association with age was only significant in healthy controls. Apart from that, the only difference displayed by diabetic patients was a slightly lower stiffness index in the age group of 30-40.

Interestingly, some studies have shown unchanged or improved QUS parameters in diabetic patients. Sosa et al. (34) conducted a study with a transverse transmission QUS device at the calcaneus and DXA on postmenopausal, T2DM, and obese women and healthy controls. BMD values of the lumbar spine were elevated in diabetic patients, however, there were no differences in QUS parameters. Yamaguchi et al. (51) conducted a study to assess the risk of vertebral fractures in T2DM patients using a QUS device at the calcaneus (CM-200; Elk Corp., Osaka, Japan). Neither BMD values nor QUS parameters, such as calcaneal SOS, were associated with a prevalent vertebral fracture in T2DM. Dobnig et al. (36) applied calcaneus ultrasound as well as axial transmission at the phalanges and radius in nursing home female patients with T2DM. In diabetic patients, the QUS parameters (e.g., calcaneus stiffness, radial and phalangeal SOS) were elevated, together with lower osteocalcin and parathyroid hormone. In a more recent study by Lasschuit et al. (52), another heel QUS device (Cuba sonometer) was applied in men and women with T2DM. In the follow-up, both BMD and broadband ultrasound attenuation were increased in diabetic patients, and no increased risk of fracture could be observed. Discrimination performance by ROC for DXA and QUS parameters was similar across all groups. For example, BUA performed similarly to femoral neck BMD in predicting fractures (52). Almost all these studies have utilized devices at the calcaneus, which are known to perform similarly to DXA in fracture risk discrimination in non-diabetic cohorts (53). Conti et al. (54) assessed which factors correlated with calcaneus-QUS parameters (BUA, SI/QUI, estimated BMD), with special regard to diabetic complications, being the only study that applied a QUS device in a cohort of both T1DM and T2DM. The QUS parameters were similar in both DM types, although there was a slight reduction in the estimated BMD of T1DM patients. The main findings were an association between diabetic neuropathy and worse QUS parameters and increased QUS parameters with increasing BMI.

The calcaneus is mostly made up of trabecular bone, for which HR-pQCT studies have shown normal to improved microstructure in T2DM (28). The broadband ultrasound attenuation measured by QUS has been proposed to be more of a reflection of bone mass rather than microstructure (55), which potentially explains why the application of calcaneus devices in T2DM has shown improved or similar QUS parameters to DXA when compared to controls. Our CortBS results show moderate correlations to DXA parameters and improved fracture risk discrimination, which is inconsistent with the findings by Lasschuit et al. (52) and Yamaguchi et al. (51) using calcaneal QUS. Concerning the complexity of bone tissue microstructure alteration in patients with DM, it is reasonable to assume that not all QUS devices may be suited for improving fracture risk discrimination compared to DXA. While most bone QUS technologies measure acoustic properties, e.g. speed of sound and broadband ultrasound attenuation, which may be empirically associated with fragility fractures, CortBS measures for the first time intracortical microstructural and viscoelastic cortical tissue properties, which have previously been shown to provide superior fracture discrimination performance compared to DXA in postmenopausal women with low bone mineral density (39). This study confirmed that CortBS parameters are associated with prevalent fractures and revealed additional related to reduces bone turnover in both type 1 and type 2 diabetes.

Both type 1 and type 2 diabetes. The combination of multiple CortBS parameters by means of PLS discrimination models provided an improved fracture discrimination performance compared to DXA.

The relationship between tibial measurements and fracture risk at other sites may seem counterintuitive. However, pathological skeletal deterioration is systemic and is therefore reflected in different bones throughout the body, including bones less susceptible to osteoporotic fractures. Areal BMD measured by DXA at the femur and vertebra are well known to predict fractures beyond these specific anatomic sites. The FRAX tool, based on DXA, estimates the risk of major osteoporotic fractures, including those at the shoulder and wrist. Several studies using HR-pQCT report associations between cortical parameters of peripheral bones, including the tibia, with the prevalence of fractures or fracture risk at major sites (43, 56–66). Therefore, measuring cortical bone using ultrasound at the tibia may also reflect bone characteristics at less accessible sites like the femoral neck and vertebra. Indeed, Iori et al. have shown in ex-vivo studies that cortical thinning and the accumulation of large intracortical pores in the tibia midshaft is associated with structural deterioration of the femoral neck (67) and reduced femur strength (68). Moreover, in a previous pilot study CortBS parameters measured in postmenopausal women provided a superior discrimination performance for vertebral and non-vertebral fractures compared to DXA. Our findings are also in line with an ex-vivo study of Cirovic et al. (69), where biopsies from male donors with T2DM revealed higher cortical porosity (Ct.Po) and lower cortical thickness (Ct.Th). T2DM subjects also showed significant differences in the structure model index and lower cortical and trabecular bone microhardness compared to controls. These changes increase the risk of femoral neck fractures in T2DM. The deteriorated cortical microarchitecture and reduced bone microhardness suggest altered bone composition in the superolateral femoral neck, which is not detected by standard DXA measurements.

Several limitations should be noted. First, this study is cross-sectional, investigating prevalent, not incident fractures, regardless of fracture mechanisms (inadequate trauma vs. fragility fracture). This approach was necessary due to the overall low number of recorded vertebral fractures. Additionally, some asymptomatic vertebral fractures might have been missed, as we did not routinely perform a complete lateral vertebral assessment and morphometric analysis. Incomplete laboratory data prevented the analysis of bone turnover parameters in relation to cortical parameters by CortBS, limiting our ability to further support our hypothesis. Moreover, 93 screened diabetic patients were excluded due to a high BMI (> 35 kg/m²). Given the high prevalence of metabolic syndrome in this risk group, future studies are needed to investigate a population with unrestricted BMI for broader applicability. No real-time analysis of the CortBS measurement quality was possible in this study, which led to the exclusion of data from 5 subjects during the post-hoc data analysis. For clinical applications, the data quality assessment needs to be incorporated into the measurement, providing real-time feedback to the operator and the possibility to repeat the measurement, until an appropriate data quality is achieved. Finally, the study population was comparatively small for each diabetes type and heterogeneous regarding diabetes history and medication, introducing a risk of confounding, by e.g. underlying medication, bone turnover parameters, concomitant diseases and medication, and others. A potential measurement bias cannot be completely ruled out due to the lack of cross-calibration between the devices; however, this is likely minimal given the small number of patients with external scans. Future studies with larger sample size, more homogenous patient and control cohorts, a prospective study design, and an improved ultrasound data acquisition protocol, including an immediate data analysis after the measurement, should therefore be conducted to confirm our results.