A total of 296 patients (with a male-to-female ratio of 61:235) met the inclusion criteria. Among them, 60 patients (20.27%) experienced new vertebral fractures post-surgery, with 37 (61.67%) developing new compression fractures of adjacent vertebrae (AVCF).

BMI, age, and bone mineral density values showed no statistically significant differences between patients treated at ShaoGuan First People’s Hospital and Yuebei People’s Hospital. However, there was a statistically significant difference in the amount of polymethylmethacrylate (PMMA) injected between the two hospitals. The average PMMA injection volume at ShaoGuan First People’s Hospital was 3.50 ± 1.02 ml, while at Yuebei People’s Hospital, it was 4.50 ± 1.62 ml, with a significant difference (P < 0.001) and a mean difference of 0.8 ± 0.2 ml. Furthermore, a comparison of the proportion of bilateral and unilateral vertebroplasty (VP) procedures revealed discrepancies between the hospitals. At ShaoGuan First People’s Hospital, only 22.4% (11/49) of the procedures were bilateral, whereas at Yuebei People’s Hospital, this proportion was 63.2% (156/247), indicating inconsistency in surgical approaches. Despite these differences, there were no statistically significant disparities in the rates of subsequent vertebral fractures between the two hospitals. The rate of subsequent fractures at ShaoGuan First People’s Hospital was 14.3% (7/49), while at Yuebei People’s Hospital, it was 21.5% (53/247), with an independent samples t-test yielding a P-value of 0.21.

Additionally, when analyzing all patients collectively, regardless of hospital, those who underwent bilateral VP procedures had a significantly higher PMMA injection volume compared to those who underwent unilateral procedures. Among the 167 patients who underwent bilateral procedures, the average PMMA injection volume was (5.00 ± 1.55) ml, whereas among the 129 patients who underwent unilateral procedures, it was (3.25 ± 1.16) ml, with a statistically significant difference (P < 0.0001). Furthermore, the choice of anesthesia method, whether local or general, did not significantly influence the amount of PMMA injected during VP procedures. Among the 260 cases performed under local anesthesia, the average PMMA injection volume was (4.00 ± 1.6) ml, while among the 36 cases performed under general anesthesia, it was (4.09 ± 1.37) ml, with a non-significant difference (P = 0.2458).

A comparison between the new vertebral fracture group and non-new vertebral fracture group revealed no statistically significant difference in age, gender, BMI, anti-osteoporosis treatment, thoracolumbar fractures, bone cement volume, reduction rate, reduction angle, and contact with endplate (P > 0. 05). Statistical significance was observed in lumbar BMD (P = 0. 028), previous OVCF (P < 0. 001), CT values (P = 0. 005), Kummell’s sign (P = 0. 003), bone cement leakage (P < 0. 001), and Cement distribution pattern (P = 0. 001), as shown in Table 1.

Univariate analysis comparing patients with and without new vertebral fractures revealed no statistical significance for gender, age, BMI, steroid use, hospital-procedure time, surgery time, bone cement volume, number of treated vertebral, prescription of anti-osteoporosis, thoracolumbar vertebral fractures, reduction rate, reduction angle, and contact with endplate (P > 0.05). On the other hand, the univariate analysis indicated a significant association between bone cement leakage (P < 0.001), previous osteoporotic vertebral compression fractures (OVCF) (P < 0.001), and Kummell’s sign (P = 0.004) with an increased risk of new vertebral fractures. Conversely, bone cement pattern (P = 0.002), bone mineral density (BMD) (P = 0.029), and CT values (P = 0.032) were significantly associated with a decreased risk of new vertebral fractures. Multifactorial logistic regression analysis further confirmed the significant association of previous OVCF (P < 0.001), bone cement leakage (P < 0.001), and Kummell’s sign (P = 0.006) with an increased risk of new vertebral fractures, as detailed in Table 2.

Univariate analysis comparing patients with and without new AVCF indicated no statistical significance for gender, age, bone cement volume, number of treated vertebral, prescription of anti-osteoporosis, thoracolumbar vertebral fractures, reduction rate, and CT values (P > 0. 05). Conversely, univariate analysis revealed a significant association between Kummell’s sign (P = 0.019), bone cement leakage (P < 0.001), and previous osteoporotic vertebral compression fractures (OVCF) (P < 0.001) with an increased risk of developing new AVCF. Multifactorial logistic regression analysis confirmed that previous OVCF (P = 0.001), bone cement leakage (P < 0.001), and Kummell’s sign (P = 0.034) were significantly associated with an increased risk of new AVCF development, as detailed in Table 3.

Predictive Performance Analysis of Zhong et al. in 2015.

To understand the discrepancies between our data and the modeling data provided by Zhong et al. in 2015, we conducted a single-sample t-test based on the data provided in their article. Our analysis revealed slight decreases in age and CT value compared to the modeling data. However, there were no differences in gender proportions (Fig. 1a).

Predictive Performance Analysis of Zhong et al. in 2015. a Single-sample t-test revealed discrepancies in age and CT value between our data and Zhong et al. in 2015’s modeling data. No differences were found in gender proportions. *P < 0.05. b Kaplan–Meier analysis demonstrated significant predictive power of the model, dividing patients into high-risk (> 8.5) and low-risk (≤ 8.5) groups for new vertebral fractures. High-risk group: approximately 50% free from new fractures after 3 years; low-risk group: proportion exceeding 80%. c ROC curve analysis indicated better predictive ability for long-term occurrence of new vertebral fractures with 1-year AUC of 0.570, 2-year AUC of 0.617, and 3-year AUC of 0.664

In the Kaplan–Meier (KM) analysis, it is evident that the division of patients into high-risk (> 8.5) and low-risk (≤ 8.5) groups based on the model demonstrates statistical significance in predicting subsequent vertebral fractures. The high-risk group exhibits a proportion of patients free from new fractures after 3 years at approximately 50%, while the low-risk group shows a proportion exceeding 80% (Fig. 1b).

However, in the overall receiver operating characteristic (ROC) curve analysis, we observed that the 1-year AUC was 0.570, the 2-year AUC was 0.617, and the 3-year AUC was 0.664 (Fig. 1c). These results indicate a better predictive ability for long-term occurrence of new vertebral fractures.

Predictive Performance Analysis of Zhong et al. in 2016.

To further assess the disparities between our data and the modeling data provided by Zhong et al. in 2016, we conducted a single-sample t-test using the data provided in their article. Our analysis revealed a slight decrease in age compared to the modeling data, while CT value and gender proportions showed no differences (Fig. 2a).

Predictive Performance Analysis of Zhong et al. in 2016. a Single-sample t-test revealed a slight decrease in age compared to Zhong et al. in 2016’s modeling data, while CT value and gender proportions showed no differences. *P < 0.05. b Kaplan–Meier analysis demonstrates significant predictive power in stratifying patients into high-risk, intermediate-risk, and low-risk groups for adjacent new vertebral fractures. c Receiver operating characteristic (ROC) curve analysis yielded an AUC of 0.738, indicating robust predictive ability for adjacent new vertebral fractures

In the KM analysis, it is evident that the division of patients into high-risk (6 points), intermediate-risk (2 and 4 points), and low-risk (0 points) groups based on the model demonstrates statistical significance in predicting subsequent adjacent vertebral fractures. The proportion of patients free from adjacent new fractures was over 93% in the low-risk group, over 81% in the intermediate-risk group, and lower than 50% in the high-risk group (Fig. 2b).

Moreover, in the overall receiver operating characteristic (ROC) curve analysis, we observed an AUC of 0.738, indicating a strong predictive ability for adjacent new vertebral fractures (Fig. 2c).

Predictive Performance Analysis of Li et al. in 2021.

To assess the disparities between our data and the modeling data provided by Li et al. in 2021, we conducted a single-sample t-test using the data provided in their article. Our analysis revealed slight differences in various parameters between our and the modeling data. Specifically, in the Control group (no new vertebral fracture), CT value, BMI, available anti-osteoporotic treatment, bone cement leakage, contact between bone cement and endplate, and bone cement dispersion showed minor variances compared to the modeling data, while no differences were observed in age and gender proportions. Similarly, in the NVCF group (new vertebral fracture group), available anti-osteoporotic treatment, age groups of 60–70 years and over 80 years, and CT value exhibited slight differences compared to the modeling data. However, BMI, age groups under 60 and between 70 and 80 years, bone cement leakage, contact between bone cement and endplate, bone cement dispersion, and gender proportions showed no disparities (Fig. 3a).

Predictive Performance Analysis of Li et al. in 2021. a Comparison of our data with Li et al. in 2021’s modeling data revealed minor differences in parameters for both Control and NVCF groups. *P < 0.05. b The ROC curve analysis of Li et al.in 2021’s predictive model resulted in an AUC of 0.518. c The calibration curve demonstrated that the model tended to overestimate the actual number of new fractures

We observed that the predictive model developed by Li et al. in 2021 exhibited a calculated AUC of 0.518 (Fig. 3b). Additionally, upon examination of the calibration curve, it was evident that the model predicted a high probability and tended to overestimate the actual number of new fractures (Fig. 3c).

Predictive Performance Analysis of Li et al. in 2022.

To assess the disparities between our data and the modeling data provided by Li et al. in 2022, we conducted a single-sample t-test using the data provided in their article. Our analysis revealed slight differences in various parameters between our Control group (no new vertebral fracture) and the modeling data. Specifically, in the Control group, variables such as steroid use, multiple vertebral fracture, anti-osteoporosis therapy, surgery time, bone mineral density (BMD), age, and BMI showed minor variances compared to the modeling data, while no differences were observed in age, gender proportions, bone cement leakage, and injection volume. Similarly, in the NVCF group (new vertebral fracture group), variables such as steroid use, anti-osteoporosis therapy, bone cement leakage, hospitalization to surgery duration, BMD, and age exhibited slight differences compared to the modeling data. However, surgery time, BMI, injection volume, and gender proportions showed no disparities (Fig. 4a).

Predictive Performance of Li et al. in 2022. a Comparison of our data with Li et al. in 2022’s modeling data revealed minor differences in parameters for both Control and NVCF groups. *P < 0.05. b ROC curves were plotted to evaluate the predictive model developed by Li et al. in 2022, resulting in a model AUC of 0.556. c The calibration curves indicated that the model tended to predict a low probability of new fractures and consistently underestimated the actual number of occurrences

The ROC curves were constructed to evaluate the predictive model developed by Li et al. in 2022, resulting in a model AUC of 0.556 (Fig. 4b). Additionally, the calibration curves indicated that the model tended to predict a low probability of new fractures and consistently underestimated the actual number of occurrences (Fig. 4c).