Neuroprognostication characterizes the potential for functional recovery in patients who remain unconscious after cardiac arrest. Prognostic impressions heavily influence goals-of-care decisions: Patients deemed to have a favorable recovery potential are often supported through their convalescence journey whereas patients with unfavorable prognosis are transitioned to a comfortable, dignified end-of-life, whenever aligned with their values. No single prognostic test modality is reliable predicting outcomes as subjective and/or unaccounted factors challenge the prediction performance of prognostic tools; hence, multimodal approaches to neuroprognostication are essential to mitigate inaccuracies. Furthermore, understanding meaningful recovery from a patient-centered perspective is critical for personalized decision-making.
Key points
-
•
Neuroprognostication after cardiac arrest involves assessing brain injury burden, cerebral reserve, resilience, and patient values to guide clinical decisions and outcomes.
-
•
Multiple modalities—biochemical markers, neuroimaging, neurophysiologic tests, and clinical examinations—are used to evaluate structural and functional brain injury, each with specific advantages and limitations.
-
•
Challenges include variable guidelines and timing of assessments, limited acess to tools, cognitive biases, and balancing prognostic accuracy with patient values.
-
•
Effective communication with surrogates, emphasizing honesty about uncertainty and understanding patient goals, is crucial for shared decision-making.
-
•
Future directions focus on integrating advanced neuroimaging, concepts of cerebral reserve and resilience, and artificial intelligence to improve prediction accuracy and personalize neuroprognostication.
Abbreviations
| ADC | apparent diffusion coefficient |
| CPC | Cerebral Performance Category |
| CT | computed tomography |
| DoC | disorders of consciousness |
| DWI | diffusion-weighted imaging |
| EEG | electroencephalography |
| FOUR | Full Outline of Unresponsiveness Score |
| FPR | false positive rates |
| GCS | Glasgow Coma Scale |
| GFAP | glial fibrillary acidic protein |
| GWR | gray-white matter ratio |
| HIBI | hypoxic-ischemic brain injury |
| MCS | minimally conscious state |
| mRS | modified Rankin Scale |
| NfL | neurofilament light chain |
| NSE | neuron-specific enolase |
| PLR | pupillary light reflex |
| ROSC | return of spontaneous circulation |
| SSEP | somatosensory evoked potentials |
Introduction
Neurologic outcome prediction (or neuroprognostication) is a cornerstone of postcardiac arrest care, and in practice, it is the converging lens through which clinical trajectories unfold. Disorders of consciousness (DoC) in the early postcardiac arrest period are highly prevalent: It is well known that 8 in 10 patients remain unresponsive to verbal commands following return of spontaneous circulation (ROSC) and deemed to be in a comatose state. Estimating the potential for functional recovery is an essential task in comatose survivors with downstream effects on decision-making that ultimately mediate outcomes. Neuroprognostic impressions uphold decisions for continued life-support and advanced organ-targeted care in patients expected to have favorable outcomes and often guide transitions toward end-of-life care in patients thought to have devastating hypoxic-ischemic brain injury (HIBI). Truly a branching point or a bottleneck in the clinical course; patients, surrogates, and clinicians rely heavily on the accuracy of predictions to guide decisions on whether sustaining life support remains appropriate. An erroneously pessimistic prognosis can cause the death of someone who would have otherwise recovered. An erroneously optimistic impression may, on the other hand, prolong survival in a state that may not align with the patient’s wishes in terms of acceptable disability or rehabilitation journey.
Neuroprognostication is a dynamic process that begins as early as possible following ROSC. As illustrated in Fig. 1 , it involves 4 key priority pillars: (a) overall burden of brain injury, which includes damage from the initial hypoxic-ischemic insult during cardiac standstill as well as the accrued secondary injury from subsequent insults, neuroinflammation, and programmed cell death despite restored circulation; (b) cerebral reserve; (c) cerebral resilience; and (d) individual patient values regarding acceptable levels of disability and the anticipated recovery journey. Neuroprognostic impressions are formed through the interpretation of findings from all 4 pillars, often within emotionally charged circumstances and in the presence of unavoidable confounders. This is an inherently complex and nuanced process that requires synthesizing multiple data points and critically evaluating the reliability and validity of each finding.
Key pillars of neuroprognostication.
The injury burden is currently evaluated directly or indirectly by guideline-recommended neuroprognostic tools ( Fig. 2 ). Biochemical markers of neuronal injury (eg, neuron-specific enolase [NSE]) and neuroimaging assess for direct evidence of structural injury. Neurophysiologic tests and features of the clinical examination evaluate injury burden indirectly through neurologic function. Prognostic impressions reflect the interpretation of test results, considered alongside the severity of the initial insult, the extent of accumulated brain injury, and the estimated cerebral reserve. Together, these factors help outline a potential trajectory for recovery. Cerebral resilience and the potential influence of residual confounding can be appraised by noticing mismatches between the perceived severity of cerebral insults and the observed corresponding burden of injury. For instance, minimal injury burden on neuroimaging and preserved neurologic examination despite severe insults suggest a high degree of cerebral resilience. On the same token, low titers of biomarkers and minimal injury on neuroimaging accompanied by ongoing neurologic dysfunction—detected through clinical assessment or neurophysiologic tests—suggest that residual confounding factors may be contributing to the clinical picture. Finally, it is essential to reconcile the projected recovery trajectory with the patient’s individual values, as conveyed by their surrogates. These values should serve as the central factor guiding outcomes through shared decision-making. For example, if a prolonged recovery over several months toward functional independence is deemed unacceptable by the patient or their surrogate, then favorable neuroprognostic findings—such as minimal injury burden—become less meaningful, particularly in the context of limited neurologic reserve and a likely extended course of recovery.
Components of neuroprognostication.
Limitations and Challenges of Neuroprognostication
Reconciliation between empirical data and individual values
Estimating early and long-term functional outcomes is not a one-size-fits-all process but rather a dynamic synthesis of multiple factors. These include the nature of acute brain injury, the patient’s premorbid state, insights from clinical status and findings from neuroprognostic tests, and ethical considerations, with each factor weighed differently depending on the individual case. The complexity of neuroprognostication following HIBI is further compounded by the significant variability and lack of standardization in current practices despite many different available guidelines. ,,,,, While current guidelines outline the evidence supporting the recommendation of each prognostic tool and highlight their pitfalls, they do not prioritize one over another or specify a recommended sequence of use. Table 1 summarizes the 3 most referenced neuroprognostication guidelines for HIBI; note that American Heart Association 2025 updated version is expected to replace 2020 when published. In practice, clinicians are often faced with the challenge of reconciling conflicting recommendations across existing guidelines and determining how best to apply them to the specific scenario at hand. Even when presented with the same clinical scenario, experts might not agree on their interpretation for the recovery potential and in their final prognostic impressions. Each guideline has their own process for selecting, reviewing, appraising, and synthesizing evidence into a set of recommendations, which include specific norms and boundaries. Most guidelines have limited generalizability, as they are rarely applicable to every possible scenario; their recommendations must remain grounded to the boundaries of available evidence, even while recognizing the disconnect between controlled study conditions and real-world clinical practice. A central practical challenge lies in aligning patient’s values regarding acceptable levels of disability with the interpretation of the findings from neuroprognostic tools, as guided by existing recommendations. While the patient’s values ultimately define what constitutes a favorable outcome or meaningful recovery—for example, what level of disability is personally considered unacceptable, and thus, a poor outcome ?—guidelines lack the nuance to accommodate individualized interpretations on a case-by-case basis. Being confined to the available evidence, the meaning behind “poor,” “unfavorable,” “good,” “favorable,” “meaningful” in guidelines is derived from the approach chosen by investigators conducting studies that informed practice guidelines. The required pragmatic definitions for analyses, often dichotomizing outcomes from functional scales into “good” and “poor” outcomes, are agnostic of individual premorbid status, preferences, and values. In most neuroprognostic studies, a “favorable” functional outcome is defined by the ability to achieve independence (ie, Glasgow-Pittsburgh Cerebral Performance Category, CPC, scores 1–2 or modified Rankin Scale, mRS, scores 0–3), although some studies have lower thresholds characterized by recovery of consciousness (ie, CPC 1–3 or mRS 0–4). For a cardiac arrest survivor whose premorbid baseline included dependence on others for advanced activities of daily living for the last 2 decades, returning to that baseline state might be a meaningful recovery—yet, without this personalized nuance, this individual is automatically allocated to a category of “poor” outcome in most studies (CPC 3). It is, therefore, essential to recognize this practical limitation when applying recommendations in clinical practice—especially during discussions with surrogates. Rather than focusing solely on the binary question, “ Will this patient survive neurologically intact? ,” clinicians should be asking instead: “ How long will recovery take, and to what extent will specific functions be regained? ” and “ What level of disability would be considered meaningful or acceptable for this individual? .” This shift in perspective is particularly important when navigating prognostic uncertainty and reflect the bias of attribute substitution, when using a simpler, related question when trying to tackle a complex problem.
Table 1
Comparison of neuroprognostication guidelines of cardiac arrest
| Guidelines | AHA 2020 Guidelines | ERC-ESICM 2021 Guidelines 1 | NCS 2023 Guidelines |
|---|---|---|---|
| General approach | Delayed, multimodal assessment | Delayed, multimodal assessment | Delayed, multimodal assessment |
| Pupillary light reflex (PLR) | Bilaterally absent PLR at ≥72 h post-ROSC is a reliable predictor for poor outcome | Same as AHA; recommends quantitative pupillometry if available | Same as ERC-ESICM |
| Corneal reflex | Bilaterally absent corneal reflexes at ≥72 h post-ROSC/normothermia support poor prognosis | Same as AHA | Cautions against relying on corneal reflex alone |
| Myoclonus | Status myoclonus within 72 h post-ROSC supports poor prognosis; EEG recommended when myoclonus is present | Same as AHA | Cautions against relying on myoclonus alone for prognostication; EEG required for confirmation |
| MRI | Extensive diffusion restriction 2–7 d post-ROSC supports poor prognosis | Same as AHA | Specifies the diffuse pattern of restricted diffusion should occur bilaterally across anterior and posterior circulation distributions and involve both cortical and subcortical areassss |
| CT | Reduced gray-white matter ratio supports poor prognosis; no specified timing | Same as AHA, but suggests evaluating at 72 h post-ROSC | Recommends evaluating for diffuse loss of gray-white differentiation and sulcal effacement at ≥48 h post-ROSC |
| EEG | Burst-suppression is a reliable predictor; recommended ≥72 h post-ROSC | Burst-suppression is reliable; suppressed background with or without periodic discharges is moderately reliable; recommends >24 h post-ROSC | Same as AHA; also considers suppressed background (± periodic discharges) at ≥72 h a moderately reliable predictor |
| Somatosensory evoked potentials (SSEP) | Evaluate ≥24 h post-ROSC | Same as AHA | Recommends delaying SSEP evaluation until ≥48 h post-ROSC |
| Serum biomarkers (NSE) | High/increasing NSE within 72 h post-ROSC supports poor prognosis | Same as AHA | Cautions against using NSE due to lack of validated threshold and high FPR |
| Indeterminate prognosis | No specific recommendation | Recommends extended observation if aligned with patient’s goals of care | Strongly supports prolonged observation in uncertain cases |
Cognitive biases
While integrating multiple predictive modalities improves prognostic accuracy and reduces uncertainty, it also increases complexity and the risk of cognitive biases driven by heuristics. Neuroprognostication is fraught with uncertainty, from limitations of current prognostic tools, methodology used in neuroprognostic studies, and cognitive biases underlying systematic errors inherent to the process of outcome prediction. Most patients who remain comatose after cardiac arrest ultimately die from neurologic causes: approximately 10% to 20% progress to brain death, ,, while the remaining 80% die following withdrawal of life support based on the perceived poor neurologic prognosis. , Even in the context of clinical trials, deaths related to withdrawal of life support occur prematurely within 72 hours from ROSC. ,
The self-fulfilling prophecy bias is pervasive in neuroprognostic studies, perpetuating the overinflated prediction performance of tools being investigated, when the findings from such tools are used in clinical practice to inform end-of-life decisions. , Acknowledging the role of heuristic judgment is critical in outcome prediction, as withdrawing life support based on an inaccurately perceived poor prognosis can lead to irreversible consequences. Health care providers are better at predicting 6-month mortality than neurologic outcomes ; this can also reflect self-fulfilling prophecy bias, reinforced heuristic judgment, and attribute substitution bias.
Timing and availability of prognostic tools
The predictive performance of various tools for estimating favorable or unfavorable outcomes varies widely. , Ideally, neuroprognostication should incorporate several modalities of tests; however, this is far from the reality: in a study utilizing administrative data, fewer than 10% of patients had at least one and only 2% had at least 2 prognostic tests. A great threat to neuroprognostication is the limited access to recommended tools, which in turn leads to reduced familiarity and potential discomfort with their use and interpretation of findings creating a vicious cycle. The limited access to these tools, coupled with variability in how each provider conducts and interprets the available tests, presents significant risks that must be carefully considered. Future guidelines should account for variations in access to these modalities and provide alternative approaches to ensure equitable and reliable prognostication.
Another important challenge is the variability in the timing of assessments and prognostic conclusions. The timing of testing for neuroprognostic tools depends on each tool and on the intent for maximizing its yield; this is determined often by avoiding the effect of confounding (eg, temperature and sedatives effect) or neurophysiologic stunning from anoxic depolarizations. The ideal timing for testing also hinges on the intent for each specific tool: If capturing the impact of early injury from primary insults (thus, early timing for testing) or capturing the overall burden of HIBI after the cascade of secondary injury is completed (ie, delayed after 48–72 hours from cardiac arrest). Providers report favoring accuracy of predictors over timeliness of prognostic impressions; in fact, the highest false positive rate deemed acceptable among practitioners who perform neuroprognostication in hypoxic-ischemic encephalopathy is just 0.1%. Conversely, there is a concerning trend of clinicians disclosing premature prognostic impressions in practice—often before the guideline-recommended minimum of 72 hours ,,, —and relying on tools that carry an unacceptably high false positive rates (FPR; anchoring effect) or are used outside their optimal time window.
Many threats to the accuracy of neuroprognostication may coexist and have a compounding effect—patients for whom prognostic impressions were premature, are also less likely to have more than one prediction tool used, or the involvement of a neurology consultant. , These findings reinforce the importance of providers anchoring their practices on evidence-based guidelines and refraining from overinflating the value of anecdotal experiences and ingrained clinical practices. Avoiding overconfidence and identifying knowledge gaps is crucial in mitigating biases and improving neuroprognostication accuracy.
Framework of Neuroprognostication
The process of neuroprognostication builds on the priority pillars depicted in Fig. 1 to incorporate them into 5 key steps that follows ROSC as illustrated in Fig. 3 A: (1) evaluating the patient’s neurologic state; (2) assessing the extent of neuronal injury using a multimodal approach combining methods that assess for direct evidence of structural damage and surrogate markers of injury using neurophysiologic tests and clinical examination findings as summarized in Figs. 2 and 3 B; (3) determining cerebral reserve and assessing cerebral resilience; (4) understanding the definition of meaningful recovery from the perspective of the patient; and (5) effectively communicating prognostic impressions with the patient’s family. Definitive prognostic impressions should be delayed until sufficient time has elapsed to account for potential confounders and the characterization of a trajectory (eg, serial examinations are captured following mitigation of confounding factors).
( A ) Key steps of neuroprognostication. ( B ) Timeline of multimodal approach to neuroprognostication. ADC, apparent diffusion coefficient; CTH, CT head; DWI, diffusion-weighted imaging; EEG, electroencephalography; NfL, neurofilament light chain; NSE, neuron-specific enolase; SSEP, somatosensory evoked potentials.
Assessment of neurologic status
Consciousness can be defined by its 2 main components: (a) “wakefulness” which is characterized by behavioral or electrophysiological indicators of arousal; and (b) “awareness” which is identified through responses to internal or external stimuli. Demonstration of awareness is a key point in the trajectory of brain injured patients, most commonly through the ability to follow commands. DoC are conventionally classified into 3 categories.
-
1.
Comatose state: Unresponsive; complete lack of awareness.
-
2.
Unresponsive wakefulness syndrome: Wakeful but unresponsive.
-
3.
Minimally conscious state (MCS): Altered consciousness; minimal but definitive evidence of awareness of self or surroundings. MCS can be further subdivided into MCS + or MCS– based on the presence or absence of complex behaviors, with emergence from MCS, defined as the recovery of functional communication or object use.
Consciousness is primarily assessed through the neurologic examination, therefore, relying on the examiner’s expertise. Distinguishing reflexive behaviors from subtle and inconsistent intentional movements is difficult, even for experienced clinicians. In a study of 137 patients with disorder of consciousness, more than 1 in 3 patients diagnosed with a wakeful unresponsive state had in fact some preservation of awareness when assessed using the Coma Recovery Scale-Revised rendering them in a MCS. In another study, 1 in 5 patients had electrophysiologic or functional neuroimaging signs of awareness despite lack of corresponding behavioral responses, a finding now commonly referred to as cognitive motor dissociation or “covert consciousness.” Commonly used scales for evaluation of coma such as the Glasgow Coma Scale (GCS) and Full Outline of Unresponsiveness Score (FOUR) are in fact crude examinations prone to missing subtle signs of awareness. Emerging technologies provide insights into neuronal activity patterns indicative of consciousness not apparent through traditional examination alone, warranting a shift from rigid classification schemes to a dynamic continuum model of consciousness. This is important as patients with some preservation of awareness such as in cognitive motor dissociation represent a distinct subgroup with a more favorable recovery trajectory compared with those who remain in unresponsive wakefulness state.
Assessment of neuronal injury
In the seconds to minutes following cardiac arrest, a cascade of pathophysiological mechanisms triggered by anoxic depolarizations and calcium influx lead to programmed cell death that is known as “primary brain injury.” Following ROSC, ongoing “secondary brain injury” ensues due to oxidative stress and reactive oxygen and nitrogen species, mitochondrial dysfunction, impaired autoregulation, and microcirculatory failure or “no-reflow,” which can continue for hours to days. The overall burden of HIBI is comprised by both primary and secondary injury ; resilience to primary insult does not necessarily translate to resilience to secondary insults and at present, it is impossible to predict the magnitude of injury burden based on the severity of insults. Most neuroprognostic tools center on the evaluation of HIBI burden directly or indirectly (see Fig. 2 ).
Direct evidence of cerebral injury
Neuroimaging
Goal: Neuroimaging is used to demonstrate the extent of HIBI without the confounding effects from sedation, metabolic derangements, and temperature.
Computed Tomography Imaging
Current Evidence: The most studied method for HIBI quantification on computed tomography (CT) is gray-white matter ratio (GWR) ; however, evidence arises mainly from out-of-hospital cardiac arrest and predominantly within 24 hours from ROSC. The prognostic performance of GWR as a predictor of unfavorable outcomes increases when imaging is delayed due to increased sensitivity, likely from the ability to capture secondary injury. Most data on GWR are derived from retrospective observational studies with various methods for calculation, imaging protocols, timing of imaging acquisition, and thresholds for best prognostic performance. The most recent prospective study demonstrated specificities of 100% for all GWR models, and sensitivities ranging from 30% to 41%, yielding the automated method area under the curve (AUC 0.84) when using a cut-off less than 1.10.
Advantage: CT is widely available and is often used to evaluate cerebral edema in addition to gauging cerebral reserve through evaluation of parenchyma volume, areas of encephalomalacia or prior injury, burden of microvascular or white matter disease.
Disadvantage: The most reliable and objective measure of HIBI is GWR, which is not performed routinely in clinical practice. Visual assessment alone is subject to significant inter-rater variability. ,
Recommendation: Diffuse loss of gray-white differentiation, ideally using quantitative methods, is helpful characterizing HIBI burden as early as within 2 hours from ROSC; however, delayed imaging is needed to capture overall burden of injury. There is insufficient evidence on GWR to support its use in predicting patients with a favorable outcome.
MRI
Current Evidence: Conventional MRI sequences used to detect HIBI include diffusion-weighted imaging (DWI) and apparent diffusion coefficient (ADC). In DWI sequences, the degree of brain injury is characterized by the restriction in diffusion of water across the bilateral cortex and deep gray matter as represented through the image intensity. This restriction in water diffusion representing cytotoxic edema can be quantified through reporting of brain volumes with ADC below specific thresholds (650 × 10 −6 mm 2/s or 450 × 10 −6 mm 2/s). Other less studied sequences have lack validation and are not recommended by guidelines. Most data are derived from observational studies with significant heterogeneity on imaging protocols, subjective definitions for HIBI burden, different thresholds and regions of interest, and timing of assessments and outcomes.
Advantage: MRI provides more granular evaluation of structural injury with a higher sensitivity for detection than CT, particularly in the posterior fossa and brainstem.
Disadvantage: Several factors restrict the use of MRI, including incompatible metal devices and medical instability in critical care patients. Moreover, conduction of MRI is time-consuming, and it may not be readily available in many facilities. Qualitative assessments also have high inter-rater variability, and quantitative assessments are not widely performed in clinical practice.
Recommendation: Extensive areas of restriction on DWI or ADC when performed within 2 to 7 days after ROSC have been associated with unfavorable outcomes. , Reviewing areas of injury can be helpful in predicting specific corresponding deficits (eg, cortical blindness and sensory ataxia when extensive primary sensory cortex is involved) and discussing the recovery journey with surrogates. Conversely, the absence of extensive injury on MRI can be helpful in identifying patients with low HIBI burden.
Biomarker-based Assessment
Goal: Chemical biomarkers of neuronal injury are used to assess structural HIBI without the confounding effect of temperature, sedating medications, and/or toxic-metabolic insults.
Current Evidence: The quantification of serum biomarkers in HIBI has explored the role of NSE, glial fibrillary acidic protein (GFAP), neurofilament light chain (NfL), tau, ubiquitin carboxyl hydrolase L1, and serum calcium binding protein 100β (S-100B); each biomarker has distinct kinetics that influence temporal trends and optimal testing time and interpretation. The most widely studied and used in practice, NSE, is a biomarker of neuron cell-body injury, and has been demonstrated to remain elevated for up to 5 days after ROSC after its peak at 48 to 72 hours due to its half-life of 24 to 30 hours. , A systematic review of 86 biomarker studies found that NfL, a marker of white matter injury and axonal damage, had the highest predictive value for unfavorable prognosis with an AUC of 0.92 (95% CI, 0.84–0.97). However, large-scale prognostic studies are needed before their use can be adopted widely. ,,
Advantages: Chemical biomarkers lack the influence of common confounders such as sedatives, and can be easily evaluated in studies that blind the clinical team to their results (therefore, mitigating confirmation biases). Its use.
Disadvantages: Heterogeneity of studies including type of biomarker, assay used, timing of assessments, and thresholds for evaluation of prognostic performance leads to significant imprecision in data; therefore, no specific recommendations can be made on the best modality and how they should be used. Most biomarkers are not available in clinical practice outside of research realm, or whenever available, their turn-around time of several days hinder its utility during the first week of hospitalization. Additionally, NSE, the most commonly used biomarker in HIBI, is susceptible to confounders by extra-cerebral sources such as neuroendocrine tumors or hemolysis; the latter is particularly relevant in patients with extracorporeal membrane oxygenation. The most promising biomarker from recent literature, NfL, requires ultra-sensitive assays due to limited diffusion through blood-brain barrier.
Recommendation: Very high levels (insufficient data precludes specific cut-offs) of NSE and NfL within 72 hours from ROSC support significant structural HIBI. Conversely, sustained normal levels of NSE (<17 μg/L), ,, NfL, and GFAP within 72 hours from ROSC may be helpful in identifying patients without significant HIBI burden.
Indirect evidence of neuronal functional impairment
Stay updated, free articles. Join our Telegram channel
Full access? Get Clinical Tree





