Association between preoperative blood–brain barrier permeability and postoperative delirium in older patients undergoing cardiac surgery: a pilot study

Abstract

Background

Postoperative delirium (POD) is a frequent and serious complication in older adults after cardiac surgery. Blood–brain barrier (BBB) dysfunction is implicated in cognitive decline, but its preoperative role in POD remains underexplored. This pilot study aimed to investigate the association between preoperative regional BBB permeability, assessed by dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI), and POD in older patients undergoing off-pump coronary artery bypass grafting (OPCABG).

Methods

This prospective observational pilot study, registered in the Chinese Clinical Trial Registry (ChiCTR2200063774), included patients aged ≥ 65 years scheduled for OPCABG. Preoperative BBB permeability (quantified as Ktrans) in the hippocampus, thalamus, frontal lobe, and temporal lobe, along with regional brain volumes and Montreal Cognitive Assessment-Basic (MoCA-B) scores, were assessed. POD was diagnosed using the 3-Minute Diagnostic Confusion Assessment Method (3D-CAM) or CAM-ICU for 5 postoperative days. Univariable and multivariable logistic regression analyses were performed to identify predictors of POD. Correlations between Ktrans, volume, and POD severity (CAM-S) were examined.

Results

Fifty patients (mean age 69.0 ± 3.3 years) were analyzed; 19 (38%) developed POD. In univariable analysis, higher preoperative Ktrans in the hippocampus (Odds Ratio [OR] 1.350, 95%CI 1.048–1.740, P = 0.020) and thalamus (OR 1.466, 95%CI 1.017–2.113, P = 0.040), lower MoCA-B scores (P = 0.020), and smaller hippocampal (OR 0.297, 95%CI 0.131–0.672, P = 0.004) and thalamic volumes (OR 0.304, 95%CI 0.121–0.766, P = 0.012) were associated with POD. However, in multivariable logistic regression including MoCA-B, Ktrans, and volumes, only lower MoCA-B scores (OR 0.697, 95%CI 0.513–0.947, P = 0.021) and smaller hippocampal volume (OR 0.322, 95%CI 0.105–0.992, P = 0.048) remained independent predictors of POD incidence. Preoperative hippocampal Ktrans correlated significantly with POD severity (CAM-S, r = 0.673, P = 0.002).

Conclusion

In this pilot study, while increased preoperative BBB permeability in the hippocampus and thalamus was associated with POD univariably, baseline cognitive function and hippocampal volume were stronger independent preoperative predictors of POD incidence after OPCABG. Higher preoperative hippocampal BBB permeability was associated with greater POD severity, suggesting a role for pre-existing BBB vulnerability in exacerbating delirium. These preliminary and exploratory findings warrant validation in larger, adequately powered cohorts and highlight the complex interplay of pre-existing brain vulnerabilities in POD development.

Trial registration

Chinese Clinical Trial Registry (ChiCTR2200063774; registered on 09/16/2022).

Introduction

Postoperative delirium (POD), an acute, transient, and fluctuating neuropsychiatric syndrome characterized by disturbances in attention, consciousness, and cognition, represents a significant complication, particularly in older patients undergoing cardiac surgery [1]. The consequences of POD are severe, including prolonged hospital stays, escalated healthcare expenditures, increased mortality rates, and an elevated risk of long-term cognitive decline and dementia [2–5]. Despite its clinical impact, the underlying pathophysiological mechanisms of POD are not fully elucidated, hindering the development of effective preventative and therapeutic strategies [6, 7]. Recent multi-omics approaches and biomarker studies continue to explore these complex pathways [6, 7].

The blood–brain barrier (BBB) is a dynamic neurovascular interface crucial for maintaining central nervous system homeostasis by strictly regulating the passage of substances between the blood and brain parenchyma [8]. Accumulating evidence suggests that BBB breakdown is an early pathological feature and biomarker in various cognitive disorders, including Alzheimer’s disease (AD) and mild cognitive impairment (MCI) [8–10]. The pathogenesis of POD is increasingly considered to involve BBB dysfunction, which may permit the entry of neurotoxic molecules and inflammatory mediators into the brain, thereby triggering or exacerbating neuroinflammation and neuronal dysfunction [11]. Indeed, the systemic inflammatory response is a recognized key contributor to such neurological complications, including postoperative cognitive dysfunction (POCD), which shares pathophysiological links with POD; importantly, attenuating this inflammatory response has been shown to reduce the risk of POCD after cardiac surgery [12]. A compromised BBB may permit the entry of neurotoxic molecules and inflammatory mediators into the brain, thereby triggering or exacerbating neuroinflammation and neuronal dysfunction [13–16]. The pathogenesis of POD is increasingly considered to involve BBB dysfunction, which may permit the entry of neurotoxic molecules and inflammatory mediators into the brain, thereby triggering or exacerbating neuroinflammation and neuronal dysfunction [11, 13–16]. While perioperative BBB disruption has been demonstrated, it remains unclear whether pre-existing BBB vulnerabilities contribute to POD development.

Traditional methods for assessing BBB permeability, such as the cerebrospinal fluid/plasma albumin ratio, are invasive and impractical for routine perioperative use [17]. Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) has emerged as a robust, non-invasive technique for quantitatively assessing subtle, regional BBB permeability in vivo, primarily through the derived parameter Ktrans (volume transfer constant) [18–20]. Cardiac surgery, even off-pump procedures, carries a substantial risk for POD [21], potentially due to unique perioperative stressors. Off-pump coronary artery bypass grafting (OPCABG) avoids cardiopulmonary bypass, theoretically reducing the systemic inflammatory response and embolic load compared to on-pump surgery; however, POD incidence remains a concern, suggesting other patient-specific vulnerabilities may play a key role. Understanding pre-existing regional BBB integrity in older patients with coronary artery disease undergoing OPCABG is therefore critical.

This prospective observational pilot study aimed to elucidate the association between preoperative regional BBB permeability and the incidence and severity of POD in older patients undergoing OPCABG. We hypothesized that patients with higher preoperative BBB permeability in specific brain regions, particularly the hippocampus and thalamus, would have increased odds of developing POD after cardiac surgery.

Methods

Study design and participants

This prospective observational pilot study was conducted at a tertiary university hospital (The Second Hospital of Hebei Medical University) from September 2022 to June 2023, following registration in the Chinese Clinical Trial Registry (No. ChiCTR2200063774). The study protocol was approved by the Ethics Committee of The Second Hospital of Hebei Medical University (No. 2022-R707), and written informed consent for study participation was obtained from all patients or their legally authorized representatives prior to any study-related procedures. This research adheres to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines [22].

Patients aged ≥ 65 years, with an American Society of Anesthesiologists (ASA) physical status of II-III, and scheduled for elective OPCABG were consecutively screened for eligibility. Exclusion criteria were: (1) baseline Montreal Cognitive Assessment Scale-Basic (MoCA-B) score < 18; (2) pre-existing diagnosis of Parkinson's disease or epileptic disease; (3) history of previous stroke (within 6 months) or major brain surgery; (4) severe hepatic (Child–Pugh C) or renal insufficiency (eGFR < 30 mL/min/1.73m² or dialysis); (5) significant audition, vision, or language impairments precluding cognitive assessment or communication; (6) contraindications to MRI scanning (e.g., incompatible metallic implants, severe claustrophobia).

Anesthesia and perioperative management

All patients received standardized anesthetic management. In the operating room, standard monitoring included electrocardiography, pulse oximetry, end-tidal carbon dioxide, bispectral index (BIS), core body temperature (nasopharyngeal), invasive arterial blood pressure, and central venous pressure. General anesthesia was induced with etomidate (0.2–0.3 mg/kg), sufentanil (0.4–0.6 μg/kg), and rocuronium (0.9 mg/kg). Anesthesia was maintained with a combination of continuous intravenous infusion of ciprofol (a novel GABAA receptor agonist, 0.5–1.0 mg·kg⁻¹·h⁻¹) and remifentanil (0.1–0.3 µg·kg⁻¹·min⁻¹), supplemented with inhaled sevoflurane (end-tidal concentration 0.5–1.0 minimal alveolar concentration [MAC]) to ensure adequate anesthetic depth. Intravenous rocuronium was administered as needed. Parasternal nerve blocks were performed at the discretion of the attending anesthesiologist for enhanced analgesia. Mean arterial pressure (MAP) was maintained above 65 mmHg or within ± 20% of baseline values. BIS values were kept between 40 and 60. End-tidal carbon dioxide partial pressure was maintained at 35–45 mmHg, and nasopharyngeal temperature at 36–37 °C. Following median sternotomy, heparin was administered before coronary artery manipulation, and protamine was used for reversal after anastomoses. Intraoperative cell salvage was routinely employed.

Post-surgery, patients were transferred to the intensive care unit (ICU), receiving sedation and analgesia until meeting criteria for tracheal extubation. Postoperative pain was managed with patient-controlled intravenous analgesia (PCIA) using sufentanil (e.g., 2 µg/kg diluted in 200 ml saline) with a background infusion (e.g., 2 ml/h), bolus doses (e.g., 1 ml), and a lockout interval (e.g., 15 min), targeting a Numerical Rating Scale (NRS) pain score of 0–3 at rest. Supplemental analgesics (e.g., intravenous sufentanil, pentazocine, or dezocine) were administered as needed.

Assessment of POD

The primary outcome was the incidence of POD within the first five postoperative days. Trained research personnel, blinded to preoperative MRI data and intraoperative details, assessed patients for POD twice daily (06:00–08:00 and 18:00–20:00). The 3-Minute Diagnostic Confusion Assessment Method (3D-CAM) was used for non-intubated patients, and the CAM for the ICU (CAM-ICU) for intubated patients [23, 24]. POD was diagnosed according to DSM-V criteria if features of (1) acute onset or fluctuating course and (2) inattention were present, along with either (3) disorganized thinking or (4) altered level of consciousness. A diagnosis of POD was made if criteria were met at any assessment point. Comprehensive reviews of nursing notes and medical records supplemented these assessments. The severity of delirium, a secondary outcome, was assessed using the long form of the Chinese version of the CAM-Severity (CAM-S) scale, with scores ranging from 0 (no delirium) to 19 (most severe) [25]. All delirium assessors underwent standardized training and achieved perfect scores on a qualifying quiz. The MoCA-B was selected because it was developed to minimize educational bias and is therefore highly suitable for screening for cognitive impairment in older adults with varied educational backgrounds [26]. Its use in our study population is supported by validation studies in similar cohorts[27, 28], thus reducing potential cultural and educational bias. MoCA-B and MRI scans were performed 1–3 days before surgery.

Imaging acquisition and analysis

All MRI scans were performed on a 3.0 T MRI scanner (GE SIGNA Architect, GE Healthcare). The protocol included: 3D T1-weighted BRAVO, 3D T2-FLAIR CUBE, diffusion-weighted imaging (DWI), and DCE-MRI. For DCE-MRI: axial orientation, TR 4.4 ms, TE 1.6 ms, slice thickness 2.0 mm with 0.6 mm interslice gap, FOV 26.9 cm × 24 cm, flip angle 12°. A bolus of gadobutrol (0.1 mmol/kg) was injected intravenously via a power injector at a rate of 4.5 ml/s, followed by a 20 ml saline flush, at the beginning of the third dynamic scan phase. A total of 40 dynamic phases were acquired over approximately 7 min.

DCE-MRI data were processed using the GEN IQ advanced analysis module on a GE AW4.7 workstation (GE Medical Systems). This software incorporates 3D motion correction. Quantitative analysis of BBB permeability was based on the extended Tofts two-compartment pharmacokinetic model, yielding Ktrans (min⁻¹) [8, 18, 19]. An arterial input function (AIF) was automatically identified by the software from the M2 segment of the middle cerebral artery, with subsequent manual verification of appropriate placement and curve morphology by an experienced neuroradiologist. T1-mapping was performed for T1 correction. Regions of interest (ROIs) were manually delineated on co-registered 3D T1-weighted or T2-FLAIR images by two trained radiologists blinded to clinical data, following established anatomical atlases and protocols (e.g., Mai-Paxinos-Voss atlas for cortical regions, specific protocols for hippocampal and thalamic segmentation) [29]. ROIs included bilateral hippocampi, thalami, frontal lobes, and temporal lobes. Inter-rater reliability for Ktrans measurements and volume delineations in key ROIs (hippocampus, thalamus) was assessed using intraclass correlation coefficients (ICCs) on a subset of 10 scans, demonstrating good to excellent agreement (ICCs > 0.85 for Ktrans and > 0.90 for volumes); any initial disagreements in ROI delineation were resolved by consensus discussion to achieve a final delineation. Mean Ktrans values from bilateral ROIs were used. Hippocampal and thalamic volumes were normalized to total intracranial volume (TIV) to account for variations in head size, though absolute volumes are reported as per original manuscript for consistency unless specified. Patients with acute ischemic lesions on preoperative DWI were excluded.

Statistical analysis

Sample size estimation was based on prior literature [8, 30] and preliminary data suggesting Ktrans in POD patients might be substantially higher than in non-POD patients. Assuming a POD incidence of 40% [31, 32], a two-sided alpha of 0.05, and 80% power to detect a clinically meaningful difference in Ktrans (e.g., an effect size Cohen's d ≈ 0.8, reflecting the "approximately half" observation from preliminary experiments), a sample size of approximately 50 patients was calculated. Sixty patients were recruited to account for potential dropouts or unusable data.

Statistical analyses were performed using SPSS version 27.0 (IBM Corp.). Patients were dichotomized into POD and Non-POD (NPOD) groups. Normality of continuous variables was assessed using Shapiro–Wilk tests and visual inspection of histograms. Normally distributed continuous variables were presented as mean (standard deviation, SD) and compared using independent samples t-tests. Non-normally distributed variables were presented as median (interquartile range, IQR) and compared using Mann–Whitney U tests. Categorical data were presented as frequencies (percentages) and compared using Chi-squared tests or Fisher's exact test, as appropriate. Univariable logistic regression was used to assess associations between preoperative variables (including regional Ktrans values, volumes, MoCA-B) and POD incidence. Variables showing P < 0.10 in univariable analysis or deemed clinically important (e.g., age, MoCA-B, hippocampal Ktrans, hippocampal volume, thalamic Ktrans, thalamic volume) were considered for inclusion in a multivariable logistic regression model using a backward stepwise (likelihood ratio) approach to identify independent predictors of POD. Potential multicollinearity among the predictors in the multivariable model was assessed by calculating Variance Inflation Factors (VIFs). P-values for group comparisons of BBB Ktrans and brain volumes in Table 2 were adjusted for multiple comparisons using the False Discovery Rate (FDR) method by Benjamini-Hochberg. Pearson correlation coefficients (for normally distributed data) or Spearman rank correlation coefficients (for non-normally distributed data) were used to assess relationships between hippocampal Ktrans, hippocampal volume, MoCA-B score, and CAM-S scores. A two-sided P-value < 0.05 was considered statistically significant.

Table 2.

Patient Preoperative MRI Characteristics: Comparison between POD and NPOD Groups

MRI Parameter	POD (n = 19)	NPOD (n = 31)	P-value†
BBB Ktrans (× 10–3 min−1), median (IQR)
Hippocampus	5.36 (3.99, 8.39)	3.89 (3.40, 4.68)	0.048
Thalamus	4.80 (3.58, 6.62)	3.55 (3.05, 4.57)	0.034
Frontal lobe	5.12 (3.77, 5.89)	3.80 (3.13, 6.29)	0.268
Temporal lobe	3.85 (2.82, 5.11)	3.29 (2.79, 3.71)	0.227
Volume (cm3), mean (SD)
Total Brain Volume	1445 (125)	1450 (139)	0.893
Hippocampal Volume	7.05 (0.93)	8.12 (1.01)	0.003
Thalamic Volume	6.81 (0.63)	7.53 (0.95)	0.008
Leukoaraiosis (Fazekas grade ≥ 2), n (%)	3 (15.8)	8 (25.8)	0.498

Data are presented as mean (SD) or median (IQR) for continuous variables and n (%) for categorical variables. ^†P-values for group comparisons of Ktrans and Volume are adjusted for multiple comparisons using the False Discovery Rate (FDR) method. Bold P-values indicate statistical significance (P_FDR < 0.05)

POD postoperative delirium, NPOD Non-POD, MRI Magnetic Resonance Imaging, BBB blood–brain barrier, Ktrans volume transfer constant, IQR interquartile range

Results

Characteristics of individuals with and without POD

From September 2022 to June 2023, 76 patients were assessed for eligibility. Of these, 60 were enrolled, and after exclusions (detailed in Fig. 1, the STROBE flow diagram), 50 patients (mean age 69.0 ± 3.3 years) completed the study and were included in the final analysis. POD occurred in 19 of 50 patients (38.0%) within the first five postoperative days. Demographic and perioperative characteristics are presented in Table 1. Patients who developed POD had significantly lower preoperative MoCA-B scores (20.8 ± 3.6) compared to those in the NPOD group (23.1 ± 2.4; P = 0.020) (Fig. 2A). There were no significant differences between groups in age, sex, BMI, education level, smoking, alcohol consumption, history of COVID-19, anxiety (HADS-A), sleep quality (PSQI), baseline comorbidities (including hypertension, diabetes, hypercholesterolemia, severe carotid stenosis, NYHA class, LMCA stenosis), or LVEF. Anesthetic techniques, number of grafted vessels, intraoperative drug dosages, duration of surgery, estimated blood loss, extubation time, and ICU length of stay were also comparable between the POD and NPOD groups (Table 1).

Fig. 1.

STROBE diagram detailing patient recruitment and flow through the study. STROBE Strengthening the Reporting of Observational Studies in Epidemiology, POD postoperative delirium, NPOD Non-POD, MoCA-B Montreal Cognitive Assessment-Basic, MRI Magnetic Resonance Imaging, DWI diffusion-weighted imaging

Table 1.

Patient Characteristics and Perioperative Data

Demographics	POD (n = 19)	NPOD (n = 31)	P-value
Age, y, mean (SD)	68.6 (3.3)	69.2 (3.3)	0.559
Female, n (%)	5 (26.3)	9 (29.0)	0.836
BMI, kg/m2, mean (SD)	25.0 (2.9)	24.9 (2.5)	0.911
Smoking, n (%)	6 (31.6)	7 (22.6)	0.481
Drinking, n (%)	6 (31.6)	9 (29.0)	0.894
Education level, n (%)			0.938
Elementary school	2 (10.5)	4 (12.9)
Middle school	12 (63.2)	20 (64.5)
College and above	5 (26.3)	7 (22.6)
COVID-19 history, n (%)	8 (42.1)	12 (38.7)	0.812
MoCA-B, mean (SD)	20.8 (3.6)	23.1 (2.4)	0.020
Anxiety (HADS-A ≥ 8), n (%)	5 (26.3)	7 (22.6)	0.764
PSQI (> 5), mean (SD)	6.8 (3.8)	4.8 (3.8)	0.079
LVEF (%), mean (SD)	58.5 (7.2)	59.1 (6.5)	0.735
Preoperative comorbidity, n (%)
Hypertension	14 (73.7)	22 (71.0)	0.836
Diabetes	4 (21.1)	14 (45.2)	0.130
Hypercholesterolemia	5 (26.3)	8 (25.8)	0.968
Severe carotid stenosis (> 70%)	1 (5.3)	2 (6.5)	1.000a
NYHA class (II/III/IV), n (%)	6/11/2 (31.6/57.9/10.5)	10/17/4 (32.3/54.8/12.9)	0.962
LMCA stenosis (> 50%), n (%)	5 (26.3)	7 (22.6)	0.764
Types of anesthesia, n (%)			0.332
General anesthesia alone	16 (84.2)	22 (71.0)
Combined general-regional	3 (15.8)	9 (29.0)
Duration of surgery, min, mean (SD)	273 (60)	272 (49)	0.954
Numbers of CABG vessels (2/3/4), n (%)	5/8/6 (26.3/42.1/31.6)	8/15/8 (25.8/48.4/25.8)	0.886
Intraoperative drugs, mean (SD)
Ciprofol, mg	108 (32)	112 (49)	0.743
Sevoflurane, ml (cumulative)	50.5 (11.3)	48.4 (9.6)	0.478
Remifentanil, µg (total)	1650 (325)	1625 (440)	0.842
Estimated blood loss, ml, mean (SD)	395 (112)	407 (112)	0.720
Extubation time, h, mean (SD)	18.6 (6.8)	23.1 (10.3)	0.094
ICU care duration, h, mean (SD)	68.0 (20.8)	67.2 (22.8)	0.907

Data are presented as mean (SD), median (IQR) for continuous variables, or n (%) for categorical variables. ^aFisher's exact test. Bold P-values indicate statistical significance (P < 0.05)

POD postoperative delirium, NPOD Non-POD, BMI body mass index, COVID-19 Corona Virus Disease 2019, MoCA-B Montreal Cognitive Assessment-Basic, HADS-A Hospital Anxiety and Depression Scale—Anxiety subscale, PSQI Pittsburgh Sleep Quality Index, LVEF Left Ventricular Ejection Fraction, NYHA New York Heart Association, LMCA Left Main Coronary Artery, CABG coronary artery bypass graft, ICU Intensive Care Unit

Fig. 2.

Comparison of POD and NPOD groups in A MoCA-B scores, B Hippocampal and Thalamic Ktrans, and C Hippocampal and Thalamic Volume. *P < 0.05. POD postoperative delirium, NPOD Non-POD, MoCA-B Montreal Cognitive Assessment-Basic, Ktrans volume transfer constant

MRI results: BBB permeability and brain volumes

Preoperative regional BBB Ktrans values and brain volumes are shown in Table 2. After FDR correction for multiple comparisons, patients in the POD group exhibited significantly higher median Ktrans values in the hippocampus (POD: 5.36 [IQR, 3.99–8.39] × 10⁻³ min⁻¹ vs. NPOD: 3.89 [IQR, 3.40–4.68] × 10⁻³ min⁻¹; P_FDR = 0.048) and thalamus (POD: 4.80 [IQR, 3.58–6.62] × 10⁻³ min⁻¹ vs. NPOD: 3.55 [IQR, 3.05–4.57] × 10⁻³ min⁻¹; P_FDR = 0.034) compared to the NPOD group (Fig. 2B). No significant differences in Ktrans were observed in the frontal or temporal lobes.

Patients in the NPOD group had significantly larger mean hippocampal volumes (NPOD: 8.12 ± 1.01 cm³ vs. POD: 7.05 ± 0.93 cm³; P = 0.003) and thalamic volumes (NPOD: 7.53 ± 0.95 cm³ vs. POD: 6.81 ± 0.63 cm³; P = 0.008) compared to the POD group (Fig. 2C). Total brain volume and the prevalence of leukoaraiosis did not differ significantly between groups.

Association with POD incidence and severity

Univariable logistic regression analysis (Table 3) revealed that higher preoperative hippocampal Ktrans (Odds Ratio [OR] per 1 × 10⁻³ min⁻¹ increment, 1.350; 95% CI, 1.048–1.740; P = 0.020) and thalamic Ktrans (OR, 1.466; 95% CI, 1.017–2.113; P = 0.040) were associated with increased odds of POD. Lower MoCA-B scores (OR per point increment, 0.760; 95%CI, 0.604–0.957; P = 0.020), smaller hippocampal volume (OR per cm³ increment, 0.297; 95% CI, 0.131–0.672; P = 0.004), and smaller thalamic volume (OR per cm³ increment, 0.304; 95% CI, 0.121–0.766; P = 0.012) were also significantly associated with higher odds of POD.

Table 3.

Univariable Logistic Regression Analysis for Predictors of Postoperative Delirium (POD)

Variable	OR	95% CI (OR)		P-value
		Lower	Upper
MoCA-B (per point increase)	0.760	0.604	0.957	0.020
Hippocampal Ktrans (per 1 × 10−3 min−1 unit)	1.350	1.048	1.740	0.020
Thalamic Ktrans (per 1 × 10−3 min−1 unit)	1.466	1.017	2.113	0.040
Frontal lobe Ktrans (per 1 × 10−3 min−1 unit)	1.189	0.820	1.526	0.360
Temporal lobe Ktrans (per 1 × 10−3 min−1 unit)	1.185	0.919	1.526	0.190
Total Brain Volume (per cm3 unit)	1.000	0.995	1.004	0.890
Hippocampal Volume (per cm3 unit)	0.297	0.131	0.672	0.004
Thalamic Volume (per cm3 unit)	0.304	0.121	0.766	0.012

(n total = 50; POD = 19). Dependent variable: POD. Bold P-values indicate statistical significance (P < 0.05)

POD postoperative delirium, OR odds ratio, CI confidence interval, MoCA-B Montreal Cognitive Assessment-Basic, Ktrans volume transfer constant

In the multivariable logistic regression model, after adjusting for other covariates (MoCA-B, hippocampal Ktrans, thalamic Ktrans, hippocampal volume, thalamic volume), only lower preoperative MoCA-B scores (adjusted OR [aOR], 0.697; 95% CI, 0.513–0.947; P = 0.021) and smaller preoperative hippocampal volume (aOR, 0.322; 95% CI, 0.105–0.992; P = 0.048) remained independent predictors of POD incidence (Table 4). No significant multicollinearity was detected among the variables included in the final model (all Variance Inflation Factors < 2.0). Neither hippocampal Ktrans nor thalamic Ktrans retained statistical significance as independent predictors of POD incidence in this adjusted model.

Table 4.

Multivariable Logistic Regression Analysis for Independent Predictors of Postoperative Delirium (POD)

Variable	Adjusted OR	95% CI (aOR)		P-value
		Lower	Upper
MoCA-B (per point increase)	0.697	0.513	0.947	0.021
Hippocampal Ktrans (per 1 × 10−3 min−1 unit)	0.837	0.540	1.296	0.424
Thalamic Ktrans (per 1 × 10−3 min−1 unit)	1.502	0.761	2.964	0.241
Hippocampal Volume (per cm3 unit)	0.322	0.105	0.992	0.048
Thalamic Volume (per cm3 unit)	0.478	0.143	1.599	0.231

(n total = 50; POD = 19). Dependent variable: POD. Model adjusted for MoCA-B, Hippocampal Ktrans, Thalamic Ktrans, Hippocampal Volume, Thalamic Volume. Bold P-values indicate statistical significance (P < 0.05)

POD postoperative delirium, aOR adjusted odds ratio, CI confidence interval, MoCA-B Montreal Cognitive Assessment-Basic, Ktrans volume transfer constant

A significant negative correlation was observed between preoperative hippocampal Ktrans and hippocampal volume (r = − 0.372, P = 0.008). Preoperative MoCA-B scores showed a moderate negative correlation with postoperative CAM-S scores (r = − 0.439, P = 0.001). Importantly, in patients who developed POD (n = 19), higher preoperative hippocampal Ktrans was strongly correlated with greater POD severity as measured by peak CAM-S scores (r = 0.673, P = 0.002) (Fig. 3A). No significant correlation was found between hippocampal volume and CAM-S scores in the POD group (r = 0.175, P = 0.474) (Fig. 3B). Representative hippocampal DCE-MRI images are shown in Fig. 4.

Fig. 3.

Correlations in patients who developed POD (n = 19). A Scatter plot showing the correlation between preoperative hippocampal Ktrans and peak postoperative CAM-S scores. B Scatter plot showing the correlation between preoperative hippocampal volume and peak postoperative CAM-S scores. CAM-S Confusion Assessment Method – Severity, POD postoperative delirium, Ktrans volume transfer constant

Fig. 4.

Representative preoperative DCE-MRI Ktrans maps of the hippocampus. A-C Images from a patient who underwent OPCABG and subsequently developed POD, showing relatively higher Ktrans values (warmer colors indicating higher permeability). D-F Images from a patient who underwent OPCABG and did not develop POD, showing relatively lower Ktrans values (cooler colors indicating lower permeability). DCE-MRI Dynamic contrast-enhanced magnetic resonance imaging, Ktrans volume transfer constant, POD postoperative delirium, OPCABG off-pump coronary artery bypass grafting.

Discussion

In this prospective observational pilot study of older patients undergoing OPCABG, we investigated the association between preoperative regional BBB permeability and POD. Our principal finding is that while higher preoperative Ktrans in the hippocampus and thalamus, indicative of increased BBB permeability, was associated with POD in univariable analyses, these associations did not persist as independent predictors of POD incidence after adjusting for baseline cognitive function (MoCA-B) and hippocampal volume. Instead, lower preoperative MoCA-B scores and smaller hippocampal volume emerged as independent preoperative risk factors for developing POD. Notably, however, higher preoperative hippocampal BBB permeability (Ktrans) was significantly correlated with greater severity of POD among those who experienced delirium.

The observation that baseline cognitive function and hippocampal volume were stronger independent predictors of POD incidence than regional Ktrans is a critical finding that warrants careful interpretation. It is plausible that BBB permeability changes represent an earlier, more subtle stage in a pathological cascade that may eventually lead to structural changes (e.g., hippocampal atrophy) and functional cognitive decline [8, 10]. In this scenario, hippocampal volume and MoCA-B scores might be more proximate and thus more powerful predictors of POD incidence when all factors are considered simultaneously in a multivariable model. The significant negative correlation we observed between hippocampal Ktrans and volume supports this notion, suggesting that chronic BBB dysfunction could contribute to neurodegenerative processes and subsequent brain atrophy over time [33, 34]. Alternatively, both increased BBB permeability and reduced hippocampal volume could be manifestations of shared underlying pathologies, such as chronic cerebrovascular disease or sustained neuroinflammation, which independently predispose to POD [35, 36].

Despite not being an independent predictor of POD incidence in the full model, the univariable association of increased hippocampal and thalamic Ktrans with POD, and particularly the strong correlation between preoperative hippocampal Ktrans and subsequent POD severity (CAM-S scores), underscores the potential clinical relevance of preoperative BBB integrity. Rather than acting as the primary trigger for the onset of delirium, a more permeable BBB may function as a vulnerability factor that dictates the “severity” of the delirious state once it is initiated by other perioperative insults. This suggests that a pre-existing compromised BBB, especially in the vulnerable hippocampus, may render the brain more susceptible to the adverse effects of perioperative stressors (e.g., anesthesia, surgical trauma, systemic inflammation, hemodynamic fluctuations), even those potentially attenuated by OPCABG [11, 16, 37]. OPCABG avoids cardiopulmonary bypass, which is a major source of systemic inflammation and embolic events, yet POD remains prevalent. Our findings suggest that a pre-weakened BBB might lower the threshold for neurotoxic substances (e.g., cytokines, damage-associated molecular patterns, or even albumin) to enter the brain, exacerbating neuroinflammatory responses and contributing to more severe delirious states when POD occurs [13, 38]. The hippocampus, crucial for learning and memory, is known to be highly susceptible to age-related BBB breakdown and is a key region implicated in cognitive dysfunction [39, 40]. The Ktrans values in our coronary artery disease cohort, particularly in the hippocampus, appear somewhat elevated compared to values typically reported in healthy older controls and are more aligned with, or slightly exceed, values seen in individuals with MCI in other cohorts [8, 40]. This elevation in our cardiac surgery candidates likely reflects the higher burden of systemic vascular risk factors (e.g., hypertension, diabetes, hypercholesterolemia) prevalent in patients with coronary artery disease, which are also known risk factors for cerebrovascular pathology and BBB dysfunction [33, 35, 36].

Our findings regarding hippocampal volume as an independent predictor of POD are consistent with some, but not all, prior research. Chen et al. [29] found hippocampal volume reduction to be a risk factor for postoperative cognitive dysfunction in elderly patients. However, studies in non-cardiac surgery by Cavallari et al. [41] and a review by Huang et al. [42] did not find a strong association between global or hippocampal atrophy and POD incidence, possibly due to differences in patient populations (e.g., better baseline cognitive function in their cohorts) or surgical types. The higher prevalence of potential MCI in our cohort might explain the observed significance of hippocampal volume. Reduced hippocampal volume can reflect underlying neurodegenerative processes or chronic vascular injury, rendering the brain less resilient to perioperative insults [43–45].

The thalamus, a critical relay station involved in attention, executive function, and memory, also showed increased preoperative BBB permeability in patients who subsequently developed POD [46]. While thalamic Ktrans and volume were significant in univariable analyses, they did not remain independent predictors in our multivariable model for POD incidence. This contrasts slightly with Fislage et al. [47], who found larger thalamic volume protective against POD, although their study did not assess BBB permeability. The complex interplay of multiple brain regions and pathological processes likely contributes to POD, and the relative importance of each may vary depending on the patient population and surgical context.

This study contributes to the understanding of POD pathophysiology by highlighting pre-existing neurovascular unit vulnerabilities. While routine preoperative DCE-MRI for Ktrans measurement is unlikely to become a widespread screening tool due to cost and accessibility, these findings support the concept that compromised BBB integrity is part of the predisposing factors for POD. Future research could focus on identifying less invasive biomarkers (e.g., blood-based markers of endothelial dysfunction or neuroinflammation) that correlate with BBB permeability and predict POD risk [6, 44, 48, 49]. Such biomarkers could help stratify patients for targeted perioperative neuroprotective interventions.

This study has several limitations. First, as a pilot study, the sample size (n = 50) is modest, which may limit statistical power, particularly for multivariable analyses, and could explain why Ktrans effects did not hold after adjustment. The initial effect size assumption for the power calculation might have been optimistic, and findings, especially the negative results in the multivariable model for Ktrans, should be interpreted with caution and require validation in larger, adequately powered multicenter studies. Second, we only assessed preoperative BBB permeability. Serial postoperative MRI scans were not performed due to logistical challenges in acutely ill, delirious older patients. Assessing dynamic postoperative BBB changes would be valuable in future studies to understand if preoperative permeability influences the magnitude or recovery trajectory of surgery-induced BBB disruption [11, 13, 50, 51]. Third, our findings are specific to older patients with a significant burden of vascular risk factors undergoing OPCABG for coronary artery disease and may not be generalizable to patients undergoing on-pump cardiac surgery, non-cardiac surgery, or to younger, healthier populations. Fourth, we did not measure plasma biomarkers of neurodegeneration (e.g., neurofilament light chain, tau) or inflammation, which could provide further mechanistic insights. Fifth, while we excluded recent stroke, the chronic impact of cerebrovascular disease on BBB integrity is complex. DWI was used to exclude acute events, but subtle chronic changes might still influence BBB status. Sixth, the inclusion criterion of a MoCA-B score ≥ 18 excluded patients with pre-existing moderate to severe cognitive impairment. This selection bias might limit the generalizability of our findings, as we did not include the population potentially at the highest risk for POD and with potentially different underlying brain pathologies. Finally, while AIF selection was standardized and motion correction applied, inherent variabilities in DCE-MRI acquisition and processing can influence Ktrans quantification; however, our use of two blinded raters with good inter-rater reliability strengthens our imaging data.

Conclusion

This prospective observational pilot study demonstrates that in older patients undergoing OPCABG, lower preoperative cognitive function (MoCA-B) and smaller hippocampal volume are independent preoperative predictors of POD incidence. Although increased preoperative BBB permeability (Ktrans) in the hippocampus and thalamus was associated with POD univariably, this association did not remain independent for POD incidence in the multivariable model. However, greater preoperative hippocampal BBB permeability was significantly correlated with increased POD severity. These preliminary and exploratory findings suggest that pre-existing neurovascular unit vulnerability, encompassing both structural and BBB integrity aspects, contributes to the development and severity of POD. This study underscores the importance of considering baseline brain health in risk stratification and provides a preliminary foundation for future, larger-scale investigations into perioperative BBB dynamics and targeted neuroprotective strategies for POD prevention.

Authors' contributions

Lichao Di: Conceptualization, Methodology, Writing- Original draft preparation. Peiying Huang: Data curation. Yeju He: Investigation, Imaging analysis. Jie Li: Supervision, Imaging analysis. Yu Liu: Resources, Patient recruitment. Liwei Chi: Data curation. Na Sun: Data curation, Validation. Lining Huang: Methodology, Writing- Reviewing and Editing All authors approved the final version of the manuscript.

Funding

This study was supported by Medical Excellent Talents Project Funded by Hebei Provincial Government in 2022 (No.303-2022-27-04) and S&T Program of Hebei (No.H2022206586). Medical Science Research Project of Hebei (No. 20240951).

Data availability

Data is provided within the manuscript files, further enquiries can be directed to the corresponding author.

Declarations

Conflict of interest

The authors declare no competing interests.

Ethical approval and consent to participate

The study protocol was approved by the Ethics Committee of The Second Hospital of Hebei Medical University (No. 2022-R707), and written informed consent for study participation was obtained from all patients or their legally authorized representatives prior to any study-related procedures. This research adheres to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines.

Consent for publication

Not applicable.

Informed Consent statement

Informed consent was obtained from all individual participants included in the study.