Safety, Efficacy, and Adjacent-level Fracture Risk Following Vertebral Augmentation and Radiofrequency Ablation for the Treatment of Spine Metastases in Patients with Cancer

Abstract

Purpose

To evaluate the safety and efficacy of vertebral augmentation (VA) and radiofrequency ablation (RFA) for treating pathologic spinal fractures in patients with cancer and adjacent fracture risk following treatment.

Materials and Methods

This single-institution retrospective study, conducted from January 2017 to September 2020, included patients with cancer who underwent percutaneous VA with or without spine RFA for pathologic spine compression fractures. The primary outcome was pain reduction, assessed using the 10-point visual analog scale before the procedure, at first follow-up, and 2–4 weeks after the procedure. Logistic regression was performed to identify factors associated with epidural cement leak.

Results

A total of 638 spinal levels in 335 patients (median age, 65 years [IQR, 58–74 years]; 147 female, 188 male) were treated. Epidural tumor and retropulsion of bone fragments were present in 15% (93 of 638) and 13% (81 of 638) of treated levels, respectively, while posterior wall erosion was observed in 30% (190 of 638). Substantial pain improvement (greater than two-point reduction) occurred in 81% (519 of 638) of cases. Factors associated with decreased risk of epidural cement leak included RFA (42% no leak vs 38% leak, P = .03) and lumbar treatments (49% no leak vs 38% leak, P = .02). Adjacent-level fractures occurred in 10.4% of patients. The total complication rate (National Cancer Institute Common Terminology Criteria for Adverse Events grade 3 or higher) was 0.6% (four of 638).

Conclusion

VA and RFA are safe and efficacious treatments for spine fractures in patients with cancer.

Keywords: Ablation Techniques, Kyphoplasty, Vertebroplasty

© RSNA, 2025

Summary

Vertebral augmentation, with or without radiofrequency ablation, safely improved pain at 4 weeks in patients with cancer with spinal fractures, including those with posterior vertebral body wall disruption.

Key Points

• ■ Vertebral augmentation, with or without radiofrequency ablation (RFA), resulted in substantial (greater than two points) pain improvement in 81% (519 of 638) of cases, and the overall major complication rate (National Cancer Institute Common Terminology Criteria for Adverse Events grade ≥ 3) was 0.6% (four of 638).

• ■ RFA and lumbar-level treatment were associated with decreased risk of epidural cement leak (P = .03 and P = .02, respectively).

• ■ Adjacent-level spine fractures occurred in 10.4% of patients.

Introduction

The spine is the most common site of osseous metastases, with an incidence of 31%–70% in patients with cancer overall and in 85% of patients with breast, lung, prostate, renal, and thyroid cancers (1,2). These metastases can result in substantial morbidity, including pain, hypercalcemia, pathologic fractures, vertebral collapse, and spinal cord or nerve compression, culminating in impaired mobility, loss of functional independence, and severely diminished quality of life (3,4).

Managing vertebral metastases involves a multidisciplinary approach, including treatment with analgesics, bone-modifying agents (bisphosphonates and osteoclast inhibitors), systemic cancer therapies, radiation therapy, and surgery. Management principles are often broadly categorized into approaches that address biologic pain (ie, radiation therapy, ablation, systemic therapies) and mechanical pain (ie, stabilization techniques). With regard to the former, radiofrequency ablation (RFA) is a well-established and safe modality for treating cancer-associated bone pain (5,6). However, the indication for this treatment option for patients with advanced-stage cancer who are also undergoing systemic cancer therapies and radiation therapy is not well understood. Likewise, with regard to stabilization techniques, percutaneous vertebral augmentation (VA) is a cornerstone in the management of painful acute compression fractures; however, safety and efficacy data in scenarios unique to the cancer population, such as sclerotic metastases, epidural tumor spread, and posterior vertebral wall erosion, are limited (7,8). For example, the OPuS One study provided high quality evidence regarding the safety and efficacy of RFA and VA for spine metastases (9,10). However, blastic lesions were excluded; moreover, while note was made that posterior wall disruption was common in the study group, the specific frequency was not reported. Similarly, the rate of epidural tumor extension at the treated level was not reported, and so the safety of RFA or VA in this context is unclear. Furthermore, most patients (81%) did not undergo radiation therapy; given the prevalence of radiation therapy for spine metastases, persistent questions regarding the safety of RFA in combination with radiation remain. Finally, the influence of stabilization in the metastatic spine and its effect on the risk of adjacent-level fractures, unlike for the osteoporotic spine, remains a knowledge gap in the literature. Therefore, while there are strong data for RFA and VA in the treatment of spine metastases, there remain unanswered questions regarding these interventions in common oncologic scenarios.

The purpose of this single-institution retrospective study was to evaluate the safety and efficacy of RFA and VA for treating spine fractures in a real-world setting for patients with advanced cancer.

Materials and Methods

Study Design

This retrospective, single-center study was performed with institutional review board approval and waiver for written informed consent. At the study institution, patient evaluation for VA includes assessment by a multidisciplinary team, including oncology, neurosurgery, and radiation oncology. Patients who are unable to tolerate analgesic, radiation, systemic, and/or surgical therapies or who refused these therapies are often referred for VA. Patients who had contraindications to percutaneous VA, including active infection, uncorrectable bleeding diathesis, or inability to tolerate procedural sedation, were not eligible for VA. Furthermore, the decision to perform VA for patients with relative contraindications such as posterior wall erosion by tumor, epidural extension of tumor, retropulsion of fracture fragments, and pre-existing neuropathies were deemed eligible after discussion of neurosurgery prior to performing VA.

For this analysis, all patients with spine fractures identified at cross-sectional imaging who underwent VA with or without concurrent RFA between January 2017 and September 2020 were included. Patients with implants such as SpineJack (Stryker) were excluded from the analysis to minimize heterogeneity among the study sample.

Procedural Technique

VA with or without concurrent RFA was performed by seven interventional radiologists with at least 5 years of experience each in spine interventions. Sedation by the anesthesia service was provided for all procedures. The choice of imaging modality (fluoroscopy or CT) was based on operator preference. Likewise, the decision to perform vertebroplasty (ie, cement delivery only) versus kyphoplasty (ie, balloon or implanted device) and whether to perform RFA concurrently was at the operator’s discretion. No patient underwent RFA without concurrent cement augmentation. Vertebroplasty devices and bone cement (polymethyl methacrylate) from a variety of manufacturers (Cook/AZA, Merit, Stryker, and Medtronic) were used during the study period. For most patients, bilateral access was obtained. Both transpedicular as well as parapedicular approaches were used, depending on anatomy. RFA was first performed using either the Medtronic OsteoCool system (255 of 264, 97% of ablated levels) or the Merit STAR system (nine of 264, 3% of ablated levels). The RFA treatment was delivered according to the prescribed manufacturer algorithm (as the OsteoCool system automatically sets the ablation time), with ablation times as determined with probe length and device protocol. Bilateral kyphoplasty balloons were used, with the balloon size selected at the operator’s discretion and based on the vertebral body size. Bone cement was then injected under image guidance until adequate fill was achieved as determined by the operator. Thermoprotective techniques such as epidural temperature monitoring or saline perfusion in the cerebrospinal fluid space were not performed for any patient.

Radiomics-based Model for Predicting Adjacent-level Fractures

CT scans for patients who underwent VA were accessed from the institution’s picture archiving and communication system server and downloaded in Digital Imaging and Communications in Medicine (DICOM) format. All CT scans performed after the VA were included in the subsequent analysis. A publicly available deep learning–based organ segmentation tool (11) was used to generate regions of interest (ROIs) for all vertebral body levels in all of the CT scans. All spinal-level volumes were calculated from segmentations, after which they were longitudinally analyzed. We defined adjacent spine-level fracture as a sustained loss of volume of at least 25% from baseline in an adjacent spinal level following cement augmentation. We defined adjacent spine levels as vertebral bodies located within two spinal levels of the level that underwent cement augmentation. The baseline volume for adjacent spinal levels was defined as the volume of that spinal level on the CT scan immediately prior to cement augmentation.

Radiomics features for adjacent spine levels were extracted in the following manner: ROIs for adjacent spine levels and the associated DICOM CT images were loaded using the pydicom package (https://pydicom.github.io/). Images from CT scans obtained immediately prior to cement augmentation were used for radiomics feature extraction. Quantitative radiomics features were then extracted from those ROIs using the pyradiomics package (https://github.com/AIM-Harvard/pyradiomics) (12). A total of 949 radiomics features were extracted, following the application of several preprocessing filters to the CT data.

A machine learning (ML) model for predicting adjacent-level fractures was then generated in the following manner: The radiomics features for all adjacent spine levels were loaded and converted to a data matrix within the R platform (version 4.0.3; The R Project for Statistical Computing, www.r-project.org). Spinal levels were classified in a binary manner as either fractured or not fractured based on the sustained decrease in volume as defined above. This matrix was then split into training and testing matrices with a 70%/30% split. An ML model using the XGBoost method was then trained with 100 boosting iterations. Given the relatively lower frequency of fractured levels compared with nonfractured levels, this imbalance was accounted for by scaling the weight given to positive (ie, fracture) cases. Overfitting was avoided by tuning the regularization term γ to 0.1. The relative importance of the radiomics features in the model was evaluated and visualized with an importance matrix. The performance of the model was evaluated by the area under the receiver operating characteristic curve (AUC) on the test dataset.

Study Outcomes

The primary outcome of this study was pain assessment with a 10-point visual analog scale performed prior to the intervention and at the first follow-up appointment scheduled 2–4 weeks following the intervention. Technical success for VA was defined as the percutaneous delivery of bone cement into the target vertebral body level. Technical success for RFA was the performance of ablation to the target vertebral body level according to the operator’s prescribed power and duration parameters. Technical adverse events such as cement leakage or intravasation were identified by evaluating both peri- and postprocedural imaging for immediate cement leakage along the injection tract (13–16). Safety was defined as a procedural complication rate of National Cancer Institute Common Terminology Criteria for Adverse Events (CTCAE) grade 3 or higher of 5% or less. Clinical efficacy was defined as reduction of pain scores by two points or more. Epidural disease was defined as tumor within the epidural space. Posterior wall erosion was defined as disruption of the cortex of the posterior wall of the vertebral body.

A Philips IntelliSpace PACS 4.4 was used to perform all analyses of cross-sectional imaging, including baseline and follow-up MRI and CT. Clinical adverse events were identified after reviewing the electronic medical record and the department’s complication database.

Statistical Analysis

Demographic and procedural details were summarized using descriptive statistics. Two-sample t tests were used to compare means of continuous variables. Fisher exact test was used for comparisons between categorical variables. The paired Wilcoxon t test was used to evaluate changes in pain scores per patient following the intervention. The cumulative incidence function was used to estimate the cumulative incidence of local tumor progression and death in the presence of competing risks, where local tumor progression and death were treated as mutually exclusive outcomes. Logistic regression was used to assess the association of demographic and clinical variables on lack of improvement in pain scores, as well as to identify predictors of epidural or neuroforaminal cement leakage; multivariable analysis was performed on variables with a P value less than .10 on univariable analysis. The variables for regression were identified through a combination of prior literature review and clinical expertise. Variables included demographic factors (age, sex) and procedural variables (multivariable analysis including variables with P < .10) on univariable analysis. All statistical analyses were performed using R statistical software version 4.0.3. A cutoff value of P less than .05 was used for statistical significance.

Results

Patient Characteristics

The study sample comprised 335 patients who underwent treatment of 638 spinal levels (Table 1). The median age was 65 years (IQR, 58–74 years); 44% (147 of 335) were female and 56% (188 of 335) were male. The median follow-up time per patient was 8.5 months (IQR, 1.7–29.4 months). Primary tumor histologies spanned the cancer spectrum (Fig 1); the most frequent were prostate cancer (22%, 74 of 335), breast cancer (16%, 54 of 335), and renal cell carcinoma (13%, 44 of 335). Forty-eight percent (160 of 335) of patients had more than one level treated in a single procedure. The median number of levels treated was one (range, one to seven). The median timing of prior radiation therapy was 216 days prior (range, 3 to 566 days).

Table 1:

Study Cohort Characteristics

Figure 1:

A distribution of primary malignancies in the study cohort. HCC = hepatocellular carcinoma, RCC = renal cell carcinoma, SCC = squamous cell carcinoma.

Technical Success Rates of Procedures

Technical success for both RFA and VA was 100%. Of all treated levels undergoing VA, RFA was performed in 264 levels (264 of 638, 41%) concurrently with VA, and kyphoplasty with balloons was performed in a minority of levels (242 of 638, 37%). A variety of bone densities were encountered at the treated spine levels; 9.4% (60 of 638) of the levels were purely osteoblastic, while the majority (396 of 638, 62%) were mixed lytic and osteoblastic. Four cervical levels (four of 638, 0.6%), 325 thoracic levels (325 of 638, 51%), and 309 lumbar levels (309 of 638, 48%) were treated, with the largest subset of treatments at the thoracolumbar junction (T10-L2) (Fig 1). The median number of levels treated per patient was three, ranging from one to seven. Posterior wall disruption involving greater than 5 mm of the posterior cortical wall was present in 191 (30%) levels; in these levels, the median length of disruption was 13 mm. Retropulsion of fracture fragments into the spinal canal (Bilsky grade 1 or greater) was present at 81 (81 of 638, 13%) levels, and extension of the tumor into the epidural space was present at 93 (93 of 638, 15%) levels (Fig 2). Prior radiation therapy had been administered for 225 (225 of 638, 35%) levels.

Figure 2:

Representative examples of common scenarios in the study cohort. Top tow: Axial CT (noncontrast) images in a 78-year-old female patient with metastatic small cell lung cancer shows a sclerotic metastasis where kyphoplasty balloons could not be inflated. Middle row: Axial CT (noncontrast) images in a 58-year-old female patient with multiple myeloma and extensive lytic destruction of the posterior vertebral wall. Cement augmentation is still achievable with careful technique in such cases. Bottom row: Axial CT (noncontrast) image (left) and axial MR image (T1-weighted) (right) demonstrate the ablation zone (blue arrowheads in bottom right image) in an 81-year-old male patient with metastatic clear cell renal cancer and substantial epidural disease. This patient underwent up-front radiofrequency ablation (RFA) and vertebral augmentation because of severe pain, which precluded stereotactic radiosurgery simulation planning. Following RFA and augmentation, the patient was successfully treated with radiation therapy.

Clinical Outcomes

The median baseline pain score before intervention was 7 ± 2.3 (SD) out of 10. Following the intervention, the median pain score at the time of the first follow-up (3 ± 2.3) was significantly lower than the baseline pain score (P < .001) (Fig 3). Clinical efficacy was achieved in 80% (268 of 335) of patients. A substantial proportion of patients experienced an improvement of four or greater points (36%, 120 of 335). At 6 months, the cumulative incidence of local tumor progression was 17.3% (95% CI: 6.55, 37.9), whereas the cumulative incidence of death was 35.3% (95% CI: 29.8, 40.8) (Fig 4).

Figure 3:

Outcomes following vertebral augmentation with or without radiofrequency ablation. The box plot shows a significant decrease in pain scores following the intervention (P < .001). Light blue (left) represents preprocedural pain, while light green (right) represents postprocedural pain. Individual points represent individual patient pain scores, displayed with jitter to demonstrate distribution. Box edges indicate quartiles, and horizontal lines inside boxes indicate medians.

Figure 4:

Cumulative incidence plot. The cumulative incidence function was used to estimate the cumulative incidence of local tumor progression and death in the presence of competing risks, where local tumor progression and death were treated as mutually exclusive outcomes. At 6 months, the cumulative incidence of local tumor progression was 17.3% (95% CI: 6.55, 37.9), whereas the cumulative incidence of death was 35.3% (95% CI: 29.8, 40.8).

On univariable analysis, several variables were significantly associated with a lack of improvement in pain scores, including older age (P < .001) and lack of kyphoplasty (P < .04) (Table 2). However, these and other clinically relevant variables were not significantly associated with a lack of response on multivariable analysis (Fig 5).

Table 2:

Univariable Analysis of Factors Associated with Lack of Improvement in Pain Following Spinal Vertebral Augmentation Intervention

Figure 5:

Multivariable analysis of predictors for pain nonresponders and epidural cement leakage. On the left, the forest plot demonstrates no significant predictors identified for pain nonresponders. Right, the forest plot demonstrates that treatment level and use of radiofrequency (RF) ablation were significantly associated with a decreased risk of epidural cement leakage. Furthermore, prior radiation therapy was significantly associated with an increased risk of epidural cement leakage. Arrows indicate CIs extending past x-axis limits. OR = odds ratio.

Regarding adverse events, no patients developed infections, prolonged bleeding, or delayed wound healing at the skin puncture sites; this was also true for patients who had previously undergone radiation therapy to the target level before the intervention.

Furthermore, there were no RFA-mediated injuries to the spinal cord, cauda equina, or exiting nerve roots. One patient with a history of renal cell carcinoma developed tumor tract seeding along the path of the vertebroplasty needle; given that the patient had extensive metastatic disease, this did not substantially alter the patient’s management and was treated with systemic therapy. Cement leakage of any kind occurred in 20% (127 of 638) of treated levels. Most of these events involved cement leakage into the intervertebral disk space and had no clinical impact. Regarding overall safety, cement leakage into the neuroforamen or epidural space occurred in 34 (34 of 638, 5.3%) levels; most of these did not result in neurologic compromise except four (four of 638, 0.6%) cases that required repeat surgical or interventional radiology intervention. Overall, there were only four events which were CTCAE grade 3 or higher (0.6%). None of these patients had thermal injuries from the cement. Following surgical decompression, there were no persistent neurologic deficits from the leakage.

Univariable analysis to identify predictors for epidural or neuroforaminal cement leakage identified lumbar level interventions (49% no leak vs 38% leak, P = .02), use of RFA (42% no leak vs 38% leak, P = .03), lack of retropulsion (11% no leak vs 32% leak when retropulsion was present, P = .002), and lack of posterior wall disruption (27% no leak vs 62% leak when posterior wall disruption was present, P < .001) as significantly associated with a decreased risk of leakage (Table 3). On multivariable analysis, RFA (odds ratio [OR], 0.7 [95% CI: 0.56, 0.87]) and the treatment levels (OR, 0.7 [95% CI: 0.56, 0.86]) were significantly associated with decreased odds of cement leakage. Furthermore, prior radiation (OR, 1.02 [95% CI: 1.01, 1.13]) was significantly associated with an increased risk of epidural cement leakage (Fig 5).

Table 3:

Univariable Analysis of Factors Associated with Epidural Cement Leakage Following Vertebral Augmentation

Prediction of Adjacent-level Fractures

After cement augmentation, evaluation for adjacent-level fractures was performed on 134 patients within the study cohort. A total of 12 936 spinal levels were segmented across 686 CT scans performed for these patients. The median imaging follow-up time, defined as the time from the VA procedure and the last available CT imaging study, was 7.3 months (IQR, 2–19.5 months). Adjacent-level fractures occurred in 10.4% of patients (14 of 134) (Fig 6). Temporally, adjacent-level fractures appeared to cluster in two time frames, early and late, with the former occurring within 6 months of the cement augmentation procedure and the latter occurring up to 3 years following the procedure. Adjacent-level fractures occurred more commonly at the thoracolumbar junction (T10–L1) compared with other spinal regions; fractures at these levels accounted for 50% (nine of 18) of adjacent-level fractures. Whether or not the metastatic spinal lesion at the treated spinal level was lytic or blastic was not significantly associated with adjacent-level fracture (P = .10).

Figure 6:

Adjacent-level fractures following cement augmentation of pathologic spine fracture. (A) Sagittal CT (noncontrast) images in a 64-year-old female patient with lung cancer and a T7 spine metastasis who underwent radiation therapy and cement augmentation (left). Two months following cement augmentation, an adjacent-level fracture developed at T8 (right). * indicates T8 spinal level. (B) Deep learning–based spinal-level segmentation of adjacent spine levels following cement augmentation revealed a cluster of early (within 6 months) and late (up to 3 years) adjacent-level fractures (orange). The green lines depict adjacent spinal levels that did not fracture. Orange line indicates line of best fit, and shading around the line indicates the CI of the line of best fit. (C) The distribution of adjacent spinal levels that fractured following cement augmentation was significantly different compared with those that did not fracture, with a greater proportion of fractured levels in the thoracolumbar junction region (T10–L1).

Multiple radiomics features were statistically significantly associated with adjacent spinal levels that developed fractures (Fig 7). The top two differential features were Contrast.log.sigma.3.0.mm.3D (P < .01) and GrayLevelVariance.wavelet.LLL (P < .001). An ML model for adjacent-level fracture based on these radiomics features demonstrated an AUC of 0.78 (95% CI: 0.72, 0.90) on a randomly selected testing cohort (70/30 test/train split, 12 936 spinal levels), with a small subset of radiomics features contributing substantially to the model.

Figure 7:

Radiomics features for predicting adjacent-level spinal fractures following vertebral augmentation. (A) Numerous radiomics parameters were found to be differentially featured in adjacent spinal levels that developed fractures (blue) compared with adjacent levels that did not (red). Jitter dots indicate individual values of each spinal level. Horizontal lines indicate quartiles. ** indicates P < .005. *** indicates P < .0005. (B) A machine learning model based on radiomics features was generated and demonstrated fair performance for predicting risk of adjacent-level spinal fractures, with an area under the receiver operating characteristic curve (AUC) of 0.78. (C) A small subset of radiomics features that contributed substantially to the model. Bar indicates absolute rank and feature importance.

Discussion

This single-institution retrospective study demonstrated that treatment of spine fractures in patients with cancer with VA, with or without RFA, was associated with improvement in pain 2–4 weeks following the intervention. Importantly, this study also found RFA to be a safe intervention, without any thermal injury complications, despite the patient cohort’s diversity of tumor histologies and bone densities. Furthermore, RFA was significantly associated with a decreased risk for epidural or neuroforaminal cement leakage (P = .03) in the subsequent VA procedure; this decrease in cement leakage was seen in both kyphoplasty and vertebroplasty procedures. Improvements in baseline pain after the intervention did not vary by bone density, with clinically significant pain relief identified in osteoblastic, lytic, and mixed-density lesions (P < .001). Furthermore, the combination of RFA with radiation therapy was found to be safe, given the overall low complication rate in this cohort.

The present study’s findings corroborate previous studies in a large patient cohort. For example, Wallace et al (17) reported on 110 spinal metastases treated with RFA combined with VA in the majority of levels and showed a significant decrease in mean and median pain scores before and after the procedure, with 70% local tumor control 1 year after treatment. Another study of 92 patients showed decreased pain that persisted 6 months after the procedure (18). Multiple other studies reported significant decreases in pain scores with minimal complications, using a 10-point visual analog score to assess patient pre- and postprocedural pain (7,19–21). These studies and the current study support the use of VA and RFA as safe treatment approaches despite previous radiation and surgery. Recently, Levy et al (9,10) reported on RFA combined with VA for treating bone metastases in 100 patients. Our data corroborate findings as reported in a recent meta-analysis analyzing 947 patients and 1163 lesions, reporting a substantial improvement in pain scores and similar local tumor progression rates (22).

There are several differences, however, in the present study’s patient cohort relative to the Levy et al study (10). Notably, patients with osteoblastic metastases were excluded in the latter study. The distribution of cancer types was different as well, with more patients with lung cancer and fewer patients with prostate cancer in the Levy et al study. Most importantly, almost all the patients in the Levy et al study did not undergo radiation therapy. Given that radiation therapy is a cornerstone in the management of spine metastases, the present study’s finding that VA and RFA can be safely performed in patients who underwent prior radiation therapy addresses an important knowledge gap in the current literature.

Spinal metastases are a common problem projected to increase in prevalence as advancements in cancer care result in improved patient survival. Radiation therapy is central to the treatment of spine metastases, but there are several clinical scenarios in which VA with or without RFA can complement radiation therapy. For example, while radiation therapy alone is often effective for addressing biologic pain generators, including the metastatic lesion itself, it does not address the mechanical instability caused by these lesions. Additionally, the eventual pain relief achieved with radiation therapy takes several weeks to manifest, with pain frequently worsening before it improves, a phenomenon termed the flair effect. This is an especially important consideration as pain is the most common indication for treating spine metastasis. Moreover, these limitations have become even more pronounced with the evolution in radiation treatments from the use of external beam radiation therapy (EBRT) to treatment with stereotactic spine radiosurgery (SSRS). This transition has been beneficial, as SSRS is associated with improved pain relief compared with EBRT and allows for a more minimally invasive operative approach with spine separation surgery rather than en bloc resection (1); however, it has also been associated with a greater incidence of vertebral fracture in 11%–39% of patients (23).

Considering these factors, some patients with severe pain are not able to tolerate radiation therapy simulation planning and can benefit from up-front cement augmentation. Otherwise, as this study indicates, VA and RFA can safely follow radiation therapy; no cases of delayed wound healing or any other negative outcomes were observed in association with this practice, even in procedures performed within 24 hours of radiation delivery. It is important to note that the procedure should not be performed between the radiation therapy simulation and the final radiation fraction, as the cement within the bone may impact the ongoing radiation treatment plan. With this in mind, RFA and VA should either occur before radiation therapy or after the completion of radiation therapy. In addition to playing an adjunct role, VA and RFA may also act as useful complementary therapies in patients with refractory pain following radiation therapy alone. This is an even more crucial consideration when approaching radioresistant lesions and patients who have already received a maximum radiation dose.

VA and RFA can likewise be safely combined with systemic therapies. At the authors’ institution, it is typical to hold targeted therapies such as tyrosine kinase inhibitors for no more than 24 hours before and after the spine intervention. Other systemic therapies, such as immunotherapy, are continued without cessation. For patients undergoing conventional chemotherapy, spine interventions are typically scheduled to avoid nadirs in blood counts. While these treatments have been associated with an increased risk of bleeding and delayed wound healing, these complications were not encountered during our study.

As all patients who underwent RFA also underwent VA, this study was not designed to determine the individual contributions of RFA versus VA. This is the standard practice at the authors’ institution, given the potential for increased fracture risk following RFA and the importance of addressing mechanical pain through cement augmentation. However, high quality data already exist regarding the efficacy of RFA for cancer-associated bone pain (5,6). Moreover, RFA contributed substantially to the pain relief seen in patients with osteoblastic metastases in whom the volume of cement delivered was limited due to the density of the bone.

One important finding from this current study is that VA was performed safely in patients with relative contraindications such as epidural disease, posterior wall disruption, and retropulsion. The decision to treat such patients should routinely involve a multidisciplinary conversation with surgery and radiation oncology teams and should be evaluated on a case-by-case basis. It is interesting to note that RFA was associated with a decreased likelihood of cement leakage into the epidural space or neuroforamina. While RFA is performed predominantly for its pain palliation effects, one can also speculate that the residual heat from the ablation zone helps cure the bone cement more quickly after injection, thus possibly accounting for the decrease in cement leakage. Similar findings have been reported in previous smaller studies (24). This finding has also been reported in the orthopedic literature with regard to cement augmentation adjacent to metallic hardware; heating of the implanted hardware significantly accelerated the cement setting time and diminished cement porosity (25). This is further supported by a recent meta-analysis by Chen and colleagues (22), which demonstrated a pooled incidence of total complications of 1% for 947 patients and 1163 metastatic lesions treated with RFA and VA. Hence, adding RFA may allow interventionalists to treat more complex vertebral compression fractures with less risk of cement leakage in the subsequent VA.

This study additionally identified a rate of adjacent-level spine fractures that parallels the frequency reported in the literature for patients with insufficiency fractures treated with cement augmentation. In a meta-analysis of 16 studies for 2549 patients who underwent cement augmentation for osteoporosis-related spine fractures, Zhang et al (26) identified an adjacent spine-level fracture incidence of 14%. Similar to the current study, thoracolumbar junction–level fractures were found to be a significant risk factor for adjacent-level fractures. Adjacent-level fracture risk has several important ramifications in the management of patients treated with cement augmentation. Counseling for this possibility prior to the procedure, as well as monitoring for it postprocedurally, are essential. For patients at high risk, such as those with fractures at the thoracolumbar junction, prophylactic treatment of spine levels, particularly "skip" levels between two cement-augmented levels, may be beneficial. These considerations, though, may not be commonly made for patients with cancer where the cause of the spine fracture is metastasis rather than osteoporosis. If the underlying bone mineralization is felt to be preserved, then the risk for adjacent-level fracture may be underestimated. Nonetheless, patients with cancer have several reasons for adjacent-level fractures, even with overall preserved bone health. Certainly, the biomechanical changes that occur with height loss and changes in kyphotic angle can still place additional stress on adjacent levels. In addition to these findings, our ML model for adjacent-level fracture based on these radiomics features demonstrated an AUC of 0.78 (70/30 test/train split, 12 936 spinal levels), with a small subset of radiomics features contributing substantially to the model. These results provide proof of concept for an entirely automated ML approach for the detection and risk estimation of adjacent-level spine fractures.

Furthermore, radiation fields that extend to levels above and below the augmented level can impact spine-level integrity. Additionally, systemic cancer therapies can include medications that can affect bone health, including steroids and hormone-based therapies. Hematologic malignancies that replace the normal marrow space additionally increase fracture risk. Thus, awareness of the risk of adjacent-level fracture is an important consideration for the patient with metastatic spine tumors, in addition to the patient with osteoporosis.

The authors acknowledge several limitations of this single-center, retrospective analysis. First, the follow-up pain assessment was only 4 weeks postoperatively, so long-term follow-up data regarding clinical response are limited. However, given the routine and frequent imaging that the cancer population undergoes, this study was able to provide long-term imaging follow-up analysis. In addition, accurately tracking pain scores in patients with cancer, particularly in a retrospective manner, can be challenging, as many patients have several causes for musculoskeletal pain, whether cancer related or age related. Given the multifocality of metastatic lesions in the study cohort, discretizing between pain from the target spinal level or levels from other pain generators can be inaccurate. Given the low event rate of 34 spinal levels with leakage, the study may be underpowered to evaluate the simultaneous effect of seven potential predictors of leakage in the multivariable analysis. Furthermore, this study did not assess changes in narcotic pain medication requirements, an important confounder in pain studies. Additionally, we did not specifically account for potential clustering of spinal levels within the same patient and instead treated each spinal level as an independent event, which aligns with our clinical practice; however, we acknowledge that other approaches (eg, generalized estimating equations) may be more appropriate and should be considered for future investigations.

In conclusion, our study demonstrates that VA and VA combined with RFA are safe and effective treatments to achieve early pain relief in patients with cancer with spine fractures. They can be safely performed together in the same procedural setting for various patients in combination with concurrent systemic therapies and recent radiation therapy. With appropriate technique, these therapies are also safe and effective for the treatment of osteoblastic lesions, as well as at vertebral levels with substantial posterior wall compromise that might otherwise pose relative contraindications to treatment. Future directions involve conducting prospective trials that combine stereotactic radiation therapy and ablation treatments to provide meaningful pain relief and limit adjacent fracture risk (ClinicalTrials.gov identifier: NCT04693377) (27).