Authors: Mariana Thedim (a Department of Anesthesia, Critical Care and Pain Medicine, Massachusetts General Hospital, Harvard Medical School, Boston, MA, United States), Duygu Aydin (b Department of Anesthesiology and Intensive Care Medicine, Technical University of Munich School of Medicine, Munich, Germany), Gerhard Schneider (b Department of Anesthesiology and Intensive Care Medicine, Technical University of Munich School of Medicine, Munich, Germany), Rajesh Kumar (c Department of Anesthesiology and Perioperative Medicine, University of California Los Angeles, Los Angeles, CA, United States), Matthias Kreuzer (b Department of Anesthesiology and Intensive Care Medicine, Technical University of Munich School of Medicine, Munich, Germany), Susana Vacas (a Department of Anesthesia, Critical Care and Pain Medicine, Massachusetts General Hospital, Harvard Medical School, Boston, MA, United States)
Categories: Article, electroencephalography, general anesthesia, oximetry, neuropsychological tests, postoperative cognitive complications, surgery
Source: Journal of clinical monitoring and computing
Authors: Mariana Thedim, Duygu Aydin, Gerhard Schneider, Rajesh Kumar, Matthias Kreuzer, Susana Vacas
To identify baseline biomarkers of delayed neurocognitive recovery (dNCR) using monitors commonly used in anesthesia.
In this sub-study of observational prospective cohorts, we evaluated adult patients submitted to general anesthesia in a tertiary academic center in the United States. Electroencephalographic (EEG) features and cerebral oximetry were assessed in the perioperative period. The primary outcome was dNCR, defined as a decrease of 2 scores in the global Montreal Cognitive Assessment (MoCA) between the baseline and postoperative period.
Forty-six adults (median [IQR] age, 65 [15]; 57% females; 65 % American Society of Anesthesiologists (ASA) 3 were analyzed. Thirty-one patients developed dNCR (67%). Baseline higher EEG power in the lower alpha band (AUC = 0.73 (95 % CI 0.48–0.93)) and lower alpha peak frequency (AUC = 0.83 (95% CI 0.48–1)), as well as lower cerebral oximetry (68 [5] vs 72 [3], p = 0.011) were associated with dNCR.
Higher EEG power in the lower alpha band, lower alpha peak frequency, and lower cerebral oximetry values can be surrogates of baseline brain vulnerability to dNCR.
Perioperative neurocognitive disorders (PND) constitute one of the most common complications from anesthesia and surgery, especially among older adults [1, 2]. Progressively, older patient populations will continue to rise in the next few decades, potentially reaching over six hundred million elderly surgical patients worldwide, with a concomitant increase in the number of patients suffering from and vulnerable to this pernicious and destructive disease [3].
Surgery and anesthesia can induce brain inflammation, leading to adverse outcomes in less resilient patients, like the older adult and other at-risk populations [4]. A subset of patients that are affected by PND return to their baseline status and sometimes worsen over time [1]. The early identification of patients susceptible to the development of PND may allow for the timely implementation of brain health measures, which might help to prevent a significant number of these cases from emerging in the first place, therefore obviating subsequent brain health deterioration [5].
Within the PND spectrum, postoperative delirium and postoperative delayed neurocognitive recovery (dNCR) can affect patients in the first month after surgery [6]. Both are characterized by postsurgical cognitive impairment and have a significant impact on patients’ lives, even up to one year after surgery [6, 7]. Postoperative delirium is assessed using the Diagnostic and Statistical Manual of Mental Disorders criteria [8], but dNCR is broadly defined by cognitive decline diagnosed up to 30 days after surgery [6]. Previously described criterion for dNCR, specifically, a decrease in two or more points between baseline value and postsurgical testing scores using the Montreal Cognitive Assessment (MoCA), has shown high sensitivity and specificity for a threshold diagnostic [2].
Redeploying widely available apparatuses and tools within the perioperative period can deepen the understanding of PND and pave the way for new diagnostics, preventive, and treatment strategies. Electroencephalogram (EEG) and cerebral oximetry monitors are commonly used intraoperatively, prompting tailored anesthetic modifications with the goal of preventing postoperative adverse neurologic outcomes. The application of these devices as feasible and noninvasive surrogates of brain vulnerability in the preoperative period is being acknowledged, but robust evidence is lacking [9, 10]. These tools have shown their efficacy in offering valuable insights into postoperative delirium risk assessment [11, 12], but similar research focused on dNCR remains scarce.
We aimed to assess the role of baseline EEG and cerebral oximetry as biomarkers of dNCR. We hypothesized that vulnerable patients predisposed to develop dNCR would demonstrate distinct EEG patterns and lower cerebral oximetry values in the preoperative period.
We assessed a subset of patients from two observational studies approved by the Institutional Review Board of the University of California Los Angeles (#19–001597 and #20–001456). Written consent was obtained from all patients before study involvement. This study adhered to the STROBE reporting checklist and was registered on clinicaltrials.gov before any study activity.
We included adult patients submitted to abdominal, urologic, and gynecological surgeries under general anesthesia between January 2020 and August 2023. Exclusion criteria have been reported previously [13], cardiovascular, hepatic failure, renal or neurologic disease, severe depression, excessive alcohol use, repeated use of opioids or other illicit substances, or body weight >125 kg. All patients received the same anesthetic regimen, including induction with propofol and fentanyl, followed by maintenance with sevoflurane. Continuous noninvasive mean arterial pressure was maintained within 20% of the patient reference value or above 65 mmHg using vasopressors, as needed. No other adjunct medications were administered.
General anesthesia was guided using a bifrontal EEG (SedLine^®^, Masimo, Irvine, CA), with a target patient state index between 20–50 to avoid patterns compatible with burst suppression. According to manufacturer instructions, electrodes were placed on the patients’ forehead five minutes before induction. The skin was prepped with alcohol before electrode placement to reduce the impedance effect. The recording was maintained intraoperatively and suspended after patient extubation. The EEG was recorded with a sampling rate of 178 Hz, stored in the .edf format and converted to MATLAB (The Mathworks, Natick, MA, USA) .mat files. A 15 s of artifact-free EEG during wakefulness (eyes closed) in the two minutes before anesthesia induction was visually identified. All episodes were filtered using a Butterworth forward-backward zero-phase bandpass routine to 0.5–45 Hz, and data points with amplitudes above a 100 μV-threshold were excluded as artifacts.
Power spectral densities (PSD) of the EEG episodes were calculated using Welch’s method with a frequency resolution of 0.7 Hz, and the median PSD of the channels Fp1 and Fp2 was used for further calculations. Absolute frequency band-powers were calculated from the PSD as the sum of power in the frequency ranges 0.5–4 Hz for delta band, 4–8 Hz for theta band, 8–13 Hz for alpha, 13–25 Hz for beta and 25–45 Hz for low gamma bands. Spectral edge frequency (SEF95) was also calculated from the PSD to assess if there are changes between the groups, as SEF95 is mostly included in anesthesia monitoring systems and, therefore, has potential as a quick and easy monitoring tool for dNCR.
Alpha peak frequency was determined using the Fitting Oscillations and One over f (FOOOF) toolbox for MATLAB [14] to achieve greater frequency resolution than afforded by Welch’s method. For the application of this toolbox, the following settings were maximum number of peaks = 3, minimum peak height = 0.3, peak width limits 2–10. Since the EEG was filtered to 0.5–45 Hz, the frequency range for the toolbox was chosen as 1–40 Hz to avoid overfitting the artifacts caused by the filters at the edges. Changes in the exponent and slope of the aperiodic (1/f-like) component of the EEG were also examined, as suggested by the toolbox authors [14].
Cerebral oximetry was assessed before induction and throughout the surgery until extubation using Masimo O3^™^ (Masimo) readings. Sensors were applied to the forehead according to the manufacturer’s instructions. Baseline values were defined as cerebral oximetry values before induction while the patient was breathing room air and before any drug administration. Intraoperative cerebral desaturation events were defined as a decrease of 20% compared with baseline values [15].
Trained clinical study team members performed neurocognitive assessment within five days of the patient’s scheduled surgery (baseline) and within one week after surgery (postoperative). The MoCA [16] and the Mini-Mental State Exam version 2 (MMSE2) [17] were performed at both time points using appropriate versions to limit learning effects.
The primary outcome of this study was the development dNCR. We measured and defined dNCR as a decrease in global MoCA by two scores between preoperative baseline and postsurgical values [2].
The confusion assessment method [18] was used to evaluate postoperative delirium upon arrival at the post-anesthesia care unit and repeated subsequently, twice per day, until patient discharge from the hospital.
Statistical analysis was performed using IBM SPSS Statistics for Windows, Version 28.0 (Armonk, NY: IBM Corp), Python 3.11.5 (Anaconda Software Distribution, Anaconda Inc., Austin, TX), and MATLAB R2023b. Continuous variables are presented in mean ± standard deviation (SD) or median [interquartile range, IQR] according to normality. Normality was assessed using the Shapiro Wilk test. Categorical variables are displayed as number of occurrences (percentage, %).
Comparison of dNCR to non-dNCR patients was achieved with Pearson chi-square, or Fisher’s exact test for categorical variables and the Mann-Whitney U-test for continuous variables. Demographic and cognitive variables were regarded as significant if p-values were <0.05.
For EEG variables, the area under the receiver operating characteristic (AUC) values with 10k-bootstrapped 95% confidence intervals (CI) or odds ratio (OR) were calculated as measures of effect size, as appropriate for variable type [19]. AUC values >0.7 were considered acceptable effect sizes indicative of a clinically relevant change [20]. Since this was a sub-study, no previous sample size estimation was conducted.
Of the 48 possible participants, two did not complete postoperative cognitive assessment, totaling 46 patients included in this analysis (median [IQR] age, 65 [15]; 57% females), Supplementary Fig. 1. Thirty-one patients (67%) developed dNCR. Participant characteristics did not differ between groups at baseline, except for the number of years of education, which was lower in the dNCR group (16 [2] vs 14 [3], p=0.049; Table 1).
None of the baseline overall cognitive scores or baseline-specific domain scores were statistically different between patients who developed dNCR when compared to the ones who did not (Table 2). No patient developed postoperative delirium.
Of the 46 patients, 34 had EEG recordings. Recordings with a sampling rate smaller than 100 Hz (n=4) and those stored under inaccurate dates by the EEG monitor (n=6) were rejected. One patient did not have an awake baseline recording. In the end, the baseline EEG of 23 patients was included, 14 of whom developed dNCR (Table 3).
Group-level PSD were compared between dNCR and non-dNCR patients (Fig. 1a). In the awake, eyes closed baseline period, a trend of a higher power in the frequencies 7–10 Hz was observed in the dNCR group (8.39 [5.64] vs. 3.42 [5.42] dB), i.e. in the lower alpha band, with an effect size of 0.73 (CI 0.48–0.93, p=0.073; (Fig. 1b)). No significant changes were observed in the remaining frequency bands or SEF95 (Table 3).
Lower alpha band change was accompanied by a lower alpha peak frequency in the dNCR group (10.01 [0.91] Hz vs. 10.93 [0.87] Hz), p=0.065, AUC=0.83, CI 0.48–1; (Fig. 1c). However, it is noteworthy that the FOOOF algorithm failed to detect an alpha peak in eight patients who developed dNCR and in four patients who did not. The OR for not having an alpha peak was 1.07 (CI 0.20–5.77). The groups had no significant differences in the aperiodic slope or offset (Table 3).
A total of 34 patients had baseline cerebral oximetry data. Patients who developed dNCR had lower baseline cerebral oximetry values (%) on the right side (68 [5] vs 72 [3], p=0.011). Within the dNCR group, seven patients (30%) had at least one episode of intraoperative cerebral desaturation, however, this was not associated with higher rates of dNCR (Table 4).
Our study revealed baseline brain vulnerability features that were associated with dNCR. In our surgical population, a trend to a higher power in the EEG frequencies between 7 to 10 Hz, lower alpha peak frequency, and lower baseline cerebral oximetry values were associated with postoperative cognitive impairment in the form of dNCR diagnoses.
A significant proportion of our cohort (67%) developed dNCR. Although this value is higher than previous reports that used the same diagnosis threshold of a 2-point decrease from baseline scores, high incidences of dNCR are not uncommon [2, 21]. This may be simply because we tested for it. In fact, dNCR was defined relatively recently [6]. Further, like other PND, cognition is not consistently tested and recorded in clinical practice, leading to potential underdiagnoses among all patient populations and across ages. This is due, in part, to the fact that neurocognitive assessments before and after are not part of the majority of preoperative or discharge hospital protocols [6]. Advances in surgical and anesthesia techniques and care drive more patients to the operating room. Early identification of vulnerable patients, those with less resilience to the anesthetic and surgical insult, could drive preventative strategies, testing protocols, and additional targeted postoperative care rather than the traditional clinical focus on addressing adverse events.
This should include intensive monitoring and targeted interventions that can be applied early in the perioperative period [1]. This approach is routinely implemented for cardiac, pulmonary, and hematologic systems. Although the central nervous system is the main target of most anesthetic procedures, no preoperative diagnostic test or perioperative monitoring standard exists for the brain. This is despite the fact that PND is a serious postoperative complication after surgery, independent of the anesthetic regimen [22]. Brain monitors, such as the EEG and cerebral oximetry, are commonly used in the intraoperative period and can provide valuable insights into PND research and clinical practice. This outlook and emphasis is even more critical and powerful given the imminent and inevitable innovations of smaller and affordable sensors and computing power within the area of brain health monitoring, even extending to the progressive popularity of connected, wearable, and individually-tailored devices.
Recommendations to evaluate preoperative cognitive reserve to further understand postoperative cognitive trajectories were ineffective as clinicians largely overlooked them [23]. Additionally, as it occurs in other physiologic systems, it is possible that a comprehensive evaluation of the brain requires complementary objective tests besides cognitive assessment. Identifying biomarkers associated with brain vulnerability can potentially deploy a holistic brain evaluation and further develop risk stratification strategies [4, 24]. Our cohort’s distinct baseline EEG features and lower baseline cerebral oximetry were observed in patients who developed dNCR, revealing a subtle, unrecognized, but important brain vulnerability. Recent guidelines support this by encouraging intraoperative EEG-based anesthetic management to improve brain health [25]. During the intraoperative period, lower alpha power was previously considered a surrogate of brain vulnerability because it was associated with a higher propensity for burst suppression [26]. Specific emergence EEG patterns, such as deviation of spindle dominant activity (alpha over delta power) and increased alpha and/or beta power, were also linked to early postoperative delirium [27, 28]. In addition, preoperative decreased beta and gamma power, lower SEF, and lower alpha attenuation were also associated with postoperative delirium or with one of its cardinal features [11, 29–31]. This increased brain vulnerability was present in the preoperative period, offering a possible physiological thread or throughline that carries its distinct brain weakness and vulnerability forward through to the more pronounced surgical or anesthetic exposure, thus further demonstrating the utility of using baseline EEG as a risk marker. This prospective observational study extends this preoperative brain vulnerability to dNCR. This harmful postoperative complication is frequently not diagnosed or recognized, despite the fact that this study shows it may be present in two out of three patients. dNCR before discharge was associated with adverse cognitive and memory sequelae even after one year of surgery, clearly indicating a deterioration of the patient’s cognitive trajectory [7]. Early identification of brain vulnerability increases awareness of dNCR and allows for the timely implementation of protective measures in the perioperative period.
Cerebral oximetry is commonly used in settings with high risk for brain ischemia. By providing real-time, noninvasive recognition of cerebral hypoperfusion events, it allows for the possible implementation of measures to restore brain oxygenation [15]. Perioperative imbalance in brain oxygen supply and demand can lead to neurovascular changes [32]. Although the threshold for intraoperative desaturation events varies with the device used, avoiding these events can decrease the incidence of postoperative complications, including PND [33–35]. Besides its valuable intraoperative usage, cerebral oximetry may also aid in preoperative risk calculations for delirium and mortality [12, 36]. Our results support the role of baseline cerebral oximetry as a potential biomarker for the advent of dNCR [37]. The fact that only the right cerebral oximetry values were associated with dNCR warrants further research. This finding shows that unrecognized brain insult can be identified in the preoperative period. The brain was previously purposed as an index organ with the potential to represent an overall vulnerability to external stressors [38]. Therefore, bilateral monitoring should be considered, even for younger adults, as it can identify hemispheric differences and provide valuable information on the overall brain status. A recent report found no association between baseline cerebral oximetry, preoperative relative alpha power, or preoperative fronto-parietal functional connectivity with postoperative cognitive function scores, which countervail the relative consensus and literature on the topic [39]. We focused on the same postoperative period, but our results were significantly different, possibly due to the approach and methods implemented. For instance, our study was focused almost exclusively on dNCR and its diagnosis. Further, our study utilized a different set of neuropsychological assessments in the preoperative period, along with EEG and bilateral cerebral oximetry sensors, uniform maintenance of anesthesia for all patients guided by EEG and avoidance of burst suppression. While it is important to challenge consensus views on the topic of PND and its possible substrata, it is equally important to reproduce the results that help form them. In addition, depending on the dNCR diagnosis threshold, significant variations might affect its prevalence, frequency, and incidence [2, 40].
Our study has several limitations. First, we recognize our study sample is small, yet we found previously unreported, potential biomarkers for dNCR that can be employed in larger, future cohorts. Second, we only evaluated dNCR within one week of the postoperative period, potentially missing a dNCR diagnosis in the later stages of the postoperative period. Nonetheless, the immediate postoperative period is frequently excluded from research studies, even after nascent efforts to implement uniform definitions for PND [6]. This period is one of the most difficult to assess due to its logistical complexity, but more studies are needed to understand its impact on long-term cognitive trajectories. Third, EEG and cerebral oximetry data could not be collected for all patients, precluding data from our study conclusions. Further, while the cerebral oximetry values in this report are statistically significant, a difference of 4% may not be physiologically or clinically significant.
In this cohort of surgical patients, distinct EEG features and lower baseline cerebral oximetry values were associated with dNCR in the immediate postoperative period. These findings highlight the positive potential for holistic risk stratification by integrating neuropsychological evaluations with tools and monitors that are commonly used in surgery and anesthesia.