This 2025 post–cardiac arrest care (PCAC) guideline from the American Heart Association (AHA) is based on the expert writing group review of the relevant International Liaison Committee on Resuscitation (ILCOR) Consensus on Science With Treatment Recommendations (CoSTR) documents and the studies included in the systematic reviews, as well as new evidence updates conducted by the writing group. The writing group discussion and evidence reviews were conducted within the context of the clinical environments in which out-of-hospital and in-hospital resuscitations occur, with special consideration for the health care professionals who use these PCAC guidelines.

The PCAC Focused Update Writing Group included a diverse group of experts with backgrounds in emergency medicine, neurocritical care, interventional and critical care cardiology, psychology, and pulmonary critical care. Group members were appointed by the AHA Emergency Cardiovascular Care Science Subcommittee and approved by the AHA Manuscript Oversight Committee. Writing group members were selected to represent diverse backgrounds in clinical medicine and research expertise as well as to form a group that was institutionally diverse.

The AHA has rigorous conflict of interest policies and procedures to minimize the risk of bias or improper influence during the development of guidelines. Before appointment, writing group members disclosed all relevant commercial relationships and other potential (including intellectual) conflicts. These procedures are described more fully in “Part 2: Evidence Evaluation and Guidelines Development” of the 2020 American Heart Association Guidelines for Cardiopulmonary Resuscitation and Emergency Cardiovascular Care.¹ Appendix 1 of this document lists the writing group members’ relevant relationships with industry.

The writing group members evaluated the current list of patient, intervention, comparison, and outcome questions included in current PCAC guidelines. Patient, intervention, comparison, and outcome questions with novel evidence were revisited by the writing group. For each targeted patient, intervention, comparison, and outcome question, writing group members created a search strategy or utilized a previously created ILCOR search strategy when available. Search strategies were internally peer reviewed. This search was executed in Medline and Excerpta Medica Database, using the Ovid search interface, and the Cochrane Central Register of Controlled Trials. Final searches were executed in November 2024. Search results were not limited by language or year. Two writing group members performed dual screening of the titles and abstracts of all articles identified from each search and identified articles for full text review. Screening conflicts were resolved between the 2 writing group members and writing group leadership before full text review. Two writing group members reviewed the full text of all selected articles and applied the information contained to develop treatment recommendations appropriate for each clinical question. Each draft recommendation was created by a group of 2 or 3 writing group members and then reviewed and refined by all writing group members during regular virtual meetings. The final manuscript was reviewed and approved by all writing group members.

As with all AHA guidelines, each recommendation in this focused update is assigned a Class of Recommendation (COR) based on the strength and consistency of the evidence, alternative treatment options, and the impact on patients and society (Table 1). The Level of Evidence (LOE) is based on the quality, quantity, relevance, and consistency of the available evidence. For each recommendation, the writing group discussed and approved specific recommendation wording and the COR and LOE assignments. In determining the COR, the writing group considered the LOE and other factors, including systems issues, economic factors, and ethical factors like equity, acceptability, and feasibility. These evidence-review methods, including specific criteria used to determine COR and LOE, are described more fully in “Part 2: Evidence Evaluation and Guidelines Development” of the 2025 guidelines.1 The writing group members had final authority over and formally approved these recommendations.

The guidelines are organized into knowledge chunks, grouped into discrete modules of information on specific topics or management issues.² Each modular knowledge chunk includes a table of recommendations that uses standard AHA nomenclature of COR and LOE. A brief synopsis is provided to put the recommendations into context with important background information and overarching management or treatment concepts. Recommendation-specific supportive text clarifies the rationale and key study data supporting the recommendations. When appropriate, additional tables are included.

This 2025 document updates the recommendations for the use of MCS, vasopressor choice, ventilatory goals, coronary angiography and percutaneous coronary intervention (PCI), temperature control, seizure management (with a new section on postarrest myoclonus), and survivorship after cardiac arrest (Figure 1). This 2025 update is based on the evidence identified in systematic reviews performed by ILCOR and this writing group addressing novel data that has been published since the formal release of the 2020 AHA advanced life support guidelines for cardiopulmonary resuscitation and the 2023 AHA Focused Update on Adult Advanced Cardiovascular Life Support.^1,3

These guidelines were submitted for blinded peer review to 5 subject matter experts nominated by the AHA. Before appointment, all peer reviewers were required to disclose relationships with industry and any other conflicts of interest, and all disclosures were reviewed by AHA staff. Peer reviewer feedback was provided for guidelines in draft format and again in final format. All guidelines were reviewed and approved for publication by the AHA Science Advisory and Coordinating Committee and the AHA Executive Committee. Comprehensive disclosure information for peer reviewers is listed in Appendix 2.

Delivery of oxygen to the brain and other vital organs is critical in resuscitation. There are theoretical risks associated with both hyperoxemia and hypoxemia.^1,2 Hyperoxemia may contribute to the development of reactive oxygen species in ischemic tissue, worsening oxidative injury and cell death.³ Hypoxemia may cause additional injury to end-organs and exacerbate the ischemic injury caused by the initial cardiac arrest.⁴ Limitations in the ability to accurately measure oxygen saturation after cardiac arrest can occur due to technological limitations in the prehospital setting and poor perfusion.⁵ Additionally, pulse oximetry may be less accurate in patients with darker skin pigmentation.^{6, 7}

Recommendation-Specific Supportive Text

1. In settings where accurate measures of SpO₂ or PaO₂ cannot be obtained, maintaining a FiO₂ of 100% until a reliable measure of oxygenation is obtained can help avoid hypoxemia. Administration of 100% oxygen immediately after ROSC until oxygen saturation can reliably be measured is supported by the EXACT (Reduction of Oxygen After Cardiac Arrest) randomized trial.⁵ This study found an increase in hypoxemic events in the prehospital setting when Fio₂ was adjusted to maintain lower as compared to a higher SpO₂ target.⁵
2. Hypoxia increases the risk of end-organ damage, and hypoxemia is a widely available surrogate for hypoxia. Observational studies are mixed, with at least one study showing worse outcomes in patients experiencing post-ROSC hypoxemia.⁸ These observational data are supported by the EXACT randomized trial, which demonstrated that lower oxygen targets in the prehospital and emergency department setting led to more hypoxemic episodes and lower survival.^5,9
3. Several clinical trials have compared different SpO₂ and PaO₂ targets after ROSC in a variety of clinical settings with no superior or inferior target identified.^5,10,11An SpO₂ target of 90% to 98% (PaO₂ target of 60–105 mm Hg) reflects the ranges studied in key randomized trials. In a prespecified subgroup of patients with hypoxic-ischemic encephalopathy in one clinical trial, patients randomized to a conservative oxygen therapy arm emphasizing protocolized reduction of FiO₂ had better outcomes than those in a usual care arm.¹² Observational studies have found an association between severe hyperoxemia (PaO₂ less than 250 mm Hg) and worse outcomes.¹³
4. Patients with darker skin pigmentation may develop “occult hypoxemia,” wherein the SpO₂ on pulse oximeter reads in the normal range but the PaO₂ is in the hypoxemic range.^7,14-16

Management of carbon dioxide during the postarrest phase is necessary given its role in regulating cerebral blood flow. Hypocapnic-induced vasoconstriction may lead to cerebral ischemia, while hypercapnia may exacerbate elevated intracranial pressure (ICP) through increased cerebral blood volume associated with vasodilation. 

Recommendation-Specific Supportive Text

1. Although a strategy of mild permissive hypercapnia (target PaCO₂, of 50–55 mm Hg) might improve brain perfusion, observational studies have demonstrated mixed results on patient outcomes.^{8, 17-19} The TAME (Targeted Therapeutic Mild Hypercapnia after Resuscitated Cardiac Arrest) randomized trial compared the maintenance of normocapnia (PaCO₂, 35–45 mm Hg) to mild hypercapnia (PaCO₂, 50–55 mm Hg) for 24 hours in patients who remained comatose after cardiac arrest and found no difference in neurologic outcome at 6 months or in other secondary outcomes.²⁰ Although there was no outcome difference overall, the intervention in the TAME trial was stopped more often in the mild hypercapnia arm, possibly in response to severe acidemia because the protocol recommended suspending the intervention if pH <7.1 and base excess <−6 mmol/L. As a result, maintaining the PaCO₂ within a normal physiologic range (generally, 35–45 mm Hg) is supported. Whether targets should be individualized based on individual patient physiology remains unknown. 
2. 2. The ability to measure PaCO₂ after arrest is limited without blood gas measurement, and observational studies^21-24 suggest that end-tidal carbon dioxide levels may be falsely low and may not accurately reflect PaCO₂ levels in post–cardiac arrest patients.

Seizures and status epilepticus are common acute neurologic complications in the post–cardiac arrest period, occurring in 10% to 35% of patients who do not follow commands after ROSC.^1-6 Cerebral hyperexcitability after cardiac arrest may exacerbate a mismatch between neuronal bioenergetic supply and demand, thereby contributing to secondary brain injury.⁵ Post–cardiac arrest hyperexcitability can manifest as a wide range of electroclinical findings, from seizures with overt clinical manifestations, such as convulsions to periodic or rhythmic discharges on EEG and other abnormal EEG patterns.⁷ Clinical or electrographic findings can manifest with or without impairment of consciousness and may or may not reach the criteria for seizures or status epilepticus (Table 2).⁸

The indications for and intensity of antiseizure medications vary in clinical practice and across studies. Although the occurrence of status epilepticus in post–cardiac arrest patients has been associated with poor outcomes in observational studies,^2,9,10 some patients survive with recovery of functional independence. Myoclonus is covered in a separate section because only a subset of patients with post–cardiac arrest myoclonus meets criteria for electroclinical seizures or status epilepticus. Marked heterogeneity in the definitions of seizures and status epilepticus across studies challenges the interpretation of available data.

1. In patients who are unable to follow commands after cardiac arrest, EEG can detect nonconvulsive seizures and status epilepticus (Table 3).^11-13 This recommendation is informed by the high prevalence of nonconvulsive seizures and other epileptiform activity in cardiac arrest survivors.⁵ However, there is no direct evidence that detection of nonconvulsive seizures with EEG improves outcomes. An ILCOR systematic review from 2023 did not specifically address the timing and method of EEG monitoring in this setting.¹⁴
2. There are no nonrandomized trials or RCTs comparing the effect of treating clinical seizures versus no seizure treatment.¹⁴ There is no widely accepted definition for clinically apparent seizure or status epilepticus after cardiac arrest, and previous literature has used a variety of clinical and EEG classifications. It remains unclear which seizures or status epilepticus represent an epiphenomenon of severe brain injury or treatment-responsive targets to prevent secondary brain injury. Despite the lack of high-certainty evidence, untreated clinically apparent seizure activity is thought to be potentially harmful to the brain; therefore, treatment of seizures or status epilepticus is recommended.¹⁵
3. Seizures or status epilepticus that have no clearly apparent clinical correlate, (ie, electrographic or nonconvulsive seizures or status epilepticus), emerge in 1% to 20% of post–cardiac arrest patients.¹⁶ The American Clinical Neurophysiology Society has developed specific criteria for electrographic or nonconvulsive seizures or status epilepticus for critically ill patients (Table 2), though prior studies have used varying definitions. Given the limited evidence to support withholding treatment of nonconvulsive seizures after arrest, a similar approach to nonconvulsive seizure treatment for other etiologies is reasonable. The TELSTAR (Treatment of Electroencephalographic Status Epilepticus after Cardiopulmonary Resuscitation) trial¹⁷ is the first RCT of protocolized tiered treatment targeting suppression of electrographic seizures, status epilepticus, or interictal patterns detected on continuous EEG in adults after arrest compared with standard care (Table 2). The trial randomized 172 subjects and the use of antiseizure medication was allowed but discouraged in the standard care group. The rate of unfavorable neurological outcome (defined as Cerebral Performance Category score of 3–5) did not differ at 3 months, but only 10% of patients in both treatment groups had electrographic seizures (the rate of clinically apparent and electroclinical seizures were not reported). Although the trial was not powered for subgroup analyses, patients with electrographic seizures (ie, periodic discharge frequencies surpassing 2.5 Hz), evolving patterns, or those with nongeneralized periodic discharges between 0.5 and 2.5 Hz had better outcomes with protocolized, tiered antiseizure treatment than standard care alone.
4. There are several approaches to EEG monitoring that vary in duration (ie, from short 20- to 40-minute recordings to continuous monitoring for several days) and electrode arrangement (ie, from full 21 electrodes to simplified 6–10 electrode montages). Seizures, epileptiform abnormalities, and clinical myoclonus may occur immediately after ROSC or emerge several days after initial resuscitation.^7,12,18 Continuous EEG after cardiac arrest, although more costly and labor intensive, increases sensitivity for seizures, status epilepticus, and epileptiform activity detection when compared with brief intermittent recordings given the variability in the timing these patterns occur and frequent lack of clinical correlation. However, use of continuous EEG was not associated with improvement in survival or functional outcome in observational cardiac arrest studies^19,20 or the CERTA (Continuous EEG Randomized Trial in Adults) trial,²¹ a multicenter pragmatic study involving critically ill patients with impaired consciousness, of whom nearly one-third had been post–cardiac arrest patients.
5. Cardiac arrest as the etiology of seizures and status epilepticus has been an exclusion criterion in RCTs specific to status epilepticus^22,23; consequently, therapeutic algorithms are extrapolated from other settings, including guidelines for generalized convulsive status epilepticus. The 2023 ILCOR CoSTR summary¹⁴ recommended that seizures be treated when diagnosed in postarrest patients. No specific antiseizure medications were recommended.
6. The ACNS defines the ictal-interictal continuum as rhythmic or periodic patterns that may be consistent with possible nonconvulsive seizures or status epilepticus even without fulfilling strict electrographic criteria (Table 2).⁸ Patients with patterns on the ictal-interictal contiuum who exhibit both electrographic and clinical response to a therapeutic trial with a loading dose of a nonsedating antiseizure medication (ie, not benzodiazepines) are considered to have electroclinical status epilepticus; thus, therapeutic trials of antiseizure medication may be considered for interictal findings. Clinical response to a therapeutic trial may not be apparent on bedside examination in patients who are unresponsive to commands after ROSC, therefore determination of an electroclinical response to a therapeutic trial may be challenging.
7. No new studies were identified in the updated ILCOR CoSTR summary from 2023.¹⁴ Seizure prophylaxis did not improve outcomes after cardiac arrest in 2 prospective RCTs^24,25 and 1 nonrandomized prospective clinical trial with historical controls.²⁶ Prophylaxis was also not effective in preventing subsequent seizures in the post–cardiac arrest period.^24-26 However, these studies lacked systematic diagnosis of seizures with EEG monitoring, seizures reported were abstracted from clinical documentation, and the sample size was insufficiently powered for this outcome. Additionally, these studies were designed as therapeutic trials of neuroprotection with thiopental, or magnesium and diazepam, or diazepam alone with the primary outcome being functional neurologic outcome with the Cerebral Performance Category scale. Therefore, their study designs are inadequate to provide definitive evidence on the role of seizure prophylaxis in the prevention of acute symptomatic seizures and later development of epilepsy following cardiac arrest. Furthermore, the chosen therapeutic agents in these studies^24-26 are not commonly used as seizure prophylaxis in clinical practice in the current era, thus precluding extrapolation of findings on seizure prophylaxis with other agents that might have a more favorable side effect profile with fewer sedating effects.

Myoclonus is a common clinical manifestation of hypoxic-ischemic brain injury, being observed in approximately 18% to 34% of post–cardiac arrest patients.^1-5 Myoclonus can present as diffuse or multifocal muscle jerks, or more subtle focal motor movement involving the face and limbs.^6,7 Up to half of patients with myoclonus are found to have electrographic seizures or status epilepticus, and the clinical exam alone is insufficient to predict which patients are having electrographic seizures.^8-10 Myoclonus with a time-locked EEG correlate is a seizure (Table 2), although it is controversial whether this is an epiphenomenon of hypoxic-ischemic brain injury or if it is a seizure pattern that should be treated to prevent secondary brain injury.^8,10,11 Independent of the presence of a time-locked EEG correlate, myoclonus may interfere with supportive care and ventilatory mechanics.^12,13

Post–cardiac arrest studies are limited by substantial heterogeneity in definitions and terminology used for describing and classifying myoclonus and its associated EEG findings, including the presence of electrographic seizures and status epilepticus or epileptiform discharges or other electrographic events time-locked to the myoclonus^8,11,14-16 (Table 3).

1. There is no specific clinical exam pattern, timing of occurrence, or myoclonus pattern that is reliably associated with the presence or absence of seizures or status epilepticus on EEG after cardiac arrest.^4,6-8 Therefore, prompt EEG monitoring is necessary to diagnose whether myoclonus is associated with seizures, status epilepticus, or the presence of other EEG correlates (Table 3).¹⁶ Back averaging of EEG and electromyography signals may be necessary to determine the precise relationship between EEG events (eg, spikes, bursts) and myoclonus.^8,17 Myoclonus can cause significant muscle artifact on EEG which can be overcome with neuromuscular blockade under close monitoring.
2. There is no randomized or nonrandomized study evaluating the role of pharmacologic interventions for the management of myoclonus without an EEG correlate after cardiac arrest. Similarly, there is no evidence implicating myoclonus without an EEG correlate in the pathogenesis of secondary brain injury after cardiac arrest.¹⁸ Myoclonus after cardiac arrest is often refractory to treatment, and multiple antiseizure medications or anesthetic agents may be needed for myoclonus control.^{2,5,7,8,13,18-24} High risk of bias related to patient selection and confounding effects, and inconsistent reporting of myoclonus and EEG characteristics in the literature preclude a specific treatment recommendation for myoclonus without an EEG correlate.¹⁴ The risk of cardiovascular, neurological, and other medication side effects outweighs the unknown benefit of suppressing myoclonus without an EEG correlate on patient outcomes.

Hypoxic-ischemic brain injury is the leading cause of morbidity and mortality in patients who achieve ROSC after OHCA and accounts for a significant portion of unfavorable outcomes after resuscitation from in-hospital cardiac arrest.^1,2 Most deaths attributable to brain injury after cardiac arrest are due to withdrawal of life support based on a predicted unfavorable neurological outcome. Prognostication is critically important as surrogates commonly make decisions relying on the perceived estimated likelihood of recovery that is shared by the treatment team. Accurate neuroprognostication is important to avoid inappropriate withdrawal of life support in patients who may otherwise achieve meaningful recovery and to avoid ineffective treatment when poor outcome is inevitable (Figure 2).³

Neuroprognostication frequently incorporates assessment of structural injury (eg, neuroimaging and biomarkers) or neurological function (eg, neurophysiologic tests and clinical examination). The ideal timing for neuroprognostic tests avoids the effects of confounding (eg, from temperature, medications, and systemic organ dysfunction). Multiple tests are incorporated to estimate the likelihood for recovery to a favorable or unfavorable state. Most neuroprognostic studies utilize dichotomized outcome scales, with a favorable functional outcome defined by the ability to achieve independence (ie, Glasgow-Pittsburgh Cerebral Performance Category scores of 1–2 or modified Rankin Scale scores of 0–3), although some studies have thresholds characterized by recovery of consciousness (ie, Cerebral Performance Category score of 1–3 or modified Rankin Scale score of 0–4). These studies do not capture granular details on the functional state of patients, nor do they consider their individual preferences or values, limiting their utility. Further guidance on best practices on the interpretation of outcomes with a patient-centered approach is included in “Part 3: Ethics.”

Neuroprognostication relies on interpreting the results of tests and correlating those results with outcome. Given that a false positive test for unfavorable neurological outcome could lead to inappropriate withdrawal of life support from a patient who otherwise would have recovered, specificity is the most important test characteristic when predicting unfavorable outcome. In some instances, prognostication and withdrawal of life support may appropriately occur earlier because of nonneurologic disease, catastrophic brain herniation, patient’s goals and wishes, or clearly nonsurvivable situations. Many of the tests considered are subject to error because of the effects of medications, organ dysfunction, and temperature. Importantly, many published neuroprognostic studies are impacted by self-fulfilling prophecy bias, given that practices of neuroprognostication are inherently part of PCAC, and in turn, outcomes are modulated by such practices. Furthermore, research studies have methodological limitations including small sample sizes, single-center design, lack of blinding, failure (or inability) to account for unmeasured factors that impact the pace of recovery, and the use of outcome at hospital discharge rather than a time point associated with maximal recovery (typically 3–6 months after arrest).³ Additionally, prognostic tests may yield discordant or inconclusive results, and in such instances, continued observation and serial assessment remains a treatment option.

Recommendation-Specific Supportive Text

1. Any single method of neuroprognostication has an intrinsic error rate and may be subject to confounding, therefore multiple modalities should be used to improve decision-making accuracy.^4-9 The overall certainty in the evidence of neuroprognostication studies is low due to inherent biases that limit the internal validity of studies and limited generalizability that challenge their external validity. Neuroprognostication approaches employing multimodal testing aim at mitigating error rates; concordant results increase certainty when predicting outcomes.^{3,5,6,8,10-19}
2. Early withdrawal of life support based on presumed poor neurological prognosis remains frequent, despite the risk for unreliable outcome prediction within 72 hours after ROSC.^20,21 Sedatives and neuromuscular blockers may be metabolized more slowly in post–cardiac arrest patients, and injured brains may be more sensitive to the depressant effects of various medications. Residual sedation or paralysis can confound the accuracy of clinical examinations.²²
3. Establishing clear and cohesive communication with families is a key aspect of PCAC. Communication should be implemented early and maintained longitudinally, addressing major needs including the uncertainty of neurological prognosis.^23,24 Difficulties in family-team communication dynamics may contribute to premature withdrawal of life support.²⁵ Discordance in goals of care between clinicians and families/surrogates has been reported in more than 25% of critically ill patients.²⁶ Regular multidisciplinary meetings may help improve communication.
4. Operationally, the timing for prognostic impressions is typically at least 5 days after ROSC for patients treated with hypothermic temperature control (approximately 72 hours after normothermia) and is conducted under conditions that minimize the confounding effects of sedating medications. Each neuroprognostic test has a specific time window in which it has optimal test performance. Prognostic impressions comprise results of tests integrated into the multimodality assessment synthesized at least 72 hours after normothermia.^11,13,27-35

Historically neuroprognostic studies evaluated findings from clinical examination that correlated with unfavorable neurological outcome. More recently, the focus of neuroprognostication after cardiac arrest has shifted to include prognostication for both favorable and unfavorable outcome. Given that a test may perform better in predicting unfavorable but not favorable outcome, or vice versa, specific recommendations are made depending on the finding and the outcome (ie, favorable versus unfavorable outcome) with which it has correlated. Serial clinical examinations in comatose survivors of cardiac arrest assess the level of consciousness, pupillary light and corneal reflexes, and best motor response to noxious stimulation. The presence of myoclonus during the first week after cardiac arrest is also noted. Myoclonus describes repetitive jerk-like involuntary movements that can be generalized, focal, multifocal, synchronous, or asynchronous in their involvement of body parts, and may or may not have an electroclinical correlate with discharges on EEG (refer to the Seizures, Myoclonus, and Other Epileptiform Activity section). Clinical examination findings that correlate with poor outcome are subject to confounding by temperature and medications.

Recent ILCOR systematic reviews included studies regardless of temperature control, and findings were correlated with neurological outcome at time points ranging from hospital discharge to 12 months after arrest.^10,36 A 2023 systematic review specifically evaluated the prognostic value of different test modalities for the prediction of favorable outcome.³⁷

1. Twenty-seven retrospective studies demonstrated that absent pupillary light reflex evaluated immediately after ROSC and up to 7 days after arrest predicted unfavorable neurological outcome with specificity ranging from 48% to 100%^{5-8,11,27,28,38-57}; these studies support the reliability of pupillary light reflex as a prognostic tool.^4,58,59 Studies with earlier time points for assessments of pupillary light reflex had lower specificity^43,46; the highest specificity was seen at time points ≥72 hours after arrest^{5,6,8,11,40,42} and when combined with absent corneal reflexes.^5,7,8,42 The false positive rates of bilaterally absent pupillary light reflex, when assessed at least 72 hours after arrest, in predicting unfavorable outcome were consistently low across studies. However, pupillary light reflex is vulnerable to the effect of medications and ambient light, subjective interpretation of the examiner, rigor of technique for assessment, individual factors, and self-fulfilling prophecy bias.
2. Quantitative pupillary light reflex of 0% is an objective finding that describes bilateral absence of pupillary light reflex as assessed by pupillometry.⁴⁹ The automated quantification of pupillary reactivity with pupillometry yields measurements of amplitude, and velocity of constriction and dilatation. Four retrospective studies evaluated quantitative pupillary light response (ie, the percentage of reaction to light),^49,60-62 and 9 evaluated NPi (reactivity index from comparison with normal responses)^{13,18,49,62-67} at time points ranging from immediately after ROSC⁶⁵ to 72 hours after arrest. Quantitative responses of 4% to 5% had higher sensitivity in predicting unfavorable neurologic outcome within 72 hours after arrest while maintaining 100% specificity. The NPi ≤2 threshold at 24 hours demonstrated 100% specificity for unfavorable neurological outcome in most studies.^{13,18,49,66,67} NPi is nonspecific and may be affected by medications; thus, an absolute NPi cutoff that predicts unfavorable outcome remains unknown.^{4,13,18,49,59,62-67}
3. Eighteen studies evaluated the bilateral absence of corneal reflexes at time points ranging from immediately after ROSC to 7 days after arrest with specificity ranging from 25% to 100%^{5,7,8,27-29,35,40,42-45,48,50,53,54,68,69}; these studies support the reliability of corneal reflex as a prognostic tool. Studies with higher specificity evaluated corneal reflexes at time points ≥72 hours after arrest or in combination with bilaterally absent pupillary light reflexes.^5,7,8,42 Corneal reflexes are subject to confounding by medications and rigor of technique of the examiner, and few studies specifically evaluated the potential of residual medication effect or described the type of stimuli used for testing.
4. Marked heterogeneity in study definition of myoclonus after cardiac arrest challenge the interpretation of this finding as a neuroprognostic tool.⁷⁰ Eleven observational studies evaluated the presence of myoclonus (not otherwise specified) within 96 hours after arrest with specificities for unfavorable neurological outcome ranging from 77.8% to 100%,^{11,13,27,29-35,68} yielding false positive rates up to 22.2% (and upper bound of 95% CI of 60%).⁴ Six studies evaluated status myoclonus within 72 hours from ROSC with specificities for unfavorable outcome ranging from 94% to 100%.^{6,7,33,53,71,72} There were methodological limitations in all studies, including a lack of standard definitions, lack of blinding, incomplete data on EEG correlates, and the inability to differentiate between myoclonus subtypes. This imprecision may result in erroneous estimates of unfavorable outcome if undifferentiated myoclonus is used as a prognostic marker. Given these limitations in the interpretation of literature and the need for characterization of electroclinical subtypes of myoclonus after cardiac arrest, an expert consensus suggested avoiding using the term status myoclonus.⁷³
5. Historically, the best motor examination in the upper extremities has been used as a prognostic tool, with extensor or absent motor responses being correlated with unfavorable outcome. Previous literature was limited by methodological concerns, including confounding for effects of hypothermic temperature control and medications, as well as self-fulfilling prophecy bias. The literature was so imprecise with higher-than-acceptable false positive rates (up to 75.1%)^4,7,58,59 that the finding of absent or extensor response as best motor response was considered potentially harmful when used alone for predicting a unfavorable neurological outcome.⁷⁴ Conversely, multimodal neuroprognostic studies using the Glasgow Coma Motor subscore as an entry point have demonstrated sensitivity ranging from 23% to 71%, while maintaining high specificity.^5-9

1. Two retrospective studies evaluated the predictive performance of present pupillary light reflex to support a prognosis for favorable neurological outcome, when evaluated alone¹² or in combination with present corneal reflexes,¹⁶ from admission¹⁶ to up to 3 days following normothermia and discontinuation of sedatives.¹² Sensitivity when testing after mitigation of confounders reached 97.3%; however, specificities only surpassed 80% when both corneal and pupillary reflexes were present early after ROSC.¹²
2. Four observational studies evaluated the prediction performance of best motor responses (Glasgow Coma Motor subscores) from admission to 96 hours after arrest to support a prognosis of favorable neurological outcome.^7,12,16,75 Sensitivity ranged from 11.6% to 79.26% with specificity ranging from 71.7% to 97.7%; higher specificities were seen with higher motor scores (withdrawal or localizing responses, Glasgow Coma Motor subscores of 4 or 5, respectively).³⁷ This recommendation is supported by a 2022 ILCOR systematic review.³⁶
3. Two retrospective studies evaluated the prediction performance of pupillometry when assessed at admission to up to 72 hours after arrest to support a prognosis for favorable neurological outcome.^12,13 In a post hoc analysis of the largest (n=456) neuroprognostic study evaluating the usefulness of pupillometry in cardiac arrest, the lowest NPi measured within 3 days after arrest in patients with a favorable outcome was 4.1.¹³ The ideal thresholds and timing of testing for pupillometry derived metrics that predict favorable outcome is unknown.
4. Two retrospective studies evaluated the prediction performance of the presence of corneal reflexes to support a prognosis for favorable neurological outcome, when evaluated alone¹² or in combination with present pupillary light reflexes,¹⁶ from admission¹⁶ to up to 3 days following normothermia and discontinuation of sedatives.¹² Sensitivity when testing after mitigation of confounders reached 93.6%; however, specificities only surpassed 80% when both corneal and pupillary reflexes were present <72 hours after ROSC.¹²

Serum biomarkers are blood-based tests that measure the concentration of proteins normally found in the central nervous system. These proteins are absorbed into blood in the setting of neurological injury, and their serum levels reflect the degree of brain injury. Limitations to their prognostic utility include variability in testing methods on the basis of site and laboratory, between-laboratory inconsistency in levels, susceptibility to additional uncertainty due to hemolysis, and potential extracerebral sources of the proteins. The 2020 and 2022 ILCOR systematic reviews evaluated studies that obtained serum biomarkers within the first 7 days after arrest and correlated serum biomarker concentrations with favorable and unfavorable neurological outcomes.^4,37 Other testing of serum biomarkers, including testing levels over serial time points after arrest, was not evaluated. The 2 most studied biomarkers were NSE and S100 calcium-binding protein B (S100B). Additional evidence for NfL in predicting favorable neurological outcomes and levels of other blood biomarkers for predicting favorable neurological outcomes has accumulated since these reviews. Prospective validation of different biomarkers and thresholds for both favorable and unfavorable neurological outcome in cohorts more broadly representative of resuscitated cardiac arrest patients is of high clinical importance.

Recommendation-Specific Supportive Text

1. For prediction of unfavorable outcome, 17 observational studies evaluated NSE collected within 72 hours after cardiac arrest,^{4,13,18,72,76-86} and threshold levels for unfavorable outcome ranged from 17 to 106 μg/L with specificity for poor outcome of 44% to 100%. The evidence is limited because of a lack of blinding, laboratory inconsistencies, a broad range of thresholds needed to achieve 100% specificity, and imprecision. Data from only 2 observational studies assessing NfL for unfavorable outcome prediction were included in the 2020 ILCOR systematic review.^87,88 In the interim, 9 additional observational studies evaluated NfL within 72 hours after arrest for predicting unfavorable outcome.^81,82,89-95 Multiple studies directly compared NfL to NSE for prognostication of unfavorable outcome and found both improved sensitivity with NfL while maintaining high specificity. Many of the included studies are secondary analyses of RCTs. Strengths of these studies include multiple sites, blinding of clinical staff, and standardized, blinded outcome assessment, but the trial populations may not reflect the overall post–cardiac arrest population. SIMOA (Quanterix) was the most used assay. Values may differ among commercially available assays.⁹⁶ Studies reported multiple thresholds and collected samples at different times. In the post arrest period, the optimal threshold and timing for both NSE and NfL levels are unknown, but very high levels may inform multimodal prognostication.
2. For predicting favorable outcome, the 2022 systematic review identified normal levels of NSE (<17 μg/l)="" between="" 24="" and="" 72="" hours="" as="" a="" predictor="" of="" favorable="" neurological="">³⁷ Specificity was >80% in the included studies. The biomarkers S100B, NfL, ubiquitin carboxyl-terminal hydrolase-L1 (UCH-L1), glial fibrillary acidic protein (GFAP), and tau were also included. NfL and GFAP had the highest specificity for favorable outcome but had relatively fewer studies compared with NSE. Interval searches identified 5 studies of NSE^{8,16,76,97,98} 1 of NfL,⁹⁰ and 1 of both GFAP and tau⁹⁹ for predicting favorable neurological outcome. Specificity for favorable neurological outcome was >80% for each biomarker. Different threshold values and sample timing were reported between studies. While the evidence for favorable outcome prediction with biomarkers is less established than for prediction of unfavorable outcome, sustained low levels of biomarkers, in particular, NSE, may be an encouraging finding. Specificity may be lower when predicting favorable neurological outcome due to competing risks of nonneurological organ dysfunction or failure.
3. In addition to 2 prospective and 2 retrospective studies identified in the 2020 ILCOR systematic review,⁴ 3 observational studies evaluated S100B levels within the first 72 hours after arrest.^84,85,93 The maximal level that correlated with unfavorable outcome ranged broadly depending on the study and the timing when it was measured after arrest. Interval studies do not clarify the optimal timing or threshold value to inform prognostication. Only 1 study of GFAP¹⁰⁰ and tau¹⁰¹ were identified. Seven observational studies investigated other biomarkers, including UCH-L1, GFAP, and serum and cerebrospinal fluid tau.^{89,93,95,99,102-104} In studies that included multiple biomarkers, performance was not consistently better than NSE. Given the low number of studies, the certainty of evidence was low and these biomarkers could not be recommended for clinical practice.

Electroencephalography is widely used in clinical practice to evaluate cortical brain activity and diagnose seizures. Its use as a neuroprognostic tool for predicting both unfavorable and favorable neurological outcome is promising, but the literature is limited by several factors, including lack of standardized terminology and definitions, relatively small sample sizes, lack of blinding, subjectivity in the interpretation, and lack of accounting for effects of medications. There is also inconsistency in definitions used to describe specific findings and patterns. The EEG patterns that were evaluated in the 2020 ILCOR systematic review for predicting unfavorable neurological outcome include unreactive EEG, epileptiform discharges, seizures, status epilepticus, burst suppression, and “highly malignant” EEG.⁴ Unfortunately, studies define “highly malignant” EEG differently or imprecisely, limiting use of this finding. Use of standardized terminology from the ACNS avoids this heterogeneity and limits potential for misclassification. Several EEG patterns were observed in a 2022 systematic review for predicting favorable neurological outcome.³⁷ While studies had several operational definitions, most classifications assessed EEG continuity, voltage, and absence of epileptiform discharges.

Somatosensory-evoked potentials (SSEPs) evaluate the integrity of the dorsal column-lemniscal sensory pathway and are subject to less interference from medications than other modalities. In the context of neuroprognostication after cardiac arrest, SSEPs are obtained by stimulating the median nerve and evaluating for the presence of cortical N20 potentials, which are generated in the primary somatosensory cortex and demonstrate integrity of the thalamocortical pathway involved in cortical arousal and consciousness. The reliability of this modality is limited by technical challenges, including operator skill, interrater agreement, and impairment of signal to noise ratio from muscle and environmental contamination. Recommended technical standards are available.¹⁰⁵

1. No additional studies have been identified since the publication of the 2020 ILCOR systematic review.⁴ The 2020 systematic review included observational studies focused on electrographic seizures, though some studies also included convulsive seizures and the definitions for seizures were nonstandardized across studies. The specificity of unequivocal seizures meeting ACNS critical care EEG criteria was 100%; however, the sensitivity of this finding was poor (0.3%–26.8%), and other studies that were not included in the review found patients with post arrest seizures who had favorable outcomes.^31,106,107 Additional methodological concerns include selection bias for which patients underwent EEG monitoring and inconsistent definitions of seizure. Further, critical care EEG terminology has been updated since the publication of these studies.¹⁰⁸ The term seizure encompasses a broad spectrum of pathologies that likely have different prognoses, ranging from a single brief electrographic seizure to refractory status epilepticus, and this imprecision justified the more limited recommendation.
2. In the 2020 ILCOR systematic review, the specificity of status epilepticus for unfavorable outcome ranged from 82.6% to 100%.⁴ No additional studies were identified that reported sensitivity or specificity of this finding. One retrospective study reported electrographic characteristics of patients with definitive or possible status epilepticus.¹⁰⁹ Of the 64 subjects included, 5 survived 30 days after arrest and 1 had a favorable outcome. The most common cause of death in decedents was withdrawal of life-sustaining therapies which occurred a median of 8 days (IQR [7–11]) after arrest. Interestingly, although status epilepticus is a severe form of seizures, the specificity of status epilepticus for unfavorable outcome was less than reported in the studies examining the seizures overall (as above). Additional concerns within the studies include the inconsistent definition of status epilepticus, lack of blinding, and the use of status epilepticus to justify withdrawal of life-sustaining therapies leading to potential self-fulfilling prophecies. The 72 hour timing is based on most studies reporting presence of status epilepticus within this period and case reports of good outcome when status epilepticus is successfully treated within this period. 
3. The term highly malignant EEG was inconsistently applied in studies included in the 2020 ILCOR systematic review.⁴ Twelve studies evaluated burst suppression or suppression on EEG monitoring.^{6,8,9,40,69,72,81,110-114} Across these studies, highly malignant EEG was classified as burst suppression or suppression according to the ACNS critical care EEG terminology.^108,115 Specificity ranged from 90.7% to 100%, and sensitivity was 15% to 51%. The potential for self-fulfilling prophecies, interrater disagreement, and lack of control for medication effects limited the ability to make a stronger recommendation, despite the overall high specificity.¹¹² Further, in a preplanned analysis of the TTM2 trial, specificity of burst suppression or suppression for unfavorable outcome was lower (93%) compared to studies with specific centralized or expert EEG review.¹¹⁴ Additional studies identifying subtypes of burst suppression, such as synchronous or identical bursts, are needed.¹¹⁶ Burst suppression can be caused by medications, so it is particularly important that clinicians have knowledge about the potential effects of medication on this prognostic tool.
4. The 2020 ILCOR systematic review considered discharges as either rhythmic/periodic or nonrhythmic/periodic.⁴ Sensitivity and specificity of rhythmic/periodic patterns for unfavorable outcome ranged from 0.5% to 50.8% and 77.7% to 100%, respectively, with wide confidence intervals. As the time from the cardiac arrest to test increased, the specificity of rhythmic/periodic discharges for unfavorable outcome improved. Nonrhythmic/periodic patterns had a similar range of sensitivity (0.5%–38.5%) and specificity (84.6%–100%). One observational study reported outcomes in patients with stimulus induced rhythmic or periodic patterns recorded greater than 36 hours after arrest.¹¹⁷ Of the 142 included patients, 20 (14%) had stimulus induced rhythmic or periodic patterns recorded on EEG monitoring. The rate of unfavorable outcome was not different between patients with (n=13, 65%) or without stimulus induced rhythmic or periodic patterns (n=88, 72%) (P=0.596). The literature of both rhythmic/periodic and nonrhythmic/periodic discharges are limited by lack of accounting for effects of medications and variable timing of EEG monitoring. Low specificity currently limits the use of these findings to support an unfavorable prognosis.
5. Bilateral absence of N20 peaks following median nerve stimulation is the most studied and well-accepted neuroprognostic predictor for SSEP testing. Twenty observational studies^{5-8,11-13,27,35,38,39,41-43,47,49,50,55,57,69,72,118-130} evaluated bilateral absent N20 peaks assessed up to 6 days after arrest with specificity ranging from 50% to 100%. Six studies had specificity below 100%,^{11,27,47,50,118,123} and additional methodological limitations included lack of blinding and potential for self-fulfilling prophecy bias. Newer studies evaluated the unilateral absence of N20 combined with amplitude thresholds which may increase sensitivity for detection of unfavorable outcome.^6,57,131,132 While studies evaluated SSEPs obtained at any time starting immediately after ROSC, there is a high likelihood of potential confounding factors early after cardiac arrest, leading to the recommendation that SSEPs should only be obtained greater than 48 hours after arrest.
6. A preplanned analysis of the TTM2 trial data reported the prognostic value of unreactive EEG between 24 hours to 14 days.¹¹⁴ Specificity for unfavorable neurological outcome was 60% with a sensitivity of 79%. The study relied on local EEG reviewers to assess reactivity rather than a centralized expert review. Specificity was lower than studies included in the 2020 ILCOR review, which ranged from 41.7% to 100% but was below 90% in most studies.⁴ One study assessed EEG reactivity during hypothermia and normothermia intending to assess agreement for reactivity between raters and sensitivity for unfavorable outcome for reactivity to different stimulations.³⁵ Percent agreement ranged from 58% to 70% and sensitivity ranged from 52% to 86% depending on stimulation type and whether assessment occurred during hypothermia or normothermia. Thus, the overall certainty of the evidence was rated as very low. There is interest in automated reactivity assessment and stimulation paradigms.

1. The 2022 systematic review for prediction of favorable outcome found a continuous EEG background without discharges within 72 hours after cardiac arrest to predict a favorable neurological outcome.³⁷ Specificity ranged from 51% to 100%, with most studies reporting a specificity >80%. Specificity was higher in earlier (12 and 24 hours) compared to later timepoints (48 and 72 hours). One additional observational study evaluated continuous normal voltage background without discharges on EEG 24 hours after cardiac arrest.¹⁶ Sensitivity and specificity for favorable neurological outcome at 6 months after arrest were 64% and 89%, respectively. When compared to unfavorable outcome, the specificity for predicting favorable outcome is overall less given the competing risk of nonneurological mortality and morbidity.
2. Quantitative characteristics of the N20 signal are of interest as prognostic indicators and there are several methods for calculating the amplitude of the N20 wave. The threshold of >4 µV when measured as the peak-to-trough difference between the peak of the N20 wave and the trough of the subsequent P25 wave has achieved specificity above 90% with sensitivity above 40% in most studies using this finding to predict favorable neurological outcome.^16,36,37,72 One study found that patients sedated with midazolam had lower amplitudes (measured as N20/P25 peak-to-peak amplitude) when compared to propofol.¹⁴ The ideal amplitude thresholds, measurement methods, and timing of N20 wave amplitude testing for SSEP that predict favorable outcome is unknown.

Neuroimaging may be helpful to detect and quantify structural brain injury after cardiac arrest. The most common modalities are CT and magnetic resonance imaging (MRI). The use of quantitative analysis modalities assessing the burden of hypoxic-ischemic brain injury in clinical practice is limited by the availability of quantitative methods outside research settings, while qualitative imaging results are often subjective and have lower sensitivity.^133,134 On CT, cytotoxic brain edema can be quantified as the gray-white ratio (GWR), defined as the ratio between the density (measured as Hounsfield units) of gray matter and white matter. Normal brain has a GWR of approximately 1.3. In the setting of cerebral edema, GWR declines, either due to decreased density of gray matter (representing cytotoxic edema) or decreased attenuation of white matter. On MRI, cytotoxic brain edema can be measured as restricted diffusion on diffusion-weighted imaging (DWI) and can be quantified by the apparent diffusion coefficient (ADC). The ADC is a sensitive measure of injury, with normal values ranging between 650 and 1000 × 10⁻⁶ mm²/s and values decreasing with injury. Brain injury findings from CT and MRI evolve over the first several days after arrest, so the timing of the imaging study of interest is particularly important as it relates to prognosis.

1. Forty-one observational studies^{6,17,55,56,68,121,125,133-166} evaluated GWR—the most commonly studied method for quantification of hypoxic-ischemic brain injury on brain CT.¹⁶⁷ Whole-brain GWR (GWR average) and GWR in specific regions were evaluated with high specificity in predicting unfavorable neurological outcome; lower prediction performance was seen when using thalamus as a region of interest and with higher cutoffs for GWR. Thirty-four of the studies evaluated CT obtained within 24 hours after arrest,^{6,17,55,56,68,121,133-139,141-143,145,147-153,155,156,158-161,163-166} and 6 studies included brain CTs obtained up to 10 days after arrest.^{125,140,146,154,157,162} In retrospective studies, the predictive performance improved when images were obtained at 24 to 72 hours^{134,140,151,155} and up to 10 days from ROSC.^154,162 There were methodological limitations, including selection bias, risk of multiple comparisons, and heterogeneity of measurement techniques, such as anatomic sites, equipment, and calculation methods. Thus, specific methods for calculation and thresholds for GWR, and ideal timing for CT acquisition that predict unfavorable prognosis with highest specificity remains unknown. Fewer studies have evaluated CT metrics as predictors of death by neurological criteria^148,150; external validation and reproducibility of methods are lacking.
2. Ten observational studies^{16,55,69,110,168-173} investigated DWI changes on MRI qualitatively within 2 to 7 days after arrest for unfavorable outcome prediction. The studies evaluated MRI qualitatively and definitions of relevant findings differed between studies, such as use of “high signal intensity” and “positive findings.” In some studies, only specific brain regions were examined, with or without using a standardized scoring system. Moreover, studies are limited by subjective thresholds and lack of blinding. The specificity ranged from 55.7% to 100%. In the context of multimodal prognostication, an extensive burden of reduced DWI on MRI may be correlated with unfavorable prognosis, but a broader recommendation could not be supported. The imprecise definition and short-term outcome in some studies led to significant uncertainty about how to use qualitative DWI MRI to predict unfavorable prognosis.
3. Ten observational studies investigated quantitative ADC on MRI within 2 to 7 days after arrest for unfavorable prognosis prediction.^{17,69,158,164,170,171,174-177} The studies used custom-made computational tools to quantitatively determine the whole brain or region of interest volume below a specific ADC threshold for poor outcome prediction. The ADC thresholds used, ADC calculation methods used, and regions included for analyses varied among studies, with cutoffs for unfavorable outcome with 100% specificity using 650 × 10⁻⁶ mm²/s or 450 × 10⁻⁶ mm²/s varying from 20.5% to 42.2% and 1.3% to 5.2%, respectively. There is no standardization of quantitative measurements of ADC in cardiac arrest and such analyses require expertise in advanced computational methods; hence, there are concerns about feasibility and reliability. In the context of multimodal prognostication, an extensive burden of decreased ADC on MRI may be correlated with unfavorable prognosis. A specific ADC threshold for global or regional brain injury that predicts unfavorable prognosis is not known.

1. Three observational studies evaluated the role of GWR on brain CTs performed within 6 hours up to 36 hours after arrest to identify patients with favorable outcomes.^56,142,158 In the largest cohort, patients who regained consciousness before hospital discharge had a median GWR 1.38. The ideal thresholds and methods of calculation for GWR, and timing for CT acquisition that predict favorable prognosis are unknown.^36,37
2. Six studies investigated the absence of qualitative DWI abnormalities within 2 to 7 days from arrest as a predictor of favorable outcome.^{16,45,168-171} These studies had a sensitivity for favorable outcome prediction of 51% to 100%, but there was heterogeneity of DWI assessment methods and which regions were evaluated. In the context of multimodal prognostication, the absence of injury using DWI on MRI may be correlated with a favorable prognosis, but imprecise definitions and criteria for brain MRI assessment led to uncertainty about how to use DWI MRI to predict favorable prognosis.

• Karen G. Hirsch, MD, Chair
• Edilberto Amorim, MD
• Patrick J. Coppler, PA-C, MSPAS
• Ian R. Drennan, ACP, PhD
• Andrea Elliott, MD
• Alexandra June Gordon, MD
• Jacob C. Jentzer, MD, MS
• Nicholas J. Johnson, MD
• Ari Moskowitz, MD, MPH
• Bryn E. Mumma, MD, MAS
• Alexander M. Presciutti, PhD, MS
• Amber J. Rodriguez, PhD
• Albert Yen, MD
• Jon C. Rittenberger, MD, MS, Vice Chair