Risk factors for cage subsidence following anterior–posterior spinal fixation in osteoporotic vertebral fractures: a multicenter retrospective study

Ethical statements

All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Declaration of Helsinki and its later amendments or comparable ethical standards. The Institutional Review Board of our university approved the research protocol (approval no., 2022-062). All study participants provided informed consent.

Study design and patient population

This multicenter retrospective cohort study included patients who underwent combined APSF using a wide-footplate expandable cage system at our affiliated hospitals between January 2015 and December 2023, with a minimum 1-year follow-up.

To be eligible for participation, patients were required to have OVFs, intravertebral or intervertebral instability, and either a neurological deficit or severe back pain. The analysis focused on patients who were followed up for at least 1 year. Patients with available data on global spinal alignment before surgery and at the final follow-up were analyzed.

Surgical indications

All included patients underwent APSF. Surgical indications for APSF were OVFs presenting with intractable back pain or transient neurological symptoms, such as radicular pain caused by residual spinal instability, provided that these symptoms improved or disappeared in the supine position. According to the osteoporotic fracture (OF) classification, only patients classified as OF3–4 were included [7]. Patients treated with vertebroplasty or balloon kyphoplasty, either alone or in combination with posterior instrumentation, were excluded to maintain a homogeneous cohort, as these procedures are generally performed in the acute phase for those without significant instability. High-energy trauma cases, such as classical burst fractures, were excluded, even when the posterior wall is involved, which in this study was categorized as OVF rather than burst fracture. Patients for whom fenestrated screws were used were also excluded.

Surgical procedure for X-Core2

The patient was positioned laterally, and a true lateral radiograph was obtained using fluoroscopy. The affected vertebral body and adjacent upper and lower disks were assessed using a transthoracic retropleural or retroperitoneal approach. After the disks above and below the affected vertebral body were removed and the segmental vessels were ligated or coagulated, corpectomy was performed using a large osteotome. The cartilaginous endplate was carefully removed with a disk knife and ring curette to prevent inadvertent endplate violation. The vertebral segment was reconstructed using an expandable titanium cage with rectangular footplates (X-Core2; NuVasive, San Diego, CA, USA). Bone grafting was performed inside and outside the cage using artificial tricalcium phosphate particles, resected vertebral bodies, and rib fragments. After the patient was repositioned, posterior percutaneous pedicle screw (PPS) fixation was performed without decompression. The range of posterior fixation was determined according to the surgeon’s preference. All surgeries were performed by experienced senior spine surgeons.

Demographic data

Patient demographic data, including age, sex, comorbidities, body mass index, dual-energy X-ray absorptiometry (DEXA), and medications for osteoporosis, were collected from medical records. Postoperative complications and surgical factors, including operative duration, blood loss, and fixation range, were also recorded. Fixation range was categorized into two groups: short fusion, defined as the fixation of one vertebra above and one below the fractured vertebra (1A1B), and long fusion, defined as the fixation of two or more vertebrae above and two or more vertebrae below the fractured vertebra (≥2A2B). These were treated as categorical variables in the analysis (0: short fusion; 1: long fusion). Patients were grouped according to fracture level (thoracolumbar [TL]: T11–L2 or lumbar [L]: L3–L5).

To evaluate clinical outcomes, patients subjectively rated the intensity of their pain using a Visual Analog Scale (VAS) to characterize the average level of back pain they experienced over the past week. Treatment effectiveness for lower back pain was assessed using the Japanese Orthopedic Association (JOA) score, which has a maximum score of 29. The improvement rate for each clinical outcome was calculated using the following formula: JOA score improvement rate (%)=100×(final JOA score−preoperative JOA score)÷(29−preoperative JOA score).

Radiographic data

Radiographic evaluation was performed for all patients using whole-spine radiography before surgery and at the final follow-up. It included analysis of sagittal alignment (sagittal vertical axis [SVA], pelvic incidence, lumbar lordosis, sacral slope, pelvic tilt, thoracic kyphosis, and T1 pelvic angle [TPA]). Bone union was evaluated using computed tomography (CT) at 1 year postoperatively.

Definition of local kyphosis

Preoperative local kyphosis of the fractured vertebra was assessed using standing (or sitting, if standing was challenging) lateral radiographs, defined as the angle between the inferior endplate of the cephalad vertebra and the superior endplate of the caudal vertebra (Fig. 1A). Preoperative supine lateral radiographs were also obtained to evaluate supine local kyphosis, and the difference between the supine and standing values was defined as Δlocal kyphosis (supine−standing). Local kyphosis was assessed at four time points: preoperative standing, preoperative supine, 1 week postoperative, and the final follow-up. All postoperative and final follow-up measurements were consistently performed on standing lateral radiographs. Two independent spine surgeons, blinded to clinical data, conducted the measurements, with discrepancies resolved by consensus.

Fig. 1

(A) Method of measuring local kyphosis (LK). (B) Evaluation of endplate injury grade. (C) Distribution of vertebrae treated with Xcore. Preop, preoperative; Postop, postoperative.

Definition of endplate injury

Portable radiography was performed immediately postoperatively to assess endplate injury. In the lateral view, a clear zone between the cage footprint and the endplate was scored as 0. If the cage footprint was in contact with the endplate, it was scored as 1. If the cage footprint subsided into the endplate, it was scored as 2. Higher scores indicated greater endplate damage (Fig. 1B).

Measurement of Hounsfield unit values

All patients were assessed using a multi-slice CT scanner as part of the routine preoperative planning for surgery. A region of interest was selected in the horizontal plane at the middle level of the vertebral body to avoid the cortical bone and the area of the posterior venous plexus. HU values were measured from the cephalad and caudal vertebrae adjacent to the fractured vertebra, and their averages were calculated [8].

Assessment of cage subsidence

Cage subsidence was defined as a descent of the cage by ≥2 mm, as observed on CT, when compared to its initial postoperative position, as delineated in prior studies [9]. All evaluations were performed using CT scans at each follow-up. Two experienced spine surgeons independently conducted measurements, and discrepancies were resolved by consensus.

Statistical analysis

Demographic data

This study included 100 patients (76.7±6.9 years), among whom subsidence was observed in 53 patients. The demographic data showed a significant difference in HU (102.6±28.3 vs. 80.0±30.6, p=0.005) and endplate injury (p<0.001). Furthermore, 1A1B fixation was significantly more common in the subsidence group (p<0.001). No significant differences were observed in other variables (Table 1). No significant differences were observed between the two groups regarding postoperative complications (Table 2). Regarding clinical outcomes, postoperative improvement rates in VAS and JOA scores were significantly lower in the subsidence group (Table 2). The distribution of vertebrae treated with X-Core2 is illustrated in Fig. 1C.

Table 1

Demographic data and operative data

Table 2

Postoperative complications and clinical outcomes

Radiographic data

No significant differences in local kyphosis were observed between the preoperative, preoperative supine, and postoperative measurements (Table 3). However, significant differences were found in Δlocal kyphosis (supine−standing) (−7.1°±9.2° vs. −14.6°±11.5°, p=0.001) and in local kyphosis at the final follow-up (0.1°±13.7° vs. 11.4°±21.4°, p=0.005) (Table 3). A significant loss of correction was observed in the subsidence group from the postoperative period to the final follow-up (Δlocal kyphosis [final−post] 2.0°±5.5° vs. 10.0°±15.1°, p=0.002). In terms of global alignment, no significant differences were observed postoperatively, except for C7−SVA (76.4±35.7 mm vs. 99.5±52.2 mm, p=0.038) and TPA (26.2°±7.5° vs. 31.6°±9.3°, p=0.012) (Table 3). Correction loss was significantly greater in the TL group than in the L group (Table 4). A significant difference in the bone union rate was observed between the subsidence and non-subsidence groups (71.6% vs. 91.4%, p=0.019) (Table 2). However, within the subsidence group, there was no significant difference between the TL and L subgroups (72% vs. 71.4%, p=0.963) (Table 4).

Table 3

Radiographic data and global alignment

Table 4

Radiographic data in non-subsidence, TL subsidence, and L subsidence

ROC

ROC analysis was conducted for Δlocal kyphosis (supine-standing) and HU, which showed significant differences in the univariate analysis. The optimal cutoff value for Δlocal kyphosis (supine−standing) was identified at −14.0 (AUC=0.706, sensitivity=77.4%, specificity=59.6%) and 87.5 for HU (AUC=0.700, sensitivity=75.0%, specificity=64.7%) (Fig. 2).

Fig. 2

(A) Receiver operating characteristic (ROC) analysis of Δlocal kyphosis (supine-standing). (B) ROC analysis of Hounsfield units (HU) values. AUC, area under the ROC curve.

Multivariate analysis

Multivariate analysis was performed to identify the predictors of subsidence. The dependent variable was “subsidence,” and the independent variables included were “Δlocal kyphosis (supine−standing),” “TL,” “HU,” “fixation range,” and “endplate injury (1/2).” For the analysis, the cutoff values of Δlocal kyphosis (supine−standing) and HU were determined according to the ROC analysis. Moreover, as previous reports indicated a higher risk of implant failure in the thoracolumbar junction, vertebrae treated with X-Core2 were dichotomized into TL (Th11–L2) and other regions.

ΔLocal kyphosis (supine−standing) (aOR, 12.8; p=0.010), HU (aOR, 8.1; p=0.033), fixation range (aOR, 8.2; p=0.020), and endplate injury (1: aOR, 4.5; p=0.042; 2: aOR, 18.8; p=0.011) were identified as significant risk factors for subsidence (Table 5).

Table 5

Multivariate analysis for the risk factors of subsidence

Subgroup analysis

A previous analysis revealed that endplate injury is the most significant risk factor for subsidence. Therefore, spine surgeons should avoid intraoperative endplate injury and conduct a more detailed investigation of patient characteristics that cause subsidence. Specifically, a subanalysis was conducted by excluding cases in which endplate injury scored 2 points. Significant differences were observed in HU (p=0.037) and endplate injuries (p<0.001) between the groups (Supplement 1). Radiographic data also showed significant differences in Δlocal kyphosis (supine−standing) and Δlocal kyphosis (final−post) (Supplement 2). Global alignment parameters are summarized in Supplement 3. In the multivariate analysis, Δlocal kyphosis (supine−standing) (aOR, 10.5; p=0.015), fixation range (aOR, 9.770; p=0.015), and HU (aOR, 7.580; p=0.035) were significant risk factors for subsidence (Supplement 4).

Decision tree analysis

We conducted a decision tree analysis to predict the risk of subsidence in this cohort. The target variable was the presence of subsidence, and the explanatory variables were Δlocal kyphosis (supine−standing) and HU. In this study, Δlocal kyphosis (supine−standing) was defined as 1 if <−14.0 and 0 if ≥−14.0, whereas HU was defined as 1 if <87.5 and 0 if ≥87.5. Decision tree analysis identified Δlocal kyphosis (supine−standing) and HU as significant predictors of subsidence risk. Notably, the highest risk was observed in patients with Δlocal kyphosis (supine–standing) <−14.0 and HU <87.5 (Fig. 3).

Fig. 3

Decision tree analysis in overall. HU, Hounsfield units.

Impact of cage subsidence

Although there was no significant difference in reoperation rates between the subsidence and non-subsidence groups, clinical outcomes in terms of VAS and JOA scores were significantly worse in the subsidence group. Global alignment parameters, particularly C7−SVA, were also significantly worse. High C7−SVA has been previously linked to low back pain [18–21], and in this study, cage subsidence appeared to increase local kyphosis, worsening sagittal alignment and contributing to poor clinical outcomes. Notably, cage subsidence was strongly associated with correction loss. Subgroup analysis revealed that this association was particularly significant in the TL (T10–L2) group but not in the L (L3–5) group, likely reflecting the biomechanical characteristics of the thoracolumbar junction, where the spinal alignment shifts from lordosis to kyphosis.

In this study, no significant difference in the reoperation rate was observed between patients with and without cage subsidence. This outcome may be primarily attributed to the advanced age of the patient population, which makes revision surgery difficult. In other words, even when clinical outcomes are suboptimal, elderly patients are often unable to undergo reoperation. Therefore, it is crucial to assess risk factors preoperatively and ensure that the initial surgery provides definitive treatment.

Treatment strategies to prevent cage subsidence

To explore strategies to prevent cage subsidence, we compared patients who underwent short fusion (1A1B) with those who underwent longer fusion (2A2B or more). As our study defined long fusion as ≥2A2B, which is shorter than the extensive constructs generally described in the literature, its impact on global alignment parameters, such as C7–SVA, was limited (Supplement 5). Subsidence was more common in the 1A1B group, indicating that longer fusion may help reduce this complication (Supplements 6, 7).

Based on this study, our recommended treatment strategy is as follows. First, operators should focus on avoiding intraoperative endplate injury and ensuring stable cage placement on the apophyseal ring, which comprises the strongest and densest vertebral endplate bones. For patients with significant vertebral instability and low HU values, bone quality can be improved using agents that promote bone formation. The importance of the preoperative use of such agents, as well as efforts to increase DEXA and HU values, has been previously reported [22,23]. In clinical situations where determining the optimal extent of spinal fixation is challenging, such as cases with fractures at the lower instrumented vertebra or upper instrumented vertebra, spinal canal stenosis, or vertebral instability, longer fusion (2A2B or greater) should be considered, especially for patients with the identified risk factors. However, in the absence of these risk factors, short-segment fixation may be sufficient, thereby enabling more efficient use of medical resources.

Experienced surgeons performed all surgeries in this study. Nevertheless, endplate injuries occurred at a certain frequency, suggesting that such injuries will still occur even when the procedure is carefully performed. Previous studies on LLIF have reported an association between intraoperative endplate injury and bone mineral density [24]. In a cohort such as ours, which consisted primarily of elderly patients, endplate injuries may be inevitable. There may be a trade-off between achieving a high rate of bone fusion and avoiding endplate injury. In elderly patients with poor bone quality, aggressive disk removal can cause intraoperative endplate injury, increasing the risk of cage subsidence. These findings suggest that overly aggressive disk removal is unnecessary in such cases.

Limitations

This study had several limitations. First, its retrospective design may have introduced selection bias. Second, the follow-up period may have been insufficient to fully assess the efficacy of this surgical strategy. Third, only midterm outcomes were evaluated; further studies are needed to clarify long-term results. Cage subsidence is multifactorial and may be influenced by global alignment, bone mineral density, and the affected vertebral level. Additionally, fixation range was determined by the operating surgeon, which may have affected outcomes. Furthermore, because all included patients had OVFs, disease-specific differences could not be evaluated. However, it is conceivable that subsidence risk varies according to the underlying pathology, including degenerative disease or spinal deformity. Finally, the study lacked sufficient power to evaluate the impact of osteoporosis medications. Many patients initiated treatment at other institutions, and the exact duration of preoperative medication could not be obtained. To address this, vertebral bone quality was assessed using HU values as a surrogate for bone mineral density. Although detailed data on diabetes and nutritional status were unavailable, we reported the prevalence of rheumatoid arthritis and chronic steroid use, both of which are strongly related to osteoporosis, and found no significant differences between the subsidence and non-subsidence groups.