Developing predictive nomogram models using quantitative electroencephalography for brain function in type a aortic dissection: a prospective observational study

Abstract

Background:

Type A aortic dissection (TAAD) remains a significant challenge in cardiac surgery, presenting high risks of adverse outcomes such as permanent neurological dysfunction and mortality despite advances in medical technology and surgical techniques. This study investigates the use of quantitative electroencephalography (QEEG) to monitor and predict neurological outcomes during the perioperative period in TAAD patients.

Methods:

This prospective observational study was conducted at the hospital, involving patients undergoing TAAD surgery from February 2022 to January 2023. QEEG parameters, including the dynamic amplitude-integrated electroencephalography (aEEG) grade, which assesses changes in brain function over time, alongside aEEG and relative band power (RBP), were monitored and analyzed to assess brain function preoperatively, intraoperatively, and within 2 hours postoperatively. A predictive nomogram model was developed using these QEEG metrics along with other clinical variables to forecast neurological outcomes.

Results:

In this study, we analyzed the factors contributing to adverse outcomes (AO) and transient neurological dysfunction (TND) following TAAD surgery. For AO, multivariable analysis identified pre-mental status (odds ratio [OR] = 4.652, 95% confidence interval [CI] = 2.316–10.074, P < 0.001), cardiopulmonary bypass time (OR = 1.014, 95% CI = 1.006–1.023, P = 0.001), and dynamic aEEG grade (OR = 9.926, 95% CI = 4.493–25.268, P < 0.001) as independent risk factors. The AO model showed high discriminative ability with an area under the curve of 0.888 (95% CI = 0.818–0.960) and good calibration (Brier score = 0.0728). For TND, significant preoperative differences included dynamic aEEG grade (P < 0.001) and Log(Post-RBP Alpha%) (6.00 vs. 4.00, P < 0.001). Multivariable analysis identified cardiopulmonary bypass time (OR = 1.014, 95% CI = 1.006–1.023, P = 0.001), Post-RBP Alpha% (OR = 0.263, 95% CI = 0.121–0.532, P < 0.001), and dynamic aEEG grade (OR = 12.444, 95% CI = 5.337–30.814, P < 0.001) as independent risk factors. The TND model had an area under the curve of 0.893 (95% CI = 0.844–0.941) and good calibration (Brier score = 0.125). These findings highlight the role of QEEG in predicting postoperative neurological dysfunction in TAAD patients.

Conclusion:

Through perioperative QEEG monitoring of TAAD patients, combined with clinical indicators such as cardiopulmonary bypass time and preoperative mental status, we developed clinical predictive models for AO and TND after surgery. These models allow for early detection of postoperative brain function impairment, as assessed by QEEG parameters monitored intraoperatively and during the first 2 hours after surgery, a period chosen based on clinical definitions of delayed awakening and supported by the findings of this study. This study provides evidence supporting postoperative brain function monitoring in TAAD patients, with potential clinical implications for improved outcomes.

Introduction

Type A aortic dissection (TAAD) remains a significant surgical challenge, despite notable advances in medical technologies over the last 50 years^[1]. Currently, deep hypothermic circulatory arrest (HCA) is the leading method used for protecting the brain during surgeries for TAAD, and it has substantially improved the success rates of these procedures. Nevertheless, complications related to postoperative brain health continue to pose significant concerns^[2,3]. The occurrence of permanent neurological dysfunction (PND) after surgery, affecting long-term survival, has been observed with frequencies ranging from 10% to 30%^[4-6]. Furthermore, a considerable proportion of patients might suffer from transient neurological dysfunction (TND), including conditions like postoperative delirium, with reported incidence rates up to 40%^[7-9]. These conditions not only prolong the time patients spend in the intensive care unit but also increase the total costs of hospitalization^[8]. Research indicates that postoperative delirium could lead to long-term cognitive decline, especially among elderly patients^[10-12].

While considerable efforts have been made to identify preoperative and intraoperative risk factors associated with TAAD surgery, limited research has explored perioperative changes in brain function^[2,13-15]. Current predictive models primarily rely on traditional clinical indicators such as operative times, cardiopulmonary bypass (CPB) times, and preoperative mental status^[13-15]. However, these models often fail to capture dynamic changes in brain function during the perioperative phase, limiting their ability to predict early neurological deterioration, including TND and PND. Although intraoperative monitoring of cerebral oxygen levels in TAAD surgery is a standard practice, its reliability in detecting postoperative neurological impairments remains uncertain^[16]. Additionally, while electroencephalography (EEG) has been proven highly sensitive in detecting cortical ischemia and other brain abnormalities, its complexity and the challenges in real-time interpretation make widespread clinical use difficult in cardiac surgery settings^[17].

Quantitative electroencephalography (QEEG) simplifies the original EEG data by condensing hours or even days of recordings into a single display through graphical representation, while applying quantitative methods to analyze the EEG signals across both the frequency and time domains. QEEG is characterized by two principal components: the amplitude-integrated electroencephalogram (aEEG), which provides an integrated overview of the background activity, and the relative band power (RBP), which delineates the proportional distribution of alpha, beta, theta, and delta waves. The aEEG has been demonstrated to be effective in predicting neurological outcomes for infants undergoing surgery for congenital heart defects or experiencing cardiac arrest ^[18-20]. In the perioperative setting, alterations in wave patterns, specifically increases in theta and delta waves coupled with decreases in alpha waves, are closely linked to TND and cognitive decline^[21,22]. The use of QEEG monitoring has been proposed for the assessment of perioperative cerebral function in TAAD patients^[23-25]. To further enhance the predictive accuracy and clinical applicability of our approach, we incorporated clinical indicators in our subsequent analysis.

Despite its potential, the prognostic capability of QEEG for neurological outcomes in TAAD patients remains unexplored. This study aims to address this gap by evaluating the use of QEEG to monitor brain activity throughout the perioperative phase and develop predictive models for neurological outcomes.

Methods

Study design, setting and participants

This study was a forward-looking observational investigation conducted. The research activities took place from February 2022 to January 2023. Before the commencement of the study, approval was secured from the institutional review board of our hospital. Additionally, the study was registered. Written informed consent was obtained from all participating patients. The work has been reported in line with the STARD (Standards for the Reporting of Diagnostic accuracy studies) criteria^[26].

Prior to their participation in the study, all patients and their families were thoroughly briefed on the potential risks, and informational sheets were provided. Written informed consent was secured before commencing the study. Our research encompassed individuals aged 18 and above who underwent open surgical procedures under general anesthesia for TAAD. To ensure the precision and thoroughness of our research, we applied stringent selection criteria for our subjects. We specifically excluded individuals diagnosed with a malignancy having a life expectancy of fewer than six months, those with a history of cerebral infarction, pregnancy, or psychiatric disorders within the previous six months because such conditions can potentially interfere with the interpretation of QEEG data, and those suffering from other severe diseases unrelated to the study that could impact their life expectancy. Moreover, we excluded participants who had not undergone an aortic computed tomography angiography or other pertinent diagnostic tests. Since these diagnostic tests are integral to obtaining baseline clinical and neurological data, the absence of such information could impair the accuracy of our predictive models. These steps were taken to uphold the integrity and reliability of our study results. Our study ultimately enrolled a total of 220 patients (Fig. 1).

Figure 1.

Consolidated standards of reporting trials diagram demonstrating selection of patients undergoing TAAD surgery repair.

Surgical procedures

All participants received venous and inhalational anesthesia, followed by endotracheal intubation, arterial puncture for blood pressure monitoring in both upper and lower limbs, and insertion of an esophageal ultrasound probe. The right axillary and femoral arteries were exposed, and a median sternotomy was performed to access the supra-arch vessels. After systemic heparinization, CPB was initiated through both the right axillary and femoral arteries, allowing venous return to the right atrium. Retrograde myocardial perfusion was routinely applied through the coronary sinus. When the bladder temperature was lowered to 24–28°C, circulatory arrest was induced, with selective cerebral perfusion (SCP) from the right axillary artery used for brain perfusion. Retrograde cerebral perfusion (RCP) was administered through the superior vena cava during CPB, using either a Y-shaped arterial connector or direct superior vena cava cannulation. Distal repair techniques were adapted to the specific clinical scenario. Typically, patients with arch dilation (≥45 mm), an intimal tear in the arch, or structural arch damage underwent total arch replacement using a quadrifurcated graft. The frozen elephant trunk technique was also utilized in conjunction with total arch replacement. Alternatively, partial arch replacement or antegrade stent grafting in the aortic arch (a technique developed by our center) was employed^[27-29]. After completing the anastomosis, cardiac temperature recovery was initiated. During the rewarming phase, aortic root repair was carried out. This involved the removal of all thrombi within the aortic root dissection, the placement of a Dacron patch shaped to fit between the adventitia and intima of the aortic root as a new media, the insertion of a Dacron felt inside the intima, and continuous suturing of the newly established four-layer aortic root.

Intraoperative brain protection and monitoring

In aortic arch surgeries, maintaining optimal brain oxygenation is paramount. Various strategies for cerebral perfusion are implemented to achieve this, including anterograde cerebral perfusion (ACP), RCP, and deep HCA. ACP is facilitated through the cannulation of either the axillary or innominate/carotid artery, starting with a flow rate of 3–5 ml/kg/min, which is then adjusted based on near-infrared spectroscopy readings. RCP, conversely, involves retrograde infusion through the superior vena cava, with careful monitoring to keep venous pressure at 30–40 mmHg for optimal perfusion efficiency. The efficacy of cerebral perfusion is assessed using near-infrared spectroscopy, with additional interventions necessary if the near-infrared spectroscopy value drops below 20% of the baseline or falls below 50%. The choice of circulatory arrest duration and the target temperature for cerebral protection depend on the specific surgical technique, with temperatures ranging from 20 to 24°C and ≥25°C employed for different procedures.

QEEG recording

Brain function in each participant was continuously assessed using bedside QEEG with a Nicolet Monitor (NicoletOne 5.9.4, Natus Neurology Incorporated). Recordings were conducted using a dual-channel setup at the scalp locations C3, P3, C4, and P4, according to the international 10–20 EEG system^[30]. The raw data from each hemisphere underwent computer-generated filtering and compression, as specified in the data recording device’s software (Supplemental Digital Content 1, available at: http://links.lww.com/JS9/D756). Data quality was rigorously evaluated based on several criteria: electrode impedance had to be below 10 kΩ, the raw data were scrutinized for motion artifacts and ECG interference, and any potential disturbances from diathermy or other electrical devices were thoroughly checked for and excluded.

Throughout the perioperative period, QEEG data collection was segmented into three distinct phases. In the initial phase (Phase 1), QEEG recordings were obtained one hour prior to surgery. The second phase (Phase 2) involved the acquisition of intraoperative amplitude-integrated EEG (aEEG) at five critical intervals: (A) at the start of anesthesia, (B) as hypothermia was induced and the target low temperature was maintained, (C) upon the commencement of selective cerebral perfusion, (D) with the beginning of the rewarming process from hypothermia, and (E) during rewarming to a body temperature of 36.5°C (refer to Fig. 2). The final phase (Phase 3) saw the gathering of postoperative QEEG data for two hours in the cardiac care unit, conducted without the administration of sedative analgesic medications.

Figure 2.

Intraoperative aEEG: (A) anaesthetic induction and commencement of surgery; (B) during cooling and maintenance of the target hypothermic temperature; (C) starting selective cerebral perfusion; (D) rewarming of CPB; (E) rewarming to 36.5°C. The yellow arrow is caused by the interference of electric knife during operation.

The categorization of aEEG activity adhered to a previously established system (as shown in Fig. 2)^[31,32], which identified background patterns as continuous normal voltage (CNV), discontinuous normal voltage (DNV), low voltage (LV), flat trace (FT), and burst suppression (BS) (Supplemental Digital Content 2, available at: http://links.lww.com/JS9/D757)^[33-35]. CNV patterns are generally associated with a good prognosis; for example, Rundgren’s prospective study found that over 85% of patients with this pattern experienced favorable outcomes. In contrast, DNV patterns typically indicate mild and reversible brain dysfunction. LV patterns, when not influenced by anesthesia or sedative medications, are frequently observed in patients with severe brain injury and suggest poor prognosis. FT patterns are most commonly seen in patients with brain death or imminent brain death, indicating an extremely poor prognosis. Similarly, BS patterns, after ruling out the effects of sedative medications, are also strongly associated with a poor prognosis.

In our study, a skilled physician, proficient in QEEG interpretation and blinded to the patients’ clinical details, conducted offline analyses of all QEEG recordings. This systematic approach, using the same established grading system (CNV, DNV, LV, FT, and BS), ensured consistency and objectivity in the classification process.

Additionally, we introduced the dynamic aEEG grade (ΔaEEG) to assess changes in aEEG activity from the end of Phase 2 (stages D and E) to Phase 3. This classification was informed by prognostic studies on aEEG during hypothermia in cardiac arrest and neonatal surgeries, leading to the creation of three distinct ΔaEEG categories, which further enhanced our ability to predict neurological outcomes (Supplemental Digital Content 3, available at: http://links.lww.com/JS9/D758)^[18-20,33].

Definition of end-points

Our study’s primary endpoint was acute brain dysfunction injury, which was defined as the occurrence of adverse outcomes (AO) and TND. AO was defined as intraoperative or postoperative death, or permanent neurological deficit (PND), which includes nerve dysfunction at discharge. PND could manifest as either local injury, such as post-stroke, or systemic injury, such as coma. TND was assessed in surgical survivors who did not exhibit permanent neurological deficits and was identified as postoperative delirium. This condition was diagnosed using the Confusion Assessment Method for the ICU (CAM-ICU)^[36] and monitored bi-daily for a week. Patients who regained consciousness were further evaluated for decannulation and neurological status using the cerebral performance categories (CPC). A neurologist conducted follow-up assessments 30 days post-discharge. During this 30-day period, a CPC score between 1 and 2 was indicative of a favorable neurological outcome, whereas a score of 3 to 5 suggested a poorer prognosis.

Statistical analysis

The sample size calculation for this study was conducted using Power Analysis & Sample Size software (version 15.0, NCSS statistical software, Kaysville, UT, USA). A thorough literature review revealed that the prevalence of ND following surgery for TAAD varies between 32.5% and 52%^[37]. For our study, a conservative prevalence estimates of 45% was chosen to improve the accuracy of our findings. To achieve 80% power to detect a difference of 0.1000 in the area under the ROC curve (AUC) from 0.750 (null hypothesis) to 0.850 (alternative hypothesis), a sample of 200 patients was deemed necessary. This calculation was based on a two-sided z-test with a significance level of 0.050^[38,39]. The data consisted of discrete (rating scale) responses, with the AUC calculated between false positive rates of 0.00 and 1.00. The ratio of the standard deviation of the responses in the negative group to those in the positive group was 1.00. To account for potential dropouts and other factors, the sample size was increased by 10%, resulting in a final sample size of 220 patients.

For this research, the normality of continuous variables was determined using the Kolmogorov–Smirnov test. Variables that adhered to a normal distribution were expressed as means ± standard deviations and evaluated using the Student’s t-test. For continuous variables not following a normal distribution, they were described as medians along with interquartile ranges (Q1–Q3) and were analyzed using the Mann-Whitney U-test. Categorical variables were represented as frequencies and percentages (n, %), and were examined through either the chi-squared test or Fisher’s exact test, depending on their appropriateness. All the statistical tests were two-tailed, with a significance threshold set at P <0.05. Additionally, to explore the association between various variables and the observed outcomes, single-variable binary logistic regression was utilized to compute odds ratios (OR) and 95% confidence intervals (CI).

multivariable

Model development, performance, and internal validation

For continuous variables that did not follow a normal distribution, normalization was applied. Subsequently, variables that demonstrated a P value less than 0.05 were incorporated into a stepwise (backward: conditional) multivariable logistic regression analysis to develop a predictive model. To optimize the model and select the most relevant predictors, we employed a stepwise backward elimination method. This process involved iteratively removing variables that did not significantly contribute to the model, based on the Akaike Information Criterion and their statistical significance. Following this, the effectiveness of the our model was assessed through internal validation utilizing the bootstrap method, which was employed to repeat the process 1000 times. This evaluation focused on two key aspects: discrimination and calibration. Discrimination was quantified by the C-statistic, corresponding to the area under the receiver operating characteristic (ROC) curve^[40]. Calibration, on the other hand, was determined by plotting calibration curves and computing the Brier score using the formula (Y − P)², where “Y” is the actual observed outcome and “P” denotes the model’s estimated probability of the positive outcome, a value between 0 and 1^[41]. All statistical analyses were conducted using SPSS version 25.0 and R version 4.1.1, with a significance level set at P <0.05.

Nomogram development and decision curve analysis

The nomogram was constructed based on a multivariable regression model. Each predictor in the model was assigned a score proportional to its contribution to the outcome variable, as determined by the magnitude of its regression coefficient. These individual scores were then summed to produce a total score. Finally, the total score was mapped to the probability of the outcome through a functional relationship, providing a predicted value for each individual case. The nomogram simplifies complex regression equations into an easy-to-interpret visual format, enhancing the readability of the prediction model and facilitating patient assessment. The nomogram was implemented using the nomogram() function from the rms package in R^[42]. Additionally, to ascertain the practical application of the nomograms in predicting postoperative neurological complications in aortic surgery, decision curve analysis (DCA) was employed^[43]. This analysis measured the clinical net benefit of the predictive model across various threshold probabilities, thus offering crucial insights into its utility in clinical settings.

Results

AO following TAAD surgery

Baseline characteristics of AO and non-AO groups

During the study period spanning February, 2022, to January, 2023, a total of 220 patients were included in our analysis. Patient selection process is outlined in Figure 1. In Table 1, the baseline characteristics of patients were compared between the AO and non-AO groups. There were no significant differences in sex distribution (male: AO 66.7%, non-AO 76.5%, P = 0.327) and body mass index (BMI: AO 27.17 ± 4.46, non-AO 26.26 ± 4.82, P = 0.314). The median age was slightly higher in the AO group (57 years) compared to the non-AO group (52 years), but this difference was not statistically significant (P = 0.083). Preoperative conditions showed that hypertension was more common in the AO group (84.8% vs. 72.2%, P = 0.189). Lower limb malperfusion (9.1% vs. 1.1%, P = 0.027) and hemopericardium (33.3% vs. 14.4%, P = 0.017) were significantly more prevalent in the AO group. Operative data indicated that the AO group had longer operative times (395 minutes vs. 380 minutes, P = 0.038) and cardiopulmonary bypass (CPB) times (217 minutes vs. 180 minutes, P < 0.001). Cross-clamp time was also longer in the AO group (153 minutes vs. 130 minutes, P = 0.005). Table 1 demonstrated a significant difference in pre-mental status between the AO and non-AO groups, with 63.6% of AO patients being vigilant, 15.2% somnolent, and 21.2% in a coma, compared to 95.2% vigilant, 1.6% somnolent, and 3.2% in a coma in the non-AO group (P < 0.001). Postoperative outcomes showed that AO group patients had longer stays in the critical care unit (9 days vs. 5 days, P < 0.001) and higher rates of severe dynamic aEEG injuries (P < 0.001).

Table 1.

Comparison of baseline and perioperative data between AO and non-AO groups

Variables	AO (N = 33)	Non-AO (N = 187)	P value
Basic characters
Sex (N, %)			0.327
Male (N, %)	143 (76.5)	22 (66.7)
Female (N, %)	44 (23.5)	11 (33.3)
Body mass index (kg/m2)a	27.17 ± 4.46	26.26 ± 4.82	0.314
Age (years)b	57.00 [46.00, 69.00]	52.00 [42.00, 62.50]	0.083
Smoking (N, %)	10 (30.3)	51 (27.3)	0.883
Alcohol (N, %)	6 (18.2)	37 (19.8)	1
Hypertension (N, %)	28 (84.8)	135 (72.2)	0.189
Coronary heart disease (N, %)	1 (3.0)	4 (2.1)	1
Diabetes (N, %)	1 (3.0)	1 (0.5)	0.691
Hepatitis (N, %)	3 (9.1)	6 (3.2)	0.273
Allergies (N, %)	3 (9.1)	12 (6.4)	0.851
Nephropathy (N, %)	0 (0.0)	9 (4.8)	0.418
Hepatic insufficiency (%)	1 (3.0)	0 (0.0)	0.326
Immune diseases (N, %)	2 (6.1)	5 (2.7)	0.628
Marfan syndrome (N, %)	1 (3.0)	5 (2.7)	1
Malignancy (N, %)	1 (3.0)	2 (1.1)	0.935
Reoperation (N, %)	2 (6.1)	2 (1.1)	0.203
Preoperative data
Coronary malperfusion (N, %)	6 (3.2)	3 (9.1)	0.273
Mesenteric malperfusion (N, %)	1 (3.0)	2 (1.1)	0.935
Lower limb malperfusion (N, %)	3 (9.1)	2 (1.1)	0.027
Hemopericardium (N, %)	11 (33.3)	27 (14.4)	0.017
Pre-mental status (N, %)			<0.001
Vigilant	21 (63.6)	178 (95.2)
Somnolent	5 (15.2)	3 (1.6)
Coma	7 (21.2)	6 (3.2)
Urgency of operation (N, %)			<0.001
Elective	2 (6.1)	32 (17.1)
Urgent	21 (63.6)	144 (77.0)
Emergent	10 (30.3)	11 (5.9)
Intraoperative data
Aortic Valve Replacement (N, %)	8 (24.2)	38 (20.3)	0.781
Coronary artery bypass grafting (N, %)	3 (9.1)	6 (3.2)	0.273
Extracorporeal membrane oxygenation (N, %)	3 (9.1)	1 (0.5)	0.007
Cerebral perfusion (N, %)			0.411
Unilateral antegrade cerebral perfusion	24 (72.7)	151 (80.7)
Retrograde cerebral perfusion	1 (3.0)	7 (3.7)
Bilateral antegrade cerebral perfusion	7 (21.2)	28 (15.0)
Deep hypothermic circulatory arrest	1 (3.0)	1 (0.5)
Lowest temperature (N, %)			0.946
20-24℃	26 (78.8)	143 (76.5)
25-28℃	7 (21.2)	44 (23.5)
Aortic arch surgery (N, %)			0.676
Hemi-arch	6 (18.2)	35 (18.7)
Fenestrated arch stent	4 (12.1)	38 (20.3)
Island-total arch replacement	10 (30.3)	55 (29.4)
Total arch replacement	13 (39.4)	59 (31.6)
Operative time (min)b	395.00 [350.00, 480.00]	380.00 [330.00, 420.00]	0.038
Cardiopulmonary bypass time (min)b	217.00 [180.00, 257.00]	180.00 [152.00, 213.00]	<0.001
Cross-clamp time (min)b	153.00 [125.00, 182.00]	130.00 [108.50, 156.50]	0.005
Hypothermic circulatory arrest (min)b	28.00 [22.00, 37.00]	28.00 [22.00, 34.00]	0.525
Dynamic aEEG grade (N, %)			<0.001
Normal (△aEEG I)	4 (12.1)	112 (59.9)
Mild injury (△aEEG II)	16 (48.5)	69 (36.9)
Severe injury (△aEEG III)	13 (39.4)	6 (3.2)
Postoperative data
Post-RBP Delta (%)a	76.94 [70.02, 80.64]	68.43 [59.17, 77.48]	0.006
Post-RBP Theta (%)b	11.47 [9.76, 15.82]	13.25 [9.28, 19.10]	0.508
Post-RBP Alpha (%)b	7.02 [5.12, 8.70]	10.73 [6.32, 15.85]	0.001
Post-RBP Beta (%)b	2.66 [1.67, 4.65]	4.05 [2.57, 6.59]	0.035
CCU day (day)	9.00 [6.00, 17.00]	5.00 [3.00, 6.00]	<0.001
Length of stay (day)	16.00 [8.00, 26.00]	17.00 [13.00, 22.00]	0.671

^ais expressed as mean ± standard deviation

^bvalues are expressed as interquartile spacing (median [¼,¾ digits]).

Risk factors for AO identified by logistic regression

In Table 2, univariate and multivariable logistic regression analyses identified significant risk factors for AO after TAAD surgery. In the univariate analysis, significant factors included limb malperfusion (OR = 9.250, 95% CI = 1.476–72.454, P = 0.017), urgency of operation (OR = 4.762, 95% CI = 2.115–11.254, P < 0.001), hemopericardium (OR = 2.963, 95% CI = 1.260–6.724, P = 0.010), pre-mental status (OR = 3.704, 95% CI = 2.094–6.824, P < 0.001), ECMO (OR = 18.6, 95% CI = 2.298–382.838, P = 0.013), operative time (OR = 1.004, 95% CI = 1.000–1.008, P = 0.029), CPB time (OR = 1.012, 95% CI = 1.001–1.003, P < 0.001), cross-clamp time (OR = 1.011, 95% CI = 1.003–1.019, P < 0.001), post-RBP Alpha percentage (OR = 0.911, 95% CI = 0.858–0.968, P < 0.001), post-RBP Beta percentage (OR = 0.943, 95% CI = 0.897–0.991, P = 0.021), post-RBP Delta percentage (OR = 1.344, 95% CI = 1.091–1.656, P < 0.001), and dynamic aEEG grade (OR = 7.850, 95% CI = 4.050–16.917, P < 0.001). In the multivariable analysis, pre-mental status (adjusted OR = 4.652, 95% CI = 2.316–10.074, P < 0.001), CPB time (adjusted OR = 1.014, 95% CI = 1.006–1.023, P = 0.001), and dynamic aEEG grade (adjusted OR = 9.926, 95% CI = 4.493–25.268, P < 0.001) remained significant independent risk factors for AO.

Table 2.

Comparison of baseline and perioperative data between TND and non-TND groups

Variables	TND (N = 74)	Non-TND (N = 113)	P value
Basic characters
Sex((N, %))			0.975
Male (N, %)	56 (75.7)	87 (77.0)
Female (N, %)	18 (24.3)	26 (23.0)
Body mass index (kg/m2)b	26.34 (4.36)	26.22 (5.13)	0.87
Age (years)b	53.00 [41.00, 68.00]	51.00 [42.00, 59.00]	0.202
Smoking (N, %)	22 (29.7)	29 (25.7)	0.658
Alcohol (N, %)	16 (21.6)	21 (18.6)	0.747
Hypertension (N, %)	58 (78.4)	77 (68.1)	0.174
Coronary heart disease (N, %)	2 (2.7)	2 (1.8)	1
Diabetes (N, %)	0 (0.0)	1 (0.9)	1
Hepatitis (N, %)	3 (4.1)	3 (2.7)	0.915
Allergies (N, %)	3 (4.1)	9 (8.0)	0.446
Nephropathy (N, %)	6 (8.1)	3 (2.7)	0.176
Myocardial infarction (N, %)	1 (1.4)	0 (0.0)	0.831
Atrial fibrillation (N, %)	1 (1.4)	0 (0.0)	0.831
Immune diseases (N, %)	5 (4.4)	0 (0.0)	0.17
Marfan syndrome (N, %)	3 (2.7)	2 (2.7)	1
Malignancy (N, %)	1 (0.9)	1 (1.4)	1
Reoperation (N, %)	0 (0.0)	2 (1.8)	0.672
Preoperative data
Coronary malperfusion (N, %)	1 (1.4)	1 (0.9)	1
Mesenteric malperfusion (N, %)	1 (1.4)	1 (0.9)	1
Lower limb malperfusion (N, %)	2 (2.8)	0 (0.0)	0.303
Hemopericardium (N, %)	16 (21.6)	11 (9.7)	0.04
Pre-mental status (N, %)			0.926
Vigilant	71 (95.9)	107 (94.7)
Somnolent	1 (1.4)	2 (1.8)
Coma	2 (2.7)	4 (3.5)
Urgency of operation (N, %)			0.572
Elective	12 (16.2)	20 (17.7)
Urgent	56 (75.7)	88 (77.9)
Emergent	6 (8.1)	5 (4.4)
Intraoperative data
Aortic Valve Replacement (N, %)	11 (14.9)	27 (23.9)	0.189
Coronary artery bypass grafting (N, %)	5 (6.8)	1 (0.9)	0.071
Extracorporeal membrane oxygenation (N, %)	1 (1.4)	0 (0.0)	0.831
Cerebral perfusion (N, %)			0.168
Unilateral antegrade cerebral perfusion	62 (83.8)	89 (78.8)
Retrograde cerebral perfusion	4 (5.4)	3 (2.7)
Bilateral antegrade cerebral perfusion	7 (9.5)	21 (18.6)
Deep hypothermic circulatory arrest	1 (1.4)	0 (0.0)
Lowest temperature (N, %)			0.748
20-24℃	58 (78.4)	85 (75.2)
25-28℃	16 (21.6)	28 (24.8)
Aortic arch surgery (N, %)			0.306
Hemi-arch	17 (23.0)	18 (15.9)
Fenestrated arch stent	18 (24.3)	20 (17.7)
Island-total arch replacement	18 (24.3)	37 (32.7)
Total arch replacement	21 (28.4)	38 (33.6)
Operative time (min)b	380.00 [330.00, 420.00]	380.00 [330.00, 410.00]	0.714
Cardiopulmonary bypass time (min)b	187.00 [163.25, 218.75]	172.00 [143.00, 211.00]	0.04
Cross-clamp time (min)b	134.00 [109.00, 155.50]	126.00 [108.00, 157.00]	0.36
Hypothermic circulatory arrest (min)b	28.50 [23.00, 35.00]	28.00 [21.00, 33.00]	0.21
Postoperative data
Dynamic aEEG grade (N, %)			<0.001
Normal (△aEEG I)	16 (21.6)	96 (85.0)
Mild injury (△aEEG II)	53 (71.6)	16 (14.2)
Severe injury (△aEEG III)	5 (6.8)	1 (0.9)
Post-RBP data
Post-RBP Alpha (%)b	74.66 [61.75, 81.59]	66.32 [58.55, 74.56]	0.001
Post-RBP Beta (%)b	14.70 [9.80, 19.68]	12.48 [9.09, 18.16]	0.089
Post-RBP Theta (%)b	6.99 [4.03, 10.54]	12.69 [9.77, 17.82]	<0.001
Post-RBP Delta (%)a	2.76 [1.76, 4.61]	4.62 [3.25, 7.80]	<0.001
CCU (day)	6.00 [4.25, 9.00]	4.00 [3.00, 5.00]	<0.001
Length of stay (day)	19.00 [15.00, 25.00]	15.00 [13.00, 20.00]	<0.001

^ais expressed as mean ± standard deviation.

^bvalues are expressed as interquartile spacing (median [¼,¾ digits]).

Predictive performance of the AO model

The predictive performance of the AO model, developed based on the significant risk factors identified by logistic regression analysis, was evaluated using several graphical methods. The AUC of the ROC for the AO model was 0.888 (95% CI = 0.818–0.960), indicating a high level of discriminative ability. The optimal cutoff point had a sensitivity of 0.788 and a specificity of 0.888 (Fig. 3). The calibration curve showed good agreement between predicted and observed probabilities, with a Brier score of 0.0728. The mean absolute error was 0.017, indicating that the model predictions are well-calibrated (Fig. 4). A nomogram was developed based on the significant predictors from the logistic regression analysis (pre-mental status, CPB time, and dynamic aEEG grade), providing a visual tool for estimating the risk of AO in patients undergoing TAAD surgery (Fig. 5). The Decision Curve Analysis (DCA) showed that the AO predictive model offers significant clinical net benefit across a range of threshold probabilities (Supplemental Digital Content 4, available at: http://links.lww.com/JS9/D759). The Clinical Impact Curve (CIC) demonstrated that the AO model effectively identifies high-risk patients at various threshold probabilities, balancing sensitivity and specificity and supporting its clinical utility (Supplemental Digital Content 5, available at: http://links.lww.com/JS9/D760).

Figure 3.

ROC curve for predictive performance of the AO model following TAAD surgery.

Figure 4.

Calibration curve for predictive accuracy of the AO model following TAAD surgery.

Figure 5.

Nomogram for predictive risk of AO following TAAD surgery.

TND following TAAD surgery

Baseline characteristics of TND and non-TND groups

In Table 3, the baseline characteristics of patients were compared between the Transient Neurological Deficit (TND) and non-TND groups. There were no significant differences in sex distribution (male: TND 75.7%, non-TND 77.0%, P = 0.975) and body mass index (BMI: TND 26.34 ± 4.36, non-TND 26.22 ± 5.13, P = 0.87). The median age was slightly higher in the TND group (53 years) compared to the non-TND group (51 years, P = 0.202). Hypertension was more common in the TND group (78.4% vs. 68.1%, P = 0.174). Coronary malperfusion was significantly more prevalent in the TND group (21.6% vs. 9.7%, P = 0.04). ECMO was used more in the TND group (6.8% vs. 0.9%, P = 0.071). Operative and CPB times were similar, with a slightly longer total arch replacement time in the TND group (187.00 vs. 172.00 minutes, P = 0.04). Postoperative outcomes showed higher rates of severe dynamic aEEG injury in the TND group (P < 0.001). Post-RBP Alpha (6.00 vs. 4.00, P < 0.001) and Beta (19.00 vs. 15.00, P < 0.001) percentages were higher in the TND group, while Theta (2.85 vs. 4.62, P < 0.001) and Delta (6.35 vs. 12.69, P < 0.001) percentages were lower. Length of stay in the critical care unit was comparable between the two groups.

Table 3.

Univariate and multivariate logistic regression analysis of risk factors for AO after TAAD surgery

Variables	Univariate analysis	P Value	Multivariate analysis	P value
Limb	9.250(1.476-72.454)	0.017
Urgency of operation	4.762(2.115-11.254)	<0.001
Hemopericardium	2.963(1.260-6.724)	0.010
Pre-mental status	3.704(2.094-6.824)	<0.001	4.652(2.316-10.074)	<0.001
ECMO	18.6(2.298-382.838)	0.013
Operative time (min)	1.004(1.000-1.008)	0.029
CPB (min)	1.012(1.001-1.003)	<0.001	1.014(1.006-1.023)	0.001
Cross clamp (min)	1.011(1.003-1.019)	<0.001
Post-RBP Alpha (%)	0.911(0.858-0.968)	<0.001
Post-RBP Beta (%)	0.943(0.897-0.991)	0.021
Post-RBP Delta (%)	1.344(1.091-1.656)	<0.001
△aEEG	7.850(4.050-16.917)	<0.001	9.926(4.493-25.268)	<0.001

ECMO: Extracorporeal Membrane Oxygenation; CPB: Cardiopulmonary Bypass;△aEEG: dynamic aEEG grade.

Risk factors for TND identified by logistic regression

In Table 4, univariate and multivariable logistic regression analyses identified significant risk factors for TND after TAAD surgery. In the univariate analysis, significant factors included limb malperfusion (OR = 9.250, 95% CI = 1.476–72.454, P = 0.017), urgency of operation (OR = 4.775, 95% CI = 2.129–11.255, P < 0.001), hemopericardium (OR = 2.963, 95% CI = 1.260–6.724, P = 0.010), pre-mental status (OR = 3.704, 95% CI = 2.094–6.824, P < 0.001), ECMO (OR = 18.6, 95% CI = 2.298–382.838, P = 0.013), operative time (OR = 1.004, 95% CI = 1.000–1.008, P = 0.029), CPB time (OR = 1.002, 95% CI = 1.000–1.003, P = 0.020), cross-clamp time (OR = 1.011, 95% CI = 1.003–1.019, P = 0.009), post-RBP Alpha percentage (OR = 0.702, 95% CI = 0.640–0.769, P < 0.001), post-RBP Beta percentage (OR = 0.848, 95% CI = 0.779–0.923, P < 0.001), post-RBP Delta percentage (OR = 1.794, 95% CI = 1.266–2.542, P < 0.001), and dynamic aEEG grade (OR = 7.850, 95% CI = 4.050–16.917, P < 0.001). In the multivariable analysis, post-RBP Alpha percentage (adjusted OR = 0.263, 95% CI = 0.121–0.532, P < 0.001), CPB time (adjusted OR = 1.014, 95% CI = 1.006–1.023, P = 0.001), and dynamic aEEG grade (adjusted OR = 112.444, 95% CI = 5.337–30.814, P < 0.001) remained significant independent risk factors for TND.

Table 4.

Univariate and multivariate logistic regression analysis of risk factors for TND after TAAD surgery

Variables	Univariate analysis	P value	Multivariate analysis	P value
Limb	9.250(1.476-72.454)	0.017
Urgency of operation	4.775(2.129-11.255)	<0.001
Hemopericardium	2.963(1.260-6.724)	0.010
Pre-mental status	3.704(2.094-6.824)	<0.001
ECMO	18.6(2.298-382.838)	0.013
Operative time (min)	1.004(1.000-1.008)	0.029
CPB (min)	1.002(1.000-1.003)	0.020	1.014(1.006-1.023)	0.001
Cross clamp (min)	1.011(1.003-1.019)	0.009
Post-RBP Alpha (%)	0.702(0.640-0.769)	<0.001	0.263(0.121-0.532)	<0.001
Post-RBP Beta (%)	0.848(0.779-0.923)	<0.001
Post-RBP Delta (%)	1.794(1.266-2.542)	<0.001
△aEEG	7.850(4.050-16.917)	<0.001	112.444(5.337-30.814)	<0.001

CPB: cardiopulmonary bypass;△aEEG: dynamic aEEG grade.

Predictive performance of the TND model

The predictive performance of the TND model, developed based on the significant risk factors identified by logistic regression analysis, was evaluated using several graphical methods. The AUC for the TND model was 0.893 (95% CI = 0.844–0.941), indicating a high level of discriminative ability (Fig. 6). This high AUC value demonstrates that the TND model is effective in distinguishing between patients who will and will not experience transient neurological deficits following TAAD surgery. The calibration curve for the TND model showed good agreement between predicted and observed probabilities, with a Brier score of 0.125. The mean absolute error was 0.018, indicating that the model predictions are well-calibrated (Fig. 7). A nomogram was developed based on the significant predictors from the logistic regression analysis (CPB time, log_Post_Alpha, and dynamic aEEG grade), providing a visual tool for estimating the risk of TND in patients undergoing TAAD surgery (Fig. 8). The DCA showed that the TND predictive model offers significant clinical net benefit across a range of threshold probabilities (Supplemental Digital Content 6, available at: http://links.lww.com/JS9/D761). The CIC demonstrated that the TND model effectively identifies high-risk patients at various threshold probabilities (Supplemental Digital Content 7, available at: http://links.lww.com/JS9/D762).

Figure 6.

ROC curve for predictive performance of the TND model following TAAD surgery.

Figure 7.

Calibration curve for predictive accuracy of the TND model following TAAD surgery.

Figure 8.

Nomogram for predictive risk of TND following TAAD surgery.

Discussion

With the continuous advancement of surgical and monitoring technologies, the incidence of postoperative neurological injury following TAAD remains unchanged despite the use of cerebral protection techniques, such as hypothermia, cerebral perfusion, and blood gas management. Currently, the diagnosis of post-stroke or PND is still dependent on brain CT in clinical practice. However, for TAAD-related AO, the risk of examination is often high due to hemodynamic instability and the inability to conduct continuous monitoring of the nervous system. In this study, we report, for the first time, the use of perioperative QEEG monitoring to assess nerve function recovery in this population. Firstly, during the operation, 15% of patients experienced severe brain damage related to AO after deep HCA, and 39.6% of patients experienced mild brain damage related to TND after the surgery. Secondly, failure to recover CNV in aEEG within 2 hours postoperatively was associated with AO. Thirdly, the postoperative RBP and ΔaEEG can improve the sensitivity and accuracy for predicting TND after surgery.

During the perioperative period, ΔaEEG can provide insight into alterations in cerebral function. By means of intraoperative QEEG monitoring, it has been observed that during periods of HCA, cortical activity is suppressed. Following arch surgery and rewarming to 36.5°C, a gradual recovery of aEEG is typically observed. If recovery to CNV is achieved, it suggests an absence of significant cerebral injury during the period of HCA and arch surgery. Conversely, persistent LV, FT, and BS indicate severe cerebral injury, which are associated with AO. These findings parallel those observed in hypothermia-treated cardiac arrest patients, where ΔaEEG similarly correlate with patient prognosis^[33,35]. For ischemic stroke, magnetic resonance imaging typically requires 2–6 hours to reveal ischemic changes and brain CT 1–5 days, ΔaEEG serves as a complementary neuroimaging technique, enabling early detection of reversible injury and guiding timely clinical intervention. Notably, recovery of CNV in aEEG at 2 hours postoperatively is a critical predictor of neurological Prognosis of patients. In the AO cohort, the 95% CI of ΔaEEG in predicting the onset of AO exhibited a notable breadth, likely attributable to the limited sample size. Thus, further investigation with an expanded sample size will be imperative to substantiate the validity of our findings.

In this study, we investigated the independent risk factors for AO following aortic surgery. Our results indicated that both preoperative mental status and CPB time were significant risk factors for AO. CPB time reflected the difficulty of the entire aortic surgery. These findings were consistent with previous studies in this field. Furthermore, combining preoperative mental status, CPB time, and ΔaEEG can enhance the predictive accuracy of AO following TAAD. This study provided valuable insights into the identification of risk factors for AO and laid the foundation for future interventions to prevent and manage this condition.

Previous researches have consistently identified HCA as an independent risk factor for neurological complications following cardiac surgery^[44,45]. However, Angleitner et al^[46] suggested that HCA lasting up to 50 minutes may not pose a significant risk for AO, thanks to advancements in brain protection and surgical techniques. In our study, the median duration of HCA was 28 minutes, which did not meet the criteria for inclusion in Model AO. Age and timing of surgery have been suggested as potential risk factors for AO², but with the continuous advancements in surgical technology, advanced age is no longer considered an independent risk factor. Therefore, we did not include HCA, age, or timing of surgery in Model AO. Our findings have important implications for the identification of relevant risk factors and may inform future interventions to prevent and manage AO or PND following cardiac surgery.

Previous studies have linked preoperative coma with escalated risks of mortality and sustained postoperative neurological complications. Some experts have recommended that a coma exceeding three hours might preclude the option of central repair surgery^[47]. Our study has illustrated that a decline in preoperative mental condition is associated with an increased frequency of AO following TAAD surgery. Preoperative mental status of the patients with TAAD has wide range from full normal to somnolence even coma that could indicating the underlying neurocognitive and functional capacities, which may dictate the strategy of surgical interventions^[48]. A worse neurocognitive state increases the risk of postoperative complications and mortality which warrant more detailed intraoperative surveillance and management. Preoperative mental status is an indicator of patient brain resilience and susceptibility and preoperative evaluation is important.

Coma has traditionally considered contraindication for TAAD surgery. In our previous study, the strategy for managing acute TAAD complicated by coma involves preoperative assessments using CTA and QEEG to evaluate brain function, followed by emergency stenting and aortic replacement. In one case, the patient’s brain function improved postoperatively, although some neurological deficits persisted. While this approach shows promise, further research is needed to refine and validate the strategy^[49]. This suggests that surgical candidacy in comatose individuals ought to be considered based on multiple factors, such as ischemia duration, cerebral collateral circulation robustness, and the individual’s tolerance to ischemic conditions, rather than a singularly defined criterion^[50,51].

Upon recovery from anesthesia, patients’ brain function can be evaluated using QEEG monitoring. If the aEEG recovers to DNV at a temperature of 36.5°C, mild brain dysfunction may be suspected. However, with the metabolism of anesthetic drugs, the aEEG may recover to CNV, indicating transient brain dysfunction that is associated with TND. Previous studies^[22,52] have shown that the relative power was related to postoperative neurocognitive dysfunction. By combining ΔaEEG and RBP Alpha and CPB time, the AUROC value for TND prediction can be up to 0.893. Furthermore, the combination of ΔaEEG and RBP can be used to predict the occurrence of TND, which has not been previously reported in the literature.

TND is a prevalent and costly complication following cardiovascular surgery. Despite current research focusing on perioperative surgical factors, current delirium prediction models exhibit inadequate sensitivity and specificity. Recent studies have revealed TND to be an acute reversible cognitive dysfunction. ΔaEEG has been shown to capture this characteristic more accurately. Studies have identified an increase in the slow-wave to fast-wave ratio as an indicator of brain injury, and slow-wave dominant EEG patterns have been found to predict delirium and cognitive impairment. Our TND Model, ΔaEEG, Post-RBP Alpha, and CPB time, demonstrates excellent diagnostic efficacy for TND prediction, and could improve patient outcomes and enhance clinical decision-making for postoperative care.

The clinical significance of RBP alpha in postoperative delirium following TAAD surgery has been explored in several studies. Lower intraoperative alpha band power has been identified as a potential marker for increased brain vulnerability and the risk of developing postoperative delirium. A study by Gutiérrez et al demonstrated that patients with lower intraoperative alpha power are more likely to develop postoperative delirium, highlighting the utility of EEG monitoring in predicting postoperative neurocognitive outcomes^[53]. Another study by Khalifa et al supports this, showing that reduced intraoperative frontal alpha power correlates with a higher incidence of postoperative delirium in cardiac surgeries, including aortic dissection procedures^[54]. Additionally, our previous research has explored the use of QEEG parameters, including RBP, to assess postoperative neurological complications in patients undergoing TAAD surgery, suggesting that RBP alpha might serve as a diagnostic and prognostic tool in such cases^[24]. In contrast to our earlier studies, this time we have incorporated clinical factors into the model to enhance the prediction of postoperative neurological outcomes in TAAD patients. These studies emphasize the importance of EEG-based alpha power as a predictive marker for TND, especially in high-risk surgeries like TAAD.

In prior research, it was widely believed that TND was closely linked to perioperative clinical factors such as advanced age, total arch replacement, and postoperative hypoxemia. However, with the continued advancement of anesthesia and surgical techniques, overall surgical time has decreased, and postoperative monitoring techniques have improved. Consequently, these factors may no longer be independent risk factors for TND. To address this issue, we conducted a multivariable analysis to exclude the effects of these factors and utilized brain function testing to predict TND. Our findings suggest that previously identified clinical risk factors for TND may no longer be relevant due to improvements in perioperative care. These results have significant implications for the development of targeted interventions to address postoperative delirium and ultimately improve patient outcomes.

Our nomograms developed in our study, based on QEEG metrics such as dynamic aEEG grade and relative band power, offers a promising tool for the early identification of patients at higher risk of adverse neurological outcomes and TND following TAAD surgery. Integrating the nomogram into clinical workflows may involve incorporating QEEG monitoring as part of perioperative care, enabling clinicians to dynamically assess brain function throughout surgery and postoperatively, and potentially supporting earlier interventions to prevent or mitigate neurological complications. Our findings suggest several potential improvements to existing clinical practices, including the incorporation of QEEG monitoring to complement current neuromonitoring practices, the use of the nomogram for risk stratification to enable more personalized perioperative management strategies, and the potential for the nomogram to facilitate collaborative, multidisciplinary decision-making and improve the coordination of care. However, successful implementation would require addressing practical considerations such as training and expertise for clinical staff, availability of equipment and resources, adaptations to perioperative workflows, and further clinical validation of the nomogram across diverse patient populations and settings.

In conclusion, we demonstrated that in patients with TAAD, ΔaEEG in the perioperative could detect postoperative neurologic dysfunction in the early stages. Furthermore, postoperative RBP analysis could predict the occurrence of delirium. These findings suggest that QEEG is a valuable tool for cardiac surgeons to assess postoperative brain function and facilitate early interventions that may improve clinical outcomes.

Limitations

Nevertheless, our study faces several challenges that need to be addressed. First, QEEG is mainly effective for monitoring cortical brain functions and is less sensitive to injuries in non-cortical areas such as the cerebellum and brainstem. Second, as our single-center prospective observational study may introduce biases, such as unaccounted confounding factors due to the lack of randomization, future work should include longitudinal studies to track long-term outcomes and multicenter trials to validate the nomogram, ensuring its accuracy, generalizability, and clinical impact. Third, our study is limited by a lack of external validation; thus, expanding the sample size and conducting multi-center studies are essential to corroborate our results. Fourth, the use of a hospital-specific cohort may limit the generalizability of our findings to other institutions or healthcare settings, as variations in patient demographics, clinical protocols, and perioperative management strategies across centers could influence the applicability of our results.

Ethics approval

The Institutional Review Board approved the study. The trial was conducted in accordance with the ethical principles of the “Declaration of Helsinki,” the “Ethical Review Measures for Biomedical Research Involving Humans” of the National Health Commission of China, and other relevant national laws and regulations.

Consent

Informed consent by the study participant or a legally authorized representative was given prior to inclusion in the study.

Sources of funding

This work was supported by Jiangsu Province Capability Improvement Project through Science, Technology and Education (ZDXK202229) and supported by fundings for Clinical Trials from the Affiliated Drum Tower Hospital, Medical of School of Nanjing University (2022-LCYJ-MS-12).

Author’s contribution

Y.P.W. performed a literature review and the statistical analysis. D.J.W., J.P., Y.Q.C., Y.C. and X.L. were responsible for the conception and design of the study. Y.P.W. and Y.J. performed data collection and database management. Y.P.W., Q.J.Z. and L.M. drafted the manuscript. Y.P.W., Y.J., and Y.X.X. contributed significantly to manuscript correction and finalization. All authors contributed to the article and approved the submitted version.

Conflicts of interest disclosure

All the authors declare to have no conflicts of interest relevant to this study.

Research registration unique identifying number (UIN)

Trial registration ChiCTR2200055980 (Registered 30th January 2022). This trial was registered before the first participant was enrolled.

Guarantor

Ya-peng Wang and Dong-jin Wang.

Provenance and peer review

No, our paper was not invited.

Data availability statement

The original data supporting the conclusions of this article will be made available by the authors, without undue reservation.