Diabetes mellitus and the risk of postoperative visual and auditory hallucinations after cardiac surgery: insights from the VAACS study

Abstract

Objectives

Postoperative hallucinations are underrecognized neuropsychiatric complications of cardiac surgery. Glycemic control may influence their occurrence, particularly in patients with diabetes. This study aimed to estimate the incidence of visual and auditory hallucinations in diabetic and non-diabetic patients after cardiac surgery, and to identify factors independently associated with hallucinations within each group.

Methods

In this prospective multicenter cohort, hallucinations were assessed daily for 7 days using the Questionnaire for Psychotic Experiences. Cox regression identified predictors within diabetic and non-diabetic cohorts.

Results

Visual hallucinations occurred in 12.0% of diabetic and 11.2% of non-diabetic patients, while auditory hallucinations occurred in 7.9% and 6.0%, respectively, with no statistically significant between-group difference. Among diabetic patients, elevated preoperative HbA1c independently predicted both visual hallucinations (adjusted HR 1.29; 95% CI, 1.05–1.59) and auditory hallucinations (adjusted HR 1.37; 95% CI, 1.07–1.76). Higher preoperative blood glucose was also independently associated with increased risk of visual hallucinations (adjusted HR 1.011; 95% CI, 1.002–1.020) and auditory hallucinations (adjusted HR 1.013; 95% CI, 1.003–1.024). Morphine use showed an inverse association with auditory hallucinations in diabetic patients, although this did not reach statistical significance after adjustment (adjusted HR 0.57; 95% CI, 0.29–1.12). Longer postoperative mechanical ventilation was inversely associated with auditory hallucinations in diabetic patients (adjusted HR 0.80; 95% CI, 0.71–0.91), although this finding should be interpreted cautiously because hallucination assessment began only after extubation. Among non-diabetic patients, hemodialysis dependence independently predicted visual hallucinations (adjusted HR 6.99; 95% CI, 1.59–30.61), while critical preoperative state (adjusted HR 2.70; 95% CI, 1.23–5.93) and higher EuroSCORE II (adjusted HR 1.10; 95% CI, 1.05–1.15) independently predicted auditory hallucinations.

Conclusion

Hallucinations affect a notable prportion of patients after cardiac surgery. Higher preoperative HbA1c and blood glucose were associated with postoperative hallucination risk, particularly in diabetic patients. These findings identify dysglycemia as a clinically relevant risk marker, but do not establish that short-term glycemic optimization would directly prevent hallucinations.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s13741-026-00722-0.

Introduction

Cardiac surgery is a vital therapeutic intervention for patients with cardiovascular disease. Among the most commonly performed procedures are coronary artery bypass grafting (CABG) and valvular replacement and/or repair. Despite advances in surgical techniques and perioperative management, postoperative complications continue to affect patient outcomes significantly. These complications contribute to prolonged hospital stays and increased surgical mortality (Pahwa et al. 2021). While overall operative mortality has declined, neurological complications remain prevalent and are associated with longer ICU admissions, heightened morbidity, and long-term functional impairment (Bowden et al. 2023). In addition to well-documented complications such as stroke and delirium, increasing attention has been drawn to postoperative hallucinations. These hallucinations are frequently overlooked or misattributed to other neurological disturbances, yet they have a meaningful impact on patient recovery (Teller et al. 2022).

Postoperative hallucinations may involve visual, auditory, tactile, or olfactory experiences, as well as vivid dreams, nightmares, or delusions (Kumar et al. 2009). Risk factors include the use of sedative medications, sepsis, and extended ICU stays (Kumar et al. 2009). A meta-analysis by Liu et al. found a 44% higher risk of postoperative cognitive dysfunction (POCD) in diabetic patients compared to non-diabetic individuals (Liu et al. 2025). Moreover, even with appropriate use of oral hypoglycemic agents, diabetic patients showed a higher incidence of POD, particularly those with elevated preoperative HbA1c levels (≥ 6%) (Kotfis et al. 2019).

In this multicenter prospective cohort study, we aimed to determine and compare the cumulative incidence of auditory and visual hallucinations among diabetic and non-diabetic individuals and identify independent predictors of hallucinations within each group. While postoperative delirium is well recognized, we specifically focused on visual and auditory hallucinations because these sensory-specific experiences are clinically relevant yet often underreported after cardiac surgery. This focus is also consistent with prior prospective cardiac surgery studies that used the Questionnaire for Psychotic Experiences (QPE) to characterize the occurrence and phenomenology of postoperative hallucinations. The QPE enabled structured daily assessment of hallucination type and severity during the first postoperative week. We did not aim to diagnose delirium or to establish hallucinations as independent from delirium; rather, we examined them as postoperative neuropsychiatric symptoms that may overlap with delirium-spectrum states and merit separate descriptive evaluation. This focus allows early detection of sensory-specific neuropsychiatric changes that may otherwise be overlooked in broader delirium assessments.

Methodology

Study design, setting, and population

This prospective cohort study was conducted between September 2022 and June 2025 across 10 cardiac surgery centers in the West Bank of Palestine as part of the VAACS study (ISRCTN18368253). Adult patients (≥ 18 years) undergoing elective or semi-elective major cardiac surgery were enrolled and stratified into diabetic and non-diabetic groups. Diabetes was defined by standard WHO/ADA criteria, and preoperative HbA1c was measured in all included patients. Exclusion criteria included type 1 diabetes, major psychiatric or neurological disorders, severe sensory impairment, active delirium, or terminal illness. See Fig 1. Further details, including hospital list and diagnostic definitions, are available in Supplementary File 1.

Fig. 1.

Flow chart

Data collection instrument

Data collection was conducted using the Questionnaire for Psychotic Experiences (QPE), a validated instrument commonly employed in psychiatric research to assess auditory and visual hallucinations during the perioperative phase. Information was gathered prospectively throughout patient care by completing the questionnaire within the first week postoperatively. Quantitative approaches were applied to analyze the frequency and severity of hallucinations, while qualitative methods were used to document descriptive observations. The internal consistency of the QPE tool was evaluated using Cronbach’s alpha of 0.70–0.91, which aligns with a previous study (Teller et al. 2022).

Anesthetic and sedation protocol, and timing of hallucination assessment

Anesthetic management across participating centers followed standardized cardiac anesthesia protocols. Induction typically began with the administration of a benzodiazepine (most commonly midazolam), followed by a narcotic analgesic, a hypnotic agent, and a neuromuscular relaxant such as rocuronium. Postoperatively, morphine was used for analgesia as indicated. Benzodiazepines were withheld in patients with contraindications, including severe hypotension, advanced renal failure, known allergy, or aortic valve surgery due to aortic stenosis, in which sedative sensitivity was anticipated. Routine postoperative benzodiazepine sedation was generally avoided except when clinically justified. For patients who remained intubated postoperatively, the Questionnaire for Psychotic Experiences (QPE) was administered only after extubation, once the patient was fully awake, oriented, and hemodynamically stable. Hallucination symptoms occurring during mechanical ventilation were therefore not assessed directly. Subsequent evaluations were conducted once daily during the first seven postoperative days. This structured timing reduced confounding from residual anesthetic or sedative effects, but may also have delayed outcome ascertainment in patients requiring prolonged ventilation.

Outcomes and assessments

The primary outcomes were the incidence and severity of postoperative visual and auditory hallucinations during the first seven days after surgery. These were assessed daily using the validated Questionnaire for Psychotic Experiences (QPE), which distinguishes between hallucination types and quantifies symptom severity. The QPE was used in this study to identify and characterize postoperative visual and auditory hallucinations and their severity; it was not used as a diagnostic instrument for delirium. No parallel formal postoperative delirium adjudication using a dedicated delirium assessment tool or DSM-based criteria was performed. Accordingly, the outcome should be interpreted as postoperative hallucination symptoms rather than confirmed delirium, and potential overlap between hallucinations and delirium cannot be excluded. Hallucinations were defined as sensory perceptions that occur without external stimuli, typically while awake. Severity was scored using a four-item QPE subscale (range: 0–20). Any new hallucinations identified on days five through seven were confirmed via abbreviated QPE re-assessment. Severity scores and incidence rates were compared between diabetic and non-diabetic cohorts. Each interviewer verified alertness and comprehension before proceeding, and questionnaires with incomplete or invalid responses were excluded from the analysis. The 7-day structured follow-up minimized recall bias by capturing hallucination onset and severity in near-real time. See Supplement 1 Methods for more details.

Electronic medical records and study variables

Baseline demographic, clinical, laboratory, and perioperative variables were extracted from electronic medical records. Diabetes-specific measures included HbA1c, preoperative glucose, insulin dependence, and duration of diabetes. Surgical, intraoperative, and postoperative data were also collected. The primary endpoint was postoperative hallucinations, assessed daily for 7 days with the validated QPE tool. Further details are provided in Supplementary File 1.

Bypass strategy and off-pump procedures

Cardiac procedures in this cohort were performed either on-pump with cardiopulmonary bypass (CPB) or off-pump at the surgeon’s discretion; pump status (on vs. off) was prespecified and entered as a covariate in all bivariable and multivariable analyses. Deep hypothermic circulatory arrest (DHCA)/circulatory arrest was not employed in this cohort; therefore, it is not presented as a separate exposure. For context, off-pump cases represented a minority of operations across both diabetic and non-diabetic groups, and CPB and aortic cross-clamp times were recorded for all on-pump cases. See eTables 2–7 (Supplement 1) for the full distribution of pump status and their bivariable associations with hallucination outcomes.

Cerebrovascular disease assessment and control

Pre-existing cerebrovascular disease was defined based on a documented history of ischemic or hemorrhagic stroke, transient ischemic attack, or carotid artery stenosis ≥ 50% confirmed by imaging. This variable was extracted from preoperative medical records and EuroSCORE II entries. In statistical analyses, cerebrovascular disease was included as an independent covariate in univariate models and indirectly adjusted for in multivariable models through the EuroSCORE II composite, which incorporates cerebrovascular and peripheral vascular risk components.

Timing of assessment

To minimize confounding by residual anesthesia or sedation, the first postoperative QPE assessment was performed only after patients were fully awake and hemodynamically stable. Subsequent evaluations were conducted once daily during the first 7 postoperative days. This schedule ensured that early assessments reflected a clear state of mind while avoiding delayed recall bias associated with late postoperative interviews.

Procedure details

CABG, valve surgeries

Cardiac surgeries, including coronary artery bypass grafting (CABG) and valvular replacement and/or repair, were performed under general anesthesia using standardized operative protocols. CPB was used in applicable cases with cold cardioplegia for myocardial protection. In combined procedures, distal grafts preceded valve replacement. For isolated valve surgeries, exposure was achieved via aortotomy or atriotomy, depending on valve location. Intraoperative monitoring included transesophageal echocardiography. See eAppendix 1 in Supplementary File 1 for complete procedural descriptions and institutional practices.

Ethical considerations

This study received ethical approval from the Institutional Review Board (IRB) of An-Najah National University (Approval No.: IRB-12-09-2022, Date: 12 September 2022). Written informed consent was obtained from all participants after receiving a clear explanation of the study objectives, potential benefits, and associated risks. Participation was voluntary, and participants were informed of their right to withdraw from the study at any time without repercussions. All collected data were handled confidentially and securely stored. Formal administrative permission to conduct the study was also obtained from all collaborating academic institutions.

Statistical analysis

All analyses were performed using STATA version 17. A two-tailed P value < 0.05 was considered statistically significant. Continuous variables were summarized as means ± standard deviations or medians (Q1-Q3), and categorical variables as frequencies and percentages. For baseline comparisons, we used independent t-tests or Mann–Whitney U tests for continuous variables, and χ² or Fisher’s exact tests for categorical variables. Kaplan–Meier survival analyses were conducted to estimate the cumulative incidence of visual and auditory hallucinations over the 7-day postoperative period and were stratified by diabetes status (diabetic vs. non-diabetic cohorts). Univariate analyses were conducted to identify factors associated with the occurrence of visual or auditory hallucinations.

Covariates were pre-specified based on prior literature and clinical relevance and included demographic, comorbid, intraoperative, and postoperative factors (Ottens et al. 2020, Zhuang et al. 2023, Teller et al. 2022). Variables with a univariable P < 0.10 were then included in multivariable Cox proportional hazards regression models to determine independent predictors of postoperative hallucinations. Hazard ratios (HR) with 95% confidence intervals (CI) are reported for the multivariable models. The proportional hazards assumption was verified, and multicollinearity among covariates was assessed using variance inflation factors (VIF), with all VIF values < 2 indicating no significant collinearity. The building of the model aligned with the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD) criteria (Collins et al. 2015).

Results

Detailed comparisons of demographic, clinical, laboratory, and procedural variables with the occurrence of postoperative visual and auditory hallucinations in both diabetic and non-diabetic patients are provided in Supplement 1 File (see eTables 2 and 3). These tables illustrate key associations, including HbA1c, preoperative blood glucose, recent myocardial infarction, and other comorbidities with hallucination risk.

In Fig. 2A, 617 diabetic and 715 non-diabetic patients were included. By postoperative day 6, the number at risk decreased from 617 to 569 in the diabetic group and from 715 to 653 in the non-diabetic group. The cumulative incidence curves of visual hallucinations progressed in parallel across both cohorts, with no statistically significant difference between diabetic and non-diabetic patients (log-rank p = 0.84).

Fig. 2.

Kaplan–Meier Estimates of Postoperative Hallucinations by Diabetes Mellitus Status. A and B present Kaplan–Meier curves comparing the cumulative incidence of visual and auditory hallucinations, respectively, among patients with and without DM

In Fig. 2B, 617 diabetic and 715 non-diabetic patients were included. By postoperative day 6, the number at risk declined to 584 in the diabetic group and 678 in the non-diabetic group. The Kaplan–Meier curves for auditory hallucinations demonstrated similar trajectories between groups, with no statistically significant difference (log-rank p = 0.17), indicating no meaningful temporal variation attributable to diabetes status.

The severity of visual and auditory hallucinations stratified by diabetes status is illustrated in eFigure 1, showing a trend toward more severe auditory hallucinations in diabetic patients. Further exploratory analyses, including extended statistical correlations and subgroup comparisons, are provided in Supplementary File 1 (eTables 4–7).

Table 1 summarizes the bivariate relationships between clinical, laboratory, and procedural variables and the incidence of postoperative visual hallucinations, stratified by diabetic status. Among non-diabetic patients, significant predictors included hemodialysis dependence (5% vs. 0.6%, p < 0.001) and immunosuppressive therapy (6.3% vs. 1.1%, p = 0.001). Non-diabetic patients who developed visual hallucinations also had a significantly higher EuroSCORE II Median (Q1–Q3): 4.53 (0.54–9.60) compared to those without hallucinations Median (Q1–Q3): 3.37 (2.51–4.23)], (p < 0.001). In the diabetic cohort, recent myocardial infarction was more common in patients with visual hallucinations (29.7% vs. 19.5%, p = 0.042). Preoperative HbA1c was significantly higher among those with hallucinations (8.42 ± 1.11 vs. 8.06 ± 1.09, p = 0.007). Notably, preoperative blood glucose levels also differed significantly (188.1 ± 31.6 vs. 174.0 ± 29.3, p = < 0.001), supporting the hypothesis that chronic and acute glycemic burden may contribute to postoperative neuropsychiatric vulnerability. No significant associations were found for insulin dependence or oral diabetic drug use. To see full details, see eTable 2 in Supplementary File 1.

Table 1.

Relationship of visual hallucinations with clinical, laboratory, and procedural parameters in diabetic and non-diabetic patients

Variable	Category	Diabetic n = 617				Non-Diabetic n = 715
		Visual hallucination n = 74		No Visual hallucination n = 543	P value	Visual hallucination n = 80	No Visual hallucination n = 635	P value
Demographic variables
Age, years	Mean ± SD		64.31 ± 7.75	61.42 ± 8.03	0.933a	55.81 ± 12.80	56.24 ± 12.28	0.771a
BMI, kg/m2	Mean ± SD		24.68 ± 3.74	24.56 ± 4.38	0.828a	23.46 ± 4.46	23.47 ± 3.89	0.983a
Gender	Male		52(70.3%)	412(75.9%)	0.142b	56(70%)	481(75.7%)	0.263b
Medical History
Critical preoperative state			7(9.5%)	74(13.6%)	0.319b	10(12.5%)	71(11.2%)	0.726b
Recent myocardial infarction			22(29.7%)	106(19.5%)	0.042b	13(16.3%)	109(17.2%)	0.920b
Type of Surgery	CABG		64(86.5%)	463(85.3%)	0.291b	51(63.8%)	419(66%)	0.692b
	Valve surgery		10(13.5%)	80(14.7%)		29(36.2%)	216(34%)
Hemodialysis-dependent			0(0%)	4(0.7%)	0.459b	4(5%)	4(0.6%)	< 0.001b
Immunosuppressive drugs			2(2.7%)	4(0.7%)	0.106b	5(6.3%)	7(1.1%)	0.001b
Previous use of antibiotics			3(4.1%)	32(5.9%)	0.521b	4(5%)	33(5.2%)	0.940b
Euroscore II, Median (Q1-Q3)			4.55 (0.61–9.71)	2.61(0.38–5.60)	0.242C	4.53(0.54–9.60)	3.37(2.51–4.23)	< 0.001C
Insulin dependence Yes			39 (52.7%)	265 (48.8%)	0.519b	N\A	N\A	N\A
Oral diabetic drugs Yes			43 (58.1%)	330 (60.7%)	0.622b	N/A	N/A	N/A
Duration of diabetes (years) Mean ± SD			10.6 ± 5.1	9.8 ± 4.9	0.096a	N\A	N\A	N\A
Laboratory
Albumin, g/dl	Mean ± SD		2.24 ± 0.32	2.15 ± 0.57	0.052a	2.51 ± 0.60	2.49 ± 0.60	0.701a
Creatinine, mg/dL	Mean ± SD		1.03 ± 0.27	1.11 ± 0.75	0.354a	1.24 ± 1.06	1.04 ± 0.48	0.096a
Pre-operative HbA1c (%)	Mean ± SD		8.42 ± 1.11	8.06 ± 1.09	0.007a	5.69 ± 0.32	5.68 ± 0.35	0.893a
Pre-operative blood glucose, mg/dL	Mean ± SD		188.1 ± 31.6	174.0 ± 29.3	< 0.001a	102.7 ± 13.4	101.7 ± 12.2	0.460a
Medication use
Vasopressin			0(0%)	10(1.8%)	0.239b	3(3.8%)	5(0.8%)	0.050b
Pethidine			25(33.8%)	194(35.7%)	0.743b	30(37.5%)	187(29.4%)	0.156b
Morphine			32(43.2%)	220(40.5%)	0.654b	34(42.5%)	285(44.9%)	0.721b
Procedural parameters
Cardiopulmonary bypass time, min	Mean ± SD		119.68 ± 38.43	118.21 ± 45.02	0.763a	115.73 ± 41.17	116.68 ± 44.53	0.856a
Aortic cross clamp time, min	Mean ± SD		86.26 ± 34.72	87.77 ± 39.64	0.732a	76.02 ± 35.91	81.06 ± 36.62	0.245a
Left ventricle ejection fraction, %	Mean ± SD		49.44 ± 7.29	51.54 ± 6.68	0.021a	53.92 ± 5.55	53.26 ± 6.50	0.386a
Post-operative complication
Post-operative Ventilation, hours	Median (Q1-Q3)		5.96 (2.27–9.65)	5.60 (1.85–9.35)	0.601c	5.17 (2.17–8.17)	5.71 (1.38–10.04)	0.466c

a. tested by an independent t-test. b. tested by the Chi-square test. c. tested by Wilcoxon rank-sum (Mann–Whitney) test

Abbreviations: CABG Coronary Artery Bypass Grafting, DM Diabetes Mellitus, LVEF Left Ventricular Ejection Fraction, SD Standard Deviation

Table 2 presents a bivariate analysis of factors associated with postoperative auditory hallucinations, stratified by diabetic status. Among diabetic patients, those who developed auditory hallucinations were significantly younger (58.7 ± 7.3 vs. 61.7 ± 8.0 years, p = 0.012) and exhibited higher preoperative HbA1c (8.54 ± 1.26 vs. 8.06 ± 1.07%, p = 0.003) and higher preoperative blood glucose levels (188.3 ± 29.2 vs. 174.6 ± 29.8 mg/dL, p = 0.002). They also had shorter aortic cross-clamp time (73.2 ± 34.5 vs. 88.8 ± 39.2 min, p = 0.007). Patients who experienced hallucinations had a significantly shorter postoperative mechanical ventilation duration [Median (Q1–Q3): 3.58 (2.05–5.11) hours] compared to those without hallucinations [5.82 (1.96–9.68) hours] (p = 0.007). Previous antibiotic use was more frequent in the hallucination group (12.2% vs. 5.1%, p = 0.038), although no significant associations were observed with insulin dependence or oral diabetic therapy. In the non-diabetic cohort, patients had higher preoperative HbA1c (5.79 ± 0.29 vs. 5.68 ± 0.35%, p = 0.049) and were more likely to be in a critical preoperative state (20.9% vs. 10.7%, p = 0.040). No significant associations were found with age, gender, or immunosuppressive use. To see full details, see eTable 3 in the Supplement 1 file.

Table 2.

Relationship of auditory hallucinations with clinical, laboratory, and procedural parameters

Variable	Category	Diabetic n = 617				Non-Diabetic n = 715
		Auditory hallucination n = 49		No Auditory hallucination n = 568	P value	Auditory hallucination n = 43	No Auditory hallucination n = 672	P value
Demographic variables
Age, years	Mean ± SD		58.73 ± 7.3041	61.697 ± 8.0042	0.012a	55.186 ± 13.2472	56.167 ± 12.2647	0.613a
BMI, kg/m2	Mean ± SD		25.60 ± 3.864	24.49 ± 4.333	0.083a	22.5921 ± 3.44009	23.5302 ± 3.86433	0.121a
Gender	Male		38(77.6%)	426(75%)	0.692b	32(74.4%)	505(75.1%)	0.915b
Medical History
Critical preoperative state			6(12.2%)	75(13.2%)	0.849b	9(20.9%)	72(10.7%)	0.040b
Recent myocardial infarction			8(16.3%)	120(21.1%)	0.427b	9(20.9%)	113(16.8%)	0.487b
Type of Surgery	CABG		44(89.8%)	483(85%)	0.365b	30(69.8%)	440(65.5%)	0.565b
	Valve surgery		5(10.2%)	85(15%)		13(30.2%)	232(34.5%)
Hemodialysis-dependent			1(2%)	3(0.5%)	0.206b	1(2.3%)	7(1%)	0.438b
Immunosuppressive drugs			1(2%)	5(0.9%)	0.427b	2(4.7%)	10(1.5%)	0.118b
Previous use of antibiotics			6(12.2%)	29(5.1%)	0.038b	3(7%)	34(5.1%)	0.582b
Euroscore II, Median(Q1-Q3)			3.46 (0.92–6.00)	2.92 (2.02–3.82)	0.326C	5.62 (1.22–12.46)	3.36 (2.51–4.21)	0.153C
Insulin dependence Yes			26 (53.1%)	278 (48.9%)	0.525b	N/A	N/A	N/A
Oral diabetic drugs Yes			26 (53.1%)	315 (55.5%)	0.741b	N/A	N/A	N/A
Duration of diabetes			10.8 ± 4.7	10.0 ± 5.2	0.117a	N/A	N/A	N/A
Mean ± SD
Laboratory
Albumin, g/dl	Mean ± SD		2.26 ± 0.587	2.15 ± 0.547	0.195a	2.40 ± 0.714	2.49 ± 0.584	0.286a
Creatinine, mg/dL	Mean ± SD		1.14 ± 0.594	1.10 ± 0.724	0.702a	1.11 ± 0.372	1.06 ± 0.586	0.579a
Pre-operative HbA1c (%),	Mean ± SD		8.54 ± 1.26	8.06 ± 1.07	0.003a	5.79 ± 0.29	5.68 ± 0.35	0.049a
Pre-operative blood glucose, mg/dL	Mean ± SD		188.3 ± 29.2	174.6 ± 29.8	0.002a	102.3 ± 13.3	101.7 ± 12.3	0.785a
Medication use
Vasopressin			2(4.1%)	8(1.4%)	0.155b	0(0%)	8(1.2%)	0.472b
Pethidine			12(24.5%)	207(36.4%)	0.093b	15(34.9%)	202(30.1%)	0.505b
Morphine			14(28.6%)	238(41.9%)	0.069b	24(55.8%)	295(43.9%)	0.128b
Procedural parameters
Cardiopulmonary bypass time, min	Mean ± SD		105.10 ± 38.316	119.53 ± 44.574	0.028a	109.14 ± 42.181	117.05 ± 44.246	0.255a
Aortic cross clamp time, min	Mean ± SD		73.21 ± 34.494	88.83 ± 39.205	0.007a	77.38 ± 35.596	80.70 ± 36.632	0.565a
Left ventricle ejection fraction, %	Mean ± SD		52.32 ± 4.832	51.20 ± 6.921	0.268a	54.89 ± 6.079	53.23 ± 6.413	0.099a
Post-operative complication
Post-operative Ventilation, hours	Median(Q1-Q3)		3.58 (2.05–5.11)	5.82 (1.96–9.68)	0.007C	4.88 (1.91–7.85)	5.70 (1.44–9.97)	0.400C

^aIs tested by in-dependent t-test. b. Is tested by Chi-square test. C. It is tested by Wilcoxon rank-sum (Mann–Whitney) test

Abbreviations: DM Diabetes Mellitus, CABG Coronary Artery Bypass Grafting, LVEF Left Ventricular Ejection Fraction, SD Standard Deviation

Table 3 and Fig. 3 summarize predictors of postoperative hallucinations among diabetic patients. For visual hallucinations, most baseline factors (age, sex, BMI, surgical type, MI) were not significant. Both elevated HbA1c (adjusted HR 1.29, p = 0.014) and preoperative glucose (adjusted HR 1.011, p = 0.014) were independently associated with a higher risk. For auditory hallucinations, morphine use (adjusted HR 0.57, p = 0.103) and longer postoperative mechanical ventilation (adjusted HR 0.80, p < 0.001) were inversely associated with the outcome in the adjusted model. Conversely, elevated HbA1c (adjusted HR 1.37, p = 0.014) and preoperative glucose (adjusted HR 1.013, p = 0.011) predicted a higher risk.

Table 3.

Multivariable cox regression analysis of predictors associated with postoperative visual and auditory hallucinations among diabetic patients undergoing cardiac surgery

	Cohort A: DM
	Visual				Auditory
Predictors	Crude HR(%95CI)	Crude P VALUE	adjusted HR(%95CI)	adjusted P VALUE	Crude HR(%95CI)	Crude P VALUE	adjusted HR(%95CI)	adjusted P VALUE
Age, years	0.995 (0.967–1.024)	0.730	0.997 (0.970–1.030)	0.826	0.956 (0.925–0.989)	0.009	0.970 (0.940–1.010)	0.118
BMI, kg/m2	0.962 (0.896–1.032)	0.278	0.960 (0.890–1.030)	0.224	1.031 (0.996–1.068)	0.086	1.040 (1.000–1.090)	0.063
Gender	1.009 (0.592–1.719)	0.975	0.940 (0.530–1.650)	0.819	0.869 (0.444–1.700)	0.682	1.010 (0.490–2.120)	0.971
Recent myocardial infarction	0.838 (0.460–1.528)	0.564	0.960 (0.530–1.740)	0.890	0.738 (0.346–1.573)	0.431	0.710 (0.330–1.530)	0.383
Type of Surgery	0.945 (0.485–1.844)	0.869	1.160 (0.570–2.370)	0.688	0.662 (0.263–1.670)	0.383	0.810 (0.300–2.190)	0.683
Albumin, g/dl	0.812 (0.546–1.206)	0.301	0.830 (0.540–1.290)	0.412	1.403 (0.853–2.307)	0.183	1.390 (0.810–2.360)	0.228
Cardiopulmonary bypass time, min	0.995 (0.990–1.001)	0.105	0.990 (0.990–1.000)	0.051	0.993 (0.986–0.999)	0.028	1.000 (0.990–1.010)	0.981
Aortic cross clamp time, min	1.000 (0.994–1.006)	0.912	1.010 (1.000–1.010)	0.066	0.989 (0.980–0.997)	0.007	0.990 (0.980–1.010)	0.315
Left ventricle ejection fraction, %	1.033 (0.994–1.073)	0.099	1.020 (0.980–1.060)	0.243	1.025 (0.979–1.072)	0.290	1.020 (0.980–1.070)	0.342
Previous use of antibiotics	1.296 (0.522–3.215)	0.576	1.280 (0.510–3.240)	0.600	2.341 (0.996–5.500)	0.051	1.730 (0.710–4.190)	0.228
Pethidine	1.165 (0.725–1.871)	0.527	1.010 (0.600–1.700)	0.966	0.581 (0.303–1.115)	0.103	0.580 (0.290–1.160)	0.122
Morphine	0.802 (0.496–1.297)	0.368	0.790 (0.460–1.330)	0.372	0.571 (0.307–1.061)	0.076	0.570 (0.290–1.120)	0.103
Post-operative Ventilation, hours	0.950 (0.878–1.028)	0.204	0.960 (0.890–1.040)	0.305	0.755 (0.668–0.854)	0.000	0.800 (0.710–0.910)	< 0.001
Pre-operative HbA1c (%),	1.320 (1.070–1.620)	0.009	1.290 (1.050–1.590)	0.014	1.450 (1.130–1.870)	0.004	1.370 (1.070–1.760)	0.014
Pre-operative blood glucose, mg/dL	1.009 (1.003–1.015)	0.004	1.011 (1.002–1.020)	0.014	1.008 (1.001–1.016)	0.031	1.013 (1.003–1.024)	0.011
Insulin dependence	1.238 (0.745–2.059)	0.411	1.354 (0.776–2.362)	0.290	1.611 (0.845–3.071)	0.149	1.428 (0.732–2.786)	0.297
Oral diabetic drugs (Yes)	1.102 (0.642–1.892)	0.717	1.084 (0.601–1.956)	0.799	1.231 (0.672–2.254)	0.503	1.196 (0.618–2.315)	0.601
Duration of DM (yrs)	1.027 (0.985–1.072)	0.215	1.018 (0.973–1.065)	0.438	1.031 (0.983–1.082)	0.201	1.023 (0.964–1.085)	0.456

HR Hazard Ratio, CI Confidence Interval, DM Diabetes Mellitus, BMI Body Mass Index, CABG Coronary Artery Bypass Grafting, HbA1c Glycated Hemoglobin, LVEF Left Ventricular Ejection Fraction, CPB Cardiopulmonary Bypass

Fig. 3.

Adjusted Hazard Ratios for Postoperative Hallucinations in Diabetic Patients

Table 4 and Fig. 4 summarize the crude and multivariable Cox regression models for predictors of visual and auditory hallucinations among non-diabetic patients. In the adjusted analysis, most baseline and operative factors, including age, BMI, sex, type of surgery, creatinine level, cardiopulmonary bypass time, aortic cross-clamp time, left ventricular ejection fraction, and preoperative HbA1c, were not independently associated with visual hallucinations. Hemodialysis dependence was the only significant independent predictor of visual hallucinations, showing a markedly increased risk (adjusted HR 6.99, 95% CI 1.59–30.61; p = 0.010). For auditory hallucinations, critical preoperative state independently increased the risk (adjusted HR 2.70, 95% CI 1.23–5.93; p = 0.013), while higher EuroSCORE II was also significantly associated with greater risk (adjusted HR 1.10, 95% CI 1.05–1.15; p < 0.001).

Table 4.

Multivariate and univariate cox regression analysis of predictors of visual and auditory hallucinations in non-diabetic cardiac surgery patients

	Cohort B: Without DM
	Visual				Auditory
Predictors	Crude HR(%95CI)	Crude P VALUE	adjusted HR(%95CI)	adjusted P VALUE	Crude HR(%95CI)	Crude P VALUE	adjusted HR(%95CI)	adjusted P VALUE
Age, years	1.008 (0.990–1.027)	0.395	1.010 (0.990–1.030)	0.163	0.994 (0.971–1.017)	0.599	0.990 (0.970–1.020)	0.663
BMI, kg/m2	0.966 (0.909–1.028)	0.266	0.970 (0.910–1.030)	0.271	0.933 (0.856–1.017)	0.114	0.930 (0.850–1.030)	0.172
Gender	1.334 (0.833–2.134)	0.230	1.420 (0.850–2.370)	0.185	1.039 (0.524–2.062)	0.912	0.950 (0.420–2.110)	0.894
Type of Surgery	1.041 (0.662–1.637)	0.862	0.910 (0.530–1.570)	0.746	0.836 (0.436–1.602)	0.589	0.610 (0.270–1.380)	0.233
Hemodialysis-dependent	5.519 (2.020–15.081)	0.001	6.990 (1.590–30.610)	0.010	2.235 (0.308–16.243)	0.427	0.070 (0.000–2.960)	0.162
Critical preoperative state	0.606 (0.264–1.392)	0.238	0.640 (0.270–1.490)	0.301	2.099 (1.007–4.376)	0.048	2.700 (1.230–5.930)	0.013
Immunosuppressive therapy	3.432 (1.256–9.377)	0.016	2.070 (0.530–8.040)	0.294	3.112 (0.753–12.867)	0.117	4.150 (0.580–29.570)	0.155
Euroscore II	1.056 (1.024–1.089)	< 0.001	1.030 (0.990–1.070)	0.163	1.085 (1.051–1.120)	< 0.001	1.100 (1.050–1.150)	< 0.001
Creatinine, mg/dL	1.097 (0.813–1.480)	0.546	0.810 (0.550–1.200)	0.294	1.109 (0.757–1.627)	0.594	1.300 (0.790–2.160)	0.305
Vasopressin	1.143 (0.159–8.213)	0.894	0.730 (0.090–5.750)	0.763	N/A	N/A	N/A	N/A
Cardiopulmonary bypass time, min	0.998 (0.993–1.003)	0.459	1.000 (0.990–1.000)	0.327	0.996 (0.989–1.003)	0.262	0.990 (0.980–1.000)	0.072
Aortic cross clamp time, min	0.999 (0.993–1.005)	0.780	1.000 (0.990–1.010)	0.648	0.998 (0.989–1.006)	0.576	1.010 (1.000–1.020)	0.178
Left ventricle ejection fraction, %	0.995 (0.962–1.029)	0.769	0.990 (0.950–1.020)	0.385	1.042 (0.993–1.093)	0.094	1.020 (0.970–1.070)	0.403
Pre-operative HbA1c (%)	1.040 (0.560–1.930)	0.906	0.920 (0.490–1.730)	0.793	2.350 (1.000–5.560)	0.050	2.140 (0.860–5.310)	0.101

HR Hazard ratio, CI Confidence interval, BMI Body mass index, DM Diabetes mellitus, CABG Coronary artery bypass grafting, LVEF Left ventricular ejection fraction

Fig. 4.

Adjusted hazard ratios for visual and auditory hallucinations in non-diabetic cardiac surgery patients

Discussion

In this multicenter prospective cohort study, we examined predictors of early postoperative hallucinations following cardiac surgery, with particular attention to glycemic variables. Because postoperative delirium was not formally adjudicated with a dedicated delirium instrument, these hallucinations should not be interpreted as independent from delirium. Rather, they should be viewed as sensory-specific postoperative neuropsychiatric symptoms that may overlap with delirium-spectrum states. The clinical relevance of these postoperative hallucinations lies in the fact that, even without formal delirium adjudication, they should not be regarded as benign postoperative experiences. Because formal postoperative delirium assessment was not performed, we cannot determine from the present study whether hallucinations represented an early warning sign of subsequent delirium, occurred as part of an overlapping delirium-spectrum state, or reflected a more isolated postoperative neuropsychiatric symptom. Nevertheless, in the context of recent cardiac surgery, new-onset visual or auditory hallucinations are clinically meaningful findings. At minimum, they may serve as warning symptoms of acute neuropsychiatric vulnerability and should prompt careful bedside evaluation, review of reversible contributors, and closer postoperative monitoring. As described in DSM-5, hallucinations are common in hyperactive delirium and reflect acute disruptions in cognition, attention, and perception (American 2013). The biological link between glycemic dysregulation and these neuropsychiatric events is well established, as diabetes contributes to neuroinflammation, microvascular injury, oxidative stress, and impaired cerebral glucose metabolism, all of which reduce cognitive reserve (Wang et al. 2019). Supporting this, Kowalska et al. reported a significantly higher prevalence of diabetes among delirium patients (32% vs. 16%) (Kotfis et al. 2019), while Komici et al. found 52% increased odds of delirium in diabetic individuals (Komici et al. 2025).

Despite these established biological links between glycemic dysregulation and neuropsychiatric vulnerability, our time-to-event analyses did not demonstrate a significant difference in the cumulative incidence of either visual or auditory hallucinations between diabetic and non-diabetic patients. Kaplan–Meier curves showed parallel trajectories across both groups, with no statistically significant separation over the 7-day postoperative period. This finding suggests that, while diabetes may contribute to underlying susceptibility through chronic metabolic and microvascular mechanisms, it does not appear to independently drive the short-term temporal occurrence of postoperative hallucinations following cardiac surgery.

One possible explanation is that acute perioperative factors, such as hemodynamic fluctuations, inflammatory responses, anesthetic exposure, and postoperative medications, may exert a more dominant influence on early neuropsychiatric manifestations than baseline diabetic status alone. Accordingly, the absence of a significant difference between groups highlights the multifactorial nature of postoperative hallucinations and supports the need to consider both chronic vulnerability and acute perioperative stressors when evaluating these events.

Kotfis et al. found that elevated HbA1c and acute hyperglycemia increased postoperative delirium risk, likely via oxidative stress and neuroinflammation mechanisms (Kotfis et al. 2019). Similar trends were seen even in TAVR patients, where preprocedural glucose predicted many undesired outcomes, reinforcing the concept of perioperative glycemic vulnerability (Abu Khadija et al. 2025). However, other studies have suggested that acute glucose fluctuations, rather than diabetes status alone, may be more closely associated with postoperative delirium. Our findings confirm that elevated pre-op HbA1c independently predicted visual and auditory hallucinations in diabetic patients, supporting glycemic dysregulation as a shared risk factor with delirium (Kotfis et al. 2019). This may be due to microvascular brain injury, blood–brain barrier disruption, and neuroinflammation impairing perception circuits (Li et al. 2022) (Li et al. 2023).

Although reduced LVEF has been linked in previous studies to cerebral hypoperfusion and postoperative neuropsychiatric vulnerability (Cai et al. 2020), LVEF was not independently associated with visual hallucinations in our diabetic cohort after Cox regression analysis. Therefore, while reduced cardiac reserve remains a biologically plausible contributor, our findings do not support LVEF as a significant predictor of postoperative visual hallucinations in this cohort (Sloten et al. 2020) (Sonneville 2025) (Hu et al. 2024) (Costa et al. 2014) Postoperative morphine use showed a non-significant inverse trend with auditory hallucinations in diabetic patients. However, this finding should be interpreted cautiously in view of the observational design. Morphine administration was not randomized and may have been influenced by the patient’s postoperative clinical status and suitability for routine analgesia. Accordingly, the observed association may reflect confounding by indication or residual confounding by clinical stability rather than a direct protective effect of morphine itself. Among diabetic patients, longer postoperative mechanical ventilation was inversely associated with auditory hallucinations in the adjusted model; however, this finding should be interpreted cautiously because hallucination assessments began only after extubation. Patients requiring prolonged ventilation therefore entered the observation window later, and early hallucinations during the intubated period may not have been captured. Accordingly, this association should not be interpreted as evidence of a true protective effect of prolonged ventilation.

Among non-diabetic patients, higher EuroSCORE II was associated with auditory hallucinations post-op, consistent with prior studies showing its link to delirium(Krzych et al. 2013). It reflects systemic vulnerability through composite surgical and clinical risk. In the present study, however, HbA1c remained independently significant only among diabetic patients. In Kotfis et al.’s study, patients who developed POD had significantly higher mean HbA1c values compared to those who did not (Kotfis et al. 2019). Furthermore, multivariate analysis, as confirmed by Lechowicz et al., revealed that HbA1c is an independent predictor (Lechowicz et al. 2021). These findings suggest that subclinical glucose dysregulation may contribute to vulnerability to delirium. However, as stated in Nwokedi et al.’s article, there was no clear link between elevated HbA1c (> 8% or > 10%) and postoperative complications(Nwokedi et al. 2022) Hemodialysis dependence independently predicted visual hallucinations among non-diabetic patients, likely due to uremic toxin accumulation and disrupted cerebral homeostasis, as supported by Mahmoudi et al.(Mahmoudi et al. 2025). Finally, critical preoperative state independently predicted auditory hallucinations among non-diabetic patients, whereas immunosuppressive therapy was associated only in bivariate analysis and lost significance after adjustment. Notably, dexamethasone has shown both protective and deleterious neuropsychiatric effects depending on dose and context(Mahmoudi et al. 2025).

Differential risk by diabetes status

Although crude incidence was similar between diabetic and non-diabetic patients, the drivers of risk differed by diabetes status. Among diabetic patients, both chronic dysglycemia, reflected by higher HbA1c, and acute dysglycemia, reflected by higher preoperative blood glucose, independently predicted visual and auditory hallucinations, consistent with literature linking poor glycemic control to postoperative delirium-related neuropsychiatric vulnerability(Koster et al. 2013)(Ali Hassan et al. 2025). In non-diabetics, visual hallucinations were mainly associated with hemodialysis dependence, whereas auditory hallucinations were associated with critical preoperative state and higher EuroSCORE II. These findings suggest that risk in non-diabetic patients may be driven more by renal replacement dependence, global illness severity, and physiologic stress than by glycemic markers alone(Bramley et al. 2021). The separation of visual and auditory hallucinations may also offer insight into potentially different perioperative vulnerability pathways. Visual hallucinations may be more closely linked to hemodynamic and perfusion-related factors, particularly in the setting of reduced cardiac reserve, impaired cerebral autoregulation, and microvascular disease. Although reduced LVEF may biologically contribute to cerebral hypoperfusion and postoperative neuropsychiatric vulnerability, the present Cox regression analysis did not identify LVEF as a significant predictor of visual hallucinations in diabetic patients. Therefore, this mechanism remains plausible but was not directly supported by our adjusted findings. In contrast, auditory hallucinations may be more closely related to metabolic instability, especially in patients with chronic and acute dysglycemia. The association of auditory hallucinations with higher HbA1c and preoperative glucose may therefore suggest that glycemic dysregulation contributes to postoperative sensory and neuroinflammatory vulnerability. These interpretations are hypothesis-generating, but they support the value of examining hallucination subtypes separately rather than combining them into a single postoperative neuropsychiatric outcome. This distinction may also be clinically relevant because different hallucination phenotypes could reflect different perioperative vulnerabilities and may warrant different monitoring priorities in future studies.

Clinical implications

Preoperative HbA1c and blood glucose were consistently associated with postoperative hallucination risk, particularly among diabetic patients. However, these findings should be interpreted as observational associations rather than evidence that short-term perioperative correction of glycemic indices would directly prevent hallucinations. Instead, they suggest that chronic and acute dysglycemia may identify patients with greater postoperative neuropsychiatric vulnerability, a hypothesis that should be examined in future interventional studies.

Strengths and limitations

This multicenter prospective study enhances generalizability across participating centers within the West Bank. Hallucinations were evaluated using the validated QPE tool through daily bedside assessments, ensuring high adherence and reducing recall bias. Rigorous multivariable Cox models adjusted for clinical, procedural, and glycemic confounders. HbA1c emerged as an independent predictor of hallucinations among diabetic patients, while risk among non-diabetic patients appeared more closely related to illness severity, EuroSCORE II, critical preoperative state, and hemodialysis dependence. Stratified analyses therefore allowed more tailored risk profiling according to diabetes status. However, being limited to West Bank hospitals may affect global generalizability. As an observational study, causality cannot be inferred, and residual confounding remains possible. An additional limitation is that postoperative delirium was not concurrently assessed using a standard diagnostic instrument. Therefore, we could not determine whether the observed hallucinations occurred as isolated symptoms or as part of delirium, and the findings should be interpreted accordingly. An additional limitation relates to the timing of hallucination assessment. Because QPE evaluations began only after extubation, patients with prolonged postoperative mechanical ventilation became eligible for assessment later than others, and hallucinations occurring during the intubated period were not measured. This may have led to underascertainment of early hallucinations in patients with longer ventilation and may have biased the observed association between ventilation duration and hallucination risk. Associations involving postoperative medication exposure, including morphine, should be interpreted cautiously because treatment administration was not randomized and may have been influenced by the patient’s evolving postoperative clinical status and suitability for routine analgesia. Therefore, confounding by indication and residual confounding cannot be fully excluded. In addition, postoperative sedation duration was not recorded as a standardized multicenter variable, which limited our ability to model sedation-related competing risks formally. Although adherence was monitored and incomplete QPEs excluded, minor recall bias may persist, particularly for short-lived or partially remembered events. The 7-day follow-up window may have missed delayed or persistent hallucinations, and reliance on patient communication could have led to conservative incidence estimates.

Conclusion

Postoperative visual and auditory hallucinations affected approximately 10% of patients within one week of cardiac surgery. Higher preoperative HbA1c and blood glucose were associated with increased hallucination risk, particularly in diabetic patients. These findings support dysglycemia as a clinically relevant risk marker, but interventional studies are needed before optimized glycemic control can be recommended as a direct preventive strategy for postoperative hallucinations.

Abbreviations

ADA

American Diabetes Association

BMI

Body Mass Index

CABG

Coronary Artery Bypass Grafting

CI

Confidence Interval

CKD

Chronic Kidney Disease

COPD

Chronic Obstructive Pulmonary Disease

CPB

Cardiopulmonary Bypass

CRP

C-reactive Protein

DM

Diabetes Mellitus

DSM-5

Diagnostic and Statistical Manual of Mental Disorders, 5th Edition

EuroSCORE II

European System for Cardiac Operative Risk Evaluation II

HbA1c

Glycated Hemoglobin

HR

Hazard Ratio

ICU

Intensive Care Unit

IRB

Institutional Review Board

KM

Kaplan–Meier

LVEF

Left Ventricular Ejection Fraction

MI

Myocardial Infarction

NYHA

New York Heart Association

OGTT

Oral Glucose Tolerance Test

OR

Odds Ratio

POCD

Postoperative Cognitive Dysfunction

POD

Postoperative Delirium

QPE

Questionnaire for Psychotic Experiences

REM

Rapid Eye Movement (sleep)

SD

Standard Deviation

STATA

Data Analysis and Statistical Software

TAVR

Transcatheter Aortic Valve Replacement

TRIPOD

Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis

US

United States

VAACS

Visual and Auditory Assessment after Cardiac Surgery Study

VIF

Variance Inflation Factor

WHO

World Health Organization

Authors’ contributions

H.A.K. and M.A. conceived and designed the study. N.A.H., A.D., D.N., A.Da., and W.O. contributed to data acquisition and coordination at participating centers.Y.Z.F., M.M., T.O., N.J., A.Sa., A.Z., O.A.K., S.M., S.A., O.Q.H., B.B., O.E., H.A.A., O.N., and Q.H. contributed to patient recruitment, data verification, and manuscript review. M.A.K. and H.A.K. performed statistical analyses. M.A. and H.A.K. drafted the manuscript. All authors critically revised the paper for intellectual content and approved the final version. All authors agree to be accountable for all aspects of the work.

Funding

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Data availability

Data supporting the findings of this study are available from the corresponding author upon reasonable\u0000request.

Declarations

Ethics approval and consent to participate

This study received ethical approval from the Institutional Review Board (IRB) of An-Najah National University (Approval No.: IRB-12-09-2022, Date: 12 September 2022). Written informed consent was obtained from all participants after receiving a clear explanation of the study objectives, potential benefits, and associated risks. Participation was voluntary, and participants were informed of their right to withdraw from the study at any time without repercussions. All collected data were handled confidentially and securely stored. Formal administrative permission to conduct the study was also obtained from all collaborating academic institutions.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.