Risk factors for early postoperative CSF leak after anterior cervical surgery
It has been previously reported that the risk factors for CSFL after anterior cervical surgery include age, sex, BMI, revision surgery, history of hypertension, OPLL, spinal stenosis, surgery levels, and intraoperative blood loss15–21. However, the conclusions are also very different in different studies, which may be related to the discrepancy between the included samples. The conclusion of this study was that age, revision history, surgical OPLL, and blood loss were independent risk factors for CSF leak.
Age is usually a risk factor for a variety of diseases due to the aging of cells and degeneration of tissues. With increasing of age, dural degeneration makes its thickness thinner, and traction resistance decreases, so dural tears are more likely to occur 22. The regression model of Kapadia et al.15 showed that the risk of CSF leak was 1.25 times higher in the age group of 55–69 years age group than in the 40–54 years age group (P = 0.038, 95% CI: 1.01, 1.55), while the risk increased to 1.64 in the age group above 70 years age group (P = 0.001, 95% CI: 1.22, 2.20). Ehresman et al.16 showed that the risk of ID/IDT increased by an average of 0.03 (P < 0.001, 95% CI: 1.012, 1.048) for per year of age. We found through the RCS curve that age and CSF leak had a linear relationship (P = 0.003), and that the risk of CSF leak rose continuously with increasing age. Moreover, the risk of CSF leak had a 1.09-fold change for every 11-years increase. When the age is over 53 years, the risk value is greater than 1, suggesting a higher risk of CSF leak after surgery for patients over 53 years.
In some patients, spinal symptoms are not relieved or recur after surgery, so revision surgery, a second operation in the same site, is necessary. Hannallah et al.20 showed that patients who underwent revision surgery had a 2.75-fold higher incidence of CSF leak than those who underwent primary anterior surgery (P = 0.05, 95% CI: 0.85, 8.93). In this study, 17.9% (5/28) of revision surgery cases developed CSF leak, with a risk of 6.41 times that of primary surgery. The scar occurred after repetitive surgeries at the same level and adhered to the spine dura mater, predisposing patients to ID/IDT when separated16. Cunningham et al.23 found in a rabbit laminectomy model that a second operation resulted in elevated TGF-β expression, which caused epidural fibrosis and aggravated adhesions, further increasing the risk of ID/IDT.
Studies have shown that the incidence of OPLL in the Asian population is high (2.4%), and the most common site is the cervical spine24. OPLL can lead to cervical spinal stenosis, causing spinal cord compression. Therefore, the incidence of OPLL is also higher in patients with DCM. The incidence of OPLL in this study was 7.5% (31/383), which may also be one of the reasons for the higher incidence of CSF leak in this sample. According to the literature, OPLL has been recognized as a risk factor for the occurrence of CSF leak in spinal surgery. Cervical OPLL usually presents with dura ossification (DO), and it is a great technical challenge to separate the ligament from the posterior longitudinal ligament once the ossified dura mater is fused with the posterior longitudinal ligament25. Hannallah et al.20 reported that the incidence of CSF leak in patients with OPLL after cervical spine surgery was 12.5%, which was 13.7 times higher than that in the group without OPLL, and the data in our survey turned out to be 4.45. Considering the high risk of CSF leak for patients with OPLL, the ossified posterior longitudinal ligament may not be forcibly removed during the operation instead of being floated using a burr or ultrasonic osteotome. Otherwise, posterior surgery, such as open-door laminoplasty, may be performed26.
A study of Desai et al.19 argued that there was a significant correlation between intraoperative blood loss and ID/IDT (534.4/288.9 ml, P < 0.001). In anterior cervical surgery, bleeding likely contributes to the complex course and rich blood supply in the cervical epidural venous plexus. Due to more blood loss, the visual field of the operation area is unclear, which more easily causes ID/IDT. This study found that there was a linear relationship between the amount of bleeding and CSF leak (P < 0.001). Each additional bleeding of 178 ml increased the risk of CSF leak by 54%. Notably, 154 ml is a cutoff of interest. Once the amount of bleeding exceeds this volume, the risk effect of CSF leak will be greater than 1.
Previous studies have been highly controversial regarding gender differences in CSF leak. Takahashi et al.27 reported in a study of risk factors for IDT in lumbar surgery that the incidence of IDT was significantly higher in women (5.6%) than in men (3%) (P < 0.05). A multicenter observational study by Ishikura et al.17 also yielded similar conclusions (OR = 1.47, P < 0.001), but neither study proposed a reasonable explanation for the gender difference. The data obtained by Fam et al.28 after analyzing skull base dural thickness in 20 cadavers indicated that dural thickness was thinner in women than in men (P = 0.06), which may be evidence to explain the former conclusion. However, disappointingly, Kwon et al.22 showed that there was no significant difference in cervical capsule thickness between women and men (P = 0.347). In summary, there is no direct evidence that gender is one of the risk factors for CSF leak.
Evaluation and significance of predictive models
The evaluation indicators of the predictive model included discrimination and calibration11. Discrimination is the ability of the model to identify an event that occurs or does not, while calibration reflects the degree of consistency between the prediction and the actuality. The ROC curves can provide more sensitive measurements of discrimination, that is, C-statistics29. Our optimal model and preoperative model both had good discrimination (C-statistics > 0.5), and the forest plot of C-statistics further illustrated that with the introduction of variables, the discrimination of the model became increasingly better. Although the Hosmer-Lemeshow test is a relatively succinct method in calibration evaluation, it is blunted to test due to the dependency on sample size. The use of calibration curves, however, can take full advantage of the graph to respond to the calibration level of models11,30.
Although it is a rare complication in anterior cervical surgery, CSF leak after cervical surgery becomes a challenge for surgeons due to its insidious occurrence and serious sequelae. Once CSF leak occurs, continuous lumbar drainage or permanent cerebrospinal fluid shunting are required, which reduces the quality of life and increases the economic burden for patients. Some may even die of severe meningitis or pneumocephalus31. Our predictive model includes an optimal model of five variables as well as a preoperative model with removal of intraoperative blood loss, which can be used for risk assessment at different stages. For example, the preoperative model can be used to guide the preoperative level of care and management. For patients with CSF leak, repair devices and materials should be prepared in advance to cope with unavoidable durotomy and tears. Microsurgery can be another selection if possible32. In another example, the removal of ossified ligaments during surgery has become a controversial topic. Some researchers believe that there may be herniated intervertebral disc substances in the ossified posterior longitudinal ligament, which should be removed together. The other view is that the posterior longitudinal ligament should be removed only when there is significant evidence20. The risk score calculated by the model can replace the subjective judgment of the surgeons and make them comprehensively consider whether to remove the posterior longitudinal ligament based on the risk and benefit.
Our study has several limitations. First, this study was not a multicenter study, and the sample size was small. There were only 17 cases of cerebrospinal fluid leakage, which could not well reduce the sampling error. Second, the data were obtained from the medical records during hospitalization, so the patient's condition after discharge was unknown which caused the incompleteness of the data in time. Furthermore, all of the patients included in this study underwent open surgery, so the timeliness of the study may be reduced as microscope-assisted surgery develops further. Finally, our model included only five variables, which may limit the prediction ability. Therefore, we need to collect more case data to expand the sample size and conduct prospective studies to improve the level of evidence. Cases undergoing microscopic surgery should also be included. Last but not least, more scientific analytical methods should be used for further data mining, and more variables that can be included in the model should be found to improve its predictive power.