Rapid Hand-held Continuous EEG has the Potential to Detect Delirium in Older Adults

Abstract

Background:

Delirium related biochemical derangements lead to electrical changes that can be detected in electroencephalographic (EEG) patterns followed by behavioral signs and symptoms. Studies using limited lead EEG show a large difference between patients with and without delirium while discriminating delirium from other causes. Hand-held rapid EEG devices may be capable of detecting delirium prior to symptom onset. Thus, providing an objective physiological method to detect delirium when it is most amenable to interventions.

Objective:

To explore the potential for rapid EEG to detect waveform pattern changes consistent with delirium status.

Methods:

This prospective exploratory pilot study used a correlational design and mixed models to explore the relationships between hand-held portable EEG data and delirium status

Results:

While being under powered minimized opportunities to detect statistical differences in EEG derived ratios using spectral density analysis, sleep/wake ratios tended to be higher in patients with delirium.

Conclusions:

Limited lead EEG may be useful in predicting adverse outcomes and risk for delirium in older critically ill patients. Although this population is at highest risk for mortality, delirium is not easily identified by current clinical assessments. Therefore, further investigation of limited lead EEG for delirium detection is warranted.

Introduction

Delirium is a serious complex acute condition that causes fluctuating disturbances in cognition, awareness, and attention (See table 1 for DSM-V criteria). In hospitalized older adults, delirium is the most common neuropsychiatric condition. Delirium occurs in as many as 80% of patients in the intensive care unit (ICU), costing more than $185 billion annually and tripling mortality.^1,2 As the brain adapts to physiological stressors, levels of neurotransmitters become increasingly imbalanced. EEG reflects these imbalances as changes in waveform patterns. Once the brain can no longer compensate for the chemical (neurotransmitter) imbalances, behavioral symptomatology begins to appear, leading to what we know as delirium or brain failure. (See Figure 1)

Table 1.

DSM-V Criteria for diagnosing delirium

All of the following must be met:

Disturbance

• in both attention and awareness

• in cognition

• develops over a short period of time (hours to days)

• tends to fluctuate in severity

• is not better explained by a pre-existing dementia

• is not the result of severely reduced level of arousal or coma

• occurs in the presence of an underlying cause(s)

Criteria are adapted from the DSM-V1

^1.American Psychiatric Association. Diagnostic and Statistical Manual of Mental Disorders: DSM-5 (5th [rev.] ed.). Washington, DC: American Psychiatric Association.

Published 2013. Accessed.

Figure 1.

Ceribell Device

Permissions received from Ceribell, Inc. 3/2/2021

www.ceribell.com

The Confusion Assessment Method-ICU (CAM-ICU) is the only delirium screening tool validated in both geriatric and critical care patient populations.^3–9 The CAM-ICU has high sensitivity (93-100%) and specificity (98-100%) when used in research and is therefore recommended by the Society of Critical Care Medicine (SCCM).^3,4 However, in clinical practice, the use of sedation and brain injury prevent accurate assessments of as many as 58% of ICU patients. Along with clinician interrater reliability (43-47%) challenges and practice drift, identification of positive delirium cases is approximately 20%.^10,11 Even when identified, delirium is often dismissed as dementia or due to another cause.¹² Therefore, an objective physiologic monitor capable of early accurate detection is desperately needed.

EEG was identified decades ago as the true gold standard for delirium detection, but its use has not been feasible due to limited resources.^10,13,14 Electrophysiologic signals characteristic of delirium are often reported as ‘diffuse slowing,’ meaning that the brain waves are of a reduced frequency. When delirium is present, EEG waveform patterns shows generalized slowing of background activity, slowing and significant decreases in the power of awake waves (alpha waves), increases in sleep wave (theta waves) and deep sleep wave (delta waves) power, and sleep wave intrusion.^15,16 As a result, ratios of sleep waves to awake waves (delta and theta waves to alpha waves) are significantly higher.^15,16 Therefore, the emergence of low-frequency waves indicates potential occurrence of delirium on EEG. Prior research has shown that these EEG characteristics distinguish delirium from non-delirious states such as sleep, sedation, dementia and other etiologies associated with behavioral changes.^{10,13,14,15–17} When combined with the current twice daily clinical assessments using a standardized tool, such as the CAM-ICU, EEG significantly increases accuracy to greater than 95% (p = .003) compared to the 20% detected with the current practice using the CAM-ICU alone.¹⁸

Over the last couple of decades, with advances in technology, limited lead EEG devices have become available.¹⁹ For example, the bispectral index monitor (BIS) uses two electrodes and a proprietary algorithm to analyze EEG data. One waveform is obtained from the fronto-temporal region and analyzed to detect changes in consciousness. Additionally, methods of analyses such as signal processing have been developed that maximize signal concentration and filter frequencies to smooth variations.²⁰ This process ‘cleans’ the EEG waveform by removing interference from other equipment such as cardiac monitors and electronic beds. Because all EEG channels can detect the same reduction in frequency, this suggests use of a small number of channels are sufficient to obtain relevant data indicative of delirium (See supplemental data for example). Prior research has shown the use of two electrodes can detect delirium but lack specificity to rule out other causes such as sleep, sedation, and dementia. ^{10,13,14,15–17}

The Ceribell is a handheld device that provides rapid continuous EEG (cEEG) monitoring that was Food and Drug Administration (FDA) approved for seizure monitoring in 2017. (See Figure 2) Unlike other limited lead devices, the Ceribell has 10 leads (8 waveforms) that are housed in a headband that circumscribe the head, allowing for monitoring of all five cerebral lobes of the brain. Because leads are housed in the headband, the Ceribell headband can be easily applied by non-experts, such as nurses. The ability to monitor the parietal and occipital lobes may improve specificity not seen in 2-lead devices such as BIS monitoring. Using the Ceribell along with advanced methods of analyses such as automatic signal processing methods capable of detecting the known changes in EEG waveform patterns associated with delirium, is possible. As a result, limited lead EEG devices with automatic signal processing may provide a physiologic method of delirium detection that is a practical and feasible nurse friendly alternative to traditional EEG.¹⁹ Additionally, these newer devices may also provide the additional benefit of earlier detection, even prior to behavioral symptom onset.^21,22 Therefore, the hypothesis that drove this study was that limited lead EEG monitoring devices could provide a physiologic method for early and possibly predictive delirium identification (See Figure 1).²³

Figure 2.

Pathoetiological Model for Delirium

Objective/Aim

While the Ceribell device has been used to monitor critically ill patients for seizures, its ability to detect EEG changes associated with delirium has not determined. Therefore, the aim of this pilot study was to explore use a rapid handheld limited lead cEEG device, the Ceribell, with automatic signal processing for detection of changes in EEG associated with delirium. If these changed can be detect, the Ceribell device may provide more timely results with a high degree of accuracy and feasibility for nurses without adverse effects.²⁴ This pilot feasibility study used a prospective design and convenience sampling to explore potential use of the Ceribell device for delirium detection in preparation for a larger study.

Methods

Setting and Sample

After receiving Institutional Review Board (IRB) approval, participant enrollment from a tertiary rural academic medical center in eastern North Carolina began. Legally authorized representatives (LAR) provided consent prior to enrollment. Enrollment included 17 critically ill older adults (age ≥50 years) in the medical, surgical, and cardiac ICUs requiring mechanical ventilation for greater than 12 hours who met inclusion/exclusion criteria between March 2019 and March 2020. Inclusion criteria included age 50 or greater, requiring mechanical ventilation for greater than 12 hours, the ability to participate in a CAM-ICU assessment and English speaking.^25,26 Demographics and past medical history were obtained from the electronic medical record. Primary diagnoses of enrolled participants included: motor vehicle collision, respiratory failure/pneumonia, congestive heart failure (CHF), ST segment elevation myocardial infarction (STEMI), gastrointestinal (GI) bleed, abdominal mass, and bowel perforation. Severity of illness scores such as Acute Physiologic Assessment and Chronic Health Evaluation (APACHE) and Sequential Organ Failure Assessment (SOFA) were not available. To control for co-morbidities, the Charlson Co-morbidity Index Scores, a 6-point weighted index used to predict 12-month mortality risk and higher resource utilization was used.

Measures

A full description of the methods used in this pilot study has been previously described in a protocol manuscript elsewhere.¹⁹ In brief, investigators placed a headband housing 10-electrodes around the subject’s head and connected it to a handheld cEEG device, the Ceribell. This device was used because it has demonstrated the ability to provide reliable data with cloud storage.²⁴ The cEEG data were recorded for 2-hours each evening (between 5-9 pm) for up to 4 consecutive days from each participant while they were in the ICU. After the first hour of cEEG monitoring, while receiving routine care, the investigator performed CAM-ICU assessments to determine delirium status. This investigator is a practicing clinical nurse specialist with advanced training in cognition and has used the CAM-ICU in prior research. To ensure accuracy of assessments during this study, online training from Vanderbilt University on their www.icudelirium.org website. Investigator assessments were subsequently validated by a co-investigator who is both a neurologist and neuropsychologist practicing in the ICU units where participants were recruited.

A member of the research team blinded to delirium status evaluated EEG data using quantitative methods, also known as spectral density analysis.²⁹ Quantitative EEG (qEEG) relies on extensive technical knowledge in the field of digital signal processing. Raw EEG data were recorded and uploaded to a cloud server then parsed in preparation for analyses in MatLab.^27,28 Raw EEG data includes both brainwave and ‘ambient noise’ or artifact from the eye and cardio-pulmonary movement of the chest as well as electrical activity from electronic devices such as beds and telemetry monitoring. Prior to EEG data analysis, noise components were filtered out of all eight channels using traditional high- and low-pass filters and advanced independent component analysis filters. Evaluation of EEG data based on delirium status was conducted using power spectral density analysis to detect the relative presence of high- and low-frequency activity in each channel. Spectral density and independent component analyses in Matlab were used to evaluate EEG signals in 5-minute epochs. Computational algorithms allowed for objective and quantitative documentation of EEG characteristics.

Statistical analysis

Summary and exploratory statistics were performed using SPSS (version 26).30 Descriptive statistics included all participant data to summarize subject demographic and clinical characteristics and determine representativeness of the sample. Frequencies and percentages were used to describe demographic data such as age, gender, race, hospital and ICU length of stay (LOS) and discharge disposition. Data analyses explored associations between limited lead EEG derived theta/alpha, delta/alpha and delta/theta ratios and clinical delirium status.

A comparison of researcher determined delirium status based on CAM-ICU assessments (dichotomized as delirium present/not present) compared with dependent variables from signal processed EEG including ratios as continuous variables. Kolmogorov-Smirnov test evaluated subject characteristics for normalcy. Mixed effects models with day nested within person were used to evaluate whether EEG derived ratios as a continuous variable could identify differences based on delirium status within and between subjects. Random and fixed (repeated measures) effects models were used to control for patient differences. Averaging all subject data based on delirium status meant each subject had one delta/alpha, one theta/alpha and one delta/theta ratio. Exploratory analyses of EEG ratios excluded one delirium negative due to the amount of artifact.

Results

Mean age for all subjects was 66.69 (SD 8.731). Nine subjects (53%) were male, 11 subjects (65%) were Caucasian. Subjects were similar in that 15 (86%) were living at home prior to hospitalization. Overall hospital LOS was 24.80 days (SD 12.80) with an ICU LOS of 17.80 days (SD 12.47). The average number of ventilator days was 12.53 days (SD 12.73). Mean Charlson Co-morbidity Index Scores were 3.73 (SD 3.052). These Index scores indicate a five-year survival rate of less than 28%.³¹ Nine subjects (52.94%) were positive for delirium during the enrollment period. See Table 2 for additional demographics.

Table 2.

Participant demographics by Delirium Status

Variables N=17	Delirium positive (n=9)	Delirium negative (n=8)
Male	7 78%	2 25%
Female	2 22%	6 75 %
Age
50-59	2 22%	3 38%
60-69	2 22%	3 38%
70-79	5 56%	1 13%
80-89	0	1 13%
Marital Status
Single- never married	1 11%	1 13%
Married –live together	4 44%	5 63%
Married- live apart	1 11%	0
Divorced	1	1 13%
Widowed	2	1 13%
Race
Caucasian	6	5 63%
African-American	3	3 38%
Living Situation
Home	7	8 100%
Institution	2	0
Smoking status
Smoker	3	3 38%
Never Smoked	1	3 38%
Ex-smoker	4	3 38%
Alcolhol Use
Yes	4	2 25%
No	5	6 75%

Association between cEEG and clinical delirium

A random effects models reflected average delta/alpha ratios across all subjects were 2.4378 (SD 0.73650) while theta/alpha ratios were 1.5262 (SD 0.2795). (See Table 3) (Additional boxplots are available in supplemental data). Delirium positive subjects had a mean delta/theta ratio of 2.5243 (SD 0.7450) while delirium negative had a ratio of 2.2171 (SD 0.7214). Theta/alpha ratios were 1.5661 (SD 0.3100) for delirium positive subjects and 1.3862 (SD 0.7451) for delirium negative subjects. After controlling for the patient as random and fixed effects (repeated), delta/alpha ratios were b = −.99, p = 0.078 and delta/theta ratios were significant b = 1.52, p = 0.087, nearing statistical significance as a predictor for delirium. Theta/alpha (deep sleep/awake) ratios were significant at b = 1.70, p = 0.048. Given this was a pilot proof of concept study and therefore not powered for statistical significance, these results were better than expected.

Table 3.

Participant Outcomes by Delirium Status

Variables N=17	Delirium positive (n=9)	Delirium negative (n=8)
Discharge Disposition
Home	1	3
Skilled Nursing	1	2
Long-term Care	3	4
Expired	2	1
Length of Stay (LOS)
Hospital LOS	23.0 (9-54)	25.67 (11-41)
ICU LOS	14.88 (5-43)	19.33 (5-41)
Ratios
Delta/Alpha Ratio	2.52428 (1.4788 -3.5678)	2.21714 (1.3004-3.2802)
Theta/Alpha Ratio	1.5661 (1.1867-2.0822)	1.4024 (1.2604-1.6195)
Delta/Theta	2.5243 (1.7793-3.2693)	2.2171 (1.4957-2.9385)

Discussion

All ratios were higher in delirious participants, when compared to non-delirious participants. A study powered to detect statistical significance would require a total of 500 delirious. Despite the small sample size, one ratio reached significance and the others approached significance. Some overlap in ranges between delirious and non-delirious participants may also partially explain lack of significance in some of the ratios.

The search for EEG-based delirium biomarkers has a strong history.³² Prior studies have suggested a distinguishable marked difference in delirium compared to normal awareness. ^{10,13,14,15–17} Since the 1940s, reports of patterns in EEG data have indicated that delirium detection is possible. However, currently available limited lead technologies are not “tuned” for delirium screening and lack a form factor appropriate for mass screening. Nevertheless, only recently, has application of computational algorithms using of spectral density analysis demonstrated that delirium may be detected using a limited lead device. Studies for this purpose have confirmed that the sensitivity and specificity of limited EEG leads are excellent and comparable to those from machines with the traditional 21 leads.^16,33,34 Previous literature supports the notion that EEG is useful for detecting delirium. Due to EEG’s objective nature, interrater reliability concerns, as seen with bedside screening tools, are not a limitation for detection. EEG can be strongly correlated with patient outcomes and may provide additional information for goals of care decisions.^29,35 As a result, screening using a device such as the Ceribell for detection of delirium is practical, feasible and greatly facilitated by using computational algorithms. However, studies attempting to identify biomarkers of delirium, such as our study, are limited by small sample sizes. Therefore, association of findings from this type of screening method are not well established with patient outcomes such as hospital LOS, discharge disposition, and mortality or the effect of intervention on these outcomes. As a result, the capacity for using a point-of-care hand-held cEEG device requires further investigation.

This study is the first to use a 10-lead hand-held point-of-care cEEG device to demonstrate an association between cEEG signal biomarkers, delirium and patient outcomes. Considering these results, the potential usefulness of cEEG derived from hand-held devices to identify patients at risk for delirium and predict patient outcomes cannot be ruled out. With additional clinical validation studies using larger sample sizes, hand-held cEEG derived biomarkers, such as those used in this study, may be an option for enabling early intervention and potentially improve care for patients at risk of developing delirium.

Delirium is particularly dangerous because the ability to recognize and manage it early is lacking. Nurses are capable of interpreting technology data to provide cues upon which they base decisions.^36,37 The simple, noninvasive objective nature makes limited lead EEG ideal for routine delirium screening with the potential to be fast and easy, similar to measuring vital signs. A positive result would provide an early alarm to trigger a more comprehensive workup. Therefore, limited lead EEG may be more clinically relevant than previously realized. In the future, EEG may also potentially provide an objective and quantitative replacement for “altered mental status” in prognostic models such as the Sequential Organ Failure Assessment (SOFA). As the aging population continues to expand rapidly, efficient modalities for delirium screening such as EEG derived from limited lead devices are predicted to be in high demand.

Limitations

The usefulness of limited lead EEG for delirium screening assumes that EEG changes are generalized or diffuse and demarcation or cut-off scores based on delirium status, not currently available, can be determined. While patients with known history of seizures or acute brain injury were excluded, patients with an unknown history of focal changes, such as seizure activity or a structural brain abnormality may confound results. Small sample size and enrollment from a single institution limit generalizability. Thus, the potential for generalizability would require additional multi-center studies with larger sample sizes. Nevertheless, we have shown that data derived from the Ceribell device is capable of detecting EEG changes presumed to differentiate delirium from a non-delirious state. These results encourage further exploration of Ceribell derived EEG data to better understand delirium’s impact on neuroelectrical changes and ways to prevent, manage, and treat it in the hopes of potentially improving patient outcomes.

Conclusion

Although critically ill older adults are at highest risk for mortality, delirium cannot be easily identified by current bedside clinical assessments. Results of this study are promising in that ratios were higher in patients who experienced delirium compared to those who did not. Determining whether point-of-care limited lead EEG may be able to predict adverse patient outcomes remains largely unknown. Therefore, further investigation of limited lead EEG for delirium detection is warranted.

Until EEG changes associated with delirium can be fully described, including demarcation of cut-points, continuing to use validated bedside clinical screening assessments should include frequent re-education and competency validation to improve detection accuracy. Results of screening using validated methods should not be discounted as dementia or another cause and warrant aggressive implementation of nursing strategies for delirium management, as these interventions have been found to be the most effective. Minimizing the use of benzodiazepines, antipsychotics and other medications known to increase delirium risk such as medications included on the Beer’s criteria, may also help to decrease the severity and duration of delirium.