Electroencephalography and delirium in the postoperative period

Abstract

Delirium commonly manifests in the postoperative period as a clinical syndrome resulting from acute brain dysfunction or encephalopathy. Delirium is characterized by acute and often fluctuating changes in attention and cognition. Emergence delirium typically presents and resolves within minutes to hours after termination of general anaesthesia. Postoperative delirium hours to days after an invasive procedure can herald poor outcomes. Easily recognized when patients are hyperactive or agitated, delirium often evades diagnosis as it most frequently presents with hypoactivity and somnolence. EEG offers objective measurements to complement clinical assessment of this complex fluctuating disorder. Although EEG features of delirium in the postoperative period remain incompletely characterized, a shift of EEG power into low frequencies is a typical finding shared among encephalopathies that manifest with delirium. In aggregate, existing data suggest that serial or continuous EEG in the postoperative period facilitates monitoring of delirium development and severity and assists in detecting epileptic aetiologies. Future studies are needed to clarify the precise EEG features that can reliably predict or diagnose delirium in the postoperative period, and to provide mechanistic insights into this pathologically diverse neurological disorder.

Editor's key points.

• •Delirium is an acute and fluctuating change in attention and cognition that is a common neurological abnormality in postoperative and intensive care unit patients.
• •Emergence delirium after general anaesthesia presents early and resolves quickly, whereas postoperative delirium presenting later can signal poor outcome.
• •Electroencephalography is a potentially useful technique for diagnosing and monitoring postoperative delirium development and severity.

Delirium is a clinical syndrome resulting from acute brain dysfunction or encephalopathy. Delirium is an acute and often fluctuating disorder of attention and cognition.¹ Overall, this costly public health problem2, 3, 4 is associated with adverse outcomes, whether it manifests in hospital wards,² intensive care units,5, 6, 7 or generally in the postoperative setting. In this review, we first propose a taxonomic framework for considering delirium in the postoperative period, given that the aetiologies and associated EEG perturbations vary with time after surgery and anaesthetic exposure. We then summarize known EEG features reported for various aetiologies of delirium, such as hepatic encephalopathy, sepsis, and non-convulsive status epilepticus. Similar markers might offer diagnostic and prognostic yield for delirium and underlying acute encephalopathies in the postoperative period.

Delirium and subsyndromal delirium in the perioperative period

Delirium

Up to half of cardiac⁸ and major non-cardiac⁹ surgeries are complicated by delirium. Postoperative delirium is associated with re-admission,¹⁰ long-term functional decline,¹¹ greater health-care utilization,12, 13, 14 and adverse outcomes.¹⁴,¹⁵ The incidence of delirium after outpatient surgical procedures is unknown.

According to Diagnostic and Statistical Manual of Mental Disorders, 5th Edition (DSM 5) criteria, delirium is an acute syndrome of inattention and disordered cognition, which tends to fluctuate over time (Supplementary material, Data 1).1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16 The unresponsiveness of coma has been de-emphasized in this recent update.¹ Distractibility, reduced awareness of the environment, and excessive task perseveration exemplify inattention. Daytime sleepiness and insomnia are frequency observed, suggesting alterations in the circadian sleep–wake cycle. Disturbances in logic and perception can manifest with hallucinations or delusions. The severity of these features varies across patients and can fluctuate over minutes to hours in a single individual.

Subsyndromal delirium

Although there are no consensus criteria, the term subsyndromal delirium typically describes patients in whom many signs of delirium are present but not all diagnostic criteria are met. For example, a patient can demonstrate acutely disorganized thinking but lack alterations in attention. Subsyndromal delirium might represent the low-severity end on a continuum of delirium phenotypes.17, 18, 19 Additionally, subsyndromal delirium could be a prodrome before a full manifestation of delirium.²⁰ There are limited data to address these two non-exclusive possibilities in the postsurgical population. A prospective investigation followed 103 hip surgery patients longitudinally, with delirium evaluated at least three times per day.²¹ Of the 21 patients who developed delirium, 62% displayed a prodrome of some features of delirium hours before delirium diagnosis. Of note, these behavioural changes were observed in 48% of those who did not develop delirium, suggesting that these changes are not specific predictors of delirium. In a post hoc analysis of a trial evaluating delirium prophylaxis with haloperidol, postoperative impairments of memory, orientation, and coherence 1–3 days after surgery predicted subsequent delirium.²² Recently, Leung and colleagues²³ performed a prospective evaluation of subsyndromal delirium in 631 patients on postoperative day 1 after non-cardiac surgery. Aside from 31% of patients with delirium, an additional 27% had subsyndromal delirium, as defined by meeting at least one criterion for delirium diagnosis.²³ Subsequent delirium on postoperative day 2 was progressively more likely according to the number of positive delirium features during the subsyndromal phase. Overall, these data support the possibility that in many patients, subsyndromal delirium can be detected before full-blown postoperative delirium.

Current evidence suggests that subsyndromal delirium prognosticates poor outcomes, but the potential of abortive treatment is uncertain. When adjusted for co-morbidities and referenced to those without delirium, the occurrence of subsyndromal delirium on postoperative day 1 after non-cardiac surgery was associated with longer hospital stays and worse functional status on activities of daily living at 1 month after surgery.²³ The rationale of preventing the progression of subsyndromal delirium to delirium has motivated interventional studies of risperidone²⁴ and haloperidol.²⁵ Hakim and colleagues²⁴ randomized 101 cardiac surgical patients in a blinded study to assess the efficacy of risperidone (0.5 mg) every 12 h in mitigating progression of subsyndromal delirium. The treated group showed a significantly lower incidence of delirium compared with the placebo group (13.7 vs 34%). In contrast, a recent placebo-controlled trial of low-dose haloperidol showed no significant effects on the conversion of subsyndromal delirium to delirium in a population of mechanically ventilated medical and surgical patients.²⁵ A recent systematic review and meta-analysis did not support the use of antipsychotics as a preventative against subsydromal delirium or delirium.²⁶ Heterogeneities in the definition, severity, timing of progression, and application of assessment instruments all contribute to the difficulty in approaching subsyndromal delirium. Objective means of predicting and assessing delirium features over time may address these sources of variance and aid in targeted therapy for mild or subsyndromal delirium.

Diagnostic tools and phenotypic classification of delirium

Delirium is a clinical syndrome, and diagnosis depends on interview-based evaluation. There is an absence of validated confirmatory laboratory, electrophysiological, or imaging tests.²⁷ Practical screening tools for delirium have been developed to allow reliable diagnosis of delirium without the need of full neuropsychiatric evaluation. The Confusion Assessment Method (CAM)²⁸ and the Delirium Symptom Interview²⁹ are two instruments that have incorporated criteria of the DSM preceding the current 5th edition. Critically ill patients, including those who are intubated, can be assessed by the Intensive Care Delirium Screening Checklist³⁰ and a modified version of the CAM, the CAM-intensive care unit (ICU) tool.³¹ Severity of delirium can be quantified by the Memorial Delirium Assessment Scale³² and the CAM-S.³³ Administration of these delirium assessments can take longer than half an hour, even for experienced personnel. An abbreviated interview instrument has been developed for rapid administration over a few minutes.³⁴ Still, the variability in delirium signs and symptoms across individuals and within the same individual over time can impede accurate diagnosis.

Delirium is classified into hyperactive, hypoactive, or mixed subtypes based on outward behaviour and level of arousal.35, 36 Agitation, excessive movements, or pressured speech define the hyperactive form,³⁶ which accounts for less than one-third of patients.37, 38, 39 The hypoactive subtype of delirium is more common but harder to detect.⁴⁰ It can be mistaken for depression41, 42 because of inactivity, psychomotor retardation, or flat affect. The ‘mixed’ description is applied to patients who demonstrate both hyperactive and hypoactive features within a short interval. Although these subtypes are useful for descriptive purposes, their significance with regard to outcomes³⁷ and underlying neural mechanisms⁴³ deserves continued investigation.

Temporal classification of delirium in the postoperative period

Are there different types of delirium after surgery?

Curiously, there has been little discussion of whether delirium in the postoperative period is distinct from delirium in other settings. Medications, inflammation, and metabolic abnormalities can presumably precipitate delirium in the same manner before or after surgery. Delirium in the postoperative period has been temporally classified based on the time interval after surgery; emergence delirium occurs earlier than postoperative delirium.⁴⁴ There is no clear consensus on the exact temporal delineation between these entities. A temporal classification could have utility if different aetiologies and mechanisms vary over subsequent time periods after surgery (e.g. anaesthesia, postoperative sleep deprivation, or infection). A disadvantage is that temporal criteria might not reflect the underlying biological processes that, as yet, remain poorly understood.

Is there a transition from emergence delirium to postoperative delirium?

Emergence delirium, emergence agitation, postanaesthetic excitement, and postanaesthetic care unit (PACU) delirium, are terms that have been used synonymously to describe psychomotor agitation in the immediate postoperative period. Emergence delirium is common in the paediatric population,⁴⁵ and occurs within hours after discontinuation of anaesthetics, particularly sevoflurane or desflurane.

There has been little discussion on when emergence delirium is better classified as postoperative delirium. Many key outcome studies of postoperative delirium have initiated evaluation at postoperative days 1¹³ or 2,⁴⁶ potentially reducing overlap with emergence delirium. The frequency of delirium over time was evaluated by the CAM-ICU in a population of 143 patients after abdominal, vascular, or thoracic surgery.⁴⁷ A peak prevalence of 47% was reported for postoperative day 2, while the prevalence on postoperative day 1 was 30%. The mean delirium duration was 4 days.⁴⁷ Likewise, a mixed population of 143 patients undergoing surgical or transcatheter aortic valve replacement was evaluated for delirium frequency using the CAM.⁴⁸ Delirium was diagnosed in 29% (41 patients) on postoperative day 1 and in 27% (39 patients) on postoperative day 2. Postoperative delirium was diagnosed in many patients after a lucid interval.20, 48

A recent study of surgical patients⁴⁹ suggests that agitation on anaesthetic emergence can continue as delirium throughout the early postoperative period. Among 400 patients, of whom 19% were agitated on anaesthetic emergence, delirium was noted by CAM criteria in 60% at PACU intake and in 8% at PACU discharge. Thus, some patients emerge from anaesthesia with delirium that can persist beyond the PACU.⁵⁰ Likewise, among all patients, delirium declined from 31% on arrival to 4% at PACU discharge. Although the presence of emergence delirium appears to decline during early recovery, another possibility is that PACU discharge might be delayed until components of delirium resolve. Nevertheless, these data reveal a high prevalence of delirium in the PACU that declines below rates of postoperative delirium reported for postoperative day 1.47, 48

To summarize, the prevalence of postoperative delirium appears to have two peaks: first, within minutes to hours after surgery; and second, days later. We are unaware of any studies that prospectively clarify the point prevalence of delirium to characterize further the incidence of this complication over time. Elucidation of different aetiologies, neural markers, and clinical outcomes is needed to justify describing these as distinct delirium entities with potentially different mechanisms. The presence of agitation does not distinguish emergence delirium and postoperative delirium.

Delirium is a clinical syndrome reflecting an acute encephalopathy

The term encephalopathy generally describes a functional or structural disorder of the brain (Supplementary material, Data 2), but no consensus diagnostic criteria exist. Genetic inborn errors of metabolism, dementia, traumatic brain injury, or anoxic/hypoxic injury are examples of chronic encephalopathies (Supplementary material, Data 2). Initially acute encephalopathies, such as those after the onset of a stroke, can persist as chronic encephalopathies. Acute encephalopathies can resolve after treatment of underlying metabolic, toxic, infectious, or other identified aetiologies.⁵¹ The distinction between delirium and an acute encephalopathy that perturbs attention and cognition are blurred, as the terms are often used interchangeably.⁵² Delirium refers to the syndrome of signs and symptoms, whereas an acute encephalopathy refers to the reversible underlying pathophysiological process. In the absence of an identifiable cause and specific treatments, delirium reflecting an acute encephalopathy typically abates with the tincture of time.

EEG changes accompany delirium and acute encephalopathies of disparate aetiologies

Delirium arising from diverse acute encephalopathies was first associated with diffuse scalp EEG slowing in a seminal report by Engel and Romano in 1944,⁵³ a time when there were no diagnostic criteria for delirium. Nevertheless, these investigators assessed delirium severity as reduced performance on a battery that probed attention, memory, arithmetical, and abstract thinking domains. Many patients did not manifest agitation. Delirium was associated with the progressive dominance of the slower θ (4–8 Hz) and δ (<4 Hz) EEG oscillations. In the most severe instances, eye opening and closure elicited no changes in the posterior dominant rhythm. These findings contrasted with Berger's⁵⁴ discovery that eye closure generally induces the replacement of the desynchronized EEG by α (8–13 Hz) waves from scalp electrodes overlying occipital (visual) cortical regions. Engel and Romano⁵⁵ also showed reversal of EEG changes and cognitive abnormalities after treatment of the underlying aetiologies, such as hypoxaemia or hypoglycaemia. It has been posited that EEG changes always accompany delirium.⁵⁶ Several gross EEG abnormalities observed during delirium in the postoperative period are also described in association with hepatic encephalopathy and sepsis-associated encephalopathy (Table 1). These data suggest a partly conserved non-specific pattern of EEG changes during delirium of diverse aetiologies.

Table 1.

EEG signatures associated with hepatic encephalopathy, sepsis-associated encephalopathy, and delirium in the postoperative period. X indicates that the EEG perturbation has been previously reported for that condition, whereas absence indicates that it has not.

EEG perturbation	Hepatic encephalopathy	Sepsis-associated encephalopathy	Emergence delirium/ postoperative delirium
Mixed α and fast oscillations	X		X
Mixed α and θ oscillations	X	X	X
θ predominance	X	X	X
Frontal intermittent δ activity (FIRDA)	X	X
Polymorphic δ activity	X	X	X
Triphasic waves	X	X
Spike-and-wave complexes		X	X

EEG slowing or acceleration

Hepatic encephalopathy provides a model in which EEG changes reflect the severity of delirium and secondary organ failure. Acute liver failure and cirrhotic liver failure both produce hepatic encephalopathy and induce similar patterns of slowing in the EEG. Parsons-Smith and colleagues⁵⁷ provided a grading of EEG changes to accompany hepatic encephalopathy of varying severity (Supplementary Data 3, Table 1). This scale was based on deviations from the posterior dominant rhythm, a measure of EEG reactivity to eye closure in the occipital EEG. While the posterior dominant rhythm for most alert individuals is typically in the α band, it is replaced by fast activity in patients with mild hepatic encephalopathy (Fig. 1A). More severe instances were associated with the emergence of lower frequency θ activity in the posterior dominant rhythm (Fig. 1B and C). Even more severe symptoms were associated with diffuse θ or slower δ waves (Fig. 1D and E).⁵⁷ Subsequent investigators proposed the 1–2 Hz slowing in the posterior dominant rhythm as an additional criterion for grading the extent of EEG abnormalities associated with underlying hepatic encephalopathy (Supplementary material, Data 3, Table 2).⁵⁸

Fig 1.

Visual grading scheme of EEG patterns observed during hepatic encephalopathy. (A) Eyes closed wakefulness produces waves in the occipital EEG derivations of most normal individuals. These ‘Berger waves’ represent the posterior dominant rhythm, here shown as 9–10 Hz oscillations in the α band (blue). The dominant frequency is altered during the delirium of hepatic encephalopathy. Grade A: posterior α waves are replaced by higher frequency β activity. (B) Grade B: ‘unstable α’, with periodic replacement with θ waves of 5–7 Hz. (C) Grade C: α with runs of θ waves 5–6 Hz. (D) Grade D: persistent θ activity despite eye opening. (E) Persistent oscillations in the δ band. Figures were modified from Parsons-Smith and colleagues, reproduced with permission, Elsevier.⁵⁷

The introduction of quantitative EEG measures⁵⁹,⁶⁰ to grade the severity of EEG abnormalities facilitated comparison and validation among different cohorts. These grading schemes have relied on the mean (or median) dominant frequency or the relative power in θ or δ frequency bands (Supplementary material, Data 3, Tables 3 and 4). The median dominant frequency divides the total power equally among higher and lower frequencies. The relative power is the proportion of total power accounted for by the power in the frequency band of interest (e.g. 4–8 Hz for θ). Combinations of the median dominant frequency and the relative power (Supplementary material, Data 3, Table 5) are correlated with liver synthetic function,⁶¹ performance on psychometric testing,⁵⁹ and long-term survival.⁶²

Sepsis-associated encephalopathy represents another treatable aetiology of delirium. Early work by Young and colleagues⁶³ associated diffuse EEG slowing with progressive morbidity and mortality in septic patients (Supplementary material, Data 3, Table 6). Similar to the changes reported for hepatic encephalopathy, patients with sepsis-associated encephalopathy also demonstrated prominent θ oscillations (Fig. 2A) and the emergence of δ oscillations (Fig. 2B and C). Recent data suggest that early changes in EEG signatures have prognostic significance, with poor outcomes heralded by the emergence of δ activity⁶⁴ or the lack of EEG reactivity to auditory or tactile stimulation.⁶⁵ Recent systematic reviews of EEG abnormalities in patients with sepsis report a broad variability in the presence of these patterns.⁶⁶,⁶⁷

Fig 2.

EEG patterns observed during sepsis-associated encephalopathy. (A) Oscillations (5–6 Hz) in the θ band of a 51-yr-old man with confusion during a Salmonella infection. (B) Frontal intermittent rhythmic δ activity superimposed on θ oscillations. These were recorded from a 62-yr-old woman infected with streptococcal pneumonia. (C) δ oscillations (0.5–1 Hz) with little background activity in a comatose 68-yr-old man afflicted with a Pseudomonas infection. (D) Triphasic waves with minimal background activity recorded from a 74-yr-old man with bacteraemia. Shared voltage and time scales are noted. Figures were modified from Young and colleagues, reproduced with permission, Clinical and Investigative Medicine (CIM).⁶³

Polymorphic δ activity

EEG patterns dominated by power in the δ frequency band include polymorphic δ activity, frontal intermittent rhythmic δ activity, and triphasic waves. The δ activity in the EEG can show localization over scalp regions, lateralization over a single hemisphere, or diffuse expression across the head. Focal δ activity is associated with tumours⁶⁸ or structural white matter defects.⁶⁹ In contrast, widespread polymorphic δ activity, where the dominant frequencies vary over time, can be observed in patients with diffuse white matter injury⁷⁰ or thalamic tumours.⁷¹ Polymorphic δ activity is not always correlated with radiological findings, however.⁷² An older series followed neurological patients with focal or lateral polymorphic δ activity and reported that roughly one-third of the cohort had normal or only non-focal abnormalities on subsequent head computed tomography imaging.⁷³ As a non-specific finding that might reflect disruption of corticothalamic or intracortical connectivity,⁷⁴ large-amplitude diffuse polymorphic δ activity can be observed not only with severe encephalopathy and coma but also during sleep dominated by slow waves.⁷⁵

Frontal intermittent rhythmic δ activity

Frontal intermittent rhythmic δ activity⁷⁶ (Fig. 2B) is characterized by sporadic recurrences of several seconds of continuous prominent δ band oscillations with preserved α components.76, 77, 78 In contrast to the slowing in polymorphic δ activity, a regular rhythmicity is apparent in frontal intermittent rhythmic δ activity. Rhythmic δ activity was originally associated with structural brain lesions in the posterior fossa or deep midline regions, but should raise suspicion of a mass when observed unilaterally.⁷⁹ Frontal intermittent δ activity manifests in delirium associated with sepsis,⁶³ uraemia,⁸⁰ and hyperglycaemia.⁸⁰ Of note, these patterns can be mistaken for a slow blink artifact,⁸¹ or be observed in normal individuals during hyperventilation.⁷⁹ Overall, frontal intermittent rhythmic δ activity remains an uncommon marker of unclear utility.

Triphasic waves

Triphasic waves show a predominance of δ band power, but have a distinctive morphology.⁸²,⁸³ These stereotyped complexes are 300–600 ms long and have three components: a large positive peak flanked by two negative deflections (Fig. 2D).⁸⁴ They may recur every 0.5–2 s, predominantly in frontal channels.⁸⁵ Very poorly formed triphasic waves can be difficult to distinguish visually from slow-wave activity and spike-and-wave complexes of epileptic aetiology.⁸⁶ Triphasic waves observed as pathological periodic discharges⁸⁷ were originally described as ‘blunted spike and wave complexes’ in patients with hepatic encephalopathy.⁸⁸ These EEG complexes can be elicited by noxious or auditory stimulation.⁸⁴ Triphasic waves have been observed in the postoperative period⁸⁹ but are more commonly observed in hepatic or renal dysfunction,90, 91, 92 severe instances of sepsis,63, 93,⁹⁴ or medication toxicity (e.g. lithium or baclofen).95, 96 Although a high mortality rate (77%) was previously associated with triphasic waves in a diverse group of critically ill patients,92, 93, 94, 95, 96, 97 a lower rate (20%) has been reported in a more recent cohort.⁹⁵ The aetiology and diagnostic significance of triphasic waves remain unclear.⁹⁸

To summarize, both hepatic encephalopathy and sepsis-associated encephalopathy can manifest by delirium with grossly similar perturbations in the EEG. Greater clinical severity is associated with the extent of generalized EEG slowing and a shift of the posterior dominant rhythm to lower frequencies. Frontal intermittent rhythmic δ activity and triphasic waves are encountered with moderate and severe delirium. These changes in the EEG reflect altered cortical neuronal activity, presumably attributable to impairment of metabolic processes.⁵³

Non-convulsive status epilepticus is a treatable aetiology of delirium that requires EEG for diagnosis

EEG is essential for diagnosis and management of non-convulsive status epilepticus (NCSE). A recent consensus statement proposes working diagnostic criteria for NCSE.⁹⁹ Epileptiform activity during NCSE can include spike-and-wave discharges, spikes, polyspikes, or sharp waves but must recur more frequently than 2.5 times per second. Less frequent EEG complexes or slow rhythmic oscillations can constitute NCSE but require additional clinical criteria.

Spike-and-wave discharges (Fig. 3A), altered sensorium or behaviour persisting beyond 30 min, and the absence of convulsive seizure activity can serve as criteria for NCSE.100, 101, 102 The spike-and-wave discharges during NCSE (Fig. 3A)¹⁰³ vary in character depending on the aetiology.¹⁰² These EEG complexes can be difficult to distinguish from triphasic waves (Fig. 2D) that generally recur at lower frequencies (<2.4 Hz).⁸⁴ However, spike-and-wave discharges generally lack slower background EEG activity and are more likely than triphasic waves to show extra spike components.⁸⁴

Fig 3.

EEG spike-and-wave complexes can be observed during non-convulsive status epilepticus. (A) Spike-and-wave complexes arise at slightly higher frequencies (2–3 Hz) than triphasic waves (Fig. 2D), as shown in frontal EEG traces. These epileptiform discharges reflect underlying non-convulsive status epilepticus in this poorly responsive patient without generalized myoclonic activity. (B) Spectrograms show the distribution of EEG power as functions of frequency and time; here, high amplitude is shown as red, intermediate as yellow or green, and low as blue. Low-frequency (∼3 Hz) power, corresponding to the frequency of the spike-and-wave complexes, is abolished (*) after administration of midazolam (▾). The white trace indicates the spectral edge frequency 95%. Figures are adapted from Hernández-Hernández and Fernández-Torre, reproduced with permission, Elsevier.¹⁰³

Non-convulsive status epilepticus can be an underappreciated cause of delirium after cardiac¹⁰⁴ and non-cardiac surgery. Delirium can manifest during epileptiform activity (ictal period) but also persist after or between electrographic seizures (postictal confusion).¹⁰⁵ A prevalence of 5% was reported for NCSE in a recent retrospective cohort study of 154 poorly responsive surgical ICU patients monitored with continuous EEG.¹⁰⁶ The aetiologies of altered mental status in these patients included sepsis-associated encephalopathy (43%), ICU delirium (22%), and metabolic encephalopathy (19%). At the time of continuous EEG, 55% were considered comatose, with 70% sedated with midazolam, propofol, or dexmedetomidine. It is not clear how many of the eight patients with NCSE showed ICU delirium.

In evaluating long EEG recordings for evidence of NCSE, the power spectrogram (Fig. 3B), or compressed spectral array, can be a useful and efficient visualization tool¹⁰⁷ for identifying intermittent epileptiform discharges.¹⁰³ The generation of an EEG power spectrogram requires the decomposition of the signal into frequency and amplitude pairs of component waves.¹⁰⁸ While a power spectrum plots the squared amplitude of each constituent wave as a function of frequency for a fixed duration of time, a power spectrogram represents these relationships as a function of time.¹⁰⁹ Spectrogram-guided evaluation of the EEG for epileptiform discharges has shown promise in expediting detection of patients with seizures but has not been shown to facilitate identification of all events.¹⁰⁷ Further investigation will be needed to assess the utility of the power spectrogram in facilitating the diagnosis of delirium in the postoperative setting.

EEG patterns associated with delirium in the postoperative period

We now review non-epileptiform EEG abnormalities associated with delirium in the postoperative period. There have been no analyses distinguishing EEG markers of emergence delirium from those of postoperative delirium. Thus, we review EEG patterns of delirium in the postoperative period based on associated hyperactive and hypoactive subtypes.

Hyperactive delirium

There is one systematic study of hyperactive delirium in the postoperative period. This recent prospective investigation reported EEG correlates associated with emergence delirium in the paediatric population¹¹⁰ using assessment tools developed within the same laboratory.¹¹¹ Of 60 patients studied, five (8.3%) exhibited emergence delirium. All four patients with simultaneous EEG recordings awoke with agitation during an ‘indeterminate state’ of diffuse mixed α and β activity, reminiscent of Parson-Smith's Grade A EEG changes during hepatic encephalopathy (Fig. 1A; Supplementary material, Data 3, Table 3).⁵⁷ In contrast, no consistent pattern was observed during the delirious episodes. While one patient showed bilateral diffuse θ activity, another manifested frontocentral α and diffuse θ activity (6 Hz). A third showed a return of the posterior dominant rhythm in the α band and frontal bilateral 13–40 Hz activity, while the fourth showed mainly diffuse low-voltage fast activity. This variance in EEG spectral content for delirium with agitation mirrors the lack of consistent spectral patterns for the hyperactive delirium across clinical settings of adult patients. In a diverse cohort of elderly patients with hyperactive delirium, spectral analysis showed lower relative α and β (∼8–20 Hz) band power and greater relative δ and θ (∼1.5–7 Hz) compared with control subjects.¹¹² In contrast, the preservation of low-voltage power in the β range (16–30 Hz) has been noted with alcohol withdrawal,113, 114 drug toxicity,⁵⁶ and drug-related delirium,115, 116, 117 including sedation states induced by propofol¹¹⁸ and sevoflurane.118, 119, 120 Additional data are needed to determine whether there are consistent EEG patterns in both paediatric and adult patients.

Although the power across different frequency bands can vary in patients with hyperactive delirium, alterations in the connectivity between frontal cortical regions or between frontal and posterior cortical areas might be a consistent marker. The strength of connectivity between regions is typically assessed by the magnitude of correlations between signals. Martin and colleagues¹¹⁰ assessed correlated EEG activity in the 5–15 Hz range by computing the global efficiency (a measure of zero-lag correlation within the electrode network) and global coherence (a measure of bivariate oscillatory phase alignment). Compared with non-delirious control subjects, four paediatric patients with emergence delirium showed increased correlated activity in the frontal EEG. In contrast, no changes in correlated activity were observed for the parietal electrode network. The overlap of connectivity changes within the θ and α frequency bands during emergence delirium are concordant with a recent report of cardiac surgery patients manifesting hypoactive, hyperactive, and mixed delirium subtypes.¹²¹ In that study, the phase lag index was used as a measure of correlation within δ, θ, α, and 13–20 Hz frequency bands. The phase lag index describes the frequency-specific distribution of phase differences between a pair of signals sampled over each window of time. The dispersion of this distribution provides a measure of synchronization such that a consistent phase difference among these time windows suggests an interaction between associated brain regions. A near-zero phase lag suggests synchronized neural activity. A consistent non-zero lag suggests propagation of signals between brain regions at a particular frequency, with directed phase lag indicating the relative timing of signals in these regions. When averaged across EEG derivations, delirious patients had lower correlated activity in the α band compared with control subjects. Accounting for the temporal relationships between signals in different regions, frontal areas lagged behind posterior regions in patients with delirium compared with control subjects.¹²¹

Hypoactive postoperative delirium

The EEG patterns⁵⁶ associated with hypoactive delirium during the postoperative period grossly resemble those observed in hepatic encephalopathy and sepsis-associated encephalopathy. The θ122, 123 and δ band oscillations122, 124 are prominent in awake patients with hypoactive delirium. To demonstrate these previously reported EEG changes, we include frontal and occipital EEG examples (Fig. 4A) from a patient with reduced α125, 126, 127, 128, 129 and predominantly δ activity.112, 123,124, 126,128, 129, 130, 131 These data were acquired during an ongoing study, the ENGAGES trial (NCT02241655), and are included for illustrative purposes only.

Fig 4.

Low-frequency EEG predominance during delirium appears to resolve after recovery. (A) During delirium in the postoperative period, high-amplitude low-frequency activity can be seen in the EEGs of frontal (F7-Cz, red) channels, particularly with eyes closed. High-frequency activity present with eyes open may be related to activity of the frontalis muscle underneath the recording electrodes. A predominance of low-frequency activity is present in the occipital EEG (O1-Cz, green) during both eyes open (top panel) and eyes closed conditions (bottom panel). (B) After recovery from delirium, low-frequency activity becomes less prominent, and higher frequency components return during eyes open in both occipital and frontal channels. (C and D). Occipital EEGs in Fig. 3A and B are shown at a magnified scale. Low-frequency EEG oscillations can be seen during delirium (C), for eyes open (left panel) and eyes closed recordings (right panel). After resolution of delirium (D), higher frequency activity dominates during eyes open conditions (left panel). Note the return of occipital α activity during eye closure (right panel). Diagnosis of delirium was evaluated using the Confusion Assessment Method.²⁸ This 62-yr-old man did not have an obvious aetiology of delirium after a mitral valve replacement and Maze procedure. Clinical details of this patient are available in Supplementary Data. All EEG signals have undergone bandpass filtering for 0.3–50 Hz. Shared voltage and time scales are noted for a and b and for c and d.

Generalized slowing in frontal and occipital EEG is illustrated for a patient experiencing hypoactive postoperative delirium (Fig. 4A), as diagnosed by the CAM.²⁸ The prominent low-frequency oscillations112, 123,124, 125,128, 129, 130, 131 appear to wane after recovery (Fig. 4B) from hypoactive delirium, verified by the CAM. Additionally, the reduction of posterior α power, normally associated with the descent into light stage 1 non-rapid eye movement sleep, has been associated with hypoactive delirium in the postoperative period.125, 126, 127, 128, 129 In this patient, eye closure failed to elicit the normally observed posterior dominant rhythm (Fig. 4C). This abnormal lack of EEG reactivity after voluntary eye closure normalized with recovery from delirium (Fig. 4D).

The relative EEG power measure is frequently used for investigating changes in oscillatory EEG activity across a population of delirious patients, as noted in a recent meta-analysis.¹³¹ Reduced relative α power,125, 126, 127, 128, 129 increases in relative θ power,112, 125,¹²⁶,¹²⁸,¹²⁹,¹³²,¹³³ and a greater proportion of relative δ power have been reported.¹¹²,¹²³,¹²⁴,¹²⁶,128, 129, 130, 131 Whether EEG changes such as these are related to the progression, severity, or prognosis for neurological recovery is unknown.¹¹² In lieu of assessing power changes within several different frequency bands, composite power ratio measures may sufficiently represent power shifts in the frequency spectrum. For example, a high θ/α ratio during delirium¹¹²,¹²⁵,¹²⁶ can be informative because it is sensitive to either an increase in θ power or a reduction in α power. Other measures combine more disparate frequencies, such as the θ/(θ+δ)¹²⁹ or (α+β)/(θ+δ) ratios.¹¹² Future investigations can probe the clinical utility of relative power ratios for detecting and monitoring delirium.

Power spectra and spectrograms demonstrate differences in relative power in individual patients and during the course of recovery from hypoactive postoperative delirium. Again, for didactic purposes, we have included data acquired in the ENGAGES trial (NCT02241655). Comparison power spectra (Fig 5A) for a patient with delirium (red) and without delirium (blue) can show marked differences.⁵⁵ Concordant with the literature, greater δ power and less α power are observed in the patient with delirium. In contrast, the patient without delirium shows a distinct peak in the α band, which correlates with the frequency of the posterior dominant rhythm. Corresponding spectrograms show a greater predominance of δ band power in the patient with delirium (Fig. 5B) than in the control patient (Fig. 5C). The evolution of relative power is apparent in an EEG acquired from a patient monitored during hypoactive delirium on postoperative day 1 and after resolution of symptoms (Fig. 5d–F). An overlay of spectra (Fig. 5D) for the two time points demonstrates a shift of relative power from a δ predominance during delirium (red) into θ, α, and higher frequency bands after recovery (blue). In the corresponding power spectrograms, a consistent peak in occipital α is not apparent either during delirium (Fig. 5E) or after the return of power at frequencies >8 Hz (Fig. 5F).

Fig 5.

Power spectrograms of eyes closed occipital EEGs during and after delirium. (A) Comparison spectra of relative power are provided for two different patients in the postoperative period. Delirium was present in one patient (red) and absent in another (blue), as diagnosed with the Confusion Assessment Method.²⁸ Relative power is the proportion of power in a given frequency range relative to the total power. This metric is commonly used in quantitative EEG evaluation for both postoperative delirium⁹⁵,106, 107, 108, 109,^111–116 and hepatic encephalopathy.55, 56 A peak (→) in occipital α (8–13 Hz) is clear in the patient without delirium but is not prominent in the patient with delirium. Comparisons of the two spectra show a greater proportion of power in the δ (<4 Hz) band during delirium. Differences in the θ (4–8 Hz) band are unclear. The EEGs of the patient with delirium also had a lower proportion of power in the α band and at frequencies >13 Hz. (B and C) The consistency of these power distributions over time is shown in the spectrograms of EEGs acquired in the presence (B) or absence of delirium (C). The rhythm is in the low α range (→) and attenuates with eye opening, and is referred to as the posterior dominant rhythm. Clinical details of these patients are available in Supplementary Data. (D) Power spectra of longitudinal EEG recordings of a third patient (same as in Fg. 3) demonstrate a shift of power during the resolution of delirium from postoperative day 1 (red); signals acquired upon recovery (blue) show a reduction in δ and increase in θ and α band relative power. (E and F) Spectrograms of relative power in the EEGs acquired while delirium was present (E) and after resolution (F). Black markers above the spectrograms show the times of traces presented in Fig. 4C and D. Despite greater power in the θ and α bands, a distinct peak is not apparent after recovery. Spectrograms were generated using Chronux Matlab subroutines⁹¹ and multitaper spectral analysis⁹² to estimate power spectra (1–20 Hz bandwidth, 10 s moving window at 1 s increments, five tapers, time bandwidth of 3 s). For purposes of presentation, the relative power has been transformed by log base 10.

Current and future use

At present, the clinical utility of the EEG for diagnosing delirium and monitoring its course is unclear. EEG may be used in evaluating patients with delirium¹³⁴ and combined with imaging to evaluate aetiologies of altered mental status.¹³⁵ Unless clinical evaluation reveals focal neurological deficits after surgery, brain imaging can be of little utility despite high cost.¹³⁶ In contrast, EEG is (relatively) inexpensive, has few contraindications, and can be useful to detect encephalopathies related to focal insults, non-convulsive epileptic activity, or diffuse brain dysfunction, such as metabolic derangements.¹³⁷,¹³⁸

Given that delirium is acute in onset and fluctuating in severity, serial or continuous EEG can provide a useful bedside monitoring tool to complement intermittent clinical assessment. EEG can provide markers of delirium deterioration, aid in tracking resolution, and detect epileptic aetiologies. The lens of clinical scrutiny can also be widened through the detection of eye movements and blinks, commonly seen as artifacts in analysis, but gaining attention as potential markers for delirium in the postoperative period.¹³⁹ Such eye movements will need to be distinguished from slow horizontal eye movements that are combined with EEG slowing to score the transition from wakefulness to non-rapid eye movement sleep.¹⁴⁰

The extensive progress in characterizing the EEG correlates of anaesthetic-induced unconsciousness¹¹⁹,¹²⁰,141, 142, 143, 144, 145 suggests the potential for associating specific signatures with different delirium aetiologies. Spectral content of EEG signatures vary by anaesthetic within the intermittent activity burst suppression¹⁴⁶ or spindle activity during general anaesthesia.¹⁴² Moreover, comparisons of EEGs acquired during sleep and anaesthetic states suggest differences in spectral content of spindles¹⁴⁷ and propagation of slow-wave activity.¹⁴⁸ Likewise, dense multielectrode (>64 channels) recordings are likely to be needed to characterize neural underpinnings of delirium that might aid in identifying subtypes useful for diagnostic and therapeutic purposes.

Methodological considerations and caveats impeding routine postoperative clinical EEG

EEG alterations in the presence of drowsiness, sleep, and analgesics are potential sources of noise that have not been addressed in any of the studies referenced above. Normal drowsiness is accompanied by a reduction in the frequency and amplitude of the posterior dominant rhythm¹⁴⁹ and the emergence of slow rolling eye movements. The transition into sleep can occur in the absence of eye closure. Given that EEG slowing can occur with delirium or during sleep, elements of polysomnographic sleep staging must be considered in future studies. Opioids, such as fentanyl and morphine, and other analgesics used in the postoperative period can induce variable effects in the EEG.¹⁵⁰

The effect of age and location of EEG recording sites must also be addressed before routine clinical implementation. Early work has showed promise in the ability of EEG to allow detection of delirium superimposed on dementia.¹³² The EEG amplitude and power vary across individuals based on factors such as genetics,¹⁵¹ skull thickness,¹⁵² and age.¹⁵³,¹⁵⁴ Usage of relative power measures can address inter-individual variability in EEG amplitude and has been previously recommended over absolute power.155, 156, 157 Recent work has addressed whether certain scalp EEG derivations might be more useful in detecting delirium.¹³⁰ These data support the possibility that monitoring based on a few limited EEG channels, as used in clinical polysomnography, might be sufficient for diagnostic purposes.

Conclusion

Delirium is a common postoperative neurological complication that occurs within minutes after emergence from anaesthesia as emergence delirium, or hours to days after surgery as postoperative delirium. Postoperative delirium is often associated with poor outcomes and should prompt consideration of treatable ailments (e.g. hypoglycaemia, alcohol withdrawal, non-convulsive seizure activity, and sepsis). As an inexpensive, low-risk technique, EEG has potential as a clinical and research tool by providing objective measures assessing acute encephalopathies underlying delirium. A review of the major EEG findings associated with delirium shows that several potential markers exist. Further development of EEG analytical methods will provide markers for early detection, intervention, and assessment of treatment, in addition to mechanistic insights into this pathologically obscure and fluctuating neurological disorder.

Authors' contributions

Design/planning: B.J.P., M.S.A.

Writing paper: B.J.P., T.S.W., M.S.A.

Revising paper and approval of final version: all authors.

Declaration of interest

Conflicts of interest: None declared. The content is solely the responsibility of the authors and does not necessarily represent the official view of the National Institutes of Health.

Funding

Washington University Institute of Clinical and Translational Sciences grant UL1TR000448, sub-award KL2TR000450 from the National Center for Advancing Translational Sciences (NCATS) of the National Institutes of Health (NIH, Bethesda, MD, USA; to B.J.P., Y.S.J.); the National Institute of Aging (NIH, Bethesda, MD, USA) [1UH2AG050312-01, 4UH3AG050312-02 (to M.S.A., T.S.W.) and 1R21AG052821-01 (to B.J.P., M.S.A.)]; the National Heart, Lung, and Blood Institute (NIH, Bethesda, MD, USA; R21HL123666 to M.S.A., T.S.W., B.J.P.) the National Institute of Neurological Disorders and Stroke (NIH, Bethesda, MD, USA; K23NS089922 to Y.S.J., 1R21NS096590 to S.C.); the Burroughs Wellcome Fund (Research Triangle Park, NC, USA; Career Award at the Scientific Interface to S.C.).

Handling editor: Hugh C. Hemmings Jr