This retrospective cohort study was approved by the Ethics Committee of Peking University International Hospital (2024-KY-0062, Beijing, China) and was conducted in accordance with the Declaration of Helsinki. The study enrolled adult patients with DoC who received SCS electrode implants between January 2021 and April 2024 and whose representatives provided written informed consent.

The patients were categorized into three subgroups based on the pathophysiological mechanisms of their conditions: anoxic ischaemic encephalopathy (AIE), cerebrovascular accident (CVA), and TBI. As these differences may influence GABAergic receptor sensitivity to propofol, subgroup analysis was planned.

The inclusion criteria were (1) patients with DoC aged 18–60 years who met the International Classification of Diseases diagnostic criteria for DoC; (2) diagnosis of DoC, including UWS and MCS; (3) patients who required implantation of SCS electrodes during the study period; and (4) patients for whom consent was obtained from the legally authorized representatives for study participation.

The exclusion criteria were (1) severe cardiac, pulmonary, hepatic, renal, or other organ dysfunction; (2) coagulation dysfunction; (3) inability to obtain consent from the legal guardian; (4) known allergy to propofol or other medications related to the study; (5) history of psychiatric disorders before DoC onset; and (6) The concurrent administration of neuroexcitatory pharmacological agents, such as dopaminergic medications, zolpidem, or amantadine, or the engagement in alternative neuromodulation therapies, including transcranial direct current stimulation, repetitive transcranial magnetic stimulation, or vagus nerve stimulation, intended to enhance wakefulness or facilitate the recovery of consciousness within a three-month period preceding the implantation of SCS.

In this study, a senior neurologist confirmed the diagnosis through a comprehensive evaluation incorporating historical data, clinical presentation, and ancillary tests. The level of consciousness of patients with DoC was assessed preoperatively and 3 months postoperatively using the Coma Recovery Scale-Revised (CRS-R). The CRS-R comprises six subscales: auditory, visual, motor, orofacial/verbal, communication, and arousal. To ensure objective and consistent assessments, the scale was administered by professionally trained personnel. The scores were used to measure baseline consciousness levels and postoperative recovery outcomes.

The perioperative management of all patients in this study adhered to a standardized anesthetic protocol. Comprehensive preoperative monitoring was performed, including assessments involving electrocardiogram, pulse oximetry (SpO₂), core temperature, and invasive radial artery blood pressure monitoring. After establishing invasive arterial access, blood samples were collected for blood gas and biochemical analysis.

For patients in a UWS, a respiratory circuit connected to a tracheal catheter was used to facilitate spontaneous ventilation. The anesthesia induction regimen included intravenous propofol (1 mg·kg⁻¹) (31), sufentanil (0.3 μg·kg⁻¹), and rocuronium bromide (0.5 mg·kg⁻¹). Mechanical ventilation support commenced as soon as spontaneous respiration ceased. PetCO₂ was monitored throughout the procedure and maintained within the target range of 35–45 mmHg.

During anesthesia maintenance, continuous infusion of propofol (1–1.5 mg·kg⁻¹·h⁻¹) and remifentanil (0.15 μg·kg⁻¹·min⁻¹) was administered via a micro pump. The propofol infusion rate was titrated to maintain a Bispectral Index (BIS) of 40–60 (BIS Vista™ Monitor, Medtronic, USA) (32). Hemodynamic parameters, including mean arterial pressure (MAP) and heart rate, were monitored in real-time via invasive arterial access. If the MAP decreased by >20% from baseline or the BIS was >60, the propofol infusion rate was increased by 0.2 mg·kg⁻¹·h⁻¹ every 5 min until the target parameters were achieved. Conversely, if BIS dropped to <40 or haemodynamic instability persisted despite vasopressor support, the infusion rate was reduced by 0.1 mg·kg⁻¹·h⁻¹. If the patient’s systolic blood pressure dropped by >20% from baseline during surgery, 3–6 mg of ephedrine was administered intravenously to raise the blood pressure. Norepinephrine infusion was initiated as needed. Based on arterial blood gas analysis, adjustments to the patient’s acid–base balance and electrolyte concentrations were made to keep them within normal physiological limits.

During surgery, patients with DoC were placed in the prone position. X-ray imaging was used to identify the T₇–T₈ gap, followed by needle puncture at 45° into the skin and epidural space. X-ray guidance facilitated electrode placement, ensuring that the upper edge of the electrode aligned with the midpoint of the C2 vertebral body in the epidural median. After confirming the normal impedance, the electrode was secured.

Following the procedure, propofol and remifentanil infusions were discontinued, and the patient was moved to the post-anesthetic recovery unit. Once spontaneous respiration returned, intravenous sugammadex sodium was administered to reverse rocuronium bromide. The patients were then transferred to the rehabilitation department for further treatment once SpO₂ levels and circulatory stability met the pre-established criteria for discharge from the recovery room.

To ensure EEG signal stability and reliability, an experienced neurologist positioned the disc electrodes in the patient’s ward 30 min before anesthesia. Disc electrodes were affixed to the patient’s scalp using a medical collodion to ensure secure adhesion. Medical mesh sleeves were used for additional fixation to minimize the risk of electrode dislodgement due to patient movement. A specialized conductive gel was also applied to maintain electrode impedance <5 kΩ to ensure signal stability and quality. EEG acquisition and monitoring were performed using a Nicolet system (Natus Medical Inc., USA), to guarantee accurate electrode placement, the neurologist meticulously followed the international 10–20 system guidelines, using anatomical landmarks and measurements to precisely position each electrode. The positioning process was carefully checked and double-checked by another expert to avoid errors. During prolonged monitoring, especially for patients with varying levels of consciousness, continuous monitoring of electrode status was conducted. Moreover, to address potential motion artifacts, several measures were taken. The medical mesh sleeves provided a tight but comfortable fit, allowing for some flexibility as the patient moved, thus reducing the risk of electrode displacement. The reference electrodes were averaged electrodes (AVE), and EEG signals were recorded using 19-lead electrodes (Figure 1). The specific lead positions included the left prefrontal (FP1), right prefrontal (FP2), left frontal (F3), right frontal (F4), frontal midline (Fz), left center (C3), right center (C4), left parietal (P3), right parietal (P4), and parietal midline (Pz). Additional electrodes were placed at the center midline (Cz), left occipital (O1), right occipital (O2), left anterior temporal (F7), right anterior temporal (F8), left mesial temporal (T3), right mesial temporal (T4), left posterior temporal (T5), and right posterior temporal (T6).

Scalp EEG recordings were acquired at four specified time points in relation to the administration of propofol: 1. Baseline (Pre-anesthesia): This was recorded 30 min after the placement of electrodes in the ward, prior to the administration of any anesthetic, with patients in a resting state. 2. Induction Phase: This phase commenced upon the completion of a propofol bolus (1 mg·kg⁻¹) and continued until the cessation of spontaneous respiration, typically occurring 30–90 s post-injection. 3. Steady-State Anesthesia: During this phase, anesthesia was maintained throughout surgical stimulation (SCS electrode implantation), with the propofol infusion adjusted to achieve a target BIS of 40–60. A 5-min artifact-free segment was extracted after at least 15 min of stable BIS (±5 units). 4. Post-Recovery: This recording was conducted 30 min following the discontinuation of the propofol infusion in the post-anesthesia care unit (PACU), after the return of spontaneous respiration and the achievement of hemodynamic stability (mean arterial pressure within 20% of baseline).

The parameters for acquiring EEG signals were as follows: A linear phase finite impulse response (FIR) low-pass filter with a cut-off frequency of 0.5 Hz and an order of 1,000 was applied to exclude signals below the physiological EEG frequency range. Additionally, a high-frequency FIR filter with a cut-off frequency of 200 Hz and an order of 500 was used to eliminate potential high-frequency noise. A 2,000 Hz sampling rate was implemented to ensure high temporal resolution. Electrode impedance was continuously monitored in real-time using a dedicated impedance meter (Natus Medical Inc., USA). Adjustments were made using a conductive gel to maintain impedance levels <5 kΩ throughout the recording process.

The EEG data were pre-processed using the EEGLAB toolbox in MATLAB (version 2022a, MathWorks Inc., USA). The specific preprocessing steps were as follows:

Elimination of baseline drift and head motion noise: The pop_eegfiltnew.m function from the EEGLAB toolbox was used to apply a high-pass FIR filter, with a cut-off frequency of 0.1 Hz, utilizing a Hamming window and an order of 3,300, to mitigate low-frequency drifts. A zero-phase bidirectional filtering technique was implemented to prevent phase distortion.

Elimination of high-frequency signal interference: A linear phase FIR filter was applied to remove signal components with frequencies >45 Hz to mitigate high-frequency noise interference. While the filter reduced most high-frequency noise such as 50/60 Hz line noise, the remaining EMG artifacts (usually >50 Hz) were handled using independent component analysis (ICA) and visual inspection, as explained in the ‘Removal of apparent noise’section below.

Downsampling: The EEG signal was downsampled from 2,000 Hz to 100 Hz to decrease the computational load while preserving signal integrity, ensuring that the essential components of the EEG were maintained.

Identification of poor-quality electrodes and problematic data segments: Electrodes with poor conductivity and anomalous data segments were identified using visual inspection and automated detection algorithms. The veg plot.m function was used to visualize the data to allow clear identification of erroneous guides and problematic segments. These segments were removed through visual inspection, and interpolation was applied to repair any signal anomalies caused by electrode detachment or damage.

Removal of apparent noise: ICA was performed using the runica algorithm, based on the Infomax approach, within the EEGLAB environment. Artefactual components, such as those arising from ocular, muscular, or electrode noise, were identified by examining topographic maps, power spectra, and temporal characteristics. Artefactual components were classified using the following criteria:

Re-referencing: To enhance the signal quality and accuracy, the processed EEG data were re-referenced to an averaged reference.

Spectral analysis was performed using the multi-taper method, implemented in the Chronux toolbox (version 2.11; http://chronux.org/, accessed 8 September 2024). Time-frequency analysis was conducted across all EEG channels throughout the anesthetic period, focusing on the frequency range of 0.1–45 Hz. The brain regions were categorized into five areas: frontal (F3, Fz, F4), central (C3, Cz, C4), parietal (P3, Pz, P4), occipital (O1, O2), and temporal (T3, T4). The mean result of the time-frequency analysis for each region was calculated. Power spectra from the time-frequency analysis were compared across anesthesia states and divided into five frequency bands: δ-wave (1–4 Hz), theta-wave (4–7 Hz), α-wave (7–12 Hz), β-wave (12–30 Hz), and low gamma (30–45 Hz), with distinct background colors for differentiation. The multi-taper spectral parameters were as follows: window length = 5 s; overlap = 2.5 s, time-bandwidth product = three, and tapers = five.

PE was calculated as follows:

In Equation 1, x is sorted by its corresponding j when identical elements exist.

where larger and smaller values indicate more complex and regular sequences, respectively. We choose m = 6 and τ = 1 as parameters to calculate the sequencing entropy (35–36).

The CRS-R scale was used to assess the potential for consciousness recovery in each patient with a DoC. Patients who met the specified criteria were considered to exhibit signs of consciousness improvement. Patients initially classified as UWS were deemed to have shown improvement if they attained MCS–, MCS+, or emerged from a minimally conscious state (EMCS). Similarly, patients initially classified as MCS– were considered to have shown improvement if they reached MCS+ or EMCS. If the clinical diagnosis after the treatment period did not indicate improvement compared with the initial assessment, the clinical outcome was classified as invalid (Table 1) (37, 38).

All statistical analyses were conducted using IBM SPSS Statistics for Windows, version 22.0 (IBM Corp., USA) and R software (version 4.4.1).¹ Before analysis, the normality of the data was assessed using the Kolmogorov–Smirnov test. For variables that did not meet the normality assumptions, such as CRS-R scores and PE values, non-parametric tests were applied. Between-group comparisons were executed using the Mann–Whitney U or Kruskal–Wallis tests, For statistical tests with multiple comparisons, the Benjamini–Hochberg procedure was used to apply false discovery rate correction to control for false-positive results, which is more flexible and reasonable than the Bonferroni correction and better preserves the power of statistical analysis, especially for exploratory research. Within-group changes, particularly those before and after anesthesia, were analyzed using the Wilcoxon signed-rank test. Categorical variables, such as diagnostic classification, were compared using Fisher’s exact test. To control for potential confounding variables, including age, sex, and baseline CRS-R scores, multivariate linear regression models were developed with delta permutation entropy (ΔPE) and CRS-R improvement as the dependent variables. The effect sizes were reported as Cohen’s d for parametric tests and rank-biserial correlation (r) for non-parametric tests. In the PE analysis, regional differences were assessed using mixed-effects models incorporating random intercepts for individual subjects. Two-way interaction analyses (group × anesthesia state) were performed, and P_FDR values were calculated. Pearson correlation was used to assess the link between ΔPE and ΔCRS-R scores, Linear regression quantified how well ΔPE explained ΔCRS-R, shown by the R² value. The results are reported as mean ± standard deviation or median (interquartile range), and p < 0.05 was considered statistically significant.

The study enrolled 40 adult patients with DoC who received SCS electrode implants during the study period. Among these 40 patients, the duration of neurological impairment before surgery ranged from 1 to 29 (median: 4) months.

The AIE, CVA, and TBI groups included 11, 14, and 15 patients, respectively. The Preoperative Coma Recovery Scale-Revised (CRS-R) total scores varied between 4 and 11 (AIE group: 4–7; CVA group: 5–11; TBI group: 6–10), with no significant differences in baseline scores across subgroups (p > 0.05).

The patients’ data are presented in Table 2. The preoperative general information did not differ significantly among the three groups (p > 0.05).

After low-dose propofol administration, all patient groups exhibited reduced β and γ EEG power in the prefrontal and frontal regions (p < 0.05). The AIE and TBI groups also showed increased δ-wave and θ-wave power (p < 0.05), while the CVA group experienced a decrease in β- and γ-wave power in the prefrontal region and only a γ-wave decrease in the frontal region, with no significant δ- and θ-wave changes. The CVA group maintained similar α-wave activity in both regions (p < 0.05). The PSD of EEG activity in the prefrontal and frontal regions of the AIE and TBI groups decreased in the high-frequency bands (β and γ) and increased in the low-frequency bands (δ and θ), particularly in the TBI group. The reduction in γ-wave power in the frontal region was especially pronounced, whereas changes in δ and θ were less marked (p > 0.05). In the CVA group, the PSDs of the β and γ were significantly reduced in the prefrontal region. The CVA group also showed EEG activity in both the prefrontal and frontal regions that resembled α-wave characteristics (Figure 2).

After low-dose propofol administration, significant β and γ power suppression occurred in prefrontal (FP1/FP2) and frontal (F3/F4/Fz) regions across all groups (all PFDR < 0.05; Figure 2). AIE group: Increased δ (p = 0.02) and θ power (p = 0.03); TBI group: Increased δ (p = 0.01) and θ power (p = 0.007); CVA group: Isolated γ suppression in frontal regions (p < 0.001) without δ/θ changes (Pδ = 0.21; Pθ = 0.41) (Figure 3).

At the whole-brain level, θ- and δ-wave power did not differ significantly before and after anesthesia among the three patient groups (p > 0.05). However, the AIE group showed significantly decreased, α-wave power in the frontal region following anesthesia (P_FDR = 0.03). Similarly, the TBI group showed reduced α-wave power in the parietal and occipital regions (p < 0.05), but no other brain regions (p > 0.05). The AIE and TBI groups demonstrated significantly decreased β-wave power across the entire brain (p < 0.001–0.05), whereas the CVA group showed a significant decrease only in the prefrontal region (p < 0.05). Except for the parietal region in the CVA group, all brain regions in the three groups showed significantly reduced γ-wave power following anesthesia (p < 0.001–0.05) (Figure 4).

In the anaesthetized state, all brain regions of patients across the three etiological subgroups exhibited decreased entropy relative to the preoperative baseline state. Notably, the AIE group demonstrated higher entropy levels than the other two groups under identical conditions. The decreased entropy was particularly pronounced in the prefrontal and frontal brain regions following anesthesia compared with the parietal and occipital regions (Figure 5).

Table 3 shows the results of the comparisons of preoperative and 3-month postoperative level of consciousness among the three patient groups.

The radar charts of standardized CRS-R sub-scores (Figures 6A,B) showed significant differences in the recovery of consciousness before and 3 months after surgery.

Postoperative assessment in the TBI group revealed significant enhancements across multiple dimensions, with marked improvements in motor (from 0.32 to 0.51), visual (from 0.29 to 0.54), and auditory (from 0.23 to 0.54) functions, with change ranges (Δs) > 0.19 and statistical significance (p < 0.05). These findings suggest robust neuroplasticity. Notably, some patients achieved full consciousness, with two individuals attaining EMCS. Their motor function scores reached the maximum value of 1.0, and the postoperative radar chart approximated a regular hexagon, indicative of a well-rounded multidimensional recovery.

Patients in the CVA group showed significant improvements in local functional recovery, particularly arousal (from 0.63 to 0.80, Δ = 0.17) and motor (from 0.35 to 0.55, Δ = 0.20) functions (p < 0.01). However, verbal communication function showed minimal change (from 0.05 to 0.15, Δ = 0.10, p = 0.12), indicating persistent language dysfunction. The postoperative radar chart exhibited significant expansion along the motor and arousal axes, whereas the language communication axis demonstrated only slight alterations.

In the AIE group, recovery progression was limited; aside from arousal function (0.48–0.61, Δ = 0.13, p = 0.08), the change in other sub-items was <0.05 (e.g., motor function: 0.26–0.29), indicating slow neurological rehabilitation. The high overlap of radar charts, with a substantial overlap rate of preoperative and postoperative radar areas, suggested stagnant overall recovery (Table 4).

Group comparisons revealed that the TBI group exhibited the most substantial overall improvement (mean Δ = 0.21), significantly surpassing those of the CVA (Δ = 0.15) and AIE (Δ = 0.03) (F[2,36] = 9.84, p < 0.001, one-way ANOVA) groups. The post-hoc Tukey test further confirmed significant differences between the TBI and CVA groups (p = 0.02), as well as the AIE group (p < 0.001).

The pronounced differences observed among the TBI, CVA, and AIE subgroups in terms of neurophysiological responses (Figures 2–5; Table 2) and clinical outcomes (Figures 6A,B; Tables 2, 3) highlight the significant etiological heterogeneity present within the DoC cohort. Statistical analyses consistently revealed that both the extent and pattern of propofol-induced EEG modulation, such as prefrontal-parietal β/γ suppression in TBI/CVA compared to reduced frontal α in AIE, along with ΔPE differences. And subsequent functional recovery, as measured by CRS-R total and sub-score improvements, were dependent on the underlying etiology. This analysis of heterogeneity confirms that the distinct pathophysiological mechanisms associated with TBI, CVA, and AIE play a critical role in shaping both the brain’s response to anesthetic challenges and its potential for recovery following neuromodulation.

In the TBI group, there was a significant negative correlation between ΔPE in the α band and ΔCRS-R scores (y = 3.16–9.86x, R² = 0.45, p = 0.006, r = −0.67). The CVA group showed a similar negative correlation (y = 1.03–6.67x, R² = 0.45, p = 0.009, r = −0.67). However, no significant correlation was found in the AIE group (y = 0.84–0.7x, R² = 0.01, p = 0.791, r = −0.1). Figure 7 illustrates these correlation results.

The mixed choice model analysis revealed significant differences in entropy change under anesthesia between etiological groups (TBI/CVA vs. AIE). The TBI/CVA group showed a 0.21 lower entropy change than the AIE group (estimate = −0.21, SE = 0.08, t = −2.63, p = 0.009, 95% CI −0.37 to −0.05), indicating a more pronounced reduction in entropy under anesthesia. Overall, entropy was significantly lower in the anesthetized state compared to baseline, with a mean change of −0.15 (estimate = −0.15, SE = 0.07, t = −2.14, p = 0.033, 95% CI −0.29 to −0.01). The interaction term analysis showed a significant difference in entropy change under anesthesia between the TBI/CVA and AIE groups, with an interaction estimate of 0.18 (SE = 0.09, t = 2.00, p = 0.046, 95% CI 0.01–0.35). This finding highlights the varied responses to anesthesia across different causes. The model fit well, showing some individual variability with a random intercept standard error of 0.25 and residuals standard error of 0.12.