Development and validation of a nomogram prediction model for assessing cage subsidence risk following anterior cervical discectomy and fusion

Abstract

Objective

At present, cage subsidence is the most prevalent hardware-related complication of anterior cervical discectomy and fusion (ACDF), which may affect postoperative spinal biomechanics and alignments. Presently, there is no established nomogram model for predicting subsidence based on preoperative parameters. Hence, our aim was to develop nomograms for predicting cage subsidence after ACDF.

Methods

The clinical data of 476 patients who underwent ACDF at our hospital from January 1, 2014, to June 30, 2023, were retrospectively analyzed. A nomogram prediction model was constructed using 238 patients in the training cohort, which was divided into the subsidence group and the non-subsidence group. Various parameters were recorded, such as age, gender, body mass index, smoking and drinking history, history of hypertension and diabetes, operation time, intraoperative blood loss, bone mineral density, visual analogue scale (VAS) score for arm and neck, Japanese Orthopaedic Association (JOA) score, Neck Disability Index (NDI), Hounsfield Units (HU) of upper and lower vertebral bodies, Modic changes, cervical lordosis, intervertebral segmental height (ISH), functional spinal unit angle (FSUA), range of motion (ROM) for C2-C7, as well as the sagittal vertical axis (SVA) and T1 slope (T1S). Multivariate logistic regression was used to identify independent predictors for developing nomograms. Finally, the predictive values of this model were validated in an external verification cohort consisting of another 238 patients.

Results

In the training cohort, 23.95% (57 of 238 patients) had cage subsidence during an average follow-up of 23.31 ± 6.14 months. Multivariate logistic regression analysis results showed that bone mineral density (OR = 0.499, 95% CI 0.356–0.677), upper HU value (OR = 0.979, 95% CI 0.971–0.986), lower HU value (OR = 0.989, 95% CI 0.982–0.996), ISH (OR = 0.504, 95% CI 0.284–0.856), and T1 slope (OR = 0.877, 95% CI 0.821–0.933) were determined as independent risk factors for postoperative cage subsidence. The predictive model exhibited excellent performance in both the training cohort (AUC = 0.885, 95% CI 0.835–0.935) and the validation cohort (AUC = 0.914, 95% CI 0.873–0.955).

Conclusion

Our predictive model is set to become a valuable and convenient tool for clinical decision-making, enabling early prediction and timely intervention for cage subsidence after ACDF. Future prospective studies could incorporate our model in different clinical scenarios, thus reducing the burden on patients and healthcare resources.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12891-025-09139-0.

Introduction

Since ACDF was initially reported by Smith and Cloward back in the 1950s, it has progressively turned into one of the leading surgical approaches for dealing with one- and two-level radiculopathy and myelopathy [1, 2]. The increased use of ACDF has triggered numerous studies on short- and long-term clinical and radiographic outcomes, which are typically reported to be superb [3–5]. However, cage subsidence after surgery may have an impact on spinal biomechanics and alignment, leading to segmental kyphosis and contributing to adjacent segment disease [6, 7]. The decreased height of the intervertebral space might result in secondary foraminal stenosis, making it more likely for nerve root impingement and radicular symptoms to recur [6–9]. According to prior literature, the occurrence rate of cage subsidence following ACDF is said to be as high as 19.3% to 42.5% [6, 7, 10]. Although previous works have put forward some risk factors related to cage subsidence after ACDF, the results from different studies are not in agreement and there is a dearth of clinical verification.

At present, a nomogram is the most frequently utilized prediction model in research. It functions as a visual mathematical instrument and is widely applied to predict patient prognosis, having significant value in clinical practice. As far as we know, a preoperative prognostic model for cage subsidence after ACDF has not been established. Hence, this study intends to explore possible risk factors and develop and validate a nomogram prediction model for cage subsidence following ACDF. The goal is to provide a predictive model for identifying patients prone to fusion subsidence, facilitating timely and targeted intervention.

Materials and methods

This retrospective study was approved by the Medical Ethical Committee of West China Hospital of Sichuan University (No.2023(416)). All patients were enrolled after giving informed consent for the analysis of their clinical and radiographic data. The inclusion criteria were: ① Patients with radiculopathy or myelopathy caused by single-level cervical disc disease, confirmed by corresponding magnetic resonance imaging findings and unresponsive to conservative treatment for more than six weeks. ② Participants aged 18 years or older, without psychological issues, who underwent one-level ACDF using the VENTURE™ anterior cervical plate system (Medtronic Sofamor Danek, Memphis, Tennessee, USA) and a poly-etherether-ketone (PEEK) interbody fusion cage filled with allograft material ranging from C3 to C7. ③ Subjects with complete preoperative and postoperative imaging and clinical data. ④ Patients with a minimum follow-up period of one year. The exclusion criteria were: ① Patients with a history of neck surgery, trauma, or tumor at the C1-C7 level. ② Patients undergoing cervical disc replacement (CDR) or hybrid surgery (CDR with ACDF), as well as those undergoing ACDF with different types of devices and multilevel surgery. ③ Patients with comorbidities such as ankylosing spondylitis, spinal deformity, developmental cervical stenosis, and local or systemic infections. ④ Patients lost to follow-up. All patients in this study were of Han nationality.

Surgical technique

All surgeries were performed by senior spinal surgeons within the department using a standard, classical right Smith–Robinson anterior cervical approach after induction of general anesthesia. A horizontal skin incision was carried out to reach the perpendicularly spreading fibres of the platysma muscle. The disc level was confirmed again by fluoroscopy. When discectomy was performed, long shaft Caspar screws for interbody retraction were inserted into the middle of the adjacent vertebral bodies. A high-speed drill was used for the preparation of the sub—chondral end—plate of each vertebral body. The posterior longitudinal ligament along with the anterior, posterior and lateral ostophytes was resected with rongeurs. After thoroughly decompressing and scraping off the cartilaginous endplate, traditional anterior plates incorporated with polyetheretherketone cage or autogenous iliac bone were used. All patients spent the first night in a recovery room due to the potential risk of cervical wound bleeding. All patients agreed to wear an orthosis, usually for 3 months after surgery. Patients underwent CT scanning at 3 months, 6 months, and 12 months post-surgery to evaluate the fusion status.

Data collection

The data collected from enrolled patients encompassed basic details (age, BMI, gender, operated segment), overall health and lifestyle aspects (smoking, alcohol consumption, hypertension, diabetes, operation duration, intraoperative blood loss, bone mineral density, intensity of arm and neck pain, neurological status), and radiological risk factors related to cage subsidence (Hounsfield Units of the vertebral body, Modic changes, cervical sagittal parameters). Basic information and preoperative data were obtained from the medical record on the HIS system. Radiological parameters were measured and extracted from the PACS system. X-rays, CT scans, and MRIs were carried out before surgery, immediately after surgery (prior to discharge), and at 2–3 months, 6 months, 12 months after surgery as well as during the last follow-up.

To reduce measurement errors, an independent panel of radiologists was established in our study. The panel consisted of three study-blinded radiologists. Among them, two radiologists were responsible for data collection while the remaining one was responsible for data analysis.

Definitions

Cage subsidence is typically defined as a reduction in intervertebral segmental height of more than 2 mm in the postoperative period after ACDF [11–13]. Bone mineral density is measured by dual-energy x-ray absorptiometry, and then the hip joint T-score is calculated. Preoperative clinical assessment is carried out by means of visual analogue scale (VAS), Japanese Orthepaedic Association (JOA) scores, and Neck Disability Index (NDI) scores. The intensity of arm and neck pain of patients is evaluated using VAS, which measures pain on a scale of 0 to 10 points, with 0 indicating no pain and 10 representing the highest degree of pain. The NDI scores are utilized to assess the preoperative function of the neck. This study employs the Chinese version of NDI proposed by Wu et al. [14], which is specifically designed for Chinese-speaking people with neck pain. It also uses a 6-point Likert scale ranging from 0 (no disability) to 5 (complete disability) for each item [15]. Disability ratings are as follows: 0 to 4, no disability; 5 to 14, mild disability; 15 to 24, moderate disability; 25 to 34, severe disability; and above 34, complete disability. If the JOA score is lower than 9 [16], it is considered severe.

As another commonly used radiometric measure for evaluating bone quality, this study incorporates the CT-based Hounsfield units (HUs) of the vertebral bodies. HU values were evaluated at the vertebrae above and below the cage placement location (for example, in C5–6 ACDF, measuring HUs at C5 and C6 vertebral bodies). Each vertebral body was individually assessed in the midsagittal, midcoronal, and midaxial planes using the radiology patient archiving and communication system (PACS) [13, 17]. The region of interest (ROI) was demarcated by drawing an elliptic shape and expanded to cover the maximum area within the vertebral body, excluding cortical bone like lateral walls or endplates. Subsequently, the final average HU value was computed (Fig. 1a-c). The preoperative Modic changes were classified into three types: Type I (low T1 and high T2), Type II (high T1 and high T2), Type III (low T1 and T2) [18]. Cervical sagittal parameters, including cervical lordosis, intervertebral segmental height (ISH), functional spinal unit angle (FSUA), and range of motion (ROM) for C2-C7, as well as the sagittal vertical axis (SVA) and T1 slope (T1S), were carefully measured before surgery. Cervical lordosis was defined as the C2-C7 Cobb angle between the lower endplate of the C2 vertebral body and the lower endplate of the C7 vertebral body. ISH was defined as the straight line between the midpoint of the lower end plate of the upper vertebra and the midpoint of the upper end plate of the lower vertebra [13]. The FSUA was measured at the index level between the upper edge of the cranial vertebral body and the lower edge of the caudal vertebral body. The ROM of C2-C7 was defined as the difference of C2-C7 in dynamic lateral X-ray images. The SVA was determined as the distance between the plumb line from the center of C2 and extending to the posterior superior aspect of C7 [19]. The T1 slope was specified as the angle between the horizontal plane and a line parallel to the superior endplate of the T1 vertebrae [20] (Fig. 1d).

Fig. 1.

Measurement method of various imaging parameters. The HU values were measured on the midsagittal (a), midcoronal (b), and midaxial (c) planes of both the upper and lower vertebrae, respectively. The average of these measured values was defined as the final HU value.The other cervical sagittal parameters were measured according to the previous studies (d)

Statistical analysis

The retrospective nature of this study led to a fixed sample size determined by the available data. As in prior literature [21, 22], no upfront power calculation for sample size was carried out. The training set was used for model training, and the validation set was utilized for model validation. For continuous variables with a normal distribution, descriptive statistics were presented as mean ± standard deviation, and between-group comparisons were made using independent sample t-tests. Variables not normally distributed were described using the median [P25, P75], and between-group comparisons were done using the Wilcoxon rank-sum test. Categorical data were presented as counts (%), and between-group comparisons were evaluated using the chi-square test.

The construction of the predictive model based on the training set proceeded as follows: first, a univariate logistic regression analysis was employed to identify potential predictors of the outcome event (P < 0.05). Then, for the selected variables, the Least Absolute Shrinkage and Selection Operator (LASSO) logistic regression algorithm was used to select significant features (non-zero coefficients). The optimal parameter configuration was determined through tenfold cross-validation, and the lambda value corresponding to the minimum deviation distance for one standard error was used to determine the coefficients of the non-zero variables. Further, for the selected variables, a multivariate logistic regression analysis (stepwise, bidirectional) was conducted. A predictive model line chart was constructed based on the variables with P < 0.05 in the stepwise method. The constructed line chart was validated using 1000 bootstrapped samples. Calibration curves were drawn to assess the calibration of the model. The Hosmer–Lemeshow (HL) test was performed to evaluate the goodness of fit, and Receiver Operating Characteristic (ROC) curve analysis, along with the calculation of metrics such as AUC, sensitivity, and specificity, was conducted to assess discriminative performance. Clinical decision curves (DCA) were generated to evaluate the clinical application value of the model and quantify the net benefit within a threshold probability range. Finally, the constructed model was validated in the validation set.

The statistical analysis for this study was performed using R software (version 4.3.1). The"rms"package was used for nomogram construction and calibration plot generation. ROC curves were plotted using the"pROC"and"ggplot2"packages. Clinical decision curve analysis was carried out using the"ggDCA"package. A two-sided test with P < 0.05 was regarded as statistically significant.

Results

In accordance with the above inclusion and exclusion criteria, a total of 476 patients who underwent anterior cervical discectomy and fusion from January 1, 2014, to June 30, 2023, were included in this study and randomly divided into a training cohort and a validation cohort, comprising 234 males and 242 females (Fig. 2). The patients were aged between 35 and 75 years, with an average of (51.16 ± 8.34) years. In the training cohort, 57 out of 238 patients (23.95%) had cage subsidence, while 181 patients (76.05%) did not experience cage subsidence after ACDF. The results of univariate logistic regression analysis revealed that bone mineral density (P < 0.001), upper HU value (P < 0.001), lower HU value (P < 0.001), ISH (P = 0.025), and T1 slope (P = 0.025) were significantly associated with postoperative cage subsidence (Table 1). Besides, the results of multivariate logistic regression analysis indicated that bone mineral density (OR = 0.499, 95% CI.356-0.677), upper HU value (OR = 0.979, 95% CI.971-0.986), lower HU value (OR = 0.989, 95% CI.982-0.996), ISH (OR = 0.504, 95% CI.284-0.856), and T1 slope (OR = 0.877, 95% CI.821–0.933) were independent risk factors for postoperative cage subsidence (Table 2). Subsequently, a nomogram was developed based on the result of logistic regression analysis (Fig. 3).

Fig. 2.

Flow chart for screening eligible patients

Table 1.

Comparison of variables between the subsidence group and non-subsidence group in the training cohort

	Total	Subsidence Group	Non-subsidence Group	P value
Number of patients	238	57	181
Age (year)	53.62 ± 8.10	55.35 ± 6.36	53.07 ± 8.52	P = 0.075
Sex (%)				P < 0.001
Male	101	25	76
Female	137	32	105
BMI (kg/m2)	23.05 ± 4.25	22.91 ± 2.91	23.10 ± 4.60	P = 0.680
Smoking (%)				P = 0.883
No	182 (76.5)	44 (77.2)	138 (76.2)
Yes	56 (23.5)	13 (22.8)	43 (23.8)
Drinking (%)				P = 0.763
No	193 (81.1)	47 (77.2)	146 (80.7)
Yes	45 (18.9)	10 (22.8)	35 (19.3)
Hypertension (%)				P = 0.724
No	171 (71.8)	42 (73.7)	129 (71.3)
Yes	67 (28.2)	15 (26.3)	52 (28.7)
Diabetes (%)				P = 0.702
No	188 (79.0)	44 (77.2)	144 (79.6)
Yes	50 (21.0)	13 (22.8)	37 (20.4)
Operation time (Mean ± SD)	102.95 ± 12.06	100.70 ± 11.92	103.65 ± 12.05	P = 0.107
Intraoperative bleeding (ml)	70.02 ± 17.54	70.81 ± 15.35	69.77 ± 18.21	P = 0.847
T-score (g/cm2)	0.09 ± 1.37	−0.65 ± 1.28	0.33 ± 1.31	P < 0.001
VAS for arm	6.38 ± 1.01	6.47 ± 0.98	6.35 ± 1.01	P = 0.339
VAS_for_neck	6.44 ± 0.98	6.35 ± 0.95	6.47 ± 0.99	P = 0.411
JOA_score	11.90 ± 1.95	11.54 ± 1.99	12.01 ± 1.94	P = 0.098
NDI_score	20.49 ± 4.44	20.13 ± 4.38	20.60 ± 4.46	P = 0.466
HU value of upper vertebral body	364.81 ± 59.21	324.07 ± 51.30	377.64 ± 55.73	P < 0.001
HU value of lower vertebral body	345.73 ± 57.02	319.84 ± 62.62	353.88 ± 52.75	P < 0.001
Modic_changes (%)				P = 0.146
0	91(38.2)	17 (29.8)	74 (40.9)
I	115(48.3)	34 (59.6)	81 (44.8)
II	32(13.4)	6 (10.5)	26 (14.4)
C2_7_angle (°)	14.16 ± 8.05	15.05 ± 8.76	13.88 ± 7.81	P = 0.877
ISH	4.88 ± 0.78	4.67 ± 0.78	4.94 ± 0.77	P = 0.007
FSUA (°)	1.77 ± 1.09	1.62 ± 0.96	1.82 ± 1.12	P = 0.406
C2_C7_ROM (°)	51.68 ± 7.37	50.14 ± 8.00	52.16 ± 7.12	P = 0.097
SVA (mm)	19.07 ± 5.97	19.47 ± 6.08	18.95 ± 5.94	P = 0.511
T1 slope (°)	27.78 ± 6.35	24.57 ± 6.45	28.79 ± 5.99	P < 0.001
C5-6 or C6-7 level surgery (%)				P = 0.739
No	71(29.8)	16 (28.1)	55 (30.4)
Yes	167 (70.2)	41(71.9)	126 (69.6)
Follow-up periods (month)	23.31 ± 6.14	22.54 ± 7.16	23.55 ± 5.78	P = 0.283

Abbreviations: BMI body mass index, VAS visual analogue scale, JOA Japanese Orthepaedic Association, NDI Neck Disability Index, ISH intervertebral segmental height, FSUA functional spinal unit angle, HU Hounsfield units, ROM range of motion, SVA sagittal vertical axis

Table 2.

Univariate and multivariate logistic regression analysis of cage subsidence after ACDF in the training cohort

	Univariate analysis OR (95% CI)	P value	Multivariate analysis OR (95% CI)	P value
Age (year)	1.036 (0.998, 1.077)	P = 0.065	NA
Sex (%)	0.996 (0.969, 1.021)	P = 0.763	NA
BMI (kg/m2)	0.990 (0.916, 1.060)	P = 0.776	NA
Smoking (%)	0.948 (0.454, 1.885)	P = 0.883	NA
Drinking (%)	0.888 (0.391, 1.872)	P = 0.763	NA
Hypertension (%)	0.886 (0.442, 1.707)	P = 0.724	NA
Diabetes (%)	1.150 (0.546, 2.312)	P = 0.702	NA
Operation time (min)	0.979 (0.954, 1.004)	P = 0.108	NA
Intraoperative bleeding (ml)	1.003 (0.987, 1.021)	P = 0.696	NA
T-score (g/cm2)	0.577 (0.451, 0.728)	P < 0.001	0.499 (0.356,0.677)	P < 0.001
VAS for arm	1.134 (0.842, 1.536)	P = 0.411	NA
VAS_for_neck	0.883 (0.650, 1.198)	P = 0.424	NA
JOA_score	0.887 (0.763, 1.031)	P = 0.117	NA
NDI_score	0.976 (0.912, 1.044)	P = 0.485	NA
Upper vertebra HU value	0.984 (0.977, 0.989)	P < 0.001	0.979 (0.971,0.986)	P < 0.001
Lower vertebra HU value	0.988 (0.982, 0.994)	P < 0.001	0.989 (0.982,0.996)	P = 0.005
Modic_changes (%)	1.827 (0.953, 3.606)	P = 0.074
C2_7_angle (°)	1.018 (0.981, 1.057)	P = 0.339	NA
ISH	0.615 (0.395, 0.927)	P = 0.025	0.504 (0.284,0.856)	P = 0.014
FSUA (°)	0.832 (0.610, 1.105)	P = 0.222	NA
C2_C7_ROM (°)	0.963 (0.924, 1.003)	P = 0.073	NA
SVA (mm)	1.015 (0.965, 1.065)	P = 0.567	NA
T1 slope (°)	0.906 (0.865, 0.949)	P < 0.001	0.877 (0.821,0.933)	P < 0.001
C5-6 or C6-7 level surgery (%)	1.119 (0.587, 2.208)	P = 0.739	NA

Abbreviations: BMI body mass index, VAS visual analogue scale, JOA Japanese Orthepaedic Association, NDI Neck Disability Index, ISH intervertebral segmental height, FSUA functional spinal unit angle, HU Hounsfield units, ROM range of motion, SVA sagittal vertical axis

Fig. 3.

A nomogram to predict the risk of subsidence

The area under the ROC curve (AUC) of the nomogram was.885 (95% CI.835-0.935, Fig. 4A), which demonstrated that our prediction model exhibited satisfactory discrimination. The calibration curve of the nomogram indicated good agreement between the observed cases and predicted cases in this dataset (Fig. 5A). Subsequently, the C-index of this model was.885 (95% CI.835-0.935) in this dataset and was determined to be.885 (95% CI.834-0.931) through bootstrapping validation (Bootstrap = 1000) (Fig. 6A). To identify the potential clinical benefit of the designed nomogram, a decision curve analysis (DCA) was conducted in this dataset. As shown in Fig. 6A, the DCA suggested that the clinical net benefit of this risk prediction nomogram within a range of risk thresholds (0.04 to.92) was higher than that of all-screening or no-screening strategies.

Fig. 4.

Receiver operating characteristic curve analyses for the nomogram model in the training and test cohort. A Training cohort. B Validation cohort

Fig. 5.

Calibration curve for predicting probability of subsidence after ACDF. A Training cohort. B Validation cohort

Fig. 6.

Decision curve analysis in prediction of subsidence after ACDF. A Training cohort. B Validation cohort

Furthermore, external validation using the other 238 patients showed an AUC of 0.914 (95% CI.873-0.955, Fig. 4B), indicating the high accuracy of the nomogram. The model demonstrated good consistency, and the calibration curve of the validation cohort closely followed the ideal diagonal line (Fig. 5B). Additionally, the C-index of this model in this dataset was.914 (95% CI.873-0.955) and was determined to be.914 (95% CI.873-0.950) through bootstrapping validation (Bootstrap = 1000). Moreover, the decision curve analysis (DCA) is shown in Fig. 6B and suggested that the clinical net benefit of this risk prediction nomogram within a range of risk thresholds (0.02 to.97) was higher than that of all-screening or no-screening strategies. These data confirm that our nomogram has significant potential for clinical decision-making.

Discussion

Currently, cage subsidence is the most prevalent hardware-related complication of ACDF, with an incidence ranging from 19.3% to 42.5% [6, 7, 10]. Subsidence can have adverse clinical and biomechanical impacts [23, 24]. In both clinical and biomechanical studies, numerous factors have been reported to be associated with cage subsidence risk, such as cage material, cage geometry, bone mineral density, device type, surgical level, bone graft, and patient age [25–35]. Nevertheless, the results of different studies are inconsistent, making it difficult to draw definite conclusions about these risk factors. Moreover, there is a lack of a predictive nomogram model to assist spine surgeons in proactively preventing cage subsidence after ACDF in clinical practice. In this study, the fusion cages used were the same material, guaranteeing the consistency of the material factor among all groups. Meanwhile, through statistical analysis and verification, it was found that there was no statistically significant difference in the comorbidities among the groups of patients. This has maximally excluded the potential interference of comorbidity differences on the research results and ensured the reliability of the research data and the accuracy of the results. Since there are numerous influencing factors for the subsidence of cervical spine fusion cages, it is extremely difficult to control all factors to be completely identical.

This study has several advantages compared to previous studies that identified risk factors for cage subsidence after ACDF. We developed and validated a nomogram for predicting cage subsidence in patients with single-level cervical disc disease who underwent ACDF surgery. As far as we know, our study is the first to create a nomogram to predict the occurrence of cage subsidence after cervical spine surgery based on preoperative parameters. Although our nomogram has only five parameters, its predictive performance was satisfactory after evaluation using calibration curves and the C-index. At the same time, all parameters included in this nomogram are easy to measure or calculate and have good reproducibility.

Another benefit of our nomogram is that it offers a more targeted intervention strategy for patients. We hold the view that risk factors are meaningful only if they can be changed through evidence-based interventions. Unlike uncontrollable parameters like age or sex, both of these parameters can be adjusted by drug therapy and functional exercise. Currently, anti-osteoporosis drugs, including bone resorption inhibiting drugs and bone formation promoting drugs, have obtained good results in clinical practice [36]. Moreover, basic lifestyle changes such as regular physical activities, reducing the risk of falls, nutritional and dietary counseling plus modification, early bone screening, and bone loss prevention have proven to be an essential management strategy for osteoporosis [37]. Additionally, a randomized controlled trial of osteoporotic females discovered that osteoanabolic exercises significantly increased bone mineral density (BMD), improving the t-value due to its high level of mechanical strain, which generates beneficial effects on the bone health of postmenopausal women [38].

In our study, multivariate logistic regression determined that the HU value of upper and lower vertebral bodies and preoperative disc height are also independent significant predictors of interbody cage subsidence after ACDF. These findings are in line with previous literature [13, 24, 39–43]. Recently, Pu et al. compared the predictive effect of HU value and T-scores in subsidence after ACDF with Zero-profile cage and found that HU value had a better predictive effect [39]. Wang et al. [13] also reported that the average cervical vertebral body HU was 385.0 ± 60.8, and they discovered that lower preoperative CT HU values are associated with cage subsidence in single-level ACDF (Mean HU: 390.6 ± 60.1 in the non-subsidence group vs 322.6 ± 27.7 in the subsidence group). In the training cohort of this study, the average upper and lower HU value was 390 (range of 340 to 421) and 336 (range of 273 to 362) in the non-subsidence group and 390 (range of 340 to 421) and 308 (range of 285 to 341) in the subsidence group. The preoperative ISH is considered one of the most important indicators of disc degeneration, and one of the main goals of ACDF is to restore the loss of disc height due to the degenerative process [44, 45]. Preoperative disc height loss may lead to adjacent level disc and facet joint degeneration in the cervical spine; with a 1-mm DH loss, the facets carry 10–40% of the total load [46]. The result of our study showed that preoperative disc height was another independent risk factor associated with postoperative cage subsidence. Although the specific mechanisms underlying the relationship between the intervertebral disc and cage subsidence after ACDF remain unclear, literature has reported that excessive micromotion at the implant-bone interface may play a non-negligible role in this process [2, 47].

In the current research context, the subsidence of intervertebral fusion cages has received significant attention. Micromotion is regarded as one of the potential mechanisms. However, existing studies have significant deficiencies in this regard. Although the role of micromotion in subsidence has been mentioned, there is a lack of comprehensive and in—depth analysis, and the complex interactions and mechanisms between micromotion and subsidence have not been fully explored. Especially for the interfacial micromotion after cervical spine surgery, how it affects the stability of the fusion cage, bone integration, and the final subsidence situation remains unclear. The relevant biomechanical mechanisms and the interactions with surrounding tissues also await further research. In addition, multiple factors such as the patient's bone quality (e.g., bone mineral density), surgical procedures (e.g., the implantation position and method of the fusion cage), and the design and material of the implant may also contribute to subsidence. The relationship between the commonly used postoperative auxiliary stabilization device, the cervical collar, and interfacial micromotion as well as the subsidence rate is still blank. The cervical collar may play a role in restricting cervical spine movement, reducing micromotion, and thus affecting the subsidence rate, but this has not been studied.

Previous research has produced inconsistent results regarding the connection between the loss of cervical sagittal malalignment and cage subsidence after ACDF. Park et al. believed that cage subsidence was not related to cervical sagittal alignments before and after surgery [48]. However, in another study, a contrary result was found for this relationship [49]. Lee et al. investigated whether postoperative subsidence could be predicted based on specific parameters on preoperative plain radiographs and discovered that a lower T1 slope (T1 slope < 28°) could be a risk factor for subsidence following ACDF [49]. This study also obtained similar results. In the training cohort of our study, the mean preoperative T1 slope was 30.81° (range from 25.17° to 32.95°) in the subsidence group and 27.86° (range from 18.91° to 30.16°) in the non-subsidence group. T1 slope is one of the most crucial determinants of overall spinal sagittal balance. A high T1 slope can be the outcome of thoracic hyper kyphosis or lumbopelvic sagittal malalignment, which can cause a compensatory anterior tilt of the cervical spine and a forward head posture [50]. Recently, Li et al. identified through multivariate logistic regression analysis that, among other factors, a higher T1 slope was associated with a lower incidence of dislocation of interbody fusion devices. They speculated that a higher T1 slope leads to the restoration of lordosis to normal values, influencing the physiological load on both the allospacer and posterior facet joints [51].

Our nomogram, which incorporates independent risk factors such as bone mineral density, upper and lower HU values, ISH, and T1 slope, has significant implications for both surgical decision—making and preoperative patient counseling.

In surgical decision—making, it enables personalized surgical planning. Surgeons can anticipate a higher likelihood of cage subsidence in patients with low bone mineral density, for example, and consider alternative techniques like augmented fixation or adjusting the fusion cage design. Additionally, it allows for risk stratification. By calculating the probability of cage subsidence, high—risk patients can be identified for more frequent postoperative follow—ups and intensive monitoring, with early imaging studies if necessary. For preoperative patient counseling, the nomogram serves as an objective and visual tool for informed consent. Surgeons can input patients'specific parameters to show individualized predicted risks, allowing patients to make more informed treatment decisions. It can also be easily integrated into the existing clinical workflow, with nurses or physician assistants collecting data and providing immediate risk scores to the surgical team during preoperative planning. Moreover, as more data becomes available, the nomogram can be continuously refined through future prospective studies, and new factors related to cage subsidence can be incorporated to enhance its predictive accuracy.

Preventing cage subsidence following cervical fusion requires a multimodal approach. Intraoperatively, meticulous endplate preservation should be implemented using micro-curettes to conservatively prepare endplates while maintaining sufficient subchondral bone integrity [25, 52], complemented by utilizing anatomically contoured PEEK cages to optimize endplate contact area [53]. This strategy emphasizes balancing adequate preparation for graft integration with structural preservation of load-bearing bone, guided by preoperative imaging assessments of bone quality and intraoperative tactile feedback.The formation of the osseous bridging bridge typically requires approximately three months. Consequently, most complications—including graft subsidence and cervical lordosis deterioration—tend to manifest predominantly within the initial 6-week postoperative period [48, 54, 55].Thus, the collar immobilization times were all set within 3 months post-surgery [56]. For patients with low CT Hounsfield unit values or decreased bone mineral density undergoing cervical spine surgery, perioperative anti-osteoporotic therapy may can enhance bone-implant stability and reduce cage subsidence.

Our study has several limitations that merit attention. First and foremost, the single—centre, retrospective, and non—randomized study design inherently restricts the strength of the evidence. Such a design may introduce selection bias as the patient population is limited to a single institution, and the data collection occurs after the events, potentially leading to recall bias. Secondly, we did not analyze some potential significant predictors due to various reasons. For example, all the cages used in this study were made of PEEK. This uniformity in material precluded the analysis of how different device types and materials might impact cage subsidence. Moreover, the exclusion of multilevel ACDF cases limits the generalizability of our findings. Multilevel procedures often involve different biomechanical stress distributions and surgical complexities compared to single—level cases, which could influence the occurrence of cage subsidence. Geographical and ethnic factors also play a role. As our study was carried out in a region with a predominantly Han population, this restriction potentially gives rise to selection bias. Future research should expand its scope to include a broader range of ethnicities to enhance the universality of the research findings. In addition, important statistical analyses such as multicollinearity tests and the LASSO regression model were not performed. These analyses could have helped in better understanding the relationships between variables and potentially improving the model's predictive power. Finally, long—term follow—up studies to assess subsidence were not available for all cases. Although previous studies suggest that cage subsidence after ACDF typically occurs within a relatively short time after surgery, long—term monitoring might reveal late—onset subsidence or its secondary effects on patient outcomes. This lack of comprehensive long—term data could affect the full understanding of the phenomenon.

Conclusion

Our predictive model is set to become a valuable and convenient tool for clinical decision-making, enabling early prediction and timely intervention for cage subsidence after ACDF. Future prospective studies could incorporate our model in different clinical scenarios, thus reducing the burden on patients and healthcare resources.

Authors’ contributions

Dongmei Zhao collected and analyzed the data, and Xiaojie Sun wrote the main manuscript text. All authors reviewed the manuscript.

Funding

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Data availability

The data are not publicly available due the privacy of patients included but are available from the corresponding author on reasonable request for academic research purpose.

Declarations

Ethics approval and consent to participate

The experimental protocol was established, according to the ethical guidelines of the Helsinki Declaration and was approved by the Medical Ethics Committee of West China Hospital, Sichuan University, Written informed consent was obtained from individual or guardian participants.

Consent for publication

Written informed consent for publication was obtained from all participants.

Competing interests

The authors declare no competing interests.