Phenotypic characteristics

A total of 153 patients with OI (102 boys and 51 girls), with a mean age of 10.18 ± 3.53 years (range, 1.33 to 18.00 years), were enrolled in this study. The median age at initial fracture was 2.7 years (IQR, 1.5, 6), and the median number of previous fractures was 4 (IQR, 3, 6). Of these patients, 40.5% (62/153) had a family history of fractures. The mean Z-scores for height and weight of patients with OI were − 1.079 ± 2.825 and 0.018 ± 1.878, respectively. The distribution of Sillence classification was as follows: type I (n = 70), type III (n = 33), type IV (n = 43), and type V (n = 7). The characteristics of patients with different clinical severity are shown in Table 1.

Table 1 Phenotypic characteristics of OI patients

All patients with OI underwent genetic testing, and the identified gene variants were as follows: COL1A1 (n = 68), COL1A2 (n = 29), IFITM5 (n = 7), WNT1 (n = 3), FKBP10 (n = 3), P3H1 (n = 3), BMP (n = 2), PLS3 (n = 2), SEC24D (n = 2), CRTAP (n = 1), PLOD2 (n = 1), P4HB (n = 1), TMEM38B (n = 1), SERPINF1 (n = 1), SERPINH1 (n = 1), and no genetic variants were detected in the remaining 28 patients (Fig. S2 and Tables S1 and S2) [29].

TBS and its correlations with phenotypes of OI

Among the 153 patients, the mean TBS Z-score was − 1.249 ± 1.653, with 28.8% (44/153) showing degraded vertebral microarchitecture (Table 1). TBS Z-scores of type I/III/IV patients with OI were all significantly less than zero, indicating that TBS values in patients with OI were lower than those in normal children and adolescents. However, the differences in TBS and its Z-score among patients with OI type I, III, and IV were not significant (P = 0.159 and P = 0.158, respectively) (Fig. 1 and Table 1), which may be attributed to the relatively small sample size.

Fig. 1

Comparison of TBS, TBS Z-score, lumbar spine aBMD, and its Z-score in OI patients with different clinical classifications. A Comparison of TBS among patients classified by OI clinical type. B Comparison of lumbar spine aBMD among patients classified by OI clinical type. C Comparison of TBS Z-score among patients classified by OI clinical type. D Comparison of lumbar spine aBMD Z-score among patients classified by OI clinical type. Statistical analysis was performed using one-way ANOVA for normally distributed data and Kruskal–Wallis test for non-normally distributed data. Data were presented as violin plots with median (solid line) and interquartile range (IQR, dotted line). TBS, trabecular bone score; aBMD, areal bone mineral density; OI, osteogenesis imperfecta. *P < 0.05, **P < 0.01, ***P < 0.001

TBS showed a weak negative correlation with age (r =  − 0.187, P < 0.05) (Fig. 2A). In addition, TBS Z-score was negatively correlated with age (r =  − 0.435, P < 0.001) (Fig. 2B), and this correlation remained significant after adjusting for sex, height, weight, and clinical classification (r =  − 0.308, P < 0.001). No correlations were found between TBS with sex, height, or weight of the patients.

Fig. 2

Relationships between TBS, TBS Z-score and age in patients with OI. A The correlation between TBS and age in patients with OI. B The correlation between TBS Z-score and age in patients with OI. Linear regression analyses were conducted to evaluate the relationships between TBS, TBS Z-score, and age. Correlations were significant for both TBS (r =  − 0.187, P < 0.05) and TBS Z-score (r =  − 0.435, P < 0.001). Simple linear regression lines are shown in each panel. TBS, trabecular bone score; OI, osteogenesis imperfecta

The Z-scores of aBMD at the lumbar spine and femoral neck were − 1.498 ± 2.131 and − 2.795 ± 2.018, respectively, and the Z-score of aBMDHAZ was − 0.839 ± 1.913 (Table 1). The differences in lumbar spine aBMD and its Z-score among patients with OI type I, III, and IV were significant (both P < 0.001) (Fig. 1). TBS Z-score was positively correlated with the Z-score of lumbar spine aBMD (r = 0.447, P < 0.001) and femoral neck aBMD (r = 0.295, P < 0.01) (Fig. S3), and this correlation remained significant after adjusting for age, sex, height, and weight (lumbar spine, r = 0.576; femoral neck, r = 0.377; both P < 0.001). The correlation between the TBS Z-score and the aBMDHAZ Z-score was also significant (r = 0.392, P < 0.001). Meanwhile, there were no significant correlations between TBS or its Z-score and serum β-CTX or ALP levels.

The annual fracture rate was considered to be an important parameter of bone fragility. However, there were no significant correlations between the annual fracture rate with Z-scores of TBS (r =  − 0.036, P = 0.489) and lumbar spine aBMD (r =  − 0.063, P = 0.476).

TBS and VCFs discrimination

VCFs were found in 45.1% (71/153) of patients with OI (Table 2), with Grade 1 of VCFs in 24 patients, Grade 2 in 17 patients, and Grade 3 in 30 patients. Thirty-five percent (25/71) of patients with VCFs did not show densitometric osteoporosis. Patients with VCFs had a significantly lower TBS Z-score than those without VCFs (− 1.778 ± 1.613 vs − 0.748 ± 1.539, P < 0.001).

Table 2 Comparison of patients with OI with and without vertebral fractures

In the entire study population, the AUCs in ROC plots for discriminating VCFs were 0.667 and 0.666 for Z-score of TBS and lumbar spine aBMD, respectively, with no significant difference between them (Fig. 3A). Interestingly, among the patients without densitometric osteoporosis according to their aBMD (n = 37), TBS Z-score showed a greater AUC for identifying VCFs than Z-score of lumbar spine aBMD (0.719 vs 0.545, P < 0.05) (Fig. 3B), indicating the higher discriminative power of TBS Z-score for VCFs.

Fig. 3

ROC curve analysis for the discriminative value of TBS Z-score, Z-score of lumbar spine aBMD and the combinations of the two measurements in identifying VCFs. A ROC curve analysis for the accuracy of TBS Z-score, Z-score of lumbar spine aBMD, and their combination in discriminating VCFs in all patients. B ROC curve analysis for the accuracy of TBS Z-score, Z-score of lumbar spine aBMD, and their combination in discriminating VCFs in patients without densitometric osteoporosis. ROC, receiving operator characteristic; TBS, trabecular bone score; aBMD, areal bone mineral density; VCF, vertebral compression fracture

Correlation of TBS and spinal deformity

Patients with OI had a median SDI of 1 (IQR, 0, 10), of which the SDI value of patients with OI type III was significantly higher than that of patients with OI type I (P < 0.05) (Table 1). Patients with VCFs (n = 71) had a median SDI value of 10 (IQR, 3, 15), which was significantly higher than those without VCFs (P < 0.001).

Negative correlations were found between SDI with the Z-score of TBS (r =  − 0.446) (Fig. 4A), lumbar spine aBMD (r =  − 0.479) (Fig. 4B), and aBMDHAZ (r =  − 0.423) (all P < 0.001). However, in patients without densitometric osteoporosis, only TBS Z-score exhibited a negative correlation with SDI (r =  − 0.358, P = 0.035) (Fig. 4C), while the Z-score of lumbar spine aBMD (r =  − 0.250, P = 0.148) (Fig. 4D) or aBMDHAZ (r =  − 0.242, P = 0.162) was not significantly correlated with SDI.

Fig. 4

Correlations between Z-scores of TBS or lumbar spine aBMD and SDI. A The correlation between TBS Z-score and SDI in all OI patients. B The correlation between Z-score of lumbar spine aBMD and SDI in all OI patients. C The correlation between TBS Z-score and SDI in OI patients without densitometric osteoporosis. D The correlation between Z-score of lumbar spine aBMD and SDI in OI patients without densitometric osteoporosis. Linear regression analyses were performed to evaluate the correlations between TBS Z-score, lumbar spine aBMD Z-score, and SDI. Simple linear regression lines and corresponding equations were shown in each panel. TBS, trabecular bone score; aBMD, areal bone mineral density; SDI, spinal deformity index

Association of TBS and genotype of OI

In this study, 104 patients with OI were categorized into the AD OI and 21 were non-AD OI (Table S3). Among the 97 patients with OI with pathogenic variants in COL1A1 or COL1A2, 34 patients carried the nonsense or frameshift variants, while 46 had glycine substitution variants in either COL1A1 or COL1A2. The remaining 17 patients with OI were identified with splicing variants of COL1A1 or COL1A2. No differences in TBS or its Z-score were found between the AD and non-AD OI (Table S3) [29]. Also, no significant difference was found in TBS or its Z-score between patients with the quantitative reduction and with structural defects of type I collagen (Table S3) [29].

Sub-group analysis of OI patients with identified pathogenic variants

A total of 93 patients with OI (60 boys and 33 girls), with a mean age of 10.27 ± 3.60 years, were found to carry pathogenic variants, including variants in COL1A1 (52.6%), COL1A2 (25.8%), ITIFM5 (7.5%), WNT1 (3.2%), PLS3 (2.1%), P3H1 (2.1%), and FKBP10, CRTAP, PLOD2, TMEM38B, SERPINF1, SERPINH1 (each 1.0%) (Supplementary Table 1). In this genetically homogeneous sub-group, the mean TBS Z-score was − 1.254 ± 1.676, and 30.1% (28/93) of patients displayed degraded vertebral microarchitecture. A negative correlation was found between age and TBS Z-score (r =  − 0.421, P < 0.001), which remained significant after adjusting for sex, height, weight, and clinical classification (r =  − 0.300, P < 0.01). Additionally, TBS Z-score was positively correlated with the aBMD Z-score of the lumbar spine (r = 0.364, P < 0.01).

Patients with VCFs had a significantly lower TBS Z-score than those without VCFs (− 1.705 ± 1.445 vs − 0.793 ± 1.784, P < 0.01). In this cohort with identified pathogenic variants, the AUC for discriminating VCFs using TBS Z-score was 0.669, and the AUC value was 0.651 using lumbar spine aBMD Z-score. Among patients without densitometric osteoporosis according to their aBMD (n = 24), TBS Z-score exhibited a higher but non-significant AUC for identifying VCFs than lumbar spine aBMD Z-score (0.774 vs 0.590, P = 0.117). According to the deformation of the spine, the median SDI of these patients with VCFs (n = 46) was 11 (IQR, 5, 14), which was significantly higher than that of patients without VCFs (P < 0.001). Negative correlations were found between SDI with TBS Z-score (r =  − 0.359, P < 0.01) and lumbar spine aBMD Z-score (r =  − 0.448, P < 0.001).