Predicting outcome after aneurysmal subarachnoid hemorrhage by exploitation of signal complexity: a prospective two-center cohort study

Aneurysmal subarachnoid hemorrhage (aSAH) remains a serious disease with often poor prognosis even after successful securing of the aneurysm [1]. Patients who survive the initial hemorrhage remain at risk for developing secondary brain injury, such as delayed cerebral ischemia (DCI) [2]. DCI is a major cause of death and disability after aSAH [3]. It is the consequence of complex interactions of neuronal activity, cerebral and systemic hemodynamics, and feedback mechanisms—neurovascular (un)coupling, cerebral autoregulation, and CO₂ reactivity [2]. Dynamic changes of multiple interacting factors including cerebral vasospasm [4], inflammatory markers [5], oxygenation [6], blood pressure, and cardiac output [7] precede DCI occurrence. The paramount goal of neurocritical care is to predict, counteract or even prevent these secondary injuries to improve patients’ outcome. Consequently, the acute period following the hemorrhage is accompanied by extensive multimodal monitoring within a neurocritical care unit (NCCU) environment. The monitoring comprises aspects of cerebral physiology and hemodynamics (incl. intracranial pressure (ICP), cerebral perfusion pressure (CPP)) integrated with systemic physiological parameters (arterial blood pressure, cardiac output, heart rate (HR), oxygenation, and ventilation) [8].

Signal complexity (i.e. entropy) describes the level of apparent disorder within a system. Low signal complexity predicts unfavorable outcome in a variety of diseases and is assumed to reflect increased rigidity of the cardio/cerebrovascular feedback/regulating system leading to (or reflecting) autoregulation failure [9‐12] This, in turn, leaves the brain susceptible to secondary injury. Physiological systems are regulated by multiple, interacting, mechanisms leading to dynamically changing biosignals across different temporal scales.¹⁴ Multiscale entropy (MSE), a version of signal complexity, estimates sample entropy over a range of increasingly downsampled (i.e. averaged) data [13, 14] In comparison to sample entropy of a single scale MSE has the following benefits: 1. It allows for the evaluation of complex physiological systems that operate across different time scales; 2. It suppresses the impact of noise on the resulting metric. In 2012 Lu et al. described the association between decreased ICP signal complexity and unfavorable outcome after traumatic brain injury [9]. Zeiler et al. validated the concepts presented in a large multi-center cohort and extended the description to include other biosignals [15]. In aSAH, a metric related to signal complexity, heart rate variability, has shown, to a degree, an association with complications and unfavorable outcome [16‐18]. However, other patho-physiological states such as sepsis decrease heart rate variability, whereby its use in clinical practice for prediction of specific complications has remained limited. We aimed to exploit the abundance of monitoring data acquired from each patient within an NCCU environment to assess the potential use of MSE as an outcome predictor after aSAH.

The study was approved by the local ethics committees Zurich and Aachen and was in accordance with the ethical standards laid down in the 2013 Declaration of Helsinki for research involving human subjects. Informed consent was received before inclusion by the patient or their legal medical representative. Data from two prospective observational cohorts (University Hospital Zurich, Switzerland; the Rheinisch-Westfälische Technische Hochschule Aachen, Germany) was analyzed. The Zurich cohort was used as the derivation cohort to establish models and analyses, while the Aachen cohort was used for external validation.

Entropy, and in particular its multiscale version, MSE, builds on the previously described concept that physiological systems are regulated by multiple, interacting, mechanisms that collectively result in dynamically changing, irregular, fluctuations of biosignals across different temporal scales [14]. Lower MSE represents higher regularity of a system. While a “stable”, “regular” system would at first glance seemingly give the impression of being ‘healthy’, in reality, such a system is more rigid, with impaired capacity to counteract ever-present, random environmental triggers. The disease course of aSAH includes a variety of pathological processes necessitating continuous and rapid adjustments to equilibrate the system [27]. Patients that survive the initial hemorrhage remain at risk for developing secondary brain injury due to numerous pathophysiological cascades and complications. As part of the early brain injury, due to the initial hemorrhage, ICP rises either immediately (due to the volume of the bleeding itself [28]) or with a certain delay (i.e. resulting hydrocephalus [29] and/or brain edema). In the worst-case scenario, either one can lead to a relevant reduction in CPP. If the system fails to counteract this decrease in cerebral perfusion, this may lead to transient cerebral hypoxia or, ultimately, even infarction. Various injury cascades (upregulation of inflammatory pathways [30], coagulopathy [31]) further damage the system in case of cerebral hypoxia. A non-reactive system, represented consequently by low MSE, has been associated with higher pressure reactivity index (PRx) values supporting the notion that MSE represents the activity level of the physiological regulation systems, including cerebral blood flow autoregulation [9, 15]. While both metrics clearly share similar mechanisms, to equate MSE with cerebrovascular reactivity (as assessed using PRx) would be an oversimplification, considering Lu et al. also showed that PRx loses its predictive value when MSE is added to multivariable regression models for outcome prediction after traumatic brain injury [9].

The main driver of secondary brain injury in aSAH is DCI [3]. Although, DCI is the consequence of complex interacting pathophysiological sequelae, a well described and potentially reversible cause is vasospasm, which describes the narrowing of cerebral vessels [4]. Depending on the severity of such narrowing, ischemia or even infarction can occur. In aSAH, intact cerebral autoregulation is essential to counteract such dynamic reductions of vessel diameter by automatically and immediately increasing CPP. Autoregulation failure is detrimental, as symptomatic treatment (using vasoactive medications or intra-arterial spasmolysis) can only be initiated with a significant delay.

In addition to cerebral complications, aSAH leads to various systemic and most prominently cardiac complications [32, 33]. Both, myocardial ischemia and neurogenic stunned myocardium are found after aSAH leading to wall motion abnormalities and consequently reduced cardiac output. In the worst case, cardiac complications coincide with (or cause) pulmonary edema leading to further impairment of cardiac function and oxygenation [34‐36]. Sufficient cardiac output is necessary to counteract impaired perfusion due to vasospasm. Cardiac output guided therapy is beneficial for managing cerebral oxygenation in patients with vasospasm [7, 37]. Myocardial injury might indeed be one cause of decreased entropy with previous reports describing good discrimination when assessing heart rate variability after aSAH and diagnosis of neurocardiogenic injury [17]. To date, the most commonly evaluated complexity related metric in aSAH is heart rate variability due to its simple determination based on short electrocardiograms [18]. Considering these results, it is not surprising that MSE hr was a predictor of outcome. However, variability and entropy cannot be used interchangeably due to the large number of metrics used for description some of which depend on distribution while others depend on specific patterns.

In aSAH, due to the complex nature and course of disease, NCCU monitoring plays a pivotal role for guiding treatment. Neurocritical care bioinformatics allow for the acquisition, integration, and synchronization of the various biosignals within the same environment, thereby permitting exploitation of advanced data-driven methods for guiding treatment and outcome prediction [38]. Yet, computational methods remain underutilized and are generally not readily available for direct bedside implementation. Commonly, single (at times arithmetic mean or worse, single snapshot) targets are used for guiding treatment, ignoring the potential benefits of complex integration. The results of this study underline the potential of advanced analytical tools for improving the understanding of the complex pathophysiology of multimodal monitoring dynamics after aSAH. However, relevant limitations exist.

Although this study included over 300 patients from 2 centres, they were all treated at highly specialized, high resource centers with relatively high patient volumes. On a similar note, patients underwent active treatment throughout their NCCU stay. It is unclear if, and to what extent interventions and complications are associated with changes in MSE. This is underlined by the results showing changes in MSE depending on various events. Either aspect, however, implies that the biosignals acquired do not represent the “natural” course of the disease, but patients undergoing active treatment. Due to the exploratory nature of this study, we did not explore time-trends of MSE, thus we cannot comment on the patterns of variability of MSE over the course of the NCCU stay (e.g. depending on intervention, medication or similar) or whether there were specific turning points (e.g. refractory ICP increase, DCI etc.). Overall, many complications and disease aspects were associated with changes in MSE and it is likely that the resulting metric represents a composite of various aspects. It is important to note that while the results were adjusted for the relevant and known outcome predictors age, clinical and imaging severity, and the important complication DCI, many other clinical variables are of importance when predicting outcome. To date, no multimodal monitoring based metric alone can or should replace clinical examination. However, metrics such as MSE should be seen as complementary allowing for additional physiology information.

This study provides the first description of MSE as an outcome predictor after aSAH. MSE metrics and thereby complexity of physiological signals are independent, internally and externally valid predictors of 12 month outcome after aSAH. MSE decreases monotonically with worse outcomes and remains a valid outcome predictor when adjusting the outcome definition to the initial severity of disease and age. The promising results of this study warrant further investigation into the cause of the resulting complexity as well as its association with important and potentially preventable complications (i.e. vasospasm and DCI). Of particular importance will be the assessment of time-trends and the evaluation of intraindividual episodes of decreased entropy and their association to specific events. Promising targets for such analysis are MSE sbp and MSE hr since neither requires continuous neuromonitoring and can therefore be applied to the whole aSAH population.