Risk of Shunting After Aneurysmal Subarachnoid Hemorrhage
A Collaborative Study and Initiation of a Consortium
Background and Purpose—Shunt dependent hydrocephalus after aneurysmal subarachnoid hemorrhage (aSAH) is a common sequela that may lead to poor neurological outcome and predisposes to various interventions, admissions, and complications. We reviewed post-aSAH shunt dependency in a population-based sample and tested the feasibility of a clinical risk score to identify subgroups of aSAH patients with increasing risk of shunting for hydrocephalus.
Methods—A total of 1533 aSAH patients from the population-based Eastern Finland Saccular Intracranial Aneurysm Database (Kuopio, Finland) were used in a recursive partitioning analysis to identify risk factors for shunting after aSAH. The risk model was built and internally validated in random split cohorts. External validation was conducted on 946 aSAH patients from the Southwestern Tertiary Aneurysm Registry (Dallas, TX) and tested using receiver-operating characteristic curves.
Results—Of all patients alive ≥14 days, 17.7% required permanent cerebrospinal fluid diversion. The recursive partitioning analysis defined 6 groups with successively increased risk for shunting. These groups also successively risk stratified functional outcome at 12 months, shunt complications, and time-to-shunt rates. The area under the curve–receiver-operating characteristic curve for the exploratory sample and internal validation sample was 0.82 and 0.78, respectively, with an external validation of 0.68.
Conclusions—Shunt dependency after aSAH is associated with higher morbidity and mortality, and prediction modeling of shunt dependency is feasible with clinically useful yields. It is important to identify and understand the factors that increase risk for shunting and to eliminate or mitigate the reversible factors. The aSAH-PARAS Consortium (Aneurysmal Subarachnoid Hemorrhage Patients’ Risk Assessment for Shunting) has been initiated to pool the collective insights and resources to address key questions in post-aSAH shunt dependency to inform future aSAH treatment guidelines.
- cerebrospinal fluid shunts
- risk factors
- subarachnoid hemorrhage
- ventriculoperitoneal shunt
Acute aneurysmal subarachnoid hemorrhage (aSAH) is a critical systemic condition, and survivors of the primary bleed require multidisciplinary neurointensive care.1 People who survive aSAH carry an increased risk for various complications including epilepsy, depression, cognitive impairment, shunt requirement for hydrocephalus, and shunt complications.1
Hydrocephalus after aSAH has been reported to occur in 6% to 67% of the cases.2–6 The acute phase can be self-limiting in some patients, whereas others will require external ventricular drainage (EVD) to alleviate hydrocephalus symptoms.7 Whereas mechanisms of hydrocephalus development have not been fully elucidated, studies have suggested deterioration of the cerebrospinal fluid (CSF) dynamics,3,8 obstructive mechanisms because of blood products,3,9 disrupted absorption at the arachnoid granulations level,3,10 or inflammation11 as possible causes. For some patients, this will continue to develop into a chronic condition requiring permanent CSF diversion, with all the associated interventions, (re)admissions, and complications including a relatively high rate of shunt failure and infections.12,13 Therefore, it is important to identify and understand the factors that increase risk for shunting and to eliminate or mitigate the reversible factors that put aSAH patients at risk. Also, recognition of predictive variables and identifying patients at risk for shunt-dependent hydrocephalus (SDHCP) could lead to optimized management with avoidance of increased neurological morbidity, impaired functional outcome and quality of life, and extended hospital stays associated with chronic hydrocephalus.3,9,14,15
The few published series on shunting after aSAH had short follow-up times,2,4,16–18 small samples,2,3,5,6,15–20 or used administrative databases,12,21–23 where there was an inherent problem of selection bias and uncontrollable factors that could influence the reported rate of shunt requirement. Case series are especially vulnerable to selection bias; studies that report on a series drawn from their patients from a particular population (eg, a hospital or clinic) may not appropriately represent the proportions in the wider population. Data from large administrative databases are also subject to errors in coding and sampling and reporting biases.12,21,22 Therefore, the incidence of permanent CSF shunting after aSAH ranges in the literature from 1% to 45%.2,5,6,8,12,15,16,20,22,24
In the current study, we report the incidence of post-aSAH shunt dependency in a population-based sample, review the morbidity and mortality associated with shunt dependency, and determine the feasibility of a prediction model to identify subgroups of aSAH patients with the highest likelihood of shunt placement for hydrocephalus. For the latter, we use recursive partitioning analysis (RPA) to identify aSAH patients at differential risk of shunt placement for hydrocephalus. In addition to an internal validation of the model, a pragmatic approach toward external validation is taken to assess feasibility and model performance in a real-world setting (different study population, practice pattern, and data schemes).
This study was conducted using the Kuopio Intracranial Aneurysm Database. This population-based database includes prospective information on all cases of unruptured and ruptured saccular intracranial aneurysms (sIAs) in the whole Eastern Finland Region. All cases of aSAH in this database were admitted to Kuopio University Hospital (KUH) for angiography and treatment if not moribund. For the external validation, the UT Southwestern cohort was abstracted from the Southwestern Tertiary Aneurysm Registry, which is a prospectively collected registry of all patients presenting to the University of Texas Southwestern Medical Center in Dallas for the evaluation or treatment of a cerebral aneurysm since 1989.
For the Kuopio cohort, patients were included if residing in the KUH catchment area at the time of first aSAH between January 1, 1990, and December 31, 2012, and admitted alive to KUH with active follow-up until December 31, 2014. In addition, only cases with sIA(s) confirmed by angiography were included. KUH indication guidelines for shunt surgery in SAH patients were based on clinical symptoms, radiological evidence of hydrocephalus, or patients not tolerating CSF drainage weaning. This concurs with the reported practice of other groups.2,15,19,20,24
The UT Southwestern cohort included aSAH patients admitted between January 1, 1990, and December 31, 2014, with at least 12 months of active follow-up. Indications for shunting were similar to that of the KUH population, although practice patterns in terms of timing of shunting, type of shunting device, and CSF drainage/weaning techniques between the centers will differ and were also subject to change during the study period.
Baseline patient characteristics comparison among the shunted and nonshunted groups were performed using the Wilcoxon test for variables that were either continuous or ordered categorical. General tests for associations were used for variables that did not have any inherent ordering. Differences were considered statistically significant at P<0.05. Potential prognostic factors were analyzed using 2 statistical approaches. All patients who were alive for the first 14 days after aSAH were included in these statistical analyses to evaluate the relationship between the predictor variables and risk of shunt placement.
The first analysis aimed at modeling risk factors for shunting after uni- and multivariate logistic regression analyses, with the odds ratios and 95% confidence intervals reported. Because of the number of variables considered, we chose to include variables in the final model only if the results were statistically highly significant at P<0.01. The multivariate regression model included a cross-product terms for the hypothesized interacting risk factors in addition to their main effects (Methods section in the online-only Data Supplement).
The second analysis used the variables identified in the previous analysis for an RPA to define the shunt risk groups (Methods section in the online-only Data Supplement). Data were randomly split into a 70% exploratory sample, and the remaining 30% were used as the internal validation sample to reduce overfitting and upward-biased estimates of the coefficients. An external validation was performed on the UT Southwestern cohort to pragmatically validate the model performance of the RPA model.
The model’s ability to risk stratify by other shunt and functional outcome measures was also assessed in the Kuopio cohort. For functional outcome, the Glasgow Outcome Scale at 1 year was used, with unfavorable outcome defined as scores 1 to 3. Time to shunt for the RPA classes was tested using the Kaplan–Meier method and log-rank test. SPSS 23.0 statistical software (SPSS, Inc, Chicago, IL) was used.
The study was approved by the Ethical Committee of the KUH. UT Southwestern Institutional Review Board approval was also obtained for the reporting of the data in this study (IRB STU 122015–056).
Between January 1, 1990, and December 31, 2012, 1850 patients had been admitted alive after aSAH to KUH. Of those 1553 were alive at 14 days, and shunting because of hydrocephalus after aSAH occurred in 275 (17.7%) of those patients. Total mean follow-up time was 8 years (SD: 7 years). Baseline characteristics of patients were stratified by shunt status, and P values were provided to indicate where patient characteristics may differ between the 2 groups (Table 1). Shunted patients were older and differed with respect to location of sIAs. Shunted patients also had higher Hunt and Hess (H&H) and Fisher grades on admission, more often required an EVD, had higher occurrence of meningitis and intraventricular hemorrhage (IVH), received endovascular intervention more often, and had signs of hydrocephalus on their admission computed tomography. Overall, the median time to shunting was 33 days (mean: 65 days). Baseline characteristics of patients requiring early shunting (≤30 days after admission) were different when compared with patients with delayed hydrocephalus requiring shunting (Table I in the online-only Data Supplement). Early shunted patients were younger, more often of female sex, had a more frequent signs of hydrocephalus on their admission computed tomography, more likely to present with IVH, and required more often an EVD.
A multivariate analysis conducted in the 1553 patients demonstrated the following variables as being independently associated with shunt placement: age at admission (P<0.001), main vascular branch (P<0.001), sIA size (P=0.003), hydrocephalus status on admission (P<0.001), the requirement for EVD (P<0.001), and presence of meningitis (P=0.003; Table 2). None of the interaction terms survived multiple testing correction (Methods section in the online-only Data Supplement). Post-hoc power analysis demonstrated that the sample size in our analysis was adequate to test the study hypotheses (Results section in the online-only Data Supplement).
Recursive Partitioning Analysis
An RPA was constructed to identify the group of variables with the largest measurable effect on shunt placement and to risk stratify patients within these groups. Within the exploratory cohort, RPA branching defined 6 risk groups with successively increased risk (Figure). The area under the curve–receiver-operating characteristic curve for the exploratory sample and internal validation sample to predict shunt dependency was 0.82 (95% confidence interval, 0.79–0.85) and 0.78 (95% confidence interval, 0.71–0.82), respectively. The external validation on the UT Southwestern cohort resulted in an area under the curve–receiver-operating characteristic curve of 0.68 (95% confidence interval, 0.64–0.71).
RPA Groups Assessment
The RPA groups were further assessed in the Kuopio cohort for their ability to predict additional shunt-related outcomes, including time-to-shunt and shunt complication rates. Time-to-event plots were used and demonstrated that time-to-shunt rates were inversely correlated with the RPA groups (Figure [B] and Table 3). Similar patterns were seen with the shunt complicates rate, with rates generally increasing across risk groups (Table 3). In addition, the RPA groups also demonstrated the ability to predict functional outcome. Mortality and unfavorable outcome were correlated with successively increased risk compared with the RPA groups (Table 3). All outcomes were statistically significantly different between the RPA groups (Table 3).
Chronic hydrocephalus after aSAH predisposes patients to worse neurological outcome and cognitive deficits.3,4,14,15,25 In addition, it is the most common cause of readmission after aSAH.13 CSF shunting is prone to short- and long-term complications, including life-threatening obstructions and infections.15 Studies have found that shunting after aSAH related to acute hydrocephalus,2,4,12,17,21,23,26,27 meningitis,4,7,21 ruptured posterior circulation aneurysms,3,12,18,23,28 ruptured anterior circulation aneurysms,7,26 increasing age,3,7,12,17,26 poor grade on admission,3,7,19,21,26 presence of IVH,3,17,21 endovascular29 or surgical26 treatment of sIA, high Fisher grade,2,3,5,17,26 mechanical ventilation,12,22,23 and female gender3,4 have all been described as independent risk factors for shunting after aSAH. Our multivariate analysis and risk model demonstrated that EVD, acute hydrocephalus on admission, meningitis, main vascular branches of sIA, increasing age, and size of sIA as independent risk factors for shunt placement after aSAH, therefore supporting several previous findings from the literature (Table 2 and Figure).
Requirement for EVD was the strongest predictor of receiving a shunt after aSAH according to our risk model (Table 2 and Figure). This association has also been reported in previous studies.6,18,22,23,27 CSF diversion is generally considered as the main management approach for (symptomatic) acute hydrocephalus and patients with altered level of consciousness after aSAH.1 Drainage is required to aid CSF flow dynamics, thought to be affected by blockage caused by blood product obstructing arachnoid granulations and the ventricular and cisternal drainage pathways.10 The precise pathogenesis of how this acute hydrocephalus condition transitions to a chronic communicating hydrocephalus after aSAH is yet to be elucidated. Although a strong correlation was found between EVD placement and shunt dependency on the multivariate analysis and RPA analysis and after adjusting for several hydrocephalus-related factors, a direct cause-and-effect relationship cannot be established. The vast majority of studies investigating risk factors for SDHCP after aSAH did not report their practice in terms of EVD indications and drainage/weaning techniques. KUH and UT Southwestern both have different practices in terms of CSF drainage. KUH used intermittent CSF drainage when intracranial pressure reaches a predefined threshold; UT Southwestern used continuous CSF drainage at set pressure thresholds. How this difference in practice influenced shunt dependency or the performance of the prediction model should be further studied through collaborative efforts involving well-characterized cohorts. The review of the sparse literature on this matter offers a broad scope for improvement and a direction for future research that could lead to decreased shunt dependency in aSAH patients, by optimizing EVD management (indications, drainage, and weaning techniques) and understanding physiological changes after EVD placement.30,31
Meningitis was an independent risk factor for shunt dependency in our series, and this association has also been found in other series.4,7,17,23 Studies have proposed several mechanisms on how inflammation predisposes to delayed hydrocephalus. It has been suggested that meningeal infection causes CSF flow disturbances,4,17 ependymal cell dysfunction, and CSF barrier cell inflammation that lead to increased secretion of CSF32 resulting in chronic hydrocephalus. It has been described that meningitis is associated with the duration of EVD,33 as CSF sampling from EVD can increase the risk of meningitis.34 Meningitis is a risk factor that could, in principle, be prevented with comprehensive management and therefore should be the focus of prevention efforts to minimize the patient’s risk for shunting. Studies have suggested that the usage of prophylactic antibiotics in aSAH patients requiring EVD,35 strict EVD management protocols,36 minimum sampling of CSF,34 and usage of antibiotic-impregnated catheters to reduce the risk for meningitis.37 One study has also proposed the usage of prophylactic anti-inflammatory drugs to reduce the development of chronic hydrocephalus and shunting after aSAH.11 The vast majority of studies did not report the criteria used for defining “meningitis” in their cohort. A uniform definition of EVD-related infections is a necessary first step to clarify the occurrence of these events and direct our research efforts when assessing risk.
Previous studies have described an increased risk of shunting for aSAH from posterior circulation3,12,18,23,28 or anterior circulation7,26 aneurysms and a lower risk for middle cerebral artery branch sIAs.23 Our study demonstrated results comparable to those reported previously; rupture from anterior cerebral artery or vertebrobasilar artery branch sIAs are at greater risk of permanent CSF diversion compared with middle cerebral artery branch sIAs. These locations were also more often associated with IVH7 and blood in the basal cisterns where the subarachnoid space around them is wide and offers little resistance to extravasation.3,17,23
Previous studies have shown advanced age to be an independent risk factor for SDHCP after aSAH.3,12,21,25,33 Our analysis demonstrated increased age to be associated with a higher risk of receiving a shunt after aSAH. van Gijn et al9 have demonstrated a quantitative correlation between increasing age and higher bicaudate indices; therefore, it could be that hydrocephalus is diagnosed more frequently with increasing age. Also, with larger subarachnoid spaces, older aSAH patients can hold larger quantities of blood, thus increasing their risk of developing SDHCP.25 Furthermore, the extent of meningeal fibrosis increases with age leading to impaired CSF circulation, secretion, and decreased CSF absorption. The ventricular compliance also decreases in the aging brain, thereby predisposing to CSF circulation disturbances as seen in idiopathic normal pressure hydrocephalus.38
Predictive modeling allows for subgroups of aSAH patients with the highest likelihood of shunt placement for hydrocephalus to be identified. The use of RPA groups allows for multiple risk factors to be condensed into one ordinal scale to produce one number that reflects the risk profile. Predictive modeling could provide clinicians with an enhanced ability to stratify patients with aSAH and identify those at the highest risk of future SDHCP. It also provides information on the relationship between risk factors and the development of SDHCP, which can be potentially targeted by optimizing management strategies. In this feasibility study, the RPA groups were built using a population-based sample with long-term follow-up to reduce biases in analyzing the incidence and factors affecting shunt placement in the aSAH population. The breakdown of early versus late shunting demonstrated the importance of having long-term follow-up because delayed hydrocephalus requiring shunting could occur many months after admission (Table I in the online-only Data Supplement). Because patients with the delayed hydrocephalus requiring shunting have different characteristics compared with the early shunting group, it would result in biased estimates of risk factors if the follow-up is not long enough to capture all post-aSAH hydrocephalus cases. Significantly, the RPA groups developed in this study also displayed the ability to stratify according to other outcomes, such as mortality, unfavorable outcome, shunt complications rates, and time-to-shunt rates. This demonstrated the potential additive clinical value of predictive modeling of shunt dependency after aSAH.
The model in this study was built and validated using randomly selected samples of patients, demonstrating a clinically useful predictive performance of 0.79 area under the curve–receiver-operating characteristic curve in the internal validation cohort. These data suggest that predictive modeling of post-aSAH shunt dependency is feasible. Furthermore, we investigated how the model performed when a pragmatic external validation was conducted in an independent data set (UT Southwestern cohort) with no further adjustments. This resulted in an area under the curve–receiver-operating characteristic curve of 0.68, indicating a relatively good performance, considering the unadjusted differences in study populations, practice patterns, and data schemes. The aim of this study was to establish the feasibility of prediction model of post-aSAH shunt dependency, and this seems to be achievable with clinically useful yields with additional standardized data elements having the potential to add power to future predictive modeling efforts.
Our study is limited by several factors inherent to secondary analysis of prospectively collected data. In addition, lacking standardized definitions in the literature (ie, EVD-related infections, shunt complications, etc) and given the variability among various hospitals, which was demonstrated when validating our model externally, our analysis must be interpreted in this context. Also, although outcome assessment in our cohort was performed in a setting to minimize ascertainment bias from the assessor, it cannot be completely excluded.
Current understanding of post-aSAH SDHCP is fragmented and standardizing the definitions and collection of investigational data will make it easier for the results of studies to be compared, contrasted, and combined. Therefore, a universal prediction model is only feasible through large-scale collaborative efforts involving well-characterized multicenter cohorts. An international working group is needed to collect, combine, and validate data on SDHCP patients after aSAH to unravel the modifiable factors that could influence shunt dependency and ultimately improve patients’ outcomes. We established the aSAH-PARAS Consortium (Aneurysmal Subarachnoid Hemorrhage Patients’ Risk Assessment for Shunting ) (http://www.asahparas.org) to develop common data elements for SDHCP after aSAH, to assess factors associated with shunt dependency, and to prospectively develop and validate a universal risk and prediction model for post-aSAH SDHCP to inform future aSAH treatment guidelines.
Shunt dependency after aSAH is associated with higher morbidity and mortality. Therefore, it is important to identify and understand the factors that increase risk for shunting and to eliminate or mitigate the reversible factors that put aSAH patients at risk. Prediction modeling of shunt dependency after aSAH is feasible with clinically useful yields. Large-scale collaborative efforts involving well-characterized, multicenter cohorts are needed to create a universal prediction model. The aSAH-PARAS Consortium has been initiated to pool the collective insights and resources to address key questions in post-aSAH shunt dependency to inform future aSAH treatment guidelines.
Sources of Funding
This study was supported by research grants from the Päivikki and Sakari Sohlberg Foundation, Kuopio University Hospital, and Academy of Finland.
Dr Welch is a consultant for Stryker Neurovascular and Covidien. The other authors report no conflicts.
The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.116.013739/-/DC1.
- Received April 22, 2016.
- Revision received July 16, 2016.
- Accepted August 5, 2016.
- © 2016 American Heart Association, Inc.
- Connolly ES Jr,
- Rabinstein AA,
- Carhuapoma JR,
- Derdeyn CP,
- Dion J,
- Higashida RT,
- et al
- Greenberg JK,
- Washington CW,
- Guniganti R,
- Dacey RG Jr,
- Derdeyn CP,
- Zipfel GJ
- Zaidi HA,
- Montoure A,
- Elhadi A,
- Nakaji P,
- McDougall CG,
- Albuquerque FC,
- et al
- de Oliveira JG,
- Beck J,
- Setzer M,
- Gerlach R,
- Vatter H,
- Seifert V,
- et al
- Hoh BL,
- Kleinhenz DT,
- Chi YY,
- Mocco J,
- Barker FG 2nd
- Yamada S,
- Ishikawa M,
- Yamamoto K,
- Ino T,
- Kimura T,
- Kobayashi S
- Klopfenstein JD,
- Kim LJ,
- Feiz-Erfan I,
- Hott JS,
- Goslar P,
- Zabramski JM,
- et al
- Harrop JS,
- Sharan AD,
- Ratliff J,
- Prasad S,
- Jabbour P,
- Evans JJ,
- et al
- Keong NC,
- Bulters DO,
- Richards HK,
- Farrington M,
- Sparrow OC,
- Pickard JD,
- et al