An Admission Bioclinical Score to Predict 1-Year Outcomes in Patients Undergoing Aneurysm Coiling
Background and Purpose—A number of scores were developed to predict outcomes after clipping for subarachnoid hemorrhages, yet there is no score for patients undergoing endovascular treatment. Our goal was to develop, compare, and validate a predictive score for 1-year outcomes in patients with coiled subarachnoid hemorrhage.
Methods—We studied 526 patients for 1 year after intensive care unit discharge. We developed an admission bioclinical score (ABC score), which integrated biomarkers such as troponin I and S100β, with the Glasgow Coma Scale. Using the receiver operating characteristic curve (95% CI), the ABC score was compared with the Glasgow Coma Scale, World Federation of Neurosurgical Societies score, and Fisher score in the derivation cohort and further validated in an independent cohort.
Results—In the derivation cohort (from 2003–2007, n=368), multivariate logistic regression analysis showed that only Glasgow Coma Scale (P<0.001), high S100β (P<0.001), and high troponin (P<0.02) were independently associated with 1-year mortality. Troponin, S100β, and Glasgow Coma Scale were thus integrated to derive the ABC score. In the derivation cohort, the ABC score reached an receiver operating characteristic curve of 0.82 (0.77–0.88, P<0.001) and was significantly greater than the receiver operating characteristic curves of the Glasgow Coma Scale, World Federation of Neurosurgical Societies, and Fisher scores for predicting 1-year mortality. In the validation cohort (from 2008–2009, n=158), the ABC score's receiver operating characteristic curve of 0.76 (0.67–0.86, P<0.001) remained superior to the 3 other scores for predicting 1-year mortality.
Conclusions—The ABC score improves 1-year outcome prediction at admission for patients with coiled subarachnoid hemorrhage. Our study provides large cohort-based evidence supporting integration of individual biomarkers and clinical characteristics to predict outcomes.
Patients with spontaneously ruptured subarachnoid hemorrhages (SAHs) continue to have a staggering mortality rate of 30% to 50%.1,2 The main treatment goal is to avoid a potentially devastating rebleeding of the aneurysm.3 The traditional approach has been surgical clipping, but because endovascular treatment has shown a 23.5% reduction in mortality and disability,4,5 this method is on the rise.6
Classifications for the prediction of SAH outcome are the Glasgow Coma Scale (GCS)7,8 and the World Federation of Neurosurgical Societies (WFNS) score, which combines GCS values with motor deficits.9,10 Another predictive tool is the Fisher score, which predicts the occurrence of vasospasm by evaluating the amount of blood present on the CT scan at admission.11 Finally, biological markers such as troponin I12–14 and S100β15–17 have also been associated with poor outcomes after SAH.
Complications of endovascular treatment are different from those of clipping procedures.3–5,18 However, the pre-existing scores have been designed for patients undergoing clipping procedures only,7,9 and the use of these scores for endovascular treatment is not well characterized. Therefore, a risk score specifically for patients with SAH receiving coiling that can be easily used in clinical practice is needed.
We hypothesized that a score combining clinical, radiological, and/or biological characteristics at admission would better predict 1-year outcomes in patients with coiled SAH than the previous scores designed for patients with clipped SAH. Thus, we sought to develop an admission score in a large prospective cohort study of patients with coiled SAH, compare it with the previous scores, and validate the new score in a second cohort.
Materials and Methods
We screened 752 and enrolled 526 consecutive adult patients who were admitted from January 1, 2003, to December 31, 2009, to the neurosurgical intensive care unit after a clinical diagnosis of SAH (Figure 1). Aneurysm SAH was angiographically confirmed and coiled with or without stents. Patients whose treating physician decided to either forego any invasive treatment (n=48) or perform open surgical clipping (n=168) were excluded. Open surgery was proposed when the anatomic structure of the aneurysm increased the risk of the coiling procedure or if a parenchymatous hematoma was present with a significant mass effect at admission. Ten of the 536 patients with coiled SAH were excluded for missing values. The 526 cases were chronologically split in the derivation cohort (n=368, 2003–2007) and the validation cohort (n=158, 2008–2009). The study was approved by the local Ethics Committee in accordance with the Declaration of Helsinki and registered with ClinicalTrials.gov.
We recorded the following clinical characteristics: age, sex, GCS, presence of motor deficit, presence of clinical seizure, and WFNS score (online-only Supplemental Table I, http://stroke.ahajournals.org).9 The GCS refers to the value at admission before any treatment with sedative drugs or treatment of hydrocephalus. We also recorded biological characteristics with routine sampling of S100β and troponin at admission (online-only Supplemental Methods). Admission S100β and troponin I levels were considered high for values >0.5 μg/L (corresponding to 5 times the maximum normal range) to be consistent with previous studies.13,17
Admission radiological characteristics were obtained from the admission CT scan, including hydrocephalus and intraventricular hemorrhage, and the initial angiogram, including aneurysm localization and lateral size. The CT scan was classified according to the modified Fisher score (online-only Supplemental Table II) by a neuroradiologist blinded to the patient's biomarker data.11
Intensive Care Unit Management
All patients with SAH were followed in the intensive care unit (for more detail, see online-only Supplemental Methods). Early (coiling complication, intraventricular hemorrhage, intracranial hypertension, hydrocephalus) and delayed (vasospasm) cerebral lesions were recorded.
The primary outcome, 1-year mortality, was systematically assessed using the Rankin outcome scale (online-only Supplemental Table III) at 1 year after intensive care unit discharge.19,20 The GCS, WFNS, and Fisher scores were compared to the admission bioclinical score (ABC score) for the prediction of 1-year mortality (Rankin 6), 1-year independent function (Rankin 0–3), and 1-year full recovery (Rankin 0–1).
Sensitivity analysis of the 1-year mortality outcomes was conducted by dividing the derivation cohort into 2 groups based on their coma status (coma: GCS <13 [WFNS 4 or 5] versus noncoma: GCS ≥13 [WFNS 1–3]).
To identify factors that were independently associated with 1-year mortality, multivariate models were evaluated using forward and backward stepwise logistic regression for: age, gender, seizure, GCS, Fisher score, intraventricular hemorrhage, hydrocephalus, high troponin I, and high S100β. The weights of the ABC score components were derived from the ORs of the most parsimonious model. We assessed calibration of the ABC score with the Hosmer-Lemeshow goodness-of-fit test.
We constructed receiver operating characteristic (ROC) curves and compared the areas under the ROC curves (AUC-ROC) of the ABC, WFNS, GCS, and Fisher scores to predict 1-year mortality, 1-year independent function (Rankin score 0–3), and 1-year full recovery (Rankin score 0–1). The integrated discrimination improvement, the net reclassification improvement, and risk stratification capacity were performed to compare the 4 scores for each end point (online-only Supplemental Methods).21–23 Sensitivity analyses of the AUC-ROC regarding coma status and early and delayed cerebral lesions were performed in the derivation cohort.
Data are expressed as a percentage (with 95% CIs) for binary variables, median (interquartiles) for discontinuous variables, and as a mean±SD for continuous variables. All tests are 2-sided and all probability values are provided uncorrected. Probability values <0.05 were considered significant. Statistical analyses were performed using STATA Version 11 (StataCorp).
The 1-year mortality of the 526 cases was 17.7% (95% CI, 14.4%–21.0%; Figure 1). Derivation and validation cohorts were similar in all main admission characteristics (Table 1). Furthermore, across the 7 years of the study, the proportions of 1-year mortality and 1-year full recovery were similar (P=0.92 and P=0.57, respectively) with no difference between the derivation and validation cohorts (Table 1).
ABC Score Development
Univariate analysis for 1-year mortality showed that age, gender, seizure, GCS, WFNS, troponin I, S100β, Fisher score, hydrocephalus, and intraventricular hemorrhage were associated with 1-year mortality (Table 2). In contrast, motor deficit without any consciousness deterioration (WFNS 3, P=0.66), aneurysmal localization (P=0.52), and lateral size (P=0.63) were not associated with 1-year mortality. The causes of both early and delayed cerebral lesions were associated with 1-year mortality (Table 2). Multivariate logistic regression followed by stepwise forward and backward stepwise analysis determined that only the GCS (P<0.001), high S100β (P<0.001), and high troponin I (P=0.02) were independently associated with 1-year mortality (Table 3). A simplified ABC score was constructed taking into account the ORs of the most parsimonious model. The ABC score combines the GCS and the admission troponin and S100β values into a single value between 0 and 6 (Table 4). The discriminative power of the ABC score was statistically significant (0.828; 0.772–0.885, P<0.001). Interestingly, the ROC-AUC of the ABC score was not statistically improved by the addition of age criteria (0.823; 0.768–0.878, P=0.42).
ABC Score Comparison
Comparisons of the ROC-AUCs showed that the GCS (0.79; 0.74–0.85) and WFNS scores (0.79; 0.73–0.84) were more powerful than the Fisher score (0.69; 0.62–0.76) for predicting 1-year mortality (P=0.002 and P=0.003, respectively).
The ROC-AUC of the ABC score was superior to the ROC-AUCs of the WFNS (P=0.001), GCS (P=0.03), and Fisher scores (P<0.001) for predicting 1-year mortality in the overall population (Figure 2A). The integrated discrimination improvement showed that the ABC score was superior to the 3 other scores (P<0.001; online-only Supplemental Figure I). The net reclassification improvement was statistically higher for the ABC score than for the 3 other scores with 2 different thresholds (P<0.05 for each comparison; online-only Supplemental Table IV). The risk stratification capacity was also superior for the ABC score (P<0.001). Moreover, in the coma subgroup, the ABC score was a significant predictor of 1-year mortality (0.70; 0.60–0.80, P<0.001) and superior to the WFNS (P=0.01), GCS (P=0.01), and Fisher scores (P=0.002; online-only Supplemental Figure II). With regard to the prediction of 1-year full recovery (Rankin 0–1), the ABC score's ROC-AUC of 0.83 (0.79–0.88) was greater than the ROC-AUC of the WFNS (P=0.004), GCS (P=0.007), and Fisher scores (P<0.001). Concerning 1-year independent function (Rankin 0–3), the ABC score's ROC-AUC was also greater than the ROC-AUC of the 3 other scores (online-only Supplemental Table V). Sensitivity analyses of the ABC score for early and delayed cerebral lesions showed that the ROC-AUC values for predicting 1-year mortality were statistically superior to the reference line (P<0.01 for each comparison; online-only Supplemental Table VI). Moreover, the ABC score presents a practical value with a greater range of probability than the 3 other scores to predict 1-year mortality, independent function, and full recovery (Table 4; Figure 2C–D).
ABC Score Validation
In the validation cohort (Table 1), the discriminative power of the ABC score to predict 1-year mortality was statistically significant (0.76; 0.67–0.86, P<0.001). The ROC-AUC of the ABC score was superior to the ROC-AUC of the WFNS (P=0.01), GCS (P=0.002), and Fisher scores (P=0.03) for predicting 1-year mortality (Figure 2B) and 1-year independent function (Rankin 0–3; online-only Supplemental Table VI).
The goal of our prospective study was to develop, compare, and validate a better admission score for predicting 1-year outcomes in patients with coiled SAH. This led to the ABC score, which combines clinical and biological measurements and which was superior to the 3 pre-existing scores for predicting 1-year outcomes in a validation cohort.
Usefulness of a Simplified Admission Score for Coiled SAH
Patients with SAH have a heterogeneous presentation at admission. Thus, an accurate predictive score is necessary for identifying the patient's risk of mortality and for tailoring therapy to the individual patient. In addition to assisting physicians, admission scores help patients and their relatives.24 Admission scores are also important in Phase III trials for discriminating among subpopulations.25
As a result of the International Subarachnoid Aneurysm Trial (ISAT),4 many neurosurgeons and neuroradiologists have decided to treat patients with SAH with coiling. However, we found no accurate score for predicting long-term outcomes in coiled patients.26–29
Relevance of ABC Score Components
In this study, we decided to compare the main clinical, biological, and radiological factors already known to be associated with long-term outcomes to generate a new score specifically dedicated to coiled SAH. We first observed in our derivation cohort that the 3 pre-existing scores had no discriminating power in the coma group (P>0.24 for each comparison). The GCS was strongly associated with 1-year mortality, although only for GCS values ≥13. Therefore, we decided to give all GCS values <13 the same weight in the ABC score.
Troponin I and S100β levels were also strongly associated with 1-year mortality. Cardiovascular biological markers like troponin I have been associated with mortality in SAH cases.12,30,31 Although the causes of cardiac abnormalities after SAH cases are controversial, experimental data indicate that catecholamine toxicity plays a key role.32–34 Although the causal links and associations between cardiac abnormalities and SAH outcomes are not clearly understood, troponin I is useful for predicting outcomes because of its association with primary brain injury and cardiac dysfunction. S100β increases with traumatic or vascular damage to brain tissue. Thus, high plasma S100β concentration is common after SAH and has been shown to predict death and other unfavorable outcomes.16,35,36
Interestingly, age was associated with 1-year mortality in the univariate but not the stepwise multivariate analyses. Elderly patients' vulnerability to poor cardiac and consciousness outcomes could explain why our 3 predictors aggregate most of the age effects to predict 1-year mortality. Thus, we confirmed a posteriori that the addition of age as a variable did not statistically improve the ABC score's prediction of 1-year mortality. Furthermore, we tested different combinations (eg, age+GCS+troponin, age+GCS+S100, age+GCS), and the AUC-ROCs were always less than the AUC-ROC of the ABC score (P<0.04 for the 3 comparisons, not shown).
Study Strengths and Limitations
It may be difficult to assess comorbidities like chronic hypertension and smoking history right at admission. Therefore, only data that were readily accessible in clinical practice at admission were studied.
We decided to construct a simplified predictive score with the fewest criteria to facilitate its use at patient admission in standard clinical practice. To simplify the risk score, the troponin I and S100β values were dichotomized. However, because the choice of threshold could affect the score's discriminating power, we reconstructed our risk score in the derivation cohort using different combinations of thresholds for troponin I and S100β, ranging from 0.25 to 1 μg/L without finding any significant AUC-ROC difference (P>0.05, data not shown). Furthermore, we stratified troponin I and S100β into 3-level variables (low for values <0.25 μg/L, intermediary, and high for values >1 μg/L), rebuilt our risk score with different combinations of 2-level or 3-level variables, and did not find any AUC-ROC differences between our original score and these newly created scores (P>0.05).
When comparing the ABC score with pre-existing scores, we decided to use the statistical methods that could identify the specific advantages of adding new predictors.21 These methods would address discrimination (change in ROC-AUC and integrated discrimination improvement) as well as reclassification (net reclassification improvement and risk stratification capacity). The sensitivity analyses for each of the major complications showed that the ABC score had also the power to predict 1-year mortality for each subgroup (online-only Supplemental Table VI).
A limitation of this study is that it was monocentric, because survival may be dependent on intensive care unit management. However, the advantage of our study was that all patients were managed in highly standardized conditions36,37 that allowed us to identify clinically relevant independent risk factors. Nevertheless, a prospective and multicenter study featuring different intensive care unit therapeutic strategies should be conducted to confirm the value of this improved score.
This is the first study to provide large cohort-based evidence supporting the integration of biomarkers and clinical characteristics for the prediction of 1-year outcomes in patients with SAH receiving aneurysm coiling. The simplified ABC score was superior to the GCS, WFNS, and Fisher scores for predicting 1-year mortality in an independent validation cohort. However, additional studies are needed to confirm these results at other centers.
Sources of Funding
Supported by Assistance Publique des Hôpitaux de Paris.
We thank the Neurosurgical Intensive Care Unit, the service of Neurosurgery, and the Neuroradiology Department of the Pitié-Salpêtrière Hospital. We thank the Center of Cerebrovascular Research at the University of California–San Francisco. We also thank Rachel Whelan from Dr Apfel's Clinical Research Core for her editorial assistance.
The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.111.638197/-/DC1.
- Received September 7, 2011.
- Revision received December 2, 2011.
- Accepted December 28, 2011.
- © 2012 American Heart Association, Inc.
- Hop JW,
- Rinkel GJE,
- Algra A,
- van Gijn J
- Molyneux A,
- Kerr R,
- Stratton I,
- Sandercock P,
- Clarke M,
- Shrimpton J,
- et al
- Teasdale GM,
- Drake CG,
- Hunt W,
- Kassell N,
- Sano K,
- Pertuiset B,
- et al
- Naidech AM,
- Kreiter KT,
- Janjua N,
- Ostapkovich ND,
- Parra A,
- Commichau C,
- et al
- Tung P,
- Kopelnik A,
- Banki N,
- Ong K,
- Ko N,
- Lawton MT,
- et al
- Dumont AS,
- Crowley RW,
- Monteith SJ,
- Ilodigwe D,
- Kassell NF,
- Mayer S,
- et al
- Rankin J
- van Swieten JC,
- Koudstaal PJ,
- Visser MC,
- Schouten HJ,
- van Gijn J
- van Heuven AW,
- Dorhout Mees SM,
- Algra A,
- Rinkel GJ
- Hravnak M,
- Frangiskakis JM,
- Crago EA,
- Chang Y,
- Tanabe M,
- Gorcsan J III.,
- et al
- Hachinski VC,
- Smith KE,
- Silver MD,
- Gibson CJ,
- Ciriello J
- Zaroff J,
- Pawlikowska L,
- Miss JC,
- Yarlagadda S,
- Ha C,
- Achrol AS,
- et al