Preoperative predictors of poor outcomes in Thai patients with aneurysmal subarachnoid hemorrhage

Objective A scoring system for aneurysmal subarachnoid hemorrhage (aSAH) is useful for guiding treatment decisions, especially in urgent-care limited settings. This study developed a simple algorithm of clinical conditions and grading to predict outcomes in patients treated by clipping or coiling. Methods Data on patients with aSAH hospitalized in a university’s neurovascular center in Thailand from 2013 to 2018 were obtained for chart review. Factors associated with poor outcomes evaluated at one year were identified using a stepwise logistic regression model. For each patient, the rounded regression coefficients of independent risk factors were linearly combined into a total score, which was assessed for its performance in predicting outcomes using receiver operating characteristic analysis. An appropriate cutoff point of the scores for poor outcomes was based on Youden’s criteria, which maximized the summation between sensitivity or true positive rate and the specificity or true negative rate. Results Patients (n, 121) with poor outcomes (modified Rankin Scale, mRS score, 4–6) had a significantly higher proportion of old age, underlying hypertension, diabetes and chronic kidney disease, high clinical severity grading, preoperative rebleeding, and hydrocephalus than those (n, 336) with good outcomes (mRS score, 0–3). Six variables, including age >70 years, diabetes mellitus, World Federation of Neurosurgical Societies (WFNS) scaling of IV-V, modified Fisher grading of 3–4, rebleeding, and hydrocephalus, were identified as independent risk factors and were assigned a score weight of 2, 1, 2, 1, 3 and 1, respectively. Among the total possible scores ranging from 0–10, the cut point at score 3 yielded the maximum Youden’s index (0.527), which resulted in a sensitivity of 77.7% and specificity of 75.0%. Conclusion A simple 0–10 scoring system on six risk factors for poor outcomes was validated for aSAH and should be advocated for use in limited resource settings.

Introduction due to a retrospective nature to obtain data from medical records. All data were fully anonymized throughout the study process.

Study settings and patients
This retrospective study was conducted in a university's neurovascular center that admitted patients typically referred from lower-level facilities located in the northeastern region of Thailand. The study site has a high volume of microscopic clipping and endovascular coiling. All patients presenting with aSAH were treated according to the center's standard protocol [6] and neurovascular care was provided by trained nurses. aSAH was confirmed by either computerized tomographic angiogram (CTA) or magnetic resonance angiography (MRA) or by cerebral digital subtraction angiography (DSA) if necessary. All patients with a Glasgow Coma Score (GCS) [22] of less than 8 were intubated. Patients' systolic blood pressures (SBPs) were controlled within the range of 120-160 millimeters of mercury (mmHg). Patients with symptomatic hydrocephalus had extraventricular drains inserted to control their intracranial pressure within 20-25 centimeters of water column at 4 degC (cmH 2 O). All patients received oral nimodipine and intravenous antiepileptic drugs as clinically indicated. The choice of aneurysmal obliteration was discussed with an interdisciplinary team and the patient or relatives. Aneurysmal treatment was implemented as soon as possible, usually within 48 hours after hospital admission or 72 hours after onset. Outpatient follow-up was pursued at 1, 3, 6 and 12 months plus yearly afterward. At each follow-up visit, the modified Rankin Score (mRS) [23] of each patient was assessed by the neurosurgeons. Patients who did not attend their visits as planned were reached for telephone interviews by trained nurses.
Patients diagnosed with CTA-, MRA-or DSA-confirmed aSAH from 2013 to 2018 who received definite treatment by either clipping or coiling were included in this study. Patients with aSAH due to traumatic aneurysms, mycotic aneurysms or arteriovenous malformations were excluded.

Study variables
The primary endpoint was poor outcomes after definite treatment, which was indicated by the combined dependency or death at one year after treatment. A dependent state was defined as an mRS score of 4-5 (in contrast with an independent state with an mRS score of 0-3), whereas an mRS score of 6 meant death.
Potential risk factors were obtained from a comprehensive review of inpatient and outpatient medical records by a well-trained nurse at the study center. The retrieved data covered patient demographics, including age and sex, and health risks, including overweight defined internationally by the World Health Organization as a body mass index of 25 kilograms per square meter (kg/m 2 ) or over, and the history of smoking and alcohol consumption. Clinical risk factors were the presence of underlying diseases, including hypertension, diabetes mellitus (DM), and chronic kidney disease. High SBP was indicated by a blood pressure measurement at initial admission of over 160 mmHg. Preoperative neurological values after neurological resuscitation were recorded using WFNS scaling and GCS. The scaling of WFNS I-III was defined as good grading and that of WFNS IV-V was defined as poor grading. Radiologic variables included hydrocephalus, modified Fisher (mFisher) grading [24], aneurysmal location and size, and the number of aneurysms. Hydrocephalus was defined according to Evan's ratio with 0.3 as the cut point [25]. The aneurysmal location was categorized as anterior cerebral artery, internal carotid artery, middle cerebral artery, and vertebrobasilar artery. The anterior cerebral artery is comprised of the anterior communicating artery and proximal and distal anterior cerebral arteries. The internal cerebral artery consists of the posterior communicating artery, anterior choroidal artery, and ophthalmic artery. Aneurysms of the posterior cerebral artery, superior cerebellar artery, anterior inferior cerebellar artery, posterior inferior cerebellar artery, vertebral artery and basilar artery were considered vertebrobasilar arteries. All radiologic variables were based on the initial neuroimaging obtained. Rebleeding was defined as a new episode of CT-confirmed, aSAH-related hemorrhage prior to treatment. The time to definite treatment was calculated from the aSAH onset to the intervention given, either surgical or endovascular.

Statistical analysis
To summarize patient characteristics and clinical conditions, continuous variables are presented as the mean and standard deviation (SD), and categorical variables are presented as the frequency and percentage. Each risk factor was compared between patient subgroups, either poor or good outcomes for statistical significance at p <0.05, using an independent t-test for the mean difference or Pearson's chi-square test for the proportion difference. Comparing allcause mortality of patients across age groups and the WFNS scale and mFisher scaling was performed using Kaplan-Meier estimates.
Factors associated with a poor outcome were determined using logistic regression, and the magnitude of the associations was measured in terms of an odds ratio (OR) with a 95% confidence interval (CI). A univariate analysis of each risk factor was performed, yielding an unadjusted OR to guide predictors of the poor outcome. Multivariate analysis was initiated with the full model containing all potential risk factors. A stepwise approach was undertaken to achieve a final model using backward elimination with a statistical criterion at p �0.05 for removal of variables. The performance of the model containing retained risk factors was determined by the area under the receiver operating characteristic (ROC) curve, in which values of 0.50-0.59, 0.60-0.69, 0.70-0.79, 0.80-0.89 and 0.90-1.00 were interpreted as no, poor, acceptable, excellent and outstanding, respectively, in discriminating between binary outcomes [26,27].
Predicted scores of the poor outcome were calculated as a linear combination of a regression coefficient (or log OR) of each predictor retained in the final model, excluding the constant term. A predictor with a positive signed coefficient (or OR >1.0) was a risk factor for a poor outcome, whereas a negative coefficient (or OR <1.0) represented a preventive factor. For practical and clinical applicability reasons, coefficients of 0.51-1.50, 1.51-2.50, 2.51-3.50, . . . were assigned scores of 1, 2, 3, . . . to an individual patient presenting with the risk factors.
The ability of a given score cutoff to correctly detect patients who had a poor outcome in contrast to those with a good outcome was a tradeoff. In our case, using the cutoff at a score of 0 would effectively cover all patients with poor outcomes (sensitivity, 100%), but it would miss all patients with good outcomes (specificity, 0% or false-positive rate, 100%). An increase in the cutoff to scores of 2, 3, 4, . . . would lower the sensitivity while raising the specificity. Youden's index that maximized the true positive rate (or sensitivity) and minimized the false-positive rate (or 1 -specificity) and was equal to "sensitivity + specificity-1" was used for selecting the appropriate cutoff over a possible score range of the study patients [28].

General characteristics
A total of 519 patients with aSAH were admitted during the study period. Sixty-two patients were excluded as they received neither microvascular clipping nor endovascular coiling. Among 457 patients included and analyzed in the present study, 14 (3.1%) had an in-hospital death and 65 (14.2%) died within one year after hospitalization. There were no losses to follow up among 378 patients who were the survivors at one year. Seventeen (4.5%) and 25 (6.6%) patients had a poor mRS score of 4 and 5, respectively; whereas 134 (35.4%), 121 (32.0%), 46 (12.2%) and 35 (9.3%) patients had a good mRS score of 0, 1, 2 and 3, respectively.
Patients with poor outcomes (mRS score, 4-6; n, 121) were significantly older and had significantly more underlying diseases than those with good outcomes (mRS score, 0-3; n, 336) ( Table 1). Furthermore, the poor outcome group had a significantly higher proportion of high-grade WFNS, high-grade mFisher, preoperative rebleeding, and hydrocephalus than the good outcome group. A higher proportion among those with poor outcomes had a higher initial SBP, larger (>4 mm) size of the aneurysm, multiple aneurysms, and time to interventions <24 hours than those with good outcomes but they did not reach statistical significance (p>0.05). An initial GCS in the poor outcome group on average seemed to be lower than that in the good outcome group.
Most (92.1%) patients received microsurgical clipping, and 7.9% received endovascular coiling. Among 121 patients with poor outcomes, 69 died from aSAH within 1 year after the treatments. None of the patients in the good outcome group died within 1 year.
Over the median follow-up of 726 days, 94 patients died from any cause (88 and 6 in the poor and good outcome groups, respectively), and 82 died from aSAH (81 and 1 in the poor and good outcome groups, respectively). Comparing across age groups, WFNS scaling and mFisher scores, patient survival in those >70 years, WFNS IV-V, and mFisher 3-4 were lower than their counterparts (S1-S3 Figs).

Factors associated with treatment outcomes at one year
When the patient's demographic and clinical characteristics were analyzed one at a time for an association with treatment outcomes, age >70 years, presence of underlying hypertension, DM and chronic kidney disease, having a high grade of WFNS (IV-V) and mFisher (3)(4), an experience of preoperative rebleeding and having hydrocephalus increased the likelihood of poor outcomes (OR, 6

Model-based scores predicting poor outcomes
Each predictor was assigned an appropriate scoring weight based on regression coefficients of the final model. The score of each factor was 2, 1, 2, 1, 3 and 1 for age >70 years, presence of DM, WFNS IV-V, mFisher 3-4, rebleeding and presence of hydrocephalus, respectively (Table 3), and resulted in a total possible score between 0 and 10.

PLOS ONE
Preoperative predictors of poor outcomes in Thai patients with aneurysmal subarachnoid hemorrhage Patients with poor outcomes tended to have a higher score (scores 7-9, 80.0, 68.4 and 57.5% of scores 6, 5, and 4, respectively) (Fig 1). Less than half of the patients with scores 0-3 had poor outcomes (32.3, 17.6, 8.3 and 2.4% of patients with scores 3, 2, 1 and 0, respectively). Therefore, an increasing threshold of the scores in discriminating a binary outcome would increase the ability of the scoring system to accurately detect patients with poor outcomes (or sensitivity) but reduce the ability to accurately detect patients with good outcomes (or specificity). The sensitivity and specificity of detecting the true outcomes that varied upon each cutoff over the 0-10 score range are summarized in S1 Table. Using the cutoff for a poor outcome at a score of 0 would cover all (100%) patients with poor outcomes, whereas it would miss all patients with good outcomes (or 100% false-positive rate or 0% specificity). Increasing the cutoff to scores of 2, 3 and 4 resulted in a monotonically lower sensitivity to 87.6, 77.7 and 60.3% and an increased specificity to 58.3, 75.0 and 88.1%, respectively.
The area under the ROC curve of the model-based scores was 0.825, indicating an excellent ability to discriminate patients with poor outcomes from those with good outcomes. Youden's index that maximized the sum of sensitivity and specificity at a given cutoff over the 0-9 score range is illustrated by the vertical distance between the ROC curve and the 45-degree, nondiscriminating line in Fig 2. The index was 0.527, representing the highest performance at the cutoff of 3, followed by 0.484 and 0.459 at the cutoffs of 4 and 2, respectively.

Discussion
Patients >70 years, having an underlying DM, and presenting with a WFNS scale of IV-V, mFisher scale of 3-4, preoperative rebleeding, and hydrocephalus had a significantly increased risk of combined dependency and death within one year after treatment for aSAH. Based on the risk score ranging from 0 to 10, which was calculated from the summation of the individual scores of 2, 1, 2, 1, 3 and 1 for the presence of the six corresponding characteristics, a score of 3 was an appropriate threshold for discriminating patients with poor outcomes from those with good outcomes. Other scoring systems have been created to predict poor outcomes in patients with aSAH using preoperative factors [9][10][11]. The risk factors that were retained in the present scoring system were in line with those reported in previous studies (Table 4). Our study was an incremental step to gain insight into a predictive model suitable for Thailand, where aneurysmal treatment is often delayed and the majority of patients have preexisting hypertension [10,11,14] that was poorly controlled [29] and poor adherence to antihypertensive drugs is common [30].
Old age is an established risk factor for poor outcomes [31]. Elderly patients are more likely to have underlying diseases that may complicate postoperative care. As elderly individuals are often frail, this may hamper proper rehabilitation and result in worsening outcomes. A study involving only patients with ruptured anterior communicating artery aneurysms reported a higher need for permanent cerebrospinal fluid diversion in patients >65 years [32]. Patients >85 years overwhelmingly had poor outcomes despite treatments [33]. However, old age itself should not hinder treatments for patients with aSAH, as there is a proportion with reasonable favorable outcomes [34].
For practical reasons, our study used a history of DM rather than measured hyperglycemia as a predictor. Diabetic patients independent of glycemic control had an increased risk of symptomatic cerebral vasospasm [35]. A recent meta-analysis showed that high glucose levels were prevalent at the time of hospital admission and were associated with poor outcomes [36]. The guidelines have suggested keeping glucose levels within normal limits to prevent overexerting brain metabolism [37]. The initial neurological status is correlated with the outcomes. However, recent clinical studies have shown that post resuscitation clinical grades tend to represent outcomes rather than risk factors [38,39]. Previous studies used the WFNS scale or Hunt-and-Hess measure to grade clinical severity. The Hunt-and-Hess grading, however, contains certain items measuring subjective characteristics, which may lead to unreliable assessments. The present study utilized WFNS scaling when a patient became stable after adequate neurological resuscitation as per the protocol [15]. Poor initial neurological status does not prohibit treatment; thus, both physicians and relatives must weigh the benefit of the procedure against the risk of a vegetative state.
To measure the SAH and associated intracranial hemorrhage, mFisher scaling, which is better than the original Fisher scaling in predicting poor outcomes, was henceforth used in our study. The Barrow Neurological Institute (BNI) grading system that measured the maximal thickness of SAH across cisterns or fissures was recently utilized in the HATCH scoring system [11]. Ninety-six percent of the study patients in the developmental process of HATCH scoring had a Fisher scale of 3; nonetheless, their BNI grades varied. The BNI grading system did not strongly predict cerebral vasospasm, delayed cerebral ischemia or unfavorable outcomes. Preoperative rebleeding, the strongest predictor in our study, has been reported by other studies as a very strong and independent predictor of mortality [12,40] and functional dependency [4,11,40]. Rebleeding that can lead to severe brain injuries was reported to have a prevalence varying from 8% to 23% [11,40,41]. A relatively low percentage (6%) of rebleeding in our study was probably due to survival bias. At local-level hospitals, a majority of cases are not transferred for definite treatment in cases presenting with rebleeding. A score of 3 was assigned to rebleeding in our study, which was equal to the cutoff for poor outcomes. Therefore, patients' relatives should be fully informed of a high risk of poor outcomes if rebleeding is present prior to definite treatment.
Hydrocephalus is often encountered with aSAH (6-67%) [1,6,11,14,42,43]. Acute hydrocephalus results in an increased intracranial pressure, which may result in a higher WFNS grading [44]. In these severe cases, the cerebrospinal fluid must be promptly drained to control the increased intracranial pressure prior to definite treatment [45]. Our study found hydrocephalus was present in 41.1% of patients, and its association with poor outcomes was comparable to that in other studies [4,11]. A third of patients with aSAH will need permanent cerebrospinal fluid diversion [43].
Certain limitations of our study exist. Our research was a retrospective analysis of electronic health records and a review of medical charts. The clinical measures, however, were unlikely to be prone to misclassification bias, as the variables were objectively measured for common events, which were mandatorily evaluated by the protocols implemented in our neurovascular center. Data collection was conducted in a university-level hospital with facilities superior to their provincial counterparts. Patients treated in the former setting tended to have a better outcome than the latter. Our reported risk factors and scoring algorithm, which is consistent with those reported in the USA and European countries, could be adopted by international communities. To expand their use most widely, this set of outcome predictors should be addressed in the aSAH treatment guidelines.
The strengths of our study include the study sample, which contained a relatively large number of patients receiving treatments under established protocols in a high-level facility. Apart from the excellent discriminative ability of the model (area under the ROC curve, 0.825), the defined predictors were simple and practical for clinical assessment. Therefore, the proposed 0-10 scoring system, in which a score of 3 was the cutoff point for poor outcomes, can be widely applied. A tool to guide decisions on treatments is necessary for countries with limited resources. In Thailand, only approximately 60 supertertiary or tertiary care hospitals are capable of treating aSAH. Our tool can be adopted by a lower-level facility for making decisions about patient transfer for proper treatment at a higher-level facility.

Conclusions
A scoring system with excellent discrimination of patients with poor outcomes was developed for aSAH. With a simple algorithm incorporating clinical conditions and severity grading, this system is useful in frontline health facilities.
Supporting information S1 Table. Ability to discriminate between poor and good outcomes by the score cutoff.