Prediction of outcomes for symptomatic spinal cavernous malformation surgery: a multicenter prospective clinical study

Clinical outcome of spinal cavernous malformation (SCM) varies because of its unclear natural history, and reliable prognostic prediction model for SCM patients is limited. The aim of the present study was to investigate potential factors that predict one-year neurological status in postoperative patients with SCM. This was a multicenter prospective observational study in consecutive patients with SCMs. SCMs treated microsurgically between January 2015 and January 2021 were included. Outcome was defined as the American Spinal Injury Association Impairment Scale (AIS) grade at one year after operation. Multivariable analyses were used to construct the best predictive model for patient outcomes. We identified 268 eligible SCM patients. Neurological outcome had worsened from preoperative baseline in 51 patients (19.0%) at one year. In the multivariable logistic regression, the best predictive model for unfavorable outcome included symptom duration ≥ 26 months (95% CI 2.80–16.96, P < 0.001), size ≤ 5 mm (95% CI 1.43–13.50, P = 0.010), complete intramedullary (95% CI 1.69–8.14, P = 0.001), subarachnoid hemorrhage (95% CI 2.92–12.57, P < 0.001), AIS B (95% CI 1.91–40.93, P = 0.005) and AIS C (95% CI 1.12–14.54, P = 0.033). Admission size of the lesion, morphology, symptom duration, AIS grade and the presence of subarachnoid hemorrhage were strong outcome predictors regarding prognostication of neurological outcome in postoperative patients with SCMs. A decision to surgically remove a symptomatic SCM should be justified by systematic analysis of all factors potentially affecting outcome.


Introduction
Spinal cavernous malformations (SCMs) belong to the category of central nervous system vascular malformations. They are low flow, which are made up of endotheliumlined chambers, accounting for 5-12% of all spinal vascular malformations [1][2][3][4][5][6]. About 40% of patients with SCMs also may have corresponding intracranial lesions [6,7]. The annual rate of hemorrhage in SCM patients has been estimated to be 1.4-6.8% per year [3][4][5][6][7][8][9]. SCMs were once thought to be rare entities of spinal cord vascular malformation. Nevertheless, more sizable series of these relatively lesions have been reported since the extensive application of MRI in routine clinical practice, especially the rising number of incidental findings has led to a significantly increasing incidence of SCM [10][11][12][13][14][15][16]. However, the opportunity and indications for operation in patients with SCMs are still controversial [9][10][11][12][13][14][15][16][17]. Several studies believe that SCMs have the potential to result in more aggressive behavior compared with their intracranial counterpart because of the low tolerance for space-occupying lesions in the spinal cord [9][10][11][12][13]. For symptomatic patients, early surgery should be performed before severe or long-lasting neurological deficits develop. On the other hand, others believe that patients harboring symptomatic SCM did not have significant, permanent neurological decline when treated with the conservative approach of observation [14][15][16][17]. SCM exists as a fairly benign lesion, often remaining clinically silent for life. Thus, because the surgery for them is straightforward and relatively simple, the most crucial task for successful management of SCM is appropriate patient selection and surgical timing. However, at what time point mild symptoms will progress was entirely based on personal clinical and surgical experience. As previously reported, although there have been a few studies on the prognostic factors of SCMs, the results obtained until now have been contradictory, and existing evidence in terms of treatment is limited mostly to retrospective studies [18][19][20][21][22]. There is continuous and increasing pressure to identify patient-specific prognostic factors for outcome that would help guide immediate surgical management. Accurate prognostic information may enhance our ability to predict outcomes. It has the potential utility of providing patients and family members with reasonable and objective expectations and to guide clinicians in the choice of the optimal therapeutic modality.
Further understanding of treatment specific patient features which correlate with outcomes will aid treatment modality selection. Therefore, we intended to investigate in a multicenter prospective cohort of symptomatic SCM individuals the prognostic value of several parameters for postoperative neurological status.

Patients and baseline procedures
This multicenter prospective cohort study was performed in four hospitals. We elected to include consecutive patients who presented with symptomatic SCM and underwent microsurgical treatment between January 2015 and January 2021. Surgical resection was endorsed when patients experienced acute onset of hemorrhage or debilitating symptoms. Diagnosis of SCM was based on pathological evidence. Magnetic resonance imaging (MRI) of brain and spinal cord was performed in all cases preoperatively and postoperatively. The exclusion criteria were neurological dysfunction caused by other diseases during follow-up. In addition to the medical history and baseline characteristics, the following information was recorded: (1) lesion size: according to the largest craniocaudal measurement on MRI. Perilesional hemosiderin rim was not included in size measurements; (2) extent of hemosiderin deposition; (3) location; (4) the presence of cerebral cavernous malformation; (5) initial symptom; (6) the presence of subarachnoid hemorrhage (SAH): lumbar puncture was performed in all patients before operation.
The following definitions were based on the recommendations previous literature: (1) morphology of lesion: complete intramedullary or exophytically growing lesion.
A lesion was classified with exophytic growth if it abutted the pial surface, as observed on MRI and intraoperatively; (2) symptom duration: the time from onset of symptoms to surgery; (3) clinical course: patients prior to presentation, classification according to the description from Ogilvy et al. [23] : Type A: a pattern of discrete and acute episodes of neurological deterioration with varying degrees of recovery between episodes, Type B: slow progression of neurological decline, Type C: acute onset of symptoms with rapid decline, Type D: acute onset of mild symptoms with subsequent gradual decline lasting weeks to months, Type E: the sudden onset of back pain, asymptomatic incidentally detected lesions. The functional status of patients was assessed with the American Spinal Injury Association (ASIA) Impairment Scale (AIS) grade [24].
The research protocol and consent procedures were approved by the ethics committees of all collaborating hospitals.

Follow-up and outcomes assessment
The patients were followed-up regularly at least 1 times a year, depending on the severity of their clinical condition, and the collection of data record forms was completed by January 1, 2022. The primary neurological outcome measure was the AIS grade at initial presentation and 12 months from surgery. Neurologic outcomes were classified as clinically improved, unchanged, or worse against preoperative baseline. Favorable outcome on follow-up was defined as improved or unchanged 12 months after operation.

Surgical treatment
Surgical access was obtained through a conventional laminectomy. A small dorsal midline or dorsolateral myelotomies was performed at the point where the SCM most closely approached the pial surface, typically evidenced by focal discoloration or hemosiderin deposition, and complete intramedullary lesions were located by ultrasonography. At the lesion site, a gliotic plane between the cavernous malformation and spinal cord usually facilitates removal. The dura was then closed in a water-tight fashion, and the lamina was reattached. All surgical procedures were performed under standard microsurgical conditions with intraoperative monitoring of somatosensory-evoked and motor-evoked potentials to minimize the risk of intraoperative neurological injury.

Statistical analysis
All statistical analyses were performed using the statistical software SPSS v20.0 (SPSS Inc, Chicago, IL). Baseline characteristics were tabulated as mean and standard 1 3 deviation (SD) in the case of continuous variables and as absolute numbers and percentages for categorical variables. The Kolmogorov-Smirnov test was used to analyze the distribution of variables. Candidate prognostic variables for outcome of interest included those already known as predictive from previous studies and several other variables we considered clinically relevant. Continuous variables were subjected to the Mann-Whitney U test while categorical variables were subjected to Fisher exact or Pearson chi-square test analysis. The cutoff points periodontal variables used in the logistic regression analysis were determined by means of sensitivity and specificity tests. Those within the most appropriate area under the receiver operating characteristic curve (AUC) were adopted (higher values of specificity and sensitivity). The risk factors that determined postoperative neurological function were investigated by univariate and multivariate analyses. Those variables that presented statistically significant at P < 0.2 in univariate analysis and those considered by the researchers to be of clinical interest were included as independent variables in the multivariable logistic regression model. In the case of elevated collinearity between two variables (Spearman or Pearson correlation coefficient > 0.6), the variable with greater clinical significance was chosen, based on the judgment of the authors. Subsequently, the following initial variables were selected: symptom duration, morphology, extent of hemosiderin deposition, size, clinical course, AIS grade and subarachnoid hemorrhage. Multivariable logistic regression analysis was performed to identify the model with the best predictors for unfavorable outcome. The forward stepwise technique (Wald test) was used, and variables with a P greater than 0.1 were removed from the final model. This left only symptom duration, morphology, size, AIS grade and subarachnoid hemorrhage in the fully adjusted model. Odds ratios (HR) and 95% confidence intervals (95% CI) for the independent variables were also calculated. All analyses were 2-tailed, and P < 0.05 was considered to indicate statistical significance. The predictive accuracy of the model was assessed by AUC.

Clinical characteristics
A total of 270 patients with SCMs were analyzed; two were excluded from the study because of neurological dysfunction caused by cerebral infarction during follow-up. Of the 268 patients remaining in the study, 127 male (47.4%) and 141 female (52.6%) were included in the analysis. The mean age at presentation was 48.3 years (range 24-72 years). The mean duration of clinical symptoms from onset to initial hospital visit was 23.3 months. The duration of symptoms varied for each category of Ogilvy type: 29.3 mo for patients with Type A, 43.4 mo for patients with Type B, 3.5 mo for patients with Type C, 17.7 mo for patients with Type D, and 49.5 mo for patients with Type E. The symptom duration was dichotomized with a cutoff of 26 months (AUC = 0.793). Average lesion size was 10.5 mm (range 3-26 mm). SCMs occur most often in the thoracic spine (56.3%). Among symptomatic patients, the most common symptoms were motor dysfunction (79.1%) and sensory deficit (67.2%), followed by pain (19.4%). Cerebral cavernous malformations coexisted in 36.2% of all patients. Neurological outcome had worsened from preoperative baseline in 51 patients (19.0%) at 1 year. No significant difference in outcome was observed between cervical and thoracic-lumbar cavernous malformation. 167 patients (62.3%) showed significant recovery of motor dysfunction at the last follow-up as compared with their preoperative state. 52 (19.4%) patients suffered from radicular pain before surgery. During follow-up, 46 (88.5%) of these 52 patients showed significant pain relief. Bladder/bowel dysfunction was registered in 48 patients (17.9%) before surgery. 11 (22.9%) patients experienced significant improvement of sphincter function after surgery. 6 (12.5%) patients suffered from urinary incontinence, increased residual urine volume, and loss of voluntary emptying of the bladder, requiring regular bladder autocatheterization daily. The baseline and clinical characteristics of the cohort overall are shown in Table 1. All lesions were totally removed under the microscope. There were no instances of cerebrospinal fluid leak or postoperative bleeding, and there were no operative deaths. Table 2 shows the differential characteristics of the group of patients who with favorable outcome (n = 217) during the follow-up and of those who with unfavorable outcome (n = 51). There was collinearity between motor function and AIS grade. Considering that AIS grade can better stratify motor impairment, and AIS grade was retained. Univariate analysis revealed that preoperative symptom duration ≥ 26 months (P = 0.03), complete intramedullary (P = 0.001), lesion size ≤ 5 mm (P = 0.01), AIS B and C (P = 0.04 and P = 0.02, respectively) and subarachnoid hemorrhage (P = 0.03) were associated with worse outcome. Only 8.2% (n = 22) of patients had severe neurological deficit with AIS A and B. SAH was found in 61 (22.8%) patients. Favorable outcome was achieved in 69.8% of the patients without SAH, compared with 15.3% of patients with SAH. Table 3 shows the unadjusted and fully adjusted logistic regression analysis. Symptom duration ≥ 26 months (adjusted OR = 6.89, 95% CI 2.80-16.96, P < 0.001), 1 3 size ≤ 5 mm (adjusted OR = 4.40, 95% CI 1.43-13.50, P = 0.010), complete intramedullary (adjusted OR = 3.71, 95% CI 1.69-8.14, P = 0.001), subarachnoid hemorrhage (adjusted OR = 7.56, 95% CI 2.92-12.57, P < 0.001), AIS B (adjusted OR = 8.85, 95% CI 1.91-40.93, P = 0.005) and AIS C (adjusted OR = 4.03, 95% CI 1.12-14.54, P = 0.033) showed an independent adverse prognostic value in the unadjusted and fully adjusted model. Using a sensitivity analysis, the inclusion of other covariables did not improve the prognostic capacity of the fully adjusted model. The fully adjusted model demonstrated good discrimination, as determined by the AUC of 0.83 (95% CI 0.77 − 0.89).

Discussion
In recent years, the optimal operative timing for a symptomatic SCM is still controversial. The decision to undertake surgical resection of any SCMs lesion should hinges on the underlying risk of neurological prognosis. Our study demonstrated that major predictors of outcome in patients with SCM have been identified as symptom duration, morphology of lesion, lesion size, AIS grade and SAH at presentation. Clinician requires an accurate and useful risk stratification system to advise patients and guide resection in this context.

Hemorrhage and SAH
SCM hemorrhage is frequently accompanied by neurological deficits. Some have argued that repeated hemorrhage is an absolute indication for surgery [25][26][27]. However, they have all had at least one prior hemorrhage when they are initially recognized because it is hemoglobin that produce the features that permit their diagnosis via MRI, and we have very little information on the future risk of hemorrhage in patients with a previously recognized SCM with a new hemorrhage. Increase in hemosiderin deposition around the lesion typically represents the hemorrhage transitioning from an acute to chronic phase, and the extent of hemosiderin deposition is related to the number of hemorrhage. Thus, we attempted to explore the relationship between the extent of hemosiderin deposition and prognosis, but unfortunately, our study showed it was not associated with an unfavorable outcome in patients with surgical resection of lesion. At the technical level, the general view is that if SCM with signs of acute hemorrhage on MRI, the microsurgical removal should be delayed for 2-6 weeks after neurological decline to allow for a gliotic plane to develop around the lesion, thereby facilitating safe removal [28,29]. Because of edema and spinal cord vulnerability due to hemorrhage may lead to postoperative neurological deterioration, immediate operation of these lesions could be harmful for an already stressed spinal cord.
SAH was strongly associated with worse outcome. Presence of preoperative pain symptoms was suggested to be predictive of poor postoperative neurological recovery [22]. SAH is a special manifestation of SCM hemorrhage. It is caused by the hemorrhage of SCM breaking into the central canal or subarachnoid space. At this time, the hemodynamics of SCM will be unstable, sometimes radiating, back or neck pain, followed by signs and symptoms of rapidly evolving nerve root or spinal cord compression. Based on the results of our data analysis, SAH was observed in 61 of 268 lesions (22.8%), most cases present with continually or stepwise progressive neurological deficits (32.8%). The risk of worse outcome in surgically treated patients with SAH was 7.56 times greater than that of patients without SAH. It shows that the presence of SAH is one of the key risk factors for worse outcome. Accordingly, prompt surgical removal of the SCM with SAH provides an excellent outcome.

Symptom duration and clinical symptoms
Nonhemorrhagic causes of neurological deterioration were most likely caused by mass effect secondary to gliosis or minor bleeding episodes that gradually increased the size of the lesion. At present, it is generally accepted that patients have recurrent or progressive neurological deficits should be recommended for resection, and observation is more selected for patients with mildly symptomatic SCM [8,16,21,30]. However, there is no consensus on the concept of "a limited neurological deficit or mildly symptomatic patient," which is mostly personal experience. In our study, combining symptom duration and AIS grade allows stratification of unfavorable outcome risk in SCMs. According to the postoperative outcome, our data indicate that the negative impact of symptom duration on surgical prognosis increased significantly after 26 months. Of the patients with symptoms duration 26 months or longer, 27.3% achieved unfavorable outcome, compared with 12.2% of patients shorter than 26 months. Similarly, according to the reported review [12,17,18], the mean preoperative symptom duration of fewer than 29 months, only 6-11% of patients were reported with deteriorated condition. This indicates that the time window of surgical intervention for patients with symptomatic SCM can be extended to 26 months. Furthermore, unlike previously reported predictors of neurological outcomes after SCM surgery [31], our study demonstrated that the prognosis of patients was not related to the clinical course. There was no difference in final neurologic outcomes between patients with acute or more chronic clinical presentations of patients. On the other hand, patients with AIS C and lower preoperatively mainly achieved an unfavorable outcome after surgery. It was dichotomized into mildly (grade D and E) and severe (grade A, B or C) symptoms as this provided a clinical distinction between patients who were independent in their activities of daily living versus those who were not. This finding is also consistent with some series reported that greater severity of preoperative neurological impairment to be associated with unfavorable outcome [15]. This observation clearly shows that recovery does not mainly depend on a critical timeframe between onset of symptoms and surgical removal, but more on the extent of neurologic symptoms at onset. Therefore, based on the above conclusions, we believe that the "a limited neurological deficit or mildly symptomatic patient" of SCM can be defined as neurological function above grade C and lasting within 26 months.

Size and morphology
Lesion size as the main indication for surgical intervention remains a significant risk factor for unfavorable outcome. However, the risk of adverse surgical outcome did not increase with the lesion enlarging; this was most prominent in lesions ≤ 5 mm in size and in complete intramedullary lesions. The correlation between lesion size ≤ 5 mm in greatest dimension and unfavorable outcome compared with preoperative clinical status showed a statistical significance. This finding is not, however, congruent with other studies showing a no correlation of outcome with preoperative lesion size [13,28]. In contrast, the risk of unfavorable outcome of resection of lesion size ≤ 5 mm was 4.4 times higher than that of lesion size ≥ 10 mm. Furthermore, it is likely that complete intramedullary lesions are considerably more surgical risks than exophytically growth lesions of a similar size. We regard the surgical exploration may easily aggravate deficiencies due to a lesion, consequently, encroachment of the lesion more anteriorly toward the motor tracts, removal of a small lesion completely located in the spinal cord, would be associated with very high surgical morbidity, as opposed to a lesser degree of parenchymal manipulation required for resection of larger lesions rising to the pial surface.

Limitations
This study has some limitations that shall be noticed. First, it is a prospective observational cohort, as with any cohort study, residual confounding due to unmeasured or unknown factors cannot be ruled out. Second, the question of how SCMs develop and progress is still not well understood. Long-term conservative treatment is still an effective method to understand the natural history of SCMs. However, due to the existence of the concept of preventive surgery, there are few cases of conservative treatment of SCMs. By clarifying the factors of surgical prognosis, this study can evaluate the timing of surgery, so as to provide basis for conservative treatment.

Conclusions
This prospective cohort study demonstrated that admission size of the lesion, morphology, symptom duration, AIS grade and the presence of subarachnoid hemorrhage were strong outcome predictors regarding prognostication of neurological outcome in postoperative patients with SCMs. A decision to surgically remove a symptomatic SCM should be justified by systematic analysis of all factors potentially affecting outcome. In these instances, the role of subjective aspects regarding optimal treatment strategy should be weakened.