KEY POINTS

Question: Is there an association between preoperative diabetes, intraoperative electroencephalogram alpha power, and postoperative delirium? Findings: We found that preoperative diabetes was associated with the incidence of postoperative delirium in patients, and intraoperative electroencephalogram alpha power partially mediated the association between preoperative diabetes and postoperative delirium. Meaning: Preoperative diabetes and low intraoperative electroencephalogram alpha power are associated with a higher chance of patients having postoperative delirium.

Postoperative delirium (POD) is a common form of postoperative brain dysfunction characterized by sudden onset and fluctuating course, with symptoms including disturbances in attention, cognition, and awareness.1 The reported incidence of POD among elderly surgical patients ranges from 5% to 35%,1–6 and this condition is associated with a higher risk of poor functional recovery, institutionalization, dementia, and death.7,8 Therefore, identifying patients at a higher risk of POD is critical for preventing and treating this condition and ultimately improving postoperative outcomes.

Diabetes, a commonly occurring metabolic disorder, is often encountered perioperatively and may increase the risk of adverse postoperative outcomes.9,10 As a result, preoperative diabetes has been suggested as a risk factor for POD in systematic reviews and meta-analyses.11,12 However, conflicting findings exist in the literature,13,14 and previous studies may be subject to various methodological limitations such as poor definition of diabetes, heterogeneous study designs, and inadequate control for confounding factors. Therefore, the relationship between preoperative diabetes and POD remains largely unclear, and further investigation is needed to illustrate the potential association between preoperative diabetes and POD. In particular, a prospective study that reduces the methodological limitations would be beneficial in shedding more light on this important topic.

Prefrontal electroencephalogram (EEG) alpha oscillations (8–12 Hz) are widely recognized as a hallmark of unconsciousness induced by anesthetics such as propofol or ether-derived agents,15,16 and are believed to be a potential biomarker for brain vulnerability.17 In recent years, there has been an increasing number of studies investigating the relationship between anesthesia-induced frontal EEG alpha power and POD,5,6,18 with low alpha power emerging as a potential marker for this condition.6 This finding has been replicated in multiple studies.5,18 Interestingly, diabetes has also been identified as a potential predictor of lower intraoperative alpha power, with a stronger association observed for insulin-dependent diabetes.19 Given the previous findings that preoperative diabetes is associated with intraoperative alpha power, and intraoperative alpha power is associated with POD, we sought to investigate whether there is a further relationship whereby alpha oscillation under anesthesia/surgery might contribute to the association between preoperative diabetes and POD.

Therefore, this study aimed to investigate whether (1) preoperative diabetes was associated with POD after elective orthopedic surgery and (2) intraoperative frontal alpha power is a mediator of the association between preoperative diabetes and POD. We hypothesized that patients with preoperative diabetes would have an increased odds of developing POD, and that this association could be mediated by low intraoperative alpha power.

METHODS Study Design

This prospective matched cohort study was conducted at the First Affiliated Hospital of Anhui Medical University, Hefei, China from March 2022 to December 2022, approved by the Clinical Medical Research Ethics Committee of the First Affiliated Hospital of Anhui Medical University (PJ2022-02-40) and registered in the Chinese Clinical Trial Registry (ChiCTR22000577277, principal investigator: X. Liu, date of registration, March 16, 2022) before patient enrollment. Written informed consent was obtained from all participants. Study reporting followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline for observational studies. This study was designed to investigate the relationship among preoperative diabetes, intraoperative alpha power, and POD. This was the primary analysis of these data.

Participants

Patients were assessed for eligibility on the day before surgery. Inclusion criteria were adults aged 60 years or more, American Society of Anesthesiologists (ASA) physical status II to III. Exclusion criteria were severe diabetes (diabetes with renal, ophthalmic, neurological, or peripheral circulatory complications), with an attempt to minimize the impact of additional potentially contributory comorbidities on the outcomes; known neuropsychiatric disorders (Parkinson’s disease, depression, or schizophrenia); preoperative cognitive dysfunction (identified using education-specific cutoff points of total Mini-Mental State Examination (MMSE) scores: ≤17 for illiteracy, ≤20 for less than primary education, and ≤23 for less than postsecondary education)20; speech, vision, or hearing impairment; and expected duration of surgery <30 minutes (concerns regarding that maintenance phase of anesthesia being too short to acquire stable EEG waveform).

The primary exposure in this study was diabetes, which was defined as either a known history of diabetes or newly diagnosed diabetes according to the criteria set by the American Diabetes Association.21 The known diabetes referred to a previously diagnosed condition confirmed by a health care professional or ongoing treatment with antidiabetic medication. Newly diagnosed diabetes referred to cases diagnosed before surgery based on fasting plasma glucose ≥7.0 mmol/L, or hemoglobin A1C (HbA1c) ≥6.5%. Diabetic patients who met the inclusion criteria were included in the “diabetes group.” Notably, the prevalence of diabetes varies across age and sex among adults living in mainland China,9 we used a matching approach to minimize potential confounding factors. Specifically, we prospectively enrolled one patient with diabetes first, and then we matched a potential nondiabetic patient from patients scheduled for surgery later. The potential control was excluded if met any exclusion criteria until the 1:1 match was successful. The successful matched control was recruited to constitute the “nondiabetes group. Matching was performed by sex, age (±5 years) and type of surgery (knee arthroplasty, hip arthroplasty, lumbar spine surgery, and extremity fracture surgery).

Intraoperative Management

All patients underwent standardized anesthetic management, including the same anesthesia induction and maintenance techniques. Anesthesia was induced with propofol (1–2 mg/kg) and sufentanil (0.2–0.5 µg/kg). Muscle relaxation was achieved by using cisatracurium (0.3 mg/kg). Maintenance of anesthesia was accomplished using propofol (50–100 µg·kg–1·min–1) and remifentanil (0.1–1.0 µg·kg–1·min–1). To minimize potential sources of bias, premedication was not permitted (such as penehyclidine hydrochloride, midazolam, or dexmedetomidine), as it may influence EEG spectrum parameters,22 or contribute to delirium.1,12 In addition, flurbiprofen axetil (50 mg) were administered for postoperative analgesia; dexamethasone was administered to prevent postoperative nausea and vomiting. Intraoperative blood pressure was targeted to remain within ±20% of the baseline value which was defined as the average of the initial 3 blood pressure measurements obtained on admission to the operating room.

Frontal Electroencephalogram Alpha Power

Frontal EEG was recorded with a sampling rate of 178 Hz using the SEDline monitor (Masimo Corporation), with an electrode sensor attached to the patient’s forehead. After cleansing the skin with alcohol, 4 electrodes (bilateral frontal and prefrontal) were positioned according to international 10/20 system at Fp1, Fp2, F7, F8, with ground electrode at Fpz, and reference electrode around Fz. EEG recordings began before anesthesia induction and continued until discharge from the operating room. The EEG amplitude and feed displayed on the monitor screen were set to 10 mV/mm and 30 mm/sec, respectively, and maintained constant throughout the study.

MATLAB R2021a (MathWorks) was used for EEG processing and analysis. We focused on the absolute alpha power during the maintenance phase of general anesthesia. For each subject, we carefully selected a 2-minute epoch around 60 minutes after induction of anesthesia where the infusion rate of anesthetics was stable for at least 10 minutes. We visually inspected the EEG waveforms to ensure that the analysis windows were free of noise and artifacts and did not contain burst suppression. The EEG signals were bandpass filtered at 1 to 40 Hz, and we equally weighted the signals from FP1, FP2, F7, and F8 to obtain the overall frontal power spectra. The power spectra were estimated using the multitaper method implemented in Chronux toolbox with the following parameters: window length T = 2 seconds with no overlap, time-bandwidth product TW = 3, and number of tapers K = 5. Group-level spectrograms were computed by taking the median across all patients. We calculated the absolute alpha power by integrating the multitaper spectrum from 8 to 12 Hz using the trapezoidal integration rule. Additionally, using the same methods, we calculated the baseline frontal alpha power for each patient from a 2-mintue EEG epoch before induction of anesthesia, during which participants were asked to keep eyes closed but stay awake. We also collected the infusion rates of propofol and remifentanil during the selected EEG epochs from the electronic anesthesia records.

Assessment of Delirium

The primary outcome was the occurrence of POD. Each patient was evaluated at baseline and on postoperative days 1 to 7 or until discharge, whichever came first. Delirium was measured once per day (18:00–20:00) using the 3-minute Diagnostic Confusion Assessment Method (3D-CAM).23 It is a validated screening tool that is completed within 3 minutes, with high sensitivity (93%), specificity (96%), and interrater agreement (95%) for delirium.23 Researchers who performed the delirium assessment were trained as per the guidelines for the 3D-CAM and were blinded to exposure status. POD was defined as any positive result on 3D-CAM, independently of the number of positive assessments. Secondary outcome was the severity of POD. The severity of delirium was assessed for all participants using the short form of the CAM-Severity (CAM-S), with scores ranging from 0 (no delirium features) to 7 (most severe).24 The maximum CAM-S score for each patient’s hospitalization was analyzed. The duration of delirium was defined as the total number of days with a positive result.

Other Postoperative Outcome and Complications

Postoperative pain and postoperative nausea and vomiting were measured during postoperative days 1 to 7 follow-up or until discharge, concurrently with assessment of POD. Adverse clinical outcomes during hospitalization, including acute heart failure, acute myocardial infarction, pneumonia, stroke, and renal failure, were collected from patients’ electronic medical records.

Statistical Analysis

Normality was assessed using histograms. Demographic characteristics and perioperative data were described by mean ± standard deviation or median [quartiles] for continuous variables, number along with percentages for categorical variables. Patient characteristics between the groups were assessed with standardized differences, defined as the absolute difference in means or proportions divided by the pooled standard deviation. The comparison between diabetic patients and controls was performed using independent sample t-test, Mann-Whitney U test, χ2 test, or Yates’ continuity corrected χ2 test as appropriate. We used logistic regression model to examine the association between preoperative diabetes and POD, and proportional odds logistic regression model to examine the association between preoperative diabetes and delirium severity, treating delirium severity as an ordinal variable (0 for no delirium, 1 for mild, 2 for moderate, and 3 to 7 for severe delirium).24 We adjusted for potential confounders including age, sex, body mass index (BMI), education level, hypertension, arrhythmia, coronary heart disease, and history of stroke. Confounders for adjustment were preexposure variables that were deemed clinically relevant based on previous studies.9,11,14,25

Mediation analysis was performed to evaluate whether and to what extend the intraoperative alpha power mediates the association between preoperative diabetes and POD based on the counterfactual framework. The assumption of associations between the variables is illustrated in Figure 1. We hypothesized that the association between preoperative diabetes and POD would be partially mediated by intraoperative alpha power. Conceptually, the direct effect is the difference of the potential delirium incidence when changing the individual’s exposure value from diabetes to nondiabetes, while holding the alpha power constant at the value that would be observed under a certain exposure status. The indirect effect is defined as the difference of the potential delirium incidence when an individual’s exposure status remained fixed, while the alpha power was modified as if individual’s exposure status had changed. The total effect is the entire effect of diabetes on POD and equals the sum of the direct effect and the indirect effect. Specifically, the estimation of these effects is based on 2 regression models. We first fitted the mediator model using general linear regression, in which the alpha power was modeled as a function of diabetes. Next, we fitted the outcome model using logistic regression, in which POD was modeled as a function of diabetes and alpha power. The total, indirect, and direct effects and their 95% confidence intervals (CIs) were derived from the nonparametric bootstrap resamples (n = 2000) using the “mediation” package.26 Specifically, we first generated a bootstrapped dataset with the same number of patients by resampling with replacement from the original dataset, which consisted of a total of 266 patients. We did not use the matching information in the bootstrapping as the different number of people ultimately included in the analysis in the diabetic and control groups. Then, for each of the bootstrapped samples, the steps were repeated including fitting the mediator model and outcome model, simulating potential mediator and outcome values based on the regression model estimates from the original data, and computing average effects of interest based on potential outcomes. After the procedure was performed 2000 times, we then obtained point estimates and the corresponding 95% CI from the 2000 versions of the effect estimates. This was a nonparametric method as the overall distribution and parameters of the bootstrapped samples were not involved in the analysis. The procedure was repeated after adjustment for the aforementioned potential confounders including age, sex, BMI, education level, hypertension, arrhythmia, coronary heart disease, and history of stroke. It is possible that the indirect effect takes different values depending on the exposure status (diabetes or nondiabetes). Thus, as suggested by Tingley et al,26 we also tested the interaction between diabetes and intraoperative alpha power.

Figure 1.:

The assumption of relationship among preexisting diabetes, intraoperative electroencephalogram alpha power, and postoperative delirium. The total effect of diabetes on postoperative delirium is decomposed into a direct effect that summarizes all associations not going through intraoperative alpha power, and an indirect effect that acts through intraoperative alpha power. Confounders refer to several factors that could be involved in the association between preoperative diabetes, intraoperative alpha power and postoperative delirium.

Three post hoc sensitivity analyses were conducted to assess the robustness of the findings regarding the association of diabetes with POD and the mediating role of alpha power. First, we respecified the prognostic model, controlling for additional intraoperative variables that have previously been associated with POD development, namely duration of anesthesia and surgery type.27 Second, to further confirm the mediating role of alpha oscillations, we repeated the analysis by replacing alpha power with alpha peak power which defined as the highest power point within 8 to 12 Hz. Third, to determine whether the decision to use alpha power obtained from averaged prefrontal oscillations as mediator rather than a single electrode impacted the findings, a sensitivity analysis was conducted with alpha power at FP1 electrode as mediator.

With an exploratory intent, we separately evaluated the associations between preoperative diabetes and intraoperative delta (1–4 Hz), theta (4–8 Hz), and beta (12–25 Hz) band power, and associations of these power with POD.

Model goodness of fit was examined using the Hosmer-Lemeshow test. The variance inflation factor was used to test collinearity for all variables in the model for intraoperative alpha power. Given that the data for primary outcomes and covariates of interest had no missing values, complete case analysis was used. A 2-tailed P < .05 was considered statistically significant. All analyses were performed using R Statistical Software version 4.0.1 (Foundation for Statistical Computing).

Sample size calculation was conducted using PASS software 15.0, based on a comparison of the incidence of POD from 2 studies.11,28 In a previous study which investigated the risk factors for delirium after total joint arthroplasty in elderly Chinese patients, the incidence of POD in nondiabetic patients was observed to be 18%, and the odds ratio (OR) for diabetes versus nondiabetes was 2.47.28 From the second study, a meta-analysis exploring the risk factors for delirium after vascular surgery, the estimated OR for diabetes was 2.15 in the setting of high-risk surgery.11 Collectively, we assumed an 18% incidence of POD in nondiabetic patients and an OR of 2.3 for increased odds of POD with diabetes, calculated as the average of the 2 aforementioned ORs. Allowing for 10% missing data due to malfunction of electrode arrays or lose of follow-up, the total required sample size to provide 80% power to detect an OR of 2.3 or greater in POD incidence at the 0.05 significance level was N = 270.

Bias

Data collection using validated assessment tools and prior training of investigators helped to minimized information bias. Selection bias was managed by strictly screening subjects against inclusion and exclusion criteria. Confounding bias was addressed by utilizing a matching approach as well as using multivariable regression analyses.

RESULTS Patient Characteristics

Initially, a total of 283 patients with preoperative diabetes and 1069 patients without preoperative diabetes were screened for eligibility. Out of these, 943 patients did not meet the inclusion or matching criteria. Among the remaining 162 patients with diabetes and 247 patients without diabetes, 126 patients were excluded because they met the exclusion criteria, and 7 patients declined to participate. The remaining patients were successfully matched in a 1:1 ratio with controls who were of similar age, sex, and underwent similar surgeries. After the enrollment, 6 patients from the diabetes group and 4 patients from the nondiabetes group were further excluded due to technical difficulties in acquiring raw EEG data. The final analysis included 132 patients in the diabetes group and 134 patients in the nondiabetes group (Figure 2).

Figure 2.:

Flow chart of cohort. Two groups of patients were recruited: those with diabetes, and a group of controls matched 1:1 for age, sex and type of surgery.

The majority of patients were female (72.6%), with a median age of 68 (interquartile range [IQR], 65–72) years (Table 1). In the diabetes cohort, the majority of patients (84.1%) were confirmed to have known history of diabetes, with the remaining being newly diagnosed diabetes before surgery (Supplemental Digital Content 1, Supplemental Table 1, https://links.lww.com/AA/E710). Patients with diabetes had a higher prevalence of ASA physical status ≥3, higher preoperative fasting blood glucose levels and slightly lower preoperative alpha power (Table 1).

Table 1. - Preoperative Patient Characteristics Diabetes
(n = 132) Nondiabetes
(n = 134) Absolute standardized differencea Age, y 69 [64, 73] 68 [65, 72] 0.007 Female sex, n (%) 97(73.5) 96(71.6) 0.042 Height, cm 160 [156, 165] 160 [156, 165] 0.035 Body mass index, kg/m2 24. 9 ± 3.4 24. 8 ± 3.5 0.042 Years of education, n (%) 0.212  ≤6 88(66.7) 102(76.1)  6–9 23(17.4) 16(11.9)  ≥9 21(15.9) 16(11.9) Preexisting conditions, n (%)  Hypertension 89(68.2) 79(59.0) 0.181  Arrhythmia 15(11.4) 10(7.5) 0.123  Coronary heart disease 10(7.6) 3(2.2) 0.202  History of stroke 8(6.1) 13(9.7) 0.153  Obesityb 8(6.1) 10(7.5) 0.059 ASA physical status ≥ III, n (%) 51(38.6) 31(23.1) 0.318 Mini-Mental State Examination 25 [23, 27] 25 [24, 26] 0.097 History of anesthesia, n (%) 49(37.1) 45(33.6) 0.073 Laboratory tests  Albumin, g/L 41. 6 ± 4.2 41. 7 ± 4.3 0.012  Creatinine, μmol/L 58.0 [48.0, 72.0] 59.0 [50.0, 71.5] 0.073  Glucose, mmol/Lc 7.3 [5.8, 8.8] 5.3 [4.9, 5.8] 0.770 Type of surgery, n (%) 0.087  Knee arthroplasty 77(58.3) 82(61.2)  Hip arthroplasty 29(22.0) 29(21.6)  Lumbar spine surgery 19(14.4) 18(13.4)  Extremity fracture surgery 7(5.3) 5(3.7) Preoperative alpha power, dB –2.6 [–4.7, −0.3] –2.1 [–3.7, 0.2] 0.159

Data are presented as no. (%), mean ± standard deviation or median [quartiles].

Abbreviations: ASA, American Society of Anesthesiologists; BMI, body mass index.

aDefined as the absolute difference in means or proportions divided by the pooled standard deviation. Absolute standardized differences >0.1 were considered inadequately balanced between the two groups.

bObesity is defined as BMI ≥ 30 kg/m2.

cThe first fasting blood glucose value after admission.

Table 2. - Intraoperative Characteristics Diabetes
(n = 132) Nondiabetes
(n = 134) Absolute standardized differencea Duration of surgery, min 77.5 [58.8, 97.5] 75.0 [65.0, 90.0] 0.033 Duration of anesthesia, min 113 [90.8, 131] 108 [98.0, 122] 0.001 Nerve block, n (%) 107(81.1) 116 (86.6) 0.141 Anesthesia maintenance agent  Total propofol, mgb 437 [341, 551] 439 [368, 520] 0.006  Propofol infusion rate, mg·kg–1·h–1c 4.0 [3.5, 4.3] 4.0 [3.6, 4.4] 0.193  Total remifentanil, µg 545 [391, 847] 520 [341, 748] 0.170  Remifentanil infusion rate, µg·kg–1·h–1c 6.3 [4.6, 8.0] 5.7 [4.2, 7.7] 0.178 Volume of crystalloid, mL 600 [500, 1000] 600 [500, 1000] 0.065 Volume of colloid, mL 440 [0, 500] 300 [0, 500] 0.086 Estimated blood loss, mL 150 [100, 200] 150 [100, 200] 0.035 Blood transfusion, n (%) 3(2.2) 1(0.7) 0.102 Intraoperative alpha power, dB 3.8 [0.6, 6.8] 5.6 [3.3, 7.9] 0.437

Data are presented as no. (%) or median [quartiles].

Abbreviations: EEG, electroencephalogram.

aDefined as the absolute difference in means or proportions divided by the pooled standard deviation. Absolute standardized differences >0.1 were considered inadequately balanced between the 2 groups.

bThe total propofol did not include the dose for induction.

cInfusion rate of propofol or remifentanil during the selected intraoperative EEG epochs.

Table 3. - Postoperative Outcomes Diabetes
(n = 132) Nondiabetes
(n = 134) P-value Postoperative delirium, n (%) 22(16.7) 8(6.0) .006 Delirium severitya 2 [1, 2] 2 [1, 2] .467 Delirium duration, db 1 [1, 1.8] 1 [1, 2.3] Not applicable ICU admission, n (%)c 1(0.7) 0(0) .994 Maximum NRS score 2 [2, 4] 3 [2, 4] .328 Postoperative nausea and vomiting, n (%) 32(24.2) 41(30.6) .192 Length of stay, dd 9 [7, 11] 9 [7, 10] .072

Data are presented as no. (%) or median [quartiles].

Abbreviations: ICU, intensive care unit; NRS, numerical rating scale scores for pain.

aDelirium severity was calculated for all patients.

bDelirium duration was calculated only for patients who experienced delirium, with 22 in the diabetes group and 8 in the nondiabetes group.

cOne diabetic patient transferred to ICU on the fifth postoperative day due to infection-related heart failure. The P-value was calculated using Yates’ continuity corrected χ2 test.

dThe number of days from admission to discharge.

Table 4. - Association of Preoperative Diabetes With the Presence and Severity of Postoperative Delirium Presence of postoperative delirium Model 1a Model 2b Adjusted odds ratio (95% CI) P-value Adjusted odds ratio (95% CI) P-value   3.2 (1.4–8.0) .009 3.2 (1.4–8.0) .010 The severity of postoperative delirium Model 1a Model 2b Adjusted odds ratio (95% CI) P-value Adjusted odds ratio (95% CI) P-value   1.2 (0.81–2.0) .280 1.3 (0.85–2.1) .209

Abbreviations: CI, confidence interval.

aModel adjusted for age, sex, body mass index, education level, hypertension, arrhythmia, coronary heart disease, and history of stroke.

bModel for sensitivity analysis, adjusted for age, sex, body mass index, education level, hypertension, arrhythmia, coronary heart disease, history of stroke, duration of anesthesia, and type of surgery.

Table 5. - Effect Sizes in Primary Mediation Analysis and Sensitivity Analyses Total effecta
(95% CI) Direct effectb
(95% CI) Indirect effectc
(95% CI) Mediated proportiond
(95% CI) Primary mediation analysis 10.4 (3.3, 18) 8.3 (1.1, 16) 2.1 (0.32, 4.0) 20 (2.6, 60) Sensitivity analyses  Sensitivity analysis 1e 10.4 (3.1, 18) 8.2 (1.1, 16) 2.1 (0.38, 5.0) 21 (3.7, 69)  Sensitivity analysis 2f 10.4 (3.4, 18) 8.3 (1.1, 16) 2.1 (0.41, 4.0) 20 (3.9, 67)  Sensitivity analysis 3g 10.5 (3.4, 18) 8.8 (1.7, 17) 1.6 (0.12, 4.0) 16 (1.0, 53)

Models adjusted for age, sex, body mass index, education level, hypertension, arrhythmia, coronary heart disease, and history of stroke.

Abbreviations: CI, confidence interval; POD, postoperative delirium.

aDefined as the entire effect of diabetes on POD and equals the sum of the direct effect and the indirect effect.

bDefined as the difference of the potential delirium incidence when changing the individual’s exposure value from diabetes to nondiabetes, while holding the alpha power constant at the value that would be observed under a certain exposure status.

cDefined as the difference of the potential delirium incidence when an individual’s exposure status remained fixed, wh