Introduction
Unruptured intracranial aneurysms (UIAs) are common, with an estimated prevalence of 1.0%–7.0%, and the increasing accuracy of imaging techniques has led to a rise in incidentally discovered asymptomatic UIAs [1,2]. Patients with these aneurysms are at risk of rupture, potentially leading to subarachnoid hemorrhage (SAH) and requiring timely intervention. Despite advances in endovascular coiling, the debate over the best treatment continues, with aneurysm clipping remaining a valuable option, especially for young patients with aneurysms in the middle cerebral artery (MCA) [3].
The neurological complication rate from clipping UIAs ranges 4.0%–10.8% that is significant given the potential for rapid progression and life-threatening outcomes [4,5]. These complications notably affect postoperative quality of life and hospital costs [6,7]. However, the risk of complications in the postoperative period is poorly understood.
Previous studies have explored postoperative risks, including scoring systems developed at Massachusetts General Hospital [8,9]. These incorporate SAH severity and mental condition (Hunt and Hess grades); however, their uniform application is challenging, especially for asymptomatic individuals’ UIAs identified through screening. Although these scoring systems are easy to apply, they are based on outdated studies and may not accurately predict risks. Recent studies suggest that beyond aneurysm size and location, neurological complications are linked to underlying diseases, previous neurological conditions, and intraoperative factors such as operation time [10,11]. Therefore, we aimed to develop and validate a comprehensive risk stratification scoring system that considers patient-, aneurysm-, and operation-specific variables to predict neurological complications in patients undergoing UIA clipping surgery.
Materials and Methods
Study participants
This study was a retrospective study using a dataset from the electronic medical records of two tertiary academic centers (Asan Medical Center [AMC] and Samsung Medical Center [SMC]). The study was approved by the Institutional Review Board of each institution (AMC 2023-1122, SMC 2023-11-141), and written informed consent was waived for retrospective data analysis. This manuscript adheres to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines.
We extracted the development cohort of patients with UIAs who underwent saccular aneurysm clipping surgery between September 2018 and October 2023. The validation cohort was extracted during the same period. Exclusion criteria were as follows: patients < 18 years of age, with subarachnoid or intracranial hemorrhage, with other vascular malformations such as arteriovenous malformations or moyamoya disease, who underwent clipping surgery for aneurysms located in the posterior circulation, and those with incomplete data or missing laboratory values.
Surgical techniques
Clipping surgery was performed by neurosurgeons with more than 10 years of experience (two specialists at the development institution and three specialists at the validation institution). Both institutions are high-volume tertiary referral centers, each performing approximately 300–350 aneurysm surgeries per year. Additionally, both institutions have dedicated neuro-critical care units, ensuring a high standard of perioperative management. UIAs were managed surgically through microsurgical aneurysm neck clipping, with additional trapping performed with or without bypass surgery as deemed necessary based on each institution’s interdisciplinary consensus [12,13]. All surgeries were performed under neurophysiological monitoring such as motor evoked and somatosensory evoked potentials. Doppler ultrasonography and indocyanine green angiography were also used to verify the flow in the relevant arteries nearby if inadvertent occlusion or narrowing was suspected.
Immediate postoperative neurological assessments consisted of evaluations of mental status, motor function, and basic cranial nerve function conducted by neurosurgeons upon the patient’s arrival in the intensive care unit. At both institutions, computed tomography (CT) or CT angiography (CTA) was performed within 24 h after surgery and on postoperative day 3. Patients’ functional status was assessed with the Glasgow Outcome Scale [14] at discharge and with the modified Rankin Scale [15] one and six months postoperatively for the evaluation of clinical outcomes.
Data collection and variable definitions
The electronic medical record system of each institution was retrospectively reviewed to obtain information on patient-, aneurysm-, and operation-specific characteristics and neurological complications. Patient characteristics included sex, age, weight, height, and body mass index. Additionally, we collected data on the history of previous neurological diseases, including cerebrovascular diseases (such as transient ischemic attack and stroke), neurodegenerative disorders (such as Parkinson’s disease), epilepsy, and neuromuscular disorders. The presence of concomitant diseases was also recorded, including hypertension, diabetes mellitus, cardiovascular diseases (e.g., coronary artery disease, heart failure), and renal disease. Aneurysm-specific factors included location, size, and multiplicity. Aneurysm location was categorized into three groups: MCA; anterior cerebral artery (ACA), anterior communicating artery (ACoA); posterior communicating artery (PCoA), anterior choroidal artery (AChA), paraclinoid [16]. In the case of multiple aneurysms, the characteristics that posed more challenging surgical access were included in the analysis. Aneurysm size was defined as the maximum diameter of the saccular aneurysm dome or the maximum length of the aneurysm on the two-dimensional image of the catheter angiogram. If measurement by catheter angiography was not feasible, CTA was used. Based on the unruptured cerebral aneurysms in a Japanese cohort study [17], aneurysms were categorized by size based on the rates of rupture (< 5 mm, 5–10 mm, 10–25 mm, > 25 mm). Although giant aneurysms (> 25 mm) are typically considered a separate category, they were excluded from the prediction model due to their extremely low prevalence (0.1%) in our cohort. Instead, aneurysms ≥ 10 mm were grouped together to ensure sufficient sample size and statistical reliability of the model. Operation-specific factors included operation time, intraoperative crystalloid and mannitol infusion, red blood cell (RBC) transfusion, and urine output. RBC transfusions were administered at the discretion of the anesthesiologist in accordance with the American Association of Blood Banks guidelines [18,19], particularly when the plasma hemoglobin concentration was below 8 g/dl or massive intraoperative bleeding was anticipated.
The primary aim of this study was to develop a model for predicting postoperative neurological complications and to evaluate the performance of the developed model through external validation. Neurological complications were defined as any new neurological sign or symptom, any radiologic finding, or a worsening of a pre-existing neurological deficit documented in the patient’s medical record during their hospital stay or the outpatient follow-up period [20]. These complications mainly included postoperative delirium and covert ischemic and hemorrhagic stroke [21]. All surgical patients were assessed for delirium using the Confusion Assessment Method for the Intensive Care Unit (CAM-ICU) while receiving care in the ICU until they were stabilized [22]. Hemorrhagic or ischemic complications were defined as the development of any newly detected intracerebral hemorrhage or postoperative low-density lesions on CT and/or CTA, respectively. Symptomatic complications were defined as either a clear neurological change or mild neurological symptoms associated with appropriate radiological findings. Additionally, other neurological changes such as aphasia, agitation, hemiplegia, and seizures were encompassed.
Statistical analysis and model development
Continuous demographic and perioperative variables were summarized as means with standard deviations or medians with interquartile ranges. Categorical variables were represented as frequencies (percentages). Demographic and perioperative variables were evaluated with Student’s t test or Mann‒Whitney U test for continuous variables and chi-square test or Fisher’s exact test for categorical variables, as appropriate. All reported P-values were two-sided, and P < 0.05 was considered significant. All statistical analyses were performed using SAS version 9.4 (SAS Institute Inc.).
To develop and validate the risk score, patients were divided into development and validation sets according to the institution where the surgery and management were performed. In the development set, univariate logistic regression analysis was performed to evaluate the prognostic ability of each variable. Based on clinical relevance and statistical significance, the variables in the scoring system were selected among all variables by using multivariate logistic regression analysis that was conducted using backward elimination with 1000-fold bootstrap resampling [23]. Variables appearing more than 500 times during bootstrap resampling were included in the final scoring system. The bootstrapping method was used again to obtain bias-corrected regression coefficients to allocate points in the scoring system. Each risk point was rescaled to designate the point with an aneurysm size of 5–10 mm as 1 point (for example, the risk point for RBC transfusion was 2 that was rounded from 1.046/0.419 = 2.496). A reference risk factor profile was chosen by selecting a base category for each risk factor that was assigned 0 points in the scoring system. The total score was the weighted sum of those predictors, with weights defined as the rounded integer value of the quotient of regression coefficients divided by the regression coefficient of the reference predictor. The discrimination and calibration of the final multivariate logistic regression model in the development and validation sets were assessed by the C-statistic and the Hosmer–Lemeshow test, respectively. This dataset enabled the replication of the observed characteristics for predicting the risk of neurological complications.
Results
Study participants
Over a five-year period, 1640 patients were screened for eligibility from the development institution and 1321 patients from the validation institution. Of these, 114 were excluded based on the exclusion criteria. Hence, 1547 and 1300 patients were included in the development and validation sets, respectively, resulting in a total of 2847 patients being included in the final analysis, as illustrated in Fig. 1. Baseline characteristics of the study population are shown in Table 1 and Supplementary Table 1.
The incidences of neurological complications were 5.7%, 5.6%, and 5.7% in the development set, validation set, and overall, respectively. Patients with neurological complications were more likely to have a history of neurological disease (P < 0.001) and hypertension (P = 0.011), undergo longer operations (P < 0.001), receive larger volumes of crystalloid infusions (P < 0.001) and RBC transfusions (P < 0.001), and exhibit a higher neutrophil count (P = 0.049) and C-reactive protein levels (P = 0.010). They were less likely to receive mannitol infusions (P = 0.045) and to present with a multiplicity of aneurysms (P = 0.012). A significant difference was observed between the neurological complication and non-complication groups in terms of aneurysm location (P = 0.006) and aneurysm size (P = 0.001) (Supplementary Table 1).
Risk score development
The associations between clinical variables and neurological outcomes following aneurysm clipping surgery, as determined by univariate and multivariate regression analysis, are presented in Table 2. Using bootstrap resampling with backward elimination, five variables were selected for inclusion in the final prediction model: previous neurological disease, categorized aneurysm location and size, categorized operation time, and RBC transfusion (Supplementary Table 2). The prediction model was developed based on the bias-corrected regression coefficients as shown in Table 3. Risk points were determined by bias-corrected regression coefficients and reference values in each category. Additionally, operation durations ranging 120–180 min were assigned a score of 1 to denote clinical significance. Thus, a new prognostic scoring system, the NEURO score (‘NEURO’logical outcome unruptured aneurysm in clipping surgery) was derived.
Validation
In the development set, ACA and ACoA aneurysms were predominant, accounting for 58.6% of the cases, whereas MCA aneurysms were predominant in the validation set, constituting 56.9% of the cases (P < 0.001, Table 1). Additionally, patients in the development set showed significantly smaller sizes of aneurysms (P = 0.033) and a larger proportion of multiplicity (P < 0.001) compared to patients in the validation set.
Discrimination and calibration abilities of the development and validation groups are presented in Table 4. The C-statistic for the NEURO score was 0.720 (95% CI [0.667–0.776]) and 0.693 (95% CI [0.631–0.754]) in the development and validation sets, respectively. Table 5 and Supplementary Fig. 1 display the performance of the final point score and risk groups according to the NEURO score in both development and validation datasets. The proportion of neurological complications in the development group ranged from 1.43% in patients with a score of 0% to 100% among those with a score of 10. Similarly, in the validation group, the rate ranged from 2.14% in patients with a score of 0% to 33.3% in those with a score of 8. Based on the estimate of the probability for neurological complication rate, the NEURO score classified the patients into three groups: low (0–3), moderate (4–6), and high (7–10) risk groups. Each group has an estimated probability of occurrence of neurological complications of 2.82%, 10.64%, and 33.81%, respectively. In the development set, neurological complications occurred in 2.75% of patients in the low-risk group, compared to 17.39% of patients in the high-risk group. In the validation set, neurological complications occurred in 3.85% of patients in the low-risk group, compared to 22.22% of patients in the high-risk group.
Discussion
The main findings of our study are as follows: (1) The incidence of neurological complications following aneurysm clipping surgery is higher in UIAs with previous neurological disease, non-MCA aneurysms, larger sizes (> 5 mm), longer operation times (> 120 min), and RBC transfusions; (2) A simple and predictive prognostic model, the NEURO score, estimates the probability of developing neurological complications in patients with UIAs who have undergone clipping surgery. This risk assessment aids physicians, patients, and families by providing an objective estimate of the likelihood of developing a neurological complication in the postoperative period. The NEURO score also allows the stratification of patients into low-risk (0–3 points), moderate-risk (4–6 points), and high-risk (7–10 points) groups, with corresponding probabilities of 2.8%, 10.6%, and 33.8%, respectively. This stratification enhances the ability to predict postoperative complications, guiding decisions regarding the duration of ICU care or hospitalization and the level of monitoring needed, while providing valuable information for tailoring individualized treatment plans.
In our study of 2847 patients, neurological complications occurred at 5.7%, 5.6%, and 5.7% in the development and validation sets and overall, respectively, aligning with the 4%–10.9% rate reported in previous meta-analyses [4,24,25].
Previous studies have addressed scoring the risk of neurological complications after intracranial aneurysm clipping surgery. Khanna et al. [8] developed a simple grading system based on aneurysm size, location, and patient age. However, this scoring system was developed using a relatively small number of patients and relied on surgical techniques from 30 years ago that may not accurately reflect current practices. Ogilvy and Carter [9] proposed a five-point grading system including age, aneurysm size, severity of SAH, and clinical conditions from a single-center study, not considering surgical factors in the prognosis [9]. The UIA treatment score model developed by Etminan et al. [26] involved 69 international experts and was multidisciplinary in nature. However, this study was based on expert consensus that typically has a low level of evidence in the hierarchy of evidence-based medicine, rather than on clinical research, and was ultimately not designed as a predictive model for UIA complications. In contrast, our study is clinically significant as it is the latest research that comprehensively considers the preoperative patient’s key information and intraoperative factors. The enhanced predictive power of the NEURO score is due to the inclusion of additional risk factors like previous neurological disease, surgery time, and intraoperative RBC transfusions. This expanded scope allows for more accurate predictions of outcomes in neurological complications compared to earlier models. Furthermore, our scoring system is specifically applicable to postoperative neurological complication outcomes in clipping surgery for patients with UIAs, and we conducted an external validation using data from other institutions to confirm its predictive accuracy. Despite significant differences between the development and validation sets in baseline characteristics, intraoperative variables, and aneurysm location and size, the model demonstrated robust predictive power, increasing its applicability to other institutions.
The NEURO score consists of five factors (size and location of the aneurysm, history of neurological disease, operation time, and RBC transfusion) in three characteristics (patient-, aneurysm-, and operation-specific) independently associated with neurological complications through a multivariate logistic regression analysis. The size and location of aneurysms are established risk factors for adverse outcomes, identified as significant ones in our scoring system [8,26,27]. Larger aneurysms, associated with mass effects, small perforating vessels, wider necks, and increased intraluminal thrombosis [8], elevate rupture risks and operative complexities, thus increasing neurological complications [28]. In our study, aneurysms > 10 mm were associated with approximately twice the risk of poor neurological outcomes after surgery compared to aneurysms < 5 mm. Understanding the specific location of the aneurysm is also crucial for assessing surgical risks. MCA aneurysms generally result in better outcomes due to their superficial location, familiar surgical approach, ease of proximal control, and minimal perforator vessels. We also observed MCA aneurysms had a significantly lower morbidity rate than the aneurysms in other locations. Within patient characteristics, a history of neurological disease showed a strong association with neurological complications, aligning with previous observations [29,30]. This increased risk is likely attributed to the higher atherosclerotic burden or vessel injuries in the surgical field [31], commonly observed in these patients. Regarding intraoperative variables, our analysis found that categorized operation time (≤ 120, 120–180, > 180 min) and the RBC transfusion were significant factors. The operation times for the development and validation cohorts in our study were 150.6 ± 48.9 min and 134.1 ± 57.6 min, respectively, aligning with ranges reported in previous studies [10,32]. Prolonged operation time may serve as an indicator of various complicating factors [10]. Anticipating and sustaining heightened vigilance for postoperative complications is crucial, especially in cases where the operation is notably complex and time-intensive [10]. In a previous study of the association of anemia and RBC transfusion with outcomes in patients with UIAs undergoing clipping surgery, each factor was independently associated with increased perioperative complications, with RBC transfusion, in particular, more than doubling the odds [33]. Each factor is independently associated with the development of neurological complications; however, their integration into a scoring system significantly enhances its predictive power.
Our study has limitations. First, its retrospective nature restricted access to certain clinical variables, such as aneurysm morphology and life expectancy that are included in the UIA treatment score [26]. These findings may not be universally applicable to individual cases of clipping surgery for UIAs, as surgical difficulty varies. Second, the predictive power for neurological complications decreases with higher scores due to the smaller number of patients with high scores. An external validation set may not include a high-risk patient population, potentially limiting its predictive capability. Third, the NEURO score demonstrated an area under the receiver operating characteristic curve of 0.720 in the development set and 0.693 in the validation set, indicating moderate predictive performance. Although this level of accuracy may not be sufficient for precise prediction at the individual patient level, it remains clinically useful, particularly in the absence of standardized models for predicting postoperative neurological complications after UIA clipping surgery. Lastly, aneurysms located in the posterior circulation were primarily treated through endovascular procedures rather than clipping surgery and therefore were not included in our scoring system, posing a challenge for application to these cases and requiring further study.
In conclusion, predicting the likelihood of neurological complications after clipping surgery for UIAs is crucial for postoperative decision-making and complication prevention. Through our large-scale data analysis, we developed a predictive scale for neurological complications after clipping surgery using five factors including history of neurological disease, aneurysm size and location, RBC transfusion, and operation time, and performed external validation of the predictive scale.
PDF Links
PubReader
ePub Link
XML Download
Full text via DOI
Download Citation
Supplement1
Print
