Patient Population

The health records of patients admitted between June 2002 and 2011 with subarachnoid hemorrhage (SAH) and an intracranial aneurysm were retrospectively reviewed at the Ottawa Hospital, a tertiary care institution and regional neurosurgical center. SAH was diagnosed by CT or CSF analysis demonstrating > 5 × 10⁶/L red blood cells in the final test tube or xanthochromia up to 2 weeks after suspected aneurysmal rupture.^20,21 Aneurysms were confirmed by CTA, MRA, or conventional digital subtraction angiography (DSA). Only patients older than 18 years presenting within the risk period for cerebral vasospasm were included. Death prior to day 4 post-aSAH, admission past day 14 post-aSAH, DCI at presentation, and aneurysms associated with arteriovenous malformations or of the mycotic or traumatic type were criteria for exclusion. Recruitment was completed using a validated text miner that searched radiology reports for keywords indicating aSAH and CSF analysis results within the Ottawa Hospital Data Warehouse.¹² Search results were verified individually to ensure accuracy and that the inclusion and exclusion criteria were satisfied.

Predictor Variables

Data were collected for each patient related to their comorbidities, presentation, and in-hospital course. The clinical variables included were age, sex, cardiovascular risk factors, previous intracranial hemorrhage, smoking status, illicit drug use, medications, family history of aneurysms or aSAH, presenting neurological grade, systolic blood pressure, body temperature, and unit of admission. Age was analyzed as a continuous variable and a categorical variable by decades, with ages 18–39 years and 70 years or older collapsed into single groups due to small numbers. Blood pressure and temperature were classified as continuous variables and dichotomized variables at 180 mm Hg and 38°C, respectively. Neurological grade was assessed using the World Federation of Neurosurgical Societies (WFNS) scale.

Radiological variables considered were modified Fisher grade¹³ (classified by a single investigator reviewing all presenting CT head imaging), ruptured aneurysm location and maximum diameter, aneurysm multiplicity, hydrocephalus, and mean flow velocities recorded by transcranial Doppler (TCD) ultrasonography on days 2–14 post-aSAH. TCD results were summarized as elevated (mean flow velocity > 120 cm/sec and Lindegaard ratio > 3), normal, or not performed. Management-related variables included the modality of aneurysm treatment; use of nimodipine, prophylactic statin, antiepileptic drugs, or nicotine replacement therapy; and required external ventricular drain (EVD) placement or endotracheal intubation within 24 hours of presentation or prior to day 4 post-aSAH. Complications entered as present or absent on day 3 post-aSAH or earlier included seizures, hyponatremia, anemia, pulmonary edema, acute coronary syndrome, arrhythmia, and heart failure. The cutoffs used for hyponatremia and anemia were a serum sodium level < 135 mmol/L and hemoglobin level ≤ 100 g/L, respectively. Pulmonary edema was verified using chest radiographs. Acute coronary syndrome was considered present if clinical symptoms (chest pain, left arm pain, diaphoresis) or electrocardiogram changes accompanied elevated cardiac enzymes. The arrhythmias of interest were ventricular tachycardia, ventricular fibrillation, and supraventricular tachycardia. An echocardiogram demonstrating an ejection fraction less than 40% was required for a diagnosis of systolic heart failure.

Definition and Assessment of DCI

DCI was defined as a focal neurological deficit attributable to a detected vascular territory of intracranial arterial narrowing (angiographic vasospasm) in the absence of alternative causes. The deficit could not be present immediately after aneurysm treatment and included hemiparesis, hemiparesthesia, aphasia, apraxia, neglect, hemianopia, and decreased level of consciousness equaling a loss of ≥ 2 points on the Glasgow Coma Scale.²⁹ Changes in neurological status were determined from chart review, and arterial narrowing was diagnosed from radiology reports documenting vasospasm of all severities on CT, MR, or DSA, or elevation in TCD velocities meeting the criteria for vasospasm. Outcome assessment was divided among 3 investigators blinded to the identity of the predictor variables. Uncertainty about the outcome was adjudicated by a fourth investigator. Interrater reliability was determined based on a pilot cohort of 25 patients using the (2,1) form of the Shrout-Fleiss intraclass correlation coefficient with each rater considered a random effect.²⁶ The results from this initial evaluation period were included in subsequent analysis.

Logistic Regression

The associations between the chosen predictor variables and DCI were characterized by univariate analysis using the chi-square or Fisher exact test and 2-sided t-test for categorical and continuous variables, respectively. Only those variables available at the time of presentation up to day 3 post-aSAH were considered. Candidate variables meeting a conservative p value < 0.20 underwent multivariate logistic regression using stepwise backward selection, and those reaching statistical significance (p < 0.05) were retained. Odds ratios with corresponding 95% confidence intervals were calculated describing the contribution of each predictor to the occurrence of DCI. Model fit was assessed using the Hosmer-Lemeshow test and its discrimination determined by receiver operating characteristic (ROC) curve analysis. The area under the ROC curve of the derived model was compared to the performance of the modified Fisher grade alone. All analyses were performed using SAS software (version 9.4, SAS Institute Inc.).

Using accepted methods, a risk score was developed based on the estimated beta coefficients from the final multivariate logistic regression model.²⁸ The smallest coefficient was assigned as the constant for the point system representing a point of 1. Values for the remaining risk factors were assigned by dividing the associated beta coefficient by the point system constant and rounding to the nearest 0.5. The estimated risk of DCI associated with each point total was calculated using the methods described by Sullivan et al. and the corresponding sensitivity and specificity for each score cutoff were determined.²⁸

Interrater Reliability

The percentage agreement among the 3 outcome assessors for the pilot aSAH cohort was 85.3%. An identical diagnosis by all assessors occurred in 20 of 25 cases. The Shrout-Fleiss intraclass correlation coefficient was 0.84 (95% CI 0.71–0.92), indicating good interrater reliability in identifying DCI.

Logistic Regression

A total of 14 variables were significantly associated with DCI at p < 0.20 (Table 1). These included age, history of connective tissue disorder, home statin use, WFNS grade, ICU admission, modified Fisher grade, aneurysm location, aneurysm diameter, early elevated TCD velocities, modality of aneurysm treatment, statin prophylaxis, EVD placement, seizures prior to admission, and early intubation. TCD ultrasonography results did not undergo further analysis, as there was a great amount of missing information, with 303 patients who did not undergo this examination.

After fitting the variables significant from univariate analysis into an initial multivariable model, only age, modified Fisher grade, and aneurysm location remained predictive of DCI (Table 2). For age, being younger than 40 years and 60–69 years did not reach significance when compared to age ≥ 70 years. Similarly, modified Fisher grade 2 in relation to grade 1 had a p value > 0.05. The Hosmer-Lemeshow test of the multivariable model did not show evidence of poor fit (p = 0.82). Discrimination was acceptable with an area under the ROC curve of 0.73 (95% CI 0.67–0.78), greater than the modified Fisher grade alone (0.66 [95% CI 0.60–0.71]).

DCI Score Derivation

The smallest estimated coefficient, 0.23 from the relationship between modified Fisher grade 2 and grade 1, was assigned a point equivalence of 1 and used as the score constant from which the point values for the remaining covariate categories were designated (Table 2). This resulted in a score containing 3 clinicoradiographic variables totaling 0–17 points and associated with an estimated risk of DCI of 1%–38% (Table 3). Figure 1 illustrates the sensitivities and specificities for all possible score cutoffs.

Sensitivity and specificity of the delayed cerebral ischemia score for all point cutoffs. Figure is available in color online only.

Factors Predictive of DCI

We derived a clinical decision rule to predict DCI using 3 predictors: age, amount of SAH on presentation, and ruptured aneurysm location. This decision rule achieves improved estimation of an individual’s DCI risk beyond the modified Fisher grade alone. A recently proposed risk chart identified age dichotomized at 55 years as an important prognostic factor.⁷ We characterize this risk between decades in a contemporary study population where endovascular coiling is increasingly prevalent. Dumont and colleagues’ artificial neural network also included age; however, the variable “age” was entered with presumed importance and its individual contribution is difficult to interpret.¹⁰ A 2013 systematic review concluded “inconsistent evidence” from studies supporting greater risk in younger patients,³ with increasing age,³⁰ or no difference with age.^8,23 Macdonald and colleagues’ description of DCI risk peaking for ages 40–59 years mirrors our findings.¹⁸ Senescent intracranial arteries become increasingly stiff with reduced elasticity from vessel wall thickening, increased thickness-to-radius ratio, and collagen fiber formation.^15,16 This, along with atherosclerosis, may render elderly patients more resistant to vasospasm with the ability to maintain luminal patency in the setting of restricted CBF. Why younger patients experienced lower rates of DCI is uncertain but perhaps reflects an increased resilience to ischemia and preservation of autoregulation. This is suggested by the loss of association between the reversibility of neurological deterioration secondary to angiographic vasospasm and severity of vessel narrowing with increasing age where cardiorespiratory comorbidities are increasingly prevalent.³¹

The amount of initial SAH has a robust association to DCI regardless of its classification.^{2,10,14,17,23} The modified Fisher grade, which we demonstrated to independently increase susceptibility to DCI with higher grade, has only been studied in other models when collapsed into thin or thick SAH despite improved prognostication by incorporating IVH status.^4,6,7,13 De Rooij and colleagues’ risk chart recognized the contribution of IVH, quantifying it using the Hijdra score.⁷ This score’s complexity comprises overall feasibility and is likely the reason it was removed in the VASOGRADE, a subsequent external validation study.⁶ A multivariate analysis of 218 aSAH patients demonstrated the modified Fisher grade to be a significant risk factor for DCI with an OR of 2.99 (95% CI 2.05–4.35).³⁰ We did not find a difference between modified Fisher grades 1 and 2, representing the absence or presence of IVH accompanying thin SAH. Only 8 cases of grade 2 aSAH were associated with DCI, resulting in insufficient power. EVD placement prior to day 4 post-SAH did significantly alter the influence of having a modified Fisher grade 4 versus grade 3 aSAH when included in the multivariable model such that the grades’ odds ratios were practically equivalent, negating the importance of IVH. If true, the deleterious effects of hemorrhagic products in CSF can be minimized by enhanced clearance through CSF drainage. Omission of EVD placement from the final prediction rule stems from the possibility that the modified Fisher grade influences EVD insertion, allowing a change of a patient’s risk score in real time, if included.

Ruptured posterior circulation aneurysms were less likely to precipitate DCI than anterior circulation aneurysms. Qureshi and colleagues’ risk index also identified anterior cerebral and internal carotid artery aneurysms as high-risk features.²³ The relationship between aneurysm location and DCI has previously been investigated. Posterior cerebral artery aneurysms were protective against DCI with an OR of 0.05 (95% CI 0.01–0.41).¹⁹ Abla et al. determined that rupture of vertebral and basilar artery aneurysms resulted in the second and third lowest mean SAH maximum thickness behind pericallosal artery aneurysms.¹ While this association with SAH thickness could explain our observed difference in DCI incidence, it is doubtful, as the proportion of aSAHs classified as modified Fisher grades 3 and 4 was comparable between locations (anterior: 56.5%, posterior: 54.3%).

Incidence of DCI and Proposed Clinical Decision Rule

The DCI score estimates the risk to be 1.2% in a ≥ 70-year-old patient with thin to absent SAH, no IVH, and a posterior circulation aneurysm. A patient 40–59 years old with thick SAH, IVH, and an anterior circulation aneurysm is at highest risk, estimated at 38.4%. This range differs from Qureshi and colleagues’ point-based risk index due to differing incidence.²³ When all risk factors are present in both scores, the probability of developing DCI almost doubles.²³ Our observed rate falls in the lower end of the range (20% to 43.2%) from previously published prediction models whose outcome definitions varied.^{2,6,7,10,14,17,23,24} Reduced diagnosis of DCI may have resulted from the requirement for attributable angiographic vasospasm. However, the artificial neural network of Dumont et al. used an identical outcome definition, resulting in 39 cases (35%).¹⁰ The lowest incidence was observed in de Rooij and colleagues’ validation cohort—20%—where radiographic evidence of infarction was required.⁷ The incidence of DCI has decreased from 32.5% in studies published prior to 1994 to 28.5% in studies published between 1994 and 2009, during which aSAH management has changed, affecting complication rates and outcome.⁹ With calcium channel blocker prophylaxis (nimodipine or nicardipine) in 32 publications, this proportion declined to 22%. Nimodipine administration in this study occurred in all but 3 patients, explaining the lower incidence as well as suggesting efficacy of this prophylactic agent. A recent study investigating the role of aneurysm location on aSAH outcomes similarly found neurological deterioration with angiographic vasospasm in 22% of patients who all received nimodipine.¹ It is recognized that our definition of DCI potentially missed cases precipitated by mechanisms other than proximal arterial narrowing, including microvascular vasospasm, microthromboembolism, and cortical spreading ischemia.

The current practice in aSAH management is to monitor all patients as high risk for DCI, particularly until after the peak onset period for vasospasm of 6–8 days after aSAH. The VASOGRADE and de Rooij and colleagues’ risk chart identified predictors (age, initial neurological grade, thickness of SAH, and IVH) available at the time of presentation with the same aim as this study for early risk stratification to tailor patient monitoring, treatment, and disposition.^6,7 Classifying age into decades and the addition of aneurysm location in our model improved discriminative performance with a greater area under the ROC curve (0.73 [current study] vs 0.63 [de Oliveira Manoel et al.⁶] vs 0.63 [de Rooij et al.⁷]). In choosing a score cutoff, clinicians are more likely to accept over-diagnosing DCI, as missed cases can lead to serious morbidity or even mortality. A DCI score of ≤ 9 is proposed to represent a low-risk group having less than half the estimated risk (< 10%) of developing DCI and high sensitivity near or at 100% for scores lower than 7. Sensitivity approaches specificity at a score of 14, where the estimated risk (28.2%) begins to exceed the population incidence, suggesting scores ≥ 15 to be a high-risk group.

Study Limitations

The major limitation of this study is its retrospective nature, which may have led to diagnosis misclassification of DCI, failure to identify important comorbidities, and precluded standardization of CBF monitoring. While this could contribute to the lower incidence of DCI observed, it remains within the expected range when considering newer studies implementing routine nimodipine administration.⁹ The proportions for cardiovascular risk factors, hypertension and diabetes specifically, were also consistent with those of previous studies (hypertension: 25%–44%,¹⁸ diabetes: 3%–9%⁸).¹¹ Patients whose altered neurological status was successfully treated medically were potentially at risk of misclassification as not experiencing DCI, which could alter the observed association of certain predictors with DCI. This situation would be conceivably rare, attributable only to transient arterial narrowing or alternate etiologies to vasospasm. Lastly, a high percentage of our cohort presented with good neurological grade, known to be associated with reduced risk of developing DCI.^6,7,23 This skewed distribution, although similarly observed in other studies investigating presenting neurological grade as a risk factor, could have reduced power in our outcome analysis and may explain why WFNS grade did not remain significant in the multivariate logistic regression model.^5,6,8 Prior to clinical use, our proposed rule requires prospective evaluation in multiple institutions to capture variability in practice and aSAH populations.

Previous Presentations

Portions of this work were previously presented in poster form at the 2017 American Association of Neurological Surgeons Annual Scientific Meeting, Los Angeles, California, April 2017.