Introduction

In recent years, the prevalence of lumbar degenerative disease (LDD) has increased, primarily due to the rapid growth of the aging population1. Unfortunately, conservative management strategies have shown limited success in alleviating symptoms or halting disease progression. Among available surgical options, spinal fusion, particularly posterior lumbar interbody fusion (PLIF) and related techniques, has emerged as the standard approach for definitive treatment2. However, both intraoperative and postoperative complications remain a challenge. One of the most frequent and clinically significant complications is cage subsidence (CS), which compromises intervertebral height and segmental stability. This can ultimately lead to fusion failure, spinal misalignment, nerve root compression, and, in some cases, the need for revision surgery, making it a critical concern for spine surgeons3. Multiple factors have been implicated in the development of CS. These include advanced age, female sex, reduced bone mineral density (BMD), endplate injury during surgery, diminished muscle mass, and interbody cage characteristics such as size, shape, and placement4,5,6,7,8.

Recent studies have highlighted a strong association between muscle mass, lumbar spine surgical outcomes, and vertebral bone mineral density9,10. Although recent research has focused on the roles of bone and muscle health in spinal pathology, reliable biomarkers that accurately reflect paraspinal muscle mass and predict the risk of CS are still lacking.

Low BMD is a well-established risk factor for CS11,12. To mitigate poor surgical outcomes, numerous studies have assessed preoperative vertebral BMD to guide early intervention strategies. Traditionally, BMD is measured using dual-energy X-ray absorptiometry (DEXA)13. However, in patients with LDD, DEXA often yields unreliable results due to structural changes that interfere with accurate assessment14. As an alternative, quantitative computed tomography (QCT) has demonstrated improved accuracy in evaluating the microarchitecture of vertebral bone in LDD patients15,16. Despite its advantages, QCT measurements can be significantly affected by surrounding calcified tissues such as the anterior and posterior longitudinal ligaments, ligamentum flavum, interspinous ligaments, and vascular structures like the aorta, which limit its utility in reliably predicting CS based solely on vertebral BMD. To address these limitations, newer models such as vertebral bone quality (VBQ) and endplate bone quality (EBQ) have been introduced to bypass the confounding effects of calcification. However, these models rely on indirect estimations involving water content and fat deposition, which can be influenced by the heterogeneous distribution of vertebral fat and inflammatory changes at the endplate, potentially affecting their accuracy17.

To address the limitations of current assessment methods, we propose a predictive model for implant subsidence following lumbar spine fusion that uses the paraspinal muscle index (PMI) and Goutallier classification (GC), both obtained through magnetic resonance imaging (MRI). This model is based on the premise that the cross-sectional area and fat infiltration of the paraspinal muscles serve as surrogate markers of vertebral bone quality, reflecting underlying muscle mass18,19. A key advantage of MRI-based parameters like PMI and GC is their insensitivity to vertebral and perivertebral artifacts that often compromise direct BMD measurements. Furthermore, these indicators eliminate radiation exposure and may reduce healthcare costs. While previous studies have investigated the interplay between muscle quality and bone mineral density, the link between muscle quality and cage subsidence remains underexplored. This study aims to evaluate the predictive value of paraspinal muscle quality for subsidence following PLIF and to compare the performance of this MRI-based model against existing predictive tools.

Materials and methods

Study population and ethical approval

This study received ethical approval from the Medical Ethics Committee of Yichang Central People’s Hospital (Approval No. 2023-164-01) and was conducted following the STROCSS reporting guidelines20. Given its retrospective design, the requirement for informed consent was waived by the Medical Ethics Committee of Yichang Central People’s Hospital. All methods were performed in accordance with relevant guidelines and regulations.

A total of 165 patients diagnosed with degenerative lumbar spine disease who underwent single-level PLIF at our institution between February 2022 and February 2024 were retrospectively reviewed. Of these, 45 patients (27.3%) developed CS within a 12-month postoperative follow-up period. All included patients met surgical indications and had not responded to previous conservative management.

The inclusion criteria were as follows:

  1. ①

    A confirmed diagnosis of degenerative lumbar conditions such as lumbar disc herniation, spinal stenosis, or spondylolisthesis by outpatient clinicians;

  2. ②

    Clinical symptoms and signs consistent with imaging findings;

  3. ③

    Failure of conservative treatment for at least three months or recurrence of symptoms requiring surgery as first-line treatment;

  4. ④

    No significant surgical contraindications;

  5. ⑤

    Completion of surgery without intraoperative complications;

  6. ⑥

    Follow-up duration of more than 12 months with complete postoperative data.

Exclusion criteria included:

  1. ①

    History of spinal surgery, particularly procedures that may compromise paraspinal muscle anatomy;

  2. ②

    Presence of lumbar infections, tuberculosis, or tumors;

  3. ③

    Incomplete or lost follow-up data;

  4. ④

    Evidence of postoperative fusion failure;

  5. ⑤

    Long-term corticosteroid use or diagnosis of immune-related disorders.

All patients underwent single-level PLIF using static polyetheretherketone (PEEK) cages. During surgery, at least 12 mL of autologous bone was grafted into the intervertebral space, followed by bilateral pedicle screw fixation. The PEEK cages manufactured by [IRENE] were specifically designed to maintain intervertebral height and provide robust support during single-level fusion procedures.

Data collection

Data collected for this study included patient demographics, perioperative imaging findings, and postoperative surgical outcomes. Demographic variables comprised age, sex, body mass index (BMI), the presence of diabetes, and the specific spinal segments operated on. Imaging data encompassed preoperative assessments of the paraspinal muscle index (PMI), Goutallier classification (GC), quantitative computed tomography (QCT) values, VBQ, and EBQ scores, along with evaluation of CS. These parameters were derived from preoperative lumbar CT and T1-weighted MRI scans, and postoperative follow-up CT imaging.

PMI and GC measurements

Preoperative MRI examinations were performed using a Philips 3.0T magnetic resonance imaging system, employing standard T1-weighted sequences. Axial images were acquired at the level of the upper endplate of the L4 vertebra, a widely recognized reference plane for evaluating the cross-sectional area (CSA) of the paraspinal muscles. The raw MRI data were imported into ImageJ software (Version 1.53t; National Institutes of Health, Bethesda, MD, USA; https://imagej.nih.gov/ij/) for scaling and calibration to ensure measurement precision. On the selected L4 reference plane, the bilateral paraspinal muscles were manually delineated to calculate CSA, carefully excluding fat, fascia, and other non-muscular tissues to ensure accurate quantification of true muscle mass. Two orthopedic physicians with more than three years of clinical experience independently performed the measurements to ensure accuracy and inter-rater reliability. The mean CSA of the bilateral paraspinal muscles was then calculated for further analysis (Fig. 1).

Furthermore, ImageJ software “image segmentation” tools were employed to distinguish between lean muscle and intramuscular fat, automating the fat-to-muscle ratio computation. This ratio was subsequently used for quantitative assessment within the Goutallier classification system, providing a more objective and standardized evaluation of muscle quality.

Given that L4/L5 and L5/S1 represent the most frequently fused segments in clinical practice and differ in terms of biomechanical loading, degenerative changes, and anatomical structure, these segmental variations may influence local paraspinal muscle condition and the risk of cage subsidence. Therefore, this study also conducted a subgroup analysis of PMI and GC at L4/L5 and L5/S1 to investigate their potential impact on segment-specific muscle quality and their association with postoperative cage subsidence.

PMI Calculation Formula:

$$\:\text{P}\text{M}\text{I}(\text{c}{\text{m}}^{2}/{\text{m}}^{2})=\frac{\text{C}\text{S}\text{A}\left(\text{c}{\text{m}}^{2}\right)}{\text{H}\text{e}\text{i}\text{g}\text{h}\text{t}\left({\text{m}}^{2}\right)}|$$

Based on the degree of fat infiltration, the Goutallier classification is divided into five grades:

Grade 0: No fat infiltration, entirely normal muscle tissue;

Grade 1: Mild fat infiltration within the muscle;

Grade 2: Fat tissue occupies less than 50% of the muscle volume;

Grade 3: Fat tissue occupies 50–75% of the muscle volume;

Grade 4: Fat tissue occupies over 75% of the muscle volume.

Fig.1
figure 1

MRI image showing the cross-section of the paraspinal muscle at the level of the upper endplate of L4; (a) The outlined area represents one side of the paraspinal muscle; (b) ImageJ software automatically identifies the lean tissue within the outlined muscle area.

Full size image

QCT measurements

Preoperative lumbar CT scans performed one month before surgery were used for quantitative analysis. Volumetric bone mineral density (vBMD) at the L1 or L2 vertebral level was assessed using Mindways QCT Pro software (Version 6.1; Mindways Software Inc., Austin, TX, USA; https://www.qct.com), employing the asynchronous calibration QCT technique to convert Hounsfield units (HU) into vBMD values21. On sagittal CT reconstructions of the lumbar spine, regions of interest (ROIs) were placed centrally within the vertebral body. The ROI placement excluded cortical bone and the posterior venous plexus to ensure measurement accuracy. Furthermore, any osseous abnormalities such as bone islands or areas of sclerosis within the vertebral body were carefully avoided. The affected vertebral segment was excluded from the analysis if such anomalies were too extensive to allow for a reliable ROI definition.

VBQ and EBQ measurements

Following the methodology outlined by Ehresman et al.22, T1-weighted sagittal MRI images were used to define ROIs within the lumbar vertebral bodies (L1 to L4) and the cerebrospinal fluid (CSF) at the L3 level for VBQ score calculation. Elliptical ROIs were placed within 3 mm of the vertebral body margin to exclude cortical bone and avoid focal anomalies such as the posterior venous plexus. In cases where accurate ROI placement was hindered, additional sagittal slices were reviewed, or alternative CSF ROIs at the L2 or L4 levels were used.

For EBQ score assessment, we adopted the approach proposed by Jones23, using axial T1-weighted MRI images to place ROIs 3 mm beneath the subchondral endplates of the surgical segments, along with a CSF reference ROI at the L3 level. When Schmorl’s nodes or obstructions were present at the CSF reference level, ROI placement was either carefully adjusted or excluded from analysis (Fig. 2).

Fig.2
figure 2

An example of signal intensity measurements for the vertebral body quality (VBQ) and endplate bone quality (EBQ) scores.

Full size image

Cage subsidence measurements

The CT is widely regarded as more accurate and sensitive than conventional X-rays for detecting implant subsidence and assessing changes in bone quality following lumbar spine fusion24 (Fig. 3). Following our institution’s standardized follow-up protocol, all patients in this study underwent routine postoperative CT imaging to monitor changes in middle vertebral body height (MIVH). Immediately after surgery, CT scans were used to identify the target vertebral level. Once the midpoints of the superior and inferior endplates were visualized, the vertical distance between them was measured and recorded as the baseline MIVH. At the 12-month follow-up, the same or a closely matched axial plane and measurement technique were employed to reassess MIVH, ensuring consistency in anatomical reference and measurement orientation.

This study defined CS as a more than 2 mm decrease in MIVH on CT scans obtained at least 12 months after surgery, relative to immediate postoperative measurements16. Based on quantitative imaging, this definition is widely used in recent literature and facilitates standardized assessment due to the consistent visualization of anatomical landmarks on CT25. Because all implanted cages were static and non-expandable, the risk of intervertebral height loss due to implant recoil was effectively minimized.

Patients were divided into CS and non-CS groups based on the presence or absence of cage subsidence. Two experienced investigators, blinded to the patients’ baseline demographic information and bone density data, independently conducted all imaging measurements. This blinding approach was implemented to minimize measurement bias and improve the objectivity and reproducibility of the findings.

Fig.3
figure 3

An example of measuring cage subsidence.

Full size image

Statistical analysis

Statistical analyses were conducted using SPSS version 26.0 (IBM Corp., USA). Continuous variables with a normal distribution (e.g., age, subsidence height, BMI, PMI, QCT, EBQ, VBQ) were expressed as mean ± standard deviation, and group comparisons were performed using independent-sample t-tests. For non-normally distributed data, the rank-sum test was applied. Categorical variables were analyzed using the chi-square test. Receiver operating characteristic (ROC) curve analysis was used to evaluate the predictive performance of each model, with calculation of the area under the curve (AUC) and the optimal cutoff value. Pearson’s correlation coefficient was used to assess the relationship between continuous variables. A two-tailed p-value < 0.05 was considered statistically significant.

Results

Table 1 presents the demographic and clinical characteristics of the study cohort. Of the 165 patients included, 45 (27.3%) developed CS, while 120 (72.7%) did not. The mean age was significantly higher in the CS group than in the non-CS group (70.4 ± 6.99 vs. 64.02 ± 8.24 years, p < 0.001). Moreover, the CS group had a significantly greater proportion of female patients (68.9%, 31/45) than the non-CS group (49.2%, 59/120, p = 0.023), indicating a potential sex-related predisposition to CS.

In terms of BMI, individuals with a BMI ≤ 25 kg/m² were less prevalent in the CS group (35.7%, 10/28) than in the non-CS group (56.1%, 64/114), with the difference reaching statistical significance (p = 0.002). No significant difference was observed in the prevalence of diabetes between the two groups (p = 0.723), suggesting a limited role in CS risk. Surgical indications, including lumbar disc herniation, spinal stenosis, and spondylolisthesis, were similarly distributed across groups, indicating no significant confounding by underlying pathology.

Regarding surgical segment involvement, L4/L5 was the most commonly fused level (66.7%, 110/165). While the incidence of CS was higher in L4/L5 fusion cases (38/110) compared to L5/S1 (6/42), this intersegmental difference was not statistically significant (p = 0.104). It is important to note that CS severity was not treated as a continuous variable in this analysis but as a binary grouping criterion.

Table 2 outlines the quantitative imaging and biomechanical parameters. The distribution of GC grades differed significantly between groups (p < 0.001). In the CS group, the majority were classified as GC grade 2 (23 cases) and grade 3 (18 cases), with fewer cases in grade 1 (3) and grade 4 (1). The non-CS group predominantly consisted of GC grade 1 (57 cases) and grade 2 (52 cases), with only a small number of patients classified as grade 3 (9) or grade 0 (2); no cases were recorded as grade 4.

Muscle quality, as assessed by the paraspinal muscle index (PMI), was significantly lower in the CS group compared to the non-CS group (6.00 ± 1.10 vs. 7.65 ± 1.34 cm²/m², p < 0.001). Bone mineral density measured by quantitative computed tomography (QCT) was also significantly reduced in the CS group (101.4 ± 13.4 vs. 145.5 ± 34.5 mg/cm³, p < 0.001). Furthermore, both the VBQ and EBQ scores were significantly higher in the CS group than in the non-CS group (VBQ: 3.29 ± 0.89 vs. 2.40 ± 0.58; EBQ: 3.49 ± 0.97 vs. 2.79 ± 0.80; both p < 0.001), indicating that poorer bone quality may be a key contributor to the development of cage subsidence.

Segmental analysis revealed significant variation in paraspinal muscle index (PMI) values across different spinal levels (Fig. 4). Correlation analysis further demonstrated a significant negative association between cage subsidence depth and PMI (r = − 0.402, p < 0.05), as well as a positive correlation between PMI and QCT-derived bone mineral density (r = 0.39, p < 0.05) (Fig. 5).

Multivariate logistic regression analysis identified reduced PMI as a significant independent risk factor for cage subsidence (OR: 0.400, 95% CI: 0.198–0.809, p = 0.011). Furthermore, higher GC scores were significantly associated with an increased risk of subsidence (OR: 4.857, 95% CI: 1.371–17.204, p = 0.014). Detailed regression coefficients for all variables are presented in Table 3.

Receiver operating characteristic (ROC) curve analysis was performed to evaluate the predictive performance of several preoperative parameters, including PMI, GC, VBQ, EBQ, and QCT scores. PMI demonstrated strong predictive value, with an AUC of 0.826 (95% CI: 0.756–0.896). GC grading yielded an AUC of 0.786 (95% CI: 0.719–0.854), with an optimal cutoff of 1.5, sensitivity of 93.3%, and specificity of 49.2%. QCT showed the highest predictive accuracy (AUC: 0.894; 95% CI: 0.844–0.943), with an optimal cutoff of 118.5 mg/cm³, sensitivity of 95.6%, and specificity of 83.3%. The EBQ score yielded an AUC of 0.719 (95% CI: 0.630–0.809), with a cutoff of 3.04, sensitivity of 64.4%, and specificity of 72.5%. VBQ score also showed good predictive power (AUC: 0.814; 95% CI: 0.739–0.889), with a cutoff of 2.375, sensitivity of 86.7%, and specificity of 65%. All tested parameters demonstrated favorable predictive performance, as illustrated in Fig. 6.

Table 1 Patient characteristics.
Full size table
Table 2 The size of various model parameters.
Full size table
Table 3 Multivariate regression analyses.
Full size table
Fig.4
figure 4

Subgroup analysis of PMI and GC in different surgical vertebral segments. (a) PMI scores differed significantly between the L4/L5 and L5/S1 segments. (b) No significant difference in Goutallier classification (GC) scores was observed between the L4/L5 and L5/S1 segments.

Full size image
Fig.5
figure 5

Linear relationship. (a) There was a positive correlation between PMI and QCT. (b) There was a negative correlation between PMI and CS.

Full size image
Fig.6
figure 6

ROC curve of QCT, GC, VBQ, EBQ, and PMI.

Full size image

Discussion

In this cohort of 165 patients undergoing single-level PLIF, CS occurred in 45 cases. The results indicate that advanced age and female sex are associated with an increased risk of CS, likely attributable to age-related bone degeneration and postmenopausal osteoporosis. Moreover, the PMI and GC emerged as independent predictors of postoperative CS. Higher GC grades and lower PMI values were significantly associated with a greater likelihood of subsidence. The predictive performance of GC and PMI was supported by AUC values of 0.786 and 0.826, respectively. These findings suggest that preoperative evaluation of GC and PMI may offer a practical and reliable means of identifying patients at elevated risk for CS after PLIF.

Previous research has identified several risk factors associated with CS, including sex, age, diminished bone mineral density, muscle degeneration, and intervertebral disc height26,27. In our multivariate analysis, female sex and older age were independently associated with a higher risk of CS. These demographic variables are often linked to decreased bone mass and weakening vertebral endplate integrity17. In particular, the sharp decline in estrogen levels following menopause contributes to accelerated bone loss, increasing the susceptibility of postmenopausal women to CS28,29. Although BMI was not an independent predictor in our analysis, patients with higher BMI demonstrated an elevated risk of CS. This may be attributed to increased mechanical loading and metabolic disturbances associated with obesity, which can intensify stress at the fusion segment and promote bone loss, especially in elderly female patients30.

Our analysis further identified the PMI and GC as independent predictors of CS. A lower PMI and a higher GC grade were significantly associated with a greater risk of CS. The areas under the ROC curves for PMI and GC were 0.826 and 0.786, respectively. Based on standard AUC interpretation criteria, values exceeding 0.8 indicate good discriminatory power31, supporting the utility of PMI and GC as reliable preoperative indicators of paraspinal muscle quality. Most lumbar fusion procedures in our cohort were performed at the L4/5 and L5/S1 levels. While significant differences in PMI were observed between levels, GC showed no such variation. This discrepancy may be attributable to anatomical factors: the maximum cross-sectional area of the paraspinal muscles is typically found at the L3/4 to L4/5 levels in the neutral spine position9, potentially influencing PMI measurements across different segments. In comparison, GC represents the degree of fatty infiltration within the muscle, which is less dependent on anatomical level and may exhibit more consistent values throughout the lumbar spine.

Regarding bone mineral density assessment, QCT yielded the highest AUC among all predictive markers in this study, underscoring its clinical utility in forecasting CS. QCT offers improved accuracy compared to conventional DEXA, particularly in patients with lumbar degenerative conditions. According to the guidelines established by the American College of Radiology32, low bone mass was defined as QCT < 120 mg/cm³. In our analysis, a QCT value below 118.5 mg/cm³ was significantly correlated with an elevated risk of postoperative CS. Despite cost and radiation exposure limitations, QCT’s three-dimensional imaging capability and high sensitivity for detecting subtle endplate changes make it a valuable tool for preoperative evaluation and risk stratification33.

The MRI has become a valuable modality for preoperative assessment in lumbar spine surgery and is increasingly employed to indirectly evaluate bone quality using imaging-based metrics such as VBQ and EBQ scores. In our study, VBQ and EBQ scores were significantly associated with CS. However, multivariate regression analysis indicated that EBQ was not an independent predictor of CS compared to the findings of a previous study23. We speculate that this inconsistency may be related to the widespread tendency for delayed healthcare-seeking behavior among patients in China, especially for chronic spinal conditions such as cervical spondylosis and lumbar disc herniation. Early symptoms are often mistaken for fatigue or minor strain, leading patients to pursue self-treatment rather than seeking timely medical care34,35. Prolonged delays may worsen endplate inflammation and bone marrow edema. Previous studies have shown that these inflammatory changes typically manifest on MRI as decreased T1 signal intensity and increased T2 signal intensity, hallmarks of Modic Type 1 changes. Such alterations could result in underestimation of EBQ scores, therefore reducing their accuracy in reflecting true bone quality and potentially explaining the weaker predictive performance of EBQ in this study36,37.

Beyond bone quality, muscle quality also plays a vital role in the pathogenesis of CS. In this study, the PMI was significantly associated with CS, indicating that diminished preoperative muscle mass may compromise postoperative segmental stability. The concept of a functional “bone–muscle unit” has been proposed in previous literature, emphasizing the dynamic interplay between muscle and bone through mechanical loading and biochemical signaling pathways38. A reduction in muscle mass can disrupt skeletal load distribution and contribute to bone loss, thus weakening the structural support of the endplate and elevating the risk of cage subsidence39. Similarly, muscle tissue influences bone metabolism via “bone–muscle crosstalk,” underscoring its essential role in maintaining skeletal health. While this study evaluated bone mineral density and muscle quality as separate predictors, it did not explore their combined effects or potential interactions. As a result, the synergistic impact of bone–muscle dynamics on CS risk may have been underestimated. Future studies should investigate the integrative mechanisms linking muscle and bone quality and consider incorporating muscle-related metrics as independent variables or covariates in multivariate models to improve the predictive accuracy and clinical applicability of CS risk assessment tools.

This study systematically evaluated various clinical and imaging factors associated with cCS, highlighting the pivotal roles of muscle quality and bone mineral density in its pathogenesis. However, several limitations should be acknowledged. First, as a single-center retrospective study with a relatively small sample size, the results may be influenced by selection bias, therefore limiting the generalizability of the findings. Second, although the L4 level is commonly used as a reference point for assessing the PMI and GC, it remains uncertain whether this anatomical level best represents overall lumbar muscle status. To address this uncertainty, future prospective studies are warranted to validate the representativeness and appropriateness of this measurement level. Establishing standardized preoperative protocols for muscle evaluation, including a unified anatomical reference level and consistent assessment criteria, will be essential. Long-term follow-up studies comparing the predictive accuracy of CS across different measurement levels are also needed to build a more robust, evidence-based evaluation framework. Lastly, some imaging-derived parameters, such as the EBQ score, are susceptible to interobserver variability and subjectivity. Future research should prioritize using multicenter datasets and incorporate automated or AI-assisted imaging analysis tools to enhance the objectivity, reproducibility, and clinical utility of such assessments.

In conclusion, both the PMI and GC were significantly correlated with the occurrence of CS and exhibited strong predictive value. A reduced PMI and elevated GC were associated with a higher risk of CS. These findings suggest that MRI-based evaluation of paraspinal muscle quality offers a reliable, non-invasive approach for identifying patients at increased risk of early cage subsidence following PLIF.