Predictive Model of Cerebrospinal Fluid Leakage After Posterior Circumferential Decompression for Thoracic Ossi cation of Posterior Longitudinal Ligament


 Background: Cerebrospinal fluid leakage(CSFL) is one of the most common complications after posterior transarticular osteotomy and circumferential decompression for thoracic ossification of posterior longitudinal ligament(OPLL). It is of great usefulness If the cerebrospinal fluid leakage can be predicted preoperatively. These predictors help to attract the attention of the surgeon in advance and warn the patient. Therefore, the aim of this study is to find out the factors that can predict the CSFL prior to operation and try to build a predictive model.Methods: A total of 61 patients with thoracic OPLL underwent posterior transarticular osteotomy and circumferential decompression from August 2015 to June 2020 in our hospital were included in this study, including 29 males and 32 females. The patients were divided into CSLF group and non-CSFL group according to whether they suffered cerebrospinal fluid leakage. Univariate analysis was used to identify possible predictors in Demographic characteristics, clinical and radiological data. A logistic regression model was developed by multivariate analyses to predict probability of CSFL. Model validation was done using the receiver operating characteristic(ROC) curve.Results: The incidence of CSFL was 31.1%, including 7 males and 12 females, with an average age of 49.8 ±11.4 years. The mean drainage indwelling time in CSFL group was 5.6±1.0 days, which was significantly higher than that in non-CSFL group (4.2±1.3 days, P < 0.001). The mean length of hospital stay was 16.3±6.3 days, slightly higher than that of the non-CSF group (15.8±6.7 days), but there was no statistical difference (P=0.77). Among them, 12 patients (63.2%) suffered low intracranial pressure manifested as headache; 1 patient (5.3%) had cerebrospinal fluid outflow from the incision, and the wound healed successfully after debridement.1 patient (5.3%) was re-admitted to the hospital due to fever after 3 weeks, considering deep wound effusion and pleural effusion. The wound effusion was found to be cured after 2 weeks of anti-infective treatment. Univariate regression analysis showed statistical differences (P<0.05) in smoking history, segment of circumferential decompression, combined with ossification of the ligamentum flavum (OLF), number of laminectomy, occupying ratio and OPLL base ratio. Multivariate regression model showed smoking history (OR=30.1, P=0.003), the upper thoracic segment (OR= 188.0, P= 0.002), the middle thoracic segment (OR= 57.4, P= 0.005) and OPLL base ratio (OR=1.3, P=0.007) were the predictors of CSFL. The ROC curve was in the upper left corner (area under the curve = 0.955, 95% CI 0.91-1.00, P< 0.001), indicating good predictability of the model.Conclusion: The predictive model established in this study has a high predictive effect. When the patients with thoracic OPLL have smoking history or the segment of circumferential decompression is located in the upper or middle thoracic spine or the OPLL has a wide base, the operator should be highly alert to the possibility of postoperative CSFL and warn the patient before surgery. Evidence level: level II-2

the pedicle will be removed when it is necessary to "penetrate" on both sides. Finally, the OPLL was pushed forward to the intervertebral space, and OPLL block was removed from the back side to complete the decompression of ventral spinal cord.
Dural defect often occurs in the two steps of the operation: when the laminectomy is performed and when the OPLL block is pushed into the intervertebral space so that it can be separated from dural mater. If there was serious adhesion between ossi cation block and dural mater, it may also lead to defects of dural. Ventral dural injury or defects was generally not repaired due to surgical eld limitations. Dorsal dural injury is rst treated with a direct suture with a 5 − 0 vessel suture. If sutures are di cult or large dural defects occur, brain cotton would be temporarily used to cover to ensure clear eld. After spinal cord decompression, a whole piece of gelatin sponge can be used to cover the dural injury part, without repair of large dural defects. Finally, the surgical incision is closed layer by layer by strict continuous suture.

Diagnostic criteria for CSFL
The patient is diagnosed as CSFL if their symptoms meet any of the following diagnostic criteria (1) Dural rupture or defect is found during operation, or cerebrospinal uid out ow is seen; (2) Large amounts (> 300 ml/days) of clear or light blood drainage uid out ow at wound drainage tube remain visible after 48 hours after surgery, and the patients develop severe headaches, nausea, vomiting and other low cranial pressure symptoms.
Postoperative management of CSFL Once a patient was found to have CSFL, a uni ed treatment process would be adopted: (1) Continuous lateral position without pillow. The wound negative pressure drainage ball was replaced by atmospheric drainage bag during 12-24 h after surgery, and the atmospheric drainage bag was placed at the same horizontal position of the surgical incision. (2) Once the patient has symptoms of low intracranial pressure such as headache, the bed tail would be raised for 10 cm, or the trendelenburg position would be adopted to alleviate the symptoms. (3) Water and electrolyte should be taken on a regular basis to strengthen nutritional support. For patients with low intracranial pressure symptoms, intravenous infusion of concentrated sodium chloride was conducted to alleviate the symptoms. (4) The patient should be aware of infection and extend the time of antibiotic use. Antibiotics can be stopped after the drainage tube is removed and the body temperature is normal for 3 consecutive days. (5) The wound drainage tube shall be kept for at least 5 days, but generally not more than 1 week. After the color of the drainage uid was basically clear, the drainage tube can be removed. One to two stitches were sutured in the whole layer of the drainage tube to avoid the out ow of cerebrospinal uid from the drainage port.

Predictor of CSFL
The cases were divided into CSFL group and non-CSFL group based on the CSFL diagnostic criteria. The predictors considered included: (1) Demographic characteristics: age, gender, body mass index; (2) Clinical data: preoperative JOA score, course of disease, complications (hypertension, diabetes), smoking history, segment of circumferential decompression (upper, middle, lower thoracic segments), and the number of laminectomy segments. (3) Radiological data: preoperative thoracic CT and MRI were performed in all enrolled cases before surgery. According to the thoracic CT sagittal image, the posterior longitudinal ligament after ossi cation was divided into three types: isolated, continuous ( at or wavy) and hybrid. The medical personnels would determine whether the segment of circumferential decompression was combined with OLF. The occupying ratio was de ned as the OPLL thickness divided by the front and rear diameter of the spinal canal, which can be measured on the sagittal CT scan. The OPLL base ratio was de ned as the OPLL width divided by the transverse diameter of the spinal canal, which can be measured on the axial CT scan (Fig. 1). The Intramedullary high signal can be de ned in the thoracic MRI T2 weighted image.

Statistical analysis
This was a retrospective cohort study, SPSS 19.0 (SPSS, USA) statistical software was adopted in this study. The measurement data were expressed with the mean ± standard deviation. Single-factor regression analysis was used to analyze whether possible predictors had statistical signi cance. Predictors that are of signi cance to statistics were included into the multivariate regression analysis. A predictive model was constructed, and a receiver operating characteristic (ROC) curve was drawn.

Incidence of CSFL
According to CSFL diagnostic criteria, 19 patients were in the CSFL group, and 42 patients were in the non-CSFL group. The incidence of CSFL of 31.1%. Among the enrolled cases, 7 are males and 12 are females, with an average age of 49.8 ± 11.4 years. Dural injury was found in 14 cases, including 10 ventral dural injury, and 4 dorsal dural injury. Although no dural injury was found in the other 5 cases, they met the diagnostic criteria of CSFL (2) ( Table 1). Complications related to CSFL 12 cases (63.2%) had symptoms of low intracranial pressure like headaches, and the symptoms were relieved after drainage tube removal; 1 patient (5.3%) had cerebrospinal uid out ow from the incision, and the wound healed successfully after debridement; 1 case (5.3%) was re-admitted to the hospital for fever 3 weeks after the operation, and the wound effusion was found to be cured after 2 weeks of antiinfective treatment; No patient had complications such as falling pneumonia, urinary tract infection, central nervous system infection or venous thrombosis of lower extremity. The average retention time of drainage tube was 5.6 ± 1.0 days, which was signi cantly higher than that of non-CSFL group (4.2 ± 1.3 days, P < 0.001); the average hospitalization days were 16.3 ± 6.3 days, which was slightly higher than those in the non-CSFL group (15.8 ± 6.7 days), but there was no statistical difference (P = 0.77).

Predictor of CSFL
Univariate regression analysis showed statistical differences (P < 0.05) in smoking history, segment of circumferential decompression, combined with OLF, number of laminectomy, occupying ratio and OPLL base ratio ( A Hosmer-Lemeshow test method was used to analyze the tting degree of the model. The results showed that the difference was not statistically signi cant (P = 0.62), indicating that the model tted well and could be used for risk prediction. The PRE was used as the target variable, the CSFL as the state variable, then a ROC curve was drawn. It was shown that the ROC curve was in the upper left corner (area under the curve = 0.955, 95% CI 0.91-1.00, P < 0.001), indicating good predictability of the model (Fig. 2). youden index = sensitivity + speci city − 1. The maximum Yorden index of this model was 0.81. This model had a sensitivity of 94.7% and a speci city of 85.7%.

Discussion
CSFL is a common complication of spinal surgery, and the incidence of different sites varies greatly.
Hannallah [] reported that the incidence of CSFL after cervical surgery was only 1%. Marcelo [] reported that the incidence of CSFL after lumbar vertebrae surgery was 3.2%. The incidence of CSFL after thoracic surgery was much higher than those of other parts of the spine, which was reported to be about 20% 40% [,,]. Moreover, the operative approach differed signi cantly. Hu [] counted 362 cases of thoracic surgery, and the overall incidence of CSFL rate was 32.3%. The incidence of anterior decompression (AD), posterior decompression (PD) and circumferential decompression (CD) were 20.5%, 31.1% and 41.8%, respectively. The incidence of CSFL of CD was much higher than other surgical approaches. Takahara [11] reported that the incidence of CSFL after CD was as high as 40%. The posterior approach of CD was a direct decompression, which required the removal of ventral compression of the spinal cord from the rear. As a result, spinal cord occlusion led to narrow space and limited visual eld, which was the reason why the incidence of CSFL of circular decompression surgery was higher than that of other approaches. The surgical method in this study is CD, and the incidence of CSFL was 28.0%, which was slightly lower than the previous result. The improvement of surgical instruments is an important factor. During the operation, we used the piezosurgery to assist in spinal cord decompression. The safety and e ciency of piezosurgery in the removal of bone structure have been recognized by academic circles. Compared with the traditional bone knife or high-speed drill, high-energy, high-frequency but low-amplitude characteristics can achieve basically no vibration when removing ossi ed blocks, and the characteristics of "cutting soft and not cutting hard" can help to avoid dural injury. As a result, the rational use of piezosurgery can effectively reduce the incidence of CSFL. But it must be admitted that even if the incidence of CSFL could be reduced through operator's surgical techniques and improved surgical instruments, nearly a third of cases were diagnosed with CSFL. Therefore, predicting the related factors that can easily lead to CSFL before surgery and screening out the high risk population in thoracic OPLL circumferential decompression was helpful to warn the operator and inform patients of the risk in advance.
At present, there are few reports of exploring predictors of CSFL after thoracic OPLL circumferential decompression, and there are no universally-acknowledged conclusions. Sun [] found that patients with more than 3-segment laminectomy (odds ratio = 2.4, P odds ratio = 0.01) have higher incidence of CSFL (odds ratio = 2.4, P < 0.01). In this study, although the number of removed vertebral plates in the CSFL group was signi cantly higher than in the non-CSFL group (5.9 VS 3.7, P = 0.01), there was no statistical difference in logistic regression analysis. This was related to differences in the etiology and surgical methods of the two studies: Sun studied the cases with OLF, the surgical method was posterior approach laminectomy and decompression, and the dural injury occurred only in the dorsal spinal cord. By contrast, in this study, the included case was OPLL, the surgical method was circumferential decompression. Dural injury occurred more often in the ventral spinal cord. As a result, the factors leading to CSFL were more complex. Nicholas [] found that patients with diabetes (OR = 2.3, P = 0.04) or smoking history (OR = 3.4, P = 0.02) were more likely to develop CSFL in the study of dural injury during lumbar surgery. The smoking history in this study is also a risk factor for the CSFL after thoracic OPLL circumferential decompression (OR = 30.1, P = 0.003), which was consistent with the study by Nicholas. It is widely believed that smoking can lead to thinning of the fascia and decreased toughness []. The dura mater, though different from the abdominal pelvic fascia, has a similar composition and structure. It can therefore partly explain why smoking increases the risk of CSFL. Hu [12] found that CSFL is more likely to occur in T5-7 circular decompression surgery than other segments. In our study, the probability of CSFL occurring in the upper thoracic segment (OR = 188.0, P = 0.002) and the middle thoracic segment (OR = 57.4, P = 0.005) was also signi cantly higher than that of the lower thoracic segment. This is mainly because the middle and upper thoracic segments were narrower and had limited surgery space, so they were likely to cause dural injury.
The shape of OPLL itself may also be the cause of postoperative CSLF. The sagittal and transverse diameters of OPLL were used to evaluate the occupying ratio and OPLL base ratio. It was found that the vertebral occupying ratio and OPLL base ratio in the CSFL group were higher than those in the non-CSFL group. The multi-factor regression analysis showed that the OPLL base ratio was a predictor of postoperative CSFL. The wider the OPLL base, the more likely postoperative CSFL (or = 1.3, P = 0.007) will happen. Du [] came to a similar conclusion in studying the risk factors of CSFL after anterior approach cervical OPLL decompression. Du collected the data of 90 patients who used anterior approach cervical decompression to treat cervical OPLL, and found that CSLF is more likely to happen during resection of wide-base OPLL (OR = 1.09, P = 0.012) than that of narrow-base OPLL. The reason was that the base at both head and tail of OPLL was cut off in advance in order to take out the OPLL ossi c block. As a result, the wide base of OPLL will increase the di culty of resection, cause more dural invasion, and develop CSLF. After cutting the base at both head and tail of OPLL, we used the "collapse method" to push the free OPLL ossi c block directly to the anterior vertebral gap without directly contacting the nal part of the OPLL ossi c block near the midline. Therefore, the OPLL sagittal diameter had little effect on the formation of CSFL, which was consistent with the results of regression analysis in this paper.
As far as I know, there is no literature on the predictive model of CSFL after thoracic circumferential decompression. This study attempted to establish a predictive model of CSFL probability after thoracic OPLL circumferential decompression. The Logistic regression analysis found that smoking history, segment of circumferential decompression and OPLL base were statistically signi cant, which can be used to predict the occurrence probability of postoperative CSFL. In general, the area under the ROC curve > 0.6 indicates that the model has good predictability. The ROC of the model was 0.955, indicating that the model can well predict the risk of CSFL after thoracic OPLL circumferential decompression. There are still some shortcomings in this study: (1) Because the focus of this study is to explore the predictive model of CSFL, patients were not followed up for long periods. So the long-term effects of CSFL on patients, such as pseudocyst formation, nerve function recovery and so on, require further follow-up studies; (2) The number of cases was limited, which was limited by the lower incidence of thoracic OPLL, and not all cases require circumferential decompression, so long-term accumulation of cases is required.

Conclusions
To sum up, the predictive model established in this study has a high predictive effect. When the patients with thoracic OPLL have smoking history or the segment of circumferential decompression is located in the upper or middle thoracic spine or the OPLL has a wide base, the surgeon should be highly alert to the possibility of postoperative CSFL and warn the patient before surgery. There was no con ict of interest by any of the authors including nancial and personal relationships with other people or organizations that inappropriately in uenced this study.
Authors' contributions ZJ collected, analyzed, and interpreted the data and wrote the manuscript. WBT, CZQ performed the surgery, designed the protocol, revised the manuscript. All the authors have read and approved the nal manuscript.

No funding
Availability of data and materials The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

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 Human Ethics Committee of Peking University International Hospital. Written informed consent was obtained from each participant.
Consent for publication Figure 1 Measurement methods for the occupying ratio and OPLL base ratio. A: The occupying ratio was de ned as the OPLL thickness divided by the front and rear diameter of the spinal canal, which can be measured on the CT scan (a/b*100%). B: The OPLL base ratio was de ned as the OPLL width divided by the transverse diameter of the spinal canal on the axial CT scan (c/d*100%)  A 60 years old man presented with progressive exacerbation of lower limb weakness and walking instability for 6 months. Pre-operative JOA score was 3. A: The CT and MRI scans both showed continuous OPLL from T1 to T3 with the worst compression at T2-3. B: The patient underwent T1-3 laminectomy and selectively T2-3 circumferential decompression. Postoperative CT scans showed that