The transition from post–cardiac arrest resuscitation to neuroprognostication is not a transition from treatment to observation. The team must continue preventing secondary injury while determining whether impaired consciousness reflects irreversible hypoxic–ischemic injury, a reversible confounder, or an injured brain whose trajectory has not yet declared itself. Neuroprognostication is therefore not a collection of abnormal tests. It is a timed clinical synthesis designed to estimate long-term neurologic function while making a falsely pessimistic prediction exceptionally unlikely. That asymmetry matters because withdrawal of life-sustaining therapy accounts for many deaths attributed to post–cardiac arrest brain injury. A falsely optimistic prediction may prolong treatment; a falsely pessimistic prediction can eliminate the recovery it incorrectly declared impossible (Hirsch et al., 2025; Rajajee et al., 2023).

Three questions must remain separate. The first is whether the patient has a treatable cause of impaired consciousness, such as residual sedation, seizures, metabolic encephalopathy, shock, infection, hypercapnia, or a new structural lesion. The second is the neurologic forecast attributable specifically to hypoxic–ischemic brain injury. The third is whether the forecasted range of outcomes would be acceptable to the patient given prior function, comorbid illness, values, and treatment preferences. A patient can have a poor overall prognosis because of refractory cardiogenic shock or advanced malignancy without having a demonstrably poor neurologic prognosis. Conversely, a patient may have severe irreversible brain injury despite recovered cardiac and pulmonary function. Conflating these questions converts a values-based treatment decision into an apparently objective neurologic conclusion.

Determination of death by neurologic criteria is also a separate pathway. A patient who develops catastrophic cerebral edema, herniation, and loss of all brain function should be evaluated under the applicable protocol once prerequisites are satisfied; there is no requirement to wait for the conventional post-arrest prognostication window. For every other comatose survivor, early examination and testing can identify emergencies, guide treatment, and establish a trajectory, but should not be converted prematurely into a definitive forecast.

The familiar “72-hour rule” is a minimum boundary, not an appointment at which prognosis automatically becomes knowable. The Neurocritical Care Society recommends deferring formal prognostication until at least 72 hours after return of spontaneous circulation in patients not treated with hypothermia and until at least 72 hours after rewarming when hypothermia was used. The 2025 American Heart Association guideline similarly recommends consolidating the multimodal impression no earlier than 72 hours after normothermia, after sedatives have been discontinued and adequately cleared. In a patient cooled below normothermia and then rewarmed, this commonly places the formal synthesis near day five or later. European guidance applies its formal algorithm at 72 hours or later after return of spontaneous circulation, provided major confounders have been excluded. Persistence of coma at that point is the indication to prognosticate; it is not itself evidence of an unfavorable outcome (Hirsch et al., 2025; Nolan et al., 2025; Rajajee et al., 2023).

The physiologic clock and the calendar clock often diverge. Drug clearance depends on the medication, dose, duration, context-sensitive half-time, active metabolites, temperature course, organ function, age, body composition, and interactions. Midazolam and its metabolites may accumulate substantially; opioids suppress arousal and alter pupillary behavior; hypothermia reduces drug metabolism; and shock impairs hepatic and renal clearance. Neuromuscular blockade invalidates motor and corneal testing. No universal number of half-lives resolves every case. The infusion history and observed trajectory should be reconstructed explicitly, with pharmacy input when clearance is uncertain. Flumazenil or naloxone should not be used indiscriminately merely to manufacture a prognostic examination because withdrawal, seizures, sympathetic activation, recurrent pain, and loss of ventilator synchrony remain possible.

Other confounders deserve the same discipline. Ongoing seizures, postictal suppression, sepsis, hepatic or uremic encephalopathy, severe dysnatremia, hypoglycemia, hypercapnia, shock, hypoxemia, hypothermia, and delirium can delay behavioral recovery. Their presence does not make all testing useless: an adequately performed somatosensory evoked potential may remain interpretable when sedation prevents a motor examination, and imaging may still demonstrate diffuse structural injury. It does mean that a modality should be interpreted according to its own susceptibility to the active confounder rather than allowing the clock alone to certify validity.

Daily examinations should begin immediately. Serial assessment can identify improvement, evolving edema, recurrent seizures, or a new focal deficit long before the formal synthesis. The examination should document spontaneous eye opening, visual fixation or tracking, reproducible command following, best motor response in each limb, pupillary size and reactivity, corneal reflexes, oculocephalic or vestibulo-ocular responses when appropriate, cough, respiratory drive, and the phenotype of any myoclonus. Temperature, hemodynamics, oxygenation, carbon dioxide, glucose, sedative exposure, neuromuscular blockade, and major metabolic abnormalities should be recorded alongside the findings. The relevant observation is not merely the best Glasgow Coma Scale score entered during a shift but whether a behavior is reproducible and demonstrates cortical processing. Reflex withdrawal can be mistaken for localization, triple flexion for purposeful movement, eyelid opening for command following, and motor impairment from peripheral or orthopedic injury for complete cerebral unresponsiveness.

Arrest duration, initial rhythm, age, lactate, pH, epinephrine exposure, and estimated time to return of spontaneous circulation establish pretest probability but cannot independently determine neurologic outcome. No validated no-flow or low-flow duration makes recovery impossible in every patient, and recorded times are frequently estimates. A nonshockable rhythm may reflect prolonged untreated arrest but may also accompany a rapidly reversible respiratory, toxic, or obstructive cause. Older age changes physiologic reserve and the distribution of achievable outcomes, but age alone is not a reliable neurologic predictor. These variables belong in the clinical context, not in place of direct assessment of brain structure and function (Rajajee et al., 2023).

Among bedside findings, bilaterally absent pupillary light responses at or beyond the appropriate 72-hour boundary are the strongest established clinical predictor of an unfavorable outcome, provided the finding is accurate, unconfounded, and consistent with the rest of the case. Manual assessment is vulnerable to ambient light, small pupils, examiner error, ophthalmic surgery, ocular trauma, topical anticholinergic exposure, and nebulized medication reaching one eye. Up to one third of pupils judged nonreactive manually in one cited cohort were reactive on quantitative testing. Quantitative pupillometry reduces measurement error but does not make interpretation autonomous.

The predefined BOX substudy prospectively evaluated 710 out-of-hospital cardiac arrest survivors. A Neurological Pupil Index of 2 or lower had a zero observed false-positive rate at the measured time points from admission through 72 hours, and a quantitative pupillary light response below 4% had a zero observed false-positive rate from 24 through 72 hours. In patients still unresponsive at 72 hours or later, adding pupillometry to neuron-specific enolase improved sensitivity while preserving a zero observed false-positive rate. These are important validation data, but they come from a selected, predominantly shockable-rhythm cohort using a proprietary device and prespecified thresholds; they do not transform one pupillometer value into a universal standalone verdict (Nyholm et al., 2024). Preserved pupillary responses are also not the mirror-image prediction: pupils frequently remain reactive in devastating cortical injury, so their presence alone does not establish favorable recovery.

Corneal reflexes require more restraint. The 2025 AHA guideline allows bilateral absence at 72 hours or later to support an unfavorable prognosis when combined with other tests, and the European algorithm incorporates combined absence of pupillary and corneal responses as one predictor. The Neurocritical Care Society concluded that absent corneal reflexes alone are not sufficiently reliable because clinically important false-positive predictions remain. Technique, residual neuromuscular blockade, topical anesthetic, ocular disease, and examiner variability all matter. An absent corneal response can strengthen concordant evidence of widespread injury; it should not independently trigger withdrawal of support.

An absent or extensor motor response at 72 hours is likewise not a reliable standalone predictor. Sedation, critical illness neuropathy or myopathy, spinal cord disease, peripheral injury, and motor-pathway damage can produce a grim motor examination without proving loss of the neural substrates required for consciousness. Withdrawal or localization is encouraging and can support the possibility of favorable recovery, but neither guarantees future independence. Reproducible command following ends the coma-prognostication pathway because the patient has awakened, yet it does not exclude substantial deficits in memory, executive function, language, vision, mood, or motor control.

Early myoclonus is a persistent source of erroneous pessimism. Generalized, multifocal, focal, synchronous, and asynchronous jerks may represent electroclinical seizures, cortical myoclonus without an evolving seizure, subcortical myoclonus without a cortical correlate, or a mixture that changes over time. Neither myoclonus within 48 to 72 hours nor the label “myoclonic status” is sufficiently reliable without EEG characterization. Some patients with early myoclonus recover meaningful function, particularly when the EEG background becomes continuous and organized. Prompt EEG is therefore essential. Suppressing movements without an EEG correlate solely to normalize the patient’s appearance provides no demonstrated neurologic benefit, although treatment may still be necessary when jerks impair ventilation, cause injury, or interfere with care (Hirsch et al., 2025).

EEG contributes two distinct streams of information: it detects treatable seizures and describes the functional state and trajectory of cortical networks. These functions should not be confused. A seizure can arise from a continuous background with meaningful recovery potential or from persistent suppression reflecting severe injury. Background continuity, voltage, organization, evolution, superimposed discharges, and the effects of temperature and medication are therefore more informative than a binary statement that seizures are present or absent.

A continuous or nearly continuous background without epileptiform discharges during the first 72 hours supports favorable recovery. Earlier return of continuity is generally more encouraging than late recovery, but no favorable EEG phenotype guarantees functional independence. Persistent suppression or burst suppression after confounders have cleared is a moderately reliable marker of unfavorable outcome. Under American Clinical Neurophysiology Society terminology, suppression means that nearly all activity remains below 10 microvolts; burst suppression consists of alternating bursts and suppression occupying a substantial proportion of the record. A discontinuous background is not automatically burst suppression. The AHA and Neurocritical Care Society place the greatest weight on these patterns when they remain present around 72 hours or later without sedative or hypothermic effects. The European framework permits a highly malignant EEG recorded after 24 hours to enter its algorithm but delays the final multimodal conclusion until at least 72 hours. These approaches should not be hybridized by adopting the earlier European timing while omitting its requirement for another concordant predictor.

The need for multimodality is reinforced by a prespecified TTM2 substudy of 845 patients whose EEGs were obtained a median of 71 hours after arrest. Highly malignant suppression or burst suppression predicted poor six-month outcome with 50% sensitivity and 93% specificity, not 100% specificity; combining the pattern with absent reactivity increased specificity to 97%. Prognostication was multimodal and withdrawal was prohibited before 96 hours, yet self-fulfilling bias could not be eliminated. The study demonstrates that real-world local interpretation of a highly malignant EEG is powerful but not infallible (Turella et al., 2024).

Absent EEG reactivity is less dependable than commonly assumed. Stimulation protocols are inconsistent, interrater agreement is imperfect, and sedatives can attenuate responses. The 2025 AHA guideline specifically advises against using absent reactivity within 72 hours to support an unfavorable prognosis. Isolated rhythmic or periodic discharges are also insufficiently specific, and status epilepticus is not uniformly fatal. Seizures emerging from a recovering continuous background require a different interpretation from discharges embedded in persistent suppression. Anesthetic-induced suppression cannot be used to prove endogenous suppression; the complete medication record must accompany prognostic EEG interpretation.

Median-nerve somatosensory evoked potentials interrogate another biological axis. Bilateral absence of cortical N20 responses, with preserved responses at Erb’s point and preserved cervical potentials, is among the most reliable predictors of unfavorable outcome. The AHA and Neurocritical Care Society favor testing at 48 hours or later; the European algorithm accepts testing from 24 hours onward. The recording must be technically adequate and reproducible. Electrical and muscle artifact, incorrect electrode placement, peripheral nerve injury, cervical disease, severe hypothermia, and failure to establish preserved subcortical conduction can create a false cortical absence. A brief neuromuscular blocker may improve signal quality when muscle artifact is prohibitive. Sedatives ordinarily affect SSEPs less than the clinical examination or EEG, making the test especially useful when drug clearance remains uncertain. Presence of an N20 response does not prove intact consciousness networks, while higher N20 amplitudes may support favorable recovery but are not sufficiently standardized to dictate prognosis.

Serum biomarkers appear objective but carry hidden subjectivity through assay selection, sampling time, specimen quality, and threshold choice. Neuron-specific enolase rises with neuronal injury, but hemolysis, extracerebral sources, and between-laboratory differences can distort the result. The obsolete claim that one value above 33 micrograms per liter invariably predicts poor outcome should not be used. European guidance retains NSE above 60 micrograms per liter at 48 or 72 hours as one adverse predictor and treats a rising trajectory as additional concern, but that threshold belongs to a specific two-predictor framework. The Neurocritical Care Society concluded that NSE measured within 72 hours should not be used alone until a consistent threshold is validated. The 2025 AHA guideline permits high NSE or neurofilament light within 72 hours to support an unfavorable prognosis only in combination with other findings and permits normal NSE to support favorable recovery, while acknowledging that universal thresholds remain unsettled (Hirsch et al., 2025; Nolan et al., 2025; Rajajee et al., 2023).

The practical synthesis is to use laboratory-specific methods, inspect hemolysis indices, repeat measurements, assess trajectory, and never allow one biomarker value to override discordant clinical, electrophysiologic, or imaging evidence. Neurofilament light may ultimately improve sensitivity, and its inclusion in the AHA guideline reflects accumulating observational evidence, but limited availability, platform-dependent values, variable sampling schedules, and insufficiently standardized thresholds still constrain routine use. The 2025 European guideline does not recommend it for routine prognostication. S100B, GFAP, tau, and UCH-L1 remain investigational for independent bedside decisions.

Imaging estimates structural injury on a different timetable. A head CT obtained immediately after return of spontaneous circulation is invaluable for identifying hemorrhage, large infarction, trauma, mass effect, or another cause of arrest, but a normal early scan does not predict favorable recovery because cytotoxic edema may not yet be visible. When CT later demonstrates diffuse loss of gray–white differentiation with sulcal effacement across bilateral vascular territories and involving both cortex and deep gray nuclei, it supports severe global injury. The Neurocritical Care Society considers this diffuse pattern on CT obtained at least 48 hours after return of spontaneous circulation a moderately reliable predictor. Quantitative gray–white ratios may improve objectivity in research, but scanner characteristics, reconstruction, regions of interest, contrast exposure, and thresholds vary enough that an unvalidated numerical cutoff should not supersede expert qualitative interpretation. If the early CT is unrevealing and the patient remains comatose, delayed repeat imaging can expose edema that had not yet evolved.

MRI is more sensitive to cytotoxic injury but is not automatically definitive. The most informative window is generally two to seven days after return of spontaneous circulation. Diffuse bilateral restricted diffusion involving cortex and deep gray structures across vascular territories is a moderately reliable adverse predictor, particularly when it agrees with malignant electrophysiology or absent brainstem reflexes. Focal or regional cortical restriction is different: seizures, embolic infarction, hypoglycemia, hyperammonemia, and other toxic-metabolic injuries can create diffusion abnormalities without representing uniform global destruction. Motion artifact and incomplete sequences also matter. Conversely, absence of substantial cortical and deep-gray diffusion restriction during the appropriate window supports the possibility of favorable recovery but cannot exclude microscopic, selectively vulnerable, or network-level injury. Quantitative apparent-diffusion-coefficient volume thresholds remain research tools rather than standardized bedside criteria (Hirsch et al., 2025; Rajajee et al., 2023).

Multimodal synthesis means more than counting abnormal reports. The 2025 European algorithm considers an unfavorable outcome likely in a patient who remains unable to follow commands at 72 hours or later, after confounders have been excluded, when at least two of six categories are present: combined absence of pupillary and corneal reflexes, bilaterally absent N20 responses, a highly malignant EEG, NSE above the specified threshold, early status myoclonus, or diffuse extensive hypoxic–ischemic injury on CT or MRI. The Neurocritical Care Society instead grades individual predictors by reliability: bilaterally absent pupils at the appropriate time and bilaterally absent N20 responses are reliable, whereas diffuse delayed CT or MRI abnormalities and a late nonconfounded suppressed or burst-suppressed EEG are moderately reliable. The AHA emphasizes concordant multimodal assessment without making one finding dispositive. These frameworks share a principle but are not interchangeable recipes.

Avoid pseudo-multimodality. Pupillary and corneal findings belong to the same examination domain, and the European algorithm counts their combined absence as one predictor. Repeated NSE values remain one biomarker domain. CT and MRI may be two views of the same structural process. Suppression and absent reactivity on one EEG are correlated rather than independent. The strongest combinations cross mechanisms: brainstem function through pupils, thalamocortical conduction through SSEP, cortical network activity through EEG, neuronal injury through biomarkers, and tissue structure through imaging.

Discordance is clinically valuable rather than inconvenient. Bilaterally absent manually assessed pupils in a patient who localizes, has a continuous EEG, preserved N20 responses, and no diffuse MRI injury should trigger repeat assessment with pupillometry and investigation for ocular or medication effects. A highly malignant EEG obtained during midazolam therapy should be repeated after clearance. Diffuse MRI restriction paired with an unexpectedly favorable examination should prompt review of timing, distribution, seizure burden, and alternative metabolic causes. One unfavorable predictor coexisting with several favorable indicators is not equivalent to two concordant unfavorable predictors. When the data conflict, the correct prognosis is usually indeterminate.

Prediction should also search actively for recovery potential rather than merely fail to prove devastation. Withdrawal or localization, a continuous EEG background without discharges, low or decreasing NSE, preserved and sometimes robust N20 responses, and absence of cortical or deep-gray diffusion restriction on appropriately timed MRI can reduce uncertainty. A systematic review found these signals predicted good outcome with greater than 80% specificity and greater than 40% sensitivity in most included studies, but definitions and thresholds varied and most studies had moderate or high risk of bias. Preserved pupils or corneal reflexes alone are insufficient favorable predictors because they commonly survive severe cortical injury (Sandroni et al., 2022).

The 72-hour boundary does not close the recovery window. In the evidence synthesized by the Neurocritical Care Society, 10% to 22% of survivors not treated with hypothermia and 10% to 19% of those treated with hypothermia awakened after the corresponding 72-hour boundary; among late awakeners, 67% to 88% achieved a good long-term functional outcome. Awakening beyond two and even four weeks has been reported. These observations are vulnerable to differences in cohort selection and definitions of awakening, but they directly refute use of persistent coma alone as an adverse predictor (Rajajee et al., 2023).

A 2026 prospective study provides useful, though not definitive, evidence with less early withdrawal bias. It followed 101 patients across eight German hospitals who remained comatose at 72 hours and did not undergo withdrawal of life-sustaining therapy during the first four weeks. Repeated assessments reduced false-positive classifications. Two or more unfavorable markers by 14 days produced no false positives for poor 12-month outcome in that cohort, but sensitivity remained below 22%, confidence intervals were wide, enrollment occurred from 2014 through 2017, and treating teams were not blinded to all results. Adding the best Coma Recovery Scale–Revised score within 14 days improved model discrimination, supporting structured behavioral trajectory as a promising adjunct in indeterminate cases. The study strengthens the rationale for time and reassessment; it does not establish a universal 14-day rule or replace current guideline algorithms (Goetze et al., 2026).

When reliable adverse predictors are absent or discordant, the correct conclusion is “indeterminate,” not “good” and not “hopeless.” Recovery, if it occurs, may unfold over days to months and may include substantial disability. If continued treatment remains consistent with the patient’s goals, an extended observation period or time-limited trial can permit further sedation clearance, serial examinations, repeat EEG, delayed imaging, and emergence of a behavioral trajectory. Tracheostomy, enteral access, or transfer to a rehabilitation-capable setting may function as bridges through uncertainty rather than declarations that indefinite life support has been chosen.

Communication should begin before formal prognostication and recur as evidence matures. Families should hear what is known, what remains confounded, which tests are planned, what each test can and cannot establish, and when synthesis will occur. At the formal meeting, distinguish independent evidence supporting recovery or nonrecovery from contextual factors and values. The usual Cerebral Performance Category and modified Rankin Scale groupings compress cognition, personality, fatigue, language, executive function, and caregiver burden into coarse bins. Predicting survival with dependence is not the same as predicting permanent unconsciousness, and neither establishes whether the outcome would be acceptable to this patient.

Finally, prognosis and decision must remain conceptually distinct. Prognostication estimates the probability and range of future neurologic states. A treatment decision integrates that estimate with the patient’s values, acceptable level of dependence, extracerebral prognosis, burdens of continued therapy, and applicable ethical and legal standards. A poor neurologic forecast does not itself issue an order to withdraw treatment, just as uncertainty does not mandate treatment without limit. Defensible neuroprognostication after cardiac arrest requires adequate time, cleared confounders, technically sound testing, genuinely independent multimodal concordance, serial trajectory, explicit uncertainty, and a values-based decision made only after those elements have been separated and then deliberately brought together.

References