Association of intraoperative dexamethasone administration with postoperative delirium and the role of hyperglycaemia: a retrospective cohort study

Summary

Background

Postoperative delirium is a frequent, serious complication triggered by various factors including systemic inflammation. Dexamethasone, an inexpensive anti-inflammatory steroid frequently administered for prophylaxis of postoperative nausea and vomiting, attenuates inflammation. We hypothesised that intraoperative dexamethasone administration is associated with a lower risk of postoperative delirium and assessed whether this is modified by the occurrence of its key side effect, hyperglycaemia.

Methods

This retrospective cohort study analysed electronic health data from adult hospitalised patients undergoing non-cardiac, non-neurosurgical, and non-transplant procedures at Beth Israel Deaconess Medical Center (Boston, MA, USA) between January 1, 2008, and January 15, 2024. Patients with missing data, preoperative delirium or glucocorticoid use, mechanical ventilation for 72 h or more, and those not expected to survive without the procedure, were excluded. The primary exposure was intraoperative administration of intravenous dexamethasone. The primary outcome was 7-day postoperative delirium, identified by keyword-triggered manual discharge note reviews, diagnostic codes, and the Confusion Assessment Method. Hyperglycaemia was defined as peak 24-h postoperative blood glucose of more than 180 mg/dL. All analyses were adjusted for 43 patient-related and procedure-related variables.

Findings

92,832 patients were included (55.8% female, median age 60 years [IQR 48–70]), of which 41,983 (45.2%) received dexamethasone at a median dose of 8 mg (IQR 4–8). 2575 (2.8%) patients developed postoperative delirium. Emergency procedures accounted for 11,970 (12.9%) of cases. Intraoperative administration of dexamethasone was associated with a lower risk of delirium (adjusted odds ratio [aOR] 0.63, 95% CI 0.56–0.70; p < 0.001; adjusted absolute risk difference −1.1%, 95% CI −1.3 to −0.8). The exploratory four-way mediation analysis suggested a 10.4% greater dexamethasone-associated reduction of postoperative delirium risk when hyperglycaemia did not occur (no hyperglycaemia aOR 0.59, 95% CI 0.51–0.67; p < 0.001; hyperglycaemia aOR 0.85, 95% CI 0.68–1.07; p = 0.17).

Interpretation

Intraoperative dexamethasone administration is associated with a lower risk of postoperative delirium, although this association was not evident in patients experiencing hyperglycaemia. Prospective studies should investigate the role of dexamethasone and optimised blood glucose control in delirium prevention.

Funding

Unrestricted philanthropic grant by Dr. J. and J. Buzen.

Research in context.

Evidence before this study

PubMed was searched on June 04, 2024 for “dexamethasone”, “steroid”, “postoperative delirium or cognitive dysfunction”. We also searched references listed in the identified publications. Five randomised controlled trials reported delirium incidences with perioperative dexamethasone compared to placebo after cardiac surgery or hip fracture repair (four reporting benefit, one no difference, none showing harm). Glucose dysregulation is a risk factor for delirium but has not been investigated in the context of delirium after dexamethasone. We aimed to fill this gap by conducting a large retrospective analysis of diverse surgical patients.

Added value of this study

In a broad surgical cohort of 92,832 patients (55.8% female, median age 60 years), this study demonstrated the association between intraoperative dexamethasone administration and a lower risk of postoperative delirium following both elective and emergency procedures. This association was stronger when hyperglycaemia, an important side effect of dexamethasone, would be prevented, as quantified by four-way mediation analysis.

Implications of all the available evidence

Dexamethasone, in doses routinely used for prophylaxis of nausea and vomiting, is associated with a lower risk of postoperative delirium when hyperglycaemia is avoided. This association as well as balanced blood glucose control for delirium prevention should now be validated in randomised trials.

Introduction

Delirium is a frequent postoperative complication, with incidences strongly varying by age, comorbidity burden, emergency status and type of procedure.¹ It is characterised by an acute onset of deficits in attention and cognition, translating into poor outcomes including long-term cognitive decline and a 50% higher mortality risk even 10 years after acute internal medicine admission.² The pathophysiology of delirium is complex and involves various precipitating factors such as systemic and neuro-inflammation or pain.¹

Dexamethasone is a potent and inexpensive glucocorticoid, frequently used during surgery to prevent postoperative nausea and vomiting (PONV).³ Dexamethasone exerts anti-neuroinflammatory and pain-modulating effects,⁴ which in turn might translate into a lowered risk of delirium. By contrast, the use of dexamethasone has been cautioned in patients suffering from diabetes mellitus, as steroid administration increases the risk of hyperglycaemia and potential surgical site infection in these patients.^5,6 A transient rise in postoperative blood glucose is the key side effect of a single low dose of intraoperative dexamethasone.⁵

Investigations in patients undergoing cardiac surgery or hip fracture repair reported equivocal effects of dexamethasone on delirium, depending on dose and type of surgery.7, 8, 9, 10, 11 Two randomised controlled trials (RCTs) in patients with hip fractures^7,9 and two RCTs in cardiac surgery reported benefit from continued administration of dexamethasone (8 mg initially, followed by 24 mg/day for 3 days)⁸ or from very high doses (1 mg/kg)¹⁰ while one RCT in cardiac surgery did not find a difference in delirium risk following dexamethasone.¹¹ As preoperative and stress-induced hyperglycaemia have been linked to a higher risk of delirium after surgery,^12,13 it is possible that dexamethasone exerts competing effects that modulate patient's risk of delirium, which could further explain previous equivocal results.

In this study, we hypothesised that the administration of intraoperative dexamethasone is associated with a lower risk of 7-day postoperative delirium and that the development of its key side effect, hyperglycaemia, as a potential intermediate in the effect of dexamethasone on delirium, modifies this association.

Methods

Study design and population

This retrospective cohort study analysed data from adult hospitalised patients who underwent general anaesthesia for non-cardiac, non-neurosurgical, and non-transplant procedures between January 1, 2008 and January 15, 2024 at Beth Israel Deaconess Medical Center, an academic tertiary-care hospital in Boston, Massachusetts, USA. Cardiac, neurosurgical, and transplant surgeries were not included due to their unique baseline risk for delirium through cardiopulmonary bypass effects, direct brain manipulation, or immunosuppression. We excluded patients with a documented diagnosis of delirium within one month prior to the index procedure, with oral or injectable corticosteroid use 1 month before surgery, with continued postoperative mechanical ventilation for 72 h or more, and those with an American Society of Anesthesiologists (ASA) physical status classification of V or VI (patients expected not to survive without the procedure or brain-dead patients). Data were initially assumed to be missing at random, and missingness patterns were assessed for confirmation (Supplemental digital content 1, Fig. S1). For the primary analysis, the complete case approach was used. Data were collected during routine clinical care, extracted from electronic hospital databases, merged into one dataset, and de-identified (for details, see Supplemental digital content 1, Section 1.1).

Ethics

This study was reviewed and approved by the Committee on Clinical Investigations, the institutional review board at Beth Israel Deaconess Medical Center (protocol number: 2024P000496), and the requirement for written informed consent was waived.

The manuscript adheres to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines and follows the Guideline for Reporting Mediation Analyses of Observational Studies (AGReMA Statement Short Form).

Definition of exposure and outcome measures

The primary exposure was intravenous bolus administration of dexamethasone between anaesthesia start and end time. The primary outcome was delirium within seven days after the index procedure. Delirium was identified from nursing and physician discharge notes using a previously published keyword-based search strategy.¹⁴ Notes containing the words “delirium” or “metabolic encephalopathy” (n = 5388) were manually reviewed by pairs of two, against each other blinded, anaesthesiologists. This approach was paired with International Classification of Diseases (9th/10th Revision, Clinical Modification [ICD-9/10-CM]) diagnostic codes (Supplemental digital content 1, Table S1),¹⁴ as well as with available Confusion Assessment Method assessments for patients with postoperative intensive care unit (ICU) admission. A secondary outcome was hyperglycaemia, defined as peak blood glucose value of more than 180 mg/dL within 24 h after surgery. For analytical purposes, hyperglycaemia was considered an intermediate between exposure and outcome that can also be an effect modifier of the exposure-primary outcome relationship. Details related to the definitions are provided in the Supplemental digital content 1, Section 1.2.

Confounder model

Adjustment for a priori-defined confounding variables, based on available literature and clinical plausibility, was applied for all analyses, as illustrated in the directed acyclic graph in the Supplemental digital content 1, Fig. S2. Confounders consisted of patient's self-reported sex, age, body mass index, and social determinants of health including zip-code based household income estimates and federal insurance status. Moreover, comorbidities diagnosed within 1 year prior to the index procedure, including diabetes mellitus, cancer, anaemia, stroke, obstructive sleep apnoea, psychosis, depression, rheumatologic disease, and tobacco, alcohol, or drug use disorder, were included.¹ Additional preoperative variables comprised the ASA physical status, the Elixhauser Comorbidity Index,¹⁵ the Apfel Score as a measure of PONV risk,¹⁶ as well as preoperatively prescribed opioids, benzodiazepines or antibiotics. Analyses were adjusted for intraoperative contributors to delirium, namely type of anaesthesia (regional, volatile-based), the occurrence of intraoperative hypotension or hypoxaemia, perioperative anaesthetic medications including vasopressors, opioid, propofol, and neuromuscular blockade doses.¹⁴ Further, the model included the perioperative use of antiemetics other than dexamethasone, fluid administration, the amount of intraoperatively packed red blood cell units transfused, and the duration of surgery. Surgery during non-regular operating room hours, relative work value units, preoperative ICU admission, admission type (same-day versus inpatient), emergency status, and year of surgery were also incorporated. Race was added post-hoc given emerging evidence of race-related disparities in PONV medication prescribing.¹⁷ The risk analysis index as measure of frailty was added post-hoc upon new evidence from the SNAP3 trial and another study by Gan et al., showing that frailty is associated with delirium independent of multimorbidity.^2,18,19 Continuous covariables were categorised in quantiles or based on clinical categories when not presenting a linear relationship with the primary outcome. Details related to all confounding variables are provided in the Supplemental digital content 1, Section 1.3.

Statistics

The primary and secondary analyses, exposure, outcome, and confounders were defined in written form and filed with the institutional review board before data were accessed for analysis. Results of multivariable logistic regression analyses are reported in adjusted odds ratios (aORs) with 95% CIs. In exploratory four-way effect decomposition analysis, excess relative risks with 95% CI are presented.^20,21 Model calibration and discrimination were assessed (Supplemental digital content 1, Fig. S3); a correlation matrix for all covariates and variance inflation factors was computed. Absolute standardised differences to quantify covariate balance between groups were calculated as the absolute difference in means or proportions divided by the pooled standard deviation.²² Analyses were performed in STATA (Version MP 18.0) and R Statistical Software (Version 4.4.1). Supporting figures were created in BioRender.

In primary logistic regression analysis, the exposure was administration of dexamethasone as a binary variable. For the co-primary analysis, dexamethasone dose was dichotomised using the median dose in our cohort (0.09 mg/kg actual bodyweight) and compared to not receiving dexamethasone. An analysis modelling the effect of dexamethasone dose in mg/kg as a continuous variable utilised a fractional polynomial model.

By including an interaction term in the primary model, we tested whether a pre-existing diagnosis of diabetes mellitus modified the primary association (Supplemental digital content 1, Section 3 and Table S2). We analysed whether dexamethasone was associated with the development of postoperative hyperglycaemia and whether hyperglycaemia modified the primary association.

After associations between dexamethasone, hyperglycaemia and delirium were known, we performed a post-hoc exploratory four-way effect decomposition analysis.^20,21 Mediation analysis was conducted based on the assumption that dexamethasone can cause hyperglycaemia which in turn potentially affects delirium. In addition, we hypothesised an interaction between the effect of the exposure (dexamethasone) and hyperglycaemia. Thus, a four-way decomposition clarifying the unique contribution of effect mediation and interaction mechanisms was carried out. This analysis provides a decomposition of the treatment effect into the component that is due to mediation through the intermediate variable, due to exposure-mediator interaction, due to both mediation and interaction and the effect that is direct, that is the effect of the treatment in the absence of the mediator. The causal interpretation of the decomposition requires the assumption of no unmeasured confounding.²¹ A detailed description of this analysis is provided in Supplemental digital content 1, Section 4.1.

We further analysed the association between dexamethasone and surgical site infections (Supplemental digital content 1, Table S3).²³

To identify groups which might benefit from dexamethasone's anti-inflammatory properties, we performed exploratory subgroup analyses in patients with pre- or postoperative leucocytosis (>11,000/μL) and high operative stress scores.²⁴ Additionally, we investigated whether the primary effect was mediated by a dexamethasone-associated reduction in PONV or delayed discharge from the post-anaesthesia care unit (PACU) due to postoperative pain (Supplemental digital content 1, Sections 4.3 and 4.4). Finally, we explored the effect of dexamethasone on different phenotypes of delirium among patients admitted to the ICU (hypoactive, hyperactive, mixed; Supplemental digital content 1, Section 4.5 and Table S6).

Potential bias introduced by patient age and comorbidity burden was addressed by employing alternative statistical approaches including propensity score and exact matching, as well as inverse probability of treatment weighting. We additionally conducted subgroup analyses in patients aged older than 60 and 80 years, stratified by the full adult age range, as well as by comorbidity load (Elixhauser Comorbidity Index >4),¹⁵ dementia status, and by frailty in those aged 65 years or more. Further subgroup analyses by patient sex, in patients undergoing emergency surgery, or same-day surgery only, and for high-risk procedures are summarised in the Supplemental digital content 1, Section 5. For the four-way effect decomposition, we conducted a sensitivity analysis using the stress-hyperglycaemia-ratio, which compares the postoperative blood glucose to the preoperative HbA1c.²⁵

Role of the funding source

This study was funded by an unrestricted philanthropic grant awarded to MSS by Dr. Jeffery and Judith Buzen. The funders had no role in study design, data collection, data analysis, data interpretation, or writing of the report. SR, TT, and MSS have accessed and verified the data. SR and MSS were responsible for the decision to submit the manuscript.

Results

113,795 patients undergoing general anaesthesia between January 1, 2008, and January 15, 2024, met the inclusion criteria. After application of exclusion criteria and exclusion of patients with missing confounder data, the final study cohort consisted of 92,832 patients (Fig. 1). 41,983 (45.2%) patients received dexamethasone at a median dose of 8 mg (IQR 4–8 mg), or 0.09 mg/kg (0.07–0.12). Overall, patients receiving dexamethasone were younger and more likely to be female. 11,970 (12.9%) patients underwent emergent, and 80,862 (87.1%) elective procedures. The median postoperative hospital length of stay was 3 days (IQR 1–5 days), and 7 days when delirium occurred (5–13 days). Detailed information about patient and procedural characteristics are provided in Table 1.

Fig. 1.

Study flow diagram. ASA, American Society of Anesthesiologists.

Table 1.

Patient and procedural characteristics by dexamethasone administration.

	No dexamethasone	Dexamethasone	Abs. Std. Diff.
	n = 50,849	n = 41,983
Demographics
Age, years	63 (52–73)	56 (43–66)	0.426
Sex, self-reported			0.412
Male	27,291 (53.7%)	14,132 (33.7%)
Female	23,558 (46.3%)	27,851 (66.3%)
Body mass index, kg/m2	27.7 (24.2–32.6)	27.5 (24.0–32.5)	0.004
Federal insurance	25,144 (49.4%)	15,678 (37.3%)	0.246
Estimated household income, United States Dollar	101,513 (78,547–125,388)	103,606 (80,424–127,748)	0.074
Race, self-reported			0.091
Asian	1674 (3.3%)	1748 (4.2%)
Black	5231 (10.3%)	4194 (10.0%)
Hispanic	2446 (4.8%)	2217 (5.3%)
Other	4209 (8.3%)	4328 (10.3%)
White	37,289 (73.3%)	29,496 (70.3%)
Preoperative characteristics
ASA physical status			0.509
I	2207 (4.3%)	3294 (7.8%)
II	17,312 (34.0%)	22,298 (53.1%)
III	27,265 (53.6%)	15,674 (37.3%)
IV	4065 (8.0%)	717 (1.7%)
Elixhauser Comorbidity Index	6 (0–16)	3 (0–10)	0.337
Frailtya			0.374
Robust	26,832 (52.8%)	29,417 (70.1%)
Normal	11,269 (22.2%)	6689 (15.9%)
Frail	7228 (14.2%)	2899 (6.9%)
Very Frail	5520 (10.9%)	2978 (7.1%)
History of comorbidity one year prior to surgery
Diabetes mellitus	14,548 (28.6%)	4936 (11.8%)	0.429
Rheumatic disease	1270 (2.5%)	791 (1.9%)	0.042
Cancer	14,051 (27.6%)	12,672 (30.2%)	0.056
Anaemia	3609 (7.1%)	2560 (6.1%)	0.040
Stroke	1352 (2.7%)	417 (1.0%)	0.125
Obstructive sleep apnoea	7275 (14.3%)	5302 (12.6%)	0.049
Psychosis	178 (0.4%)	68 (0.2%)	0.037
Depression	9354 (18.4%)	7298 (17.4%)	0.026
Smoking status	18,941 (37.2%)	12,460 (29.7%)	0.161
Alcohol abuse	2697 (5.3%)	1235 (2.9%)	0.119
Drug abuse	2065 (4.1%)	1114 (2.7%)	0.078
Intensive care unit stay one week prior to surgery	2272 (4.5%)	737 (1.8%)	0.157
Apfel score			0.460
0	477 (0.9%)	65 (0.2%)
1	19,304 (38.0%)	9242 (22.0%)
2	20,937 (41.2%)	17,431 (41.5%)
3	7708 (15.2%)	10,226 (24.4%)
4	2423 (4.8%)	5019 (12.0%)
Opioids prior to surgery	10,227 (20.1%)	8846 (21.1%)	0.024
Benzodiazepines prior to surgery	5128 (10.1%)	5064 (12.1%)	0.063
Antibiotics prior to surgery	4627 (9.1%)	4580 (10.9%)	0.060
Emergency surgery	8337 (16.4%)	3633 (8.7%)	0.236
Admission type			0.365
Inpatient	15,834 (31.1%)	6681 (15.9%)
Same-day	35,015 (68.9%)	35,302 (84.1%)
Intraoperative characteristics
Duration of surgery, min	154 (108–226)	183 (131–260)	0.264
Work relative units	16.6 (10.5–22.5)	17.3 (12.1–23.5)	0.109
Night surgery	4472 (8.8%)	1831 (4.4%)	0.180
Surgical service			0.612
Surgery at the head	921 (1.8%)	1878 (4.5%)
General surgery	10,859 (21.4%)	10,053 (23.9%)
Gynaecology surgery	2112 (4.2%)	5860 (14.0%)
Orthopaedic surgery	15,812 (31.1%)	10,795 (25.7%)
Plastic surgery	1535 (3.0%)	2979 (7.1%)
Surgical oncology	1533 (3.0%)	1646 (3.9%)
Thoracic surgery	3029 (6.0%)	1876 (4.5%)
Trauma and surgical critical care	2258 (4.4%)	1419 (3.4%)
Urology	5073 (10.0%)	3769 (9.0%)
Vascular surgery	6015 (11.8%)	1040 (2.5%)
Procedures	1702 (3.3%)	668 (1.6%)
Regional anaesthesia	7626 (15.0%)	6431 (15.3%)	0.009
Volatile based anaesthesia	48,096 (94.6%)	40,121 (95.6%)	0.045
Opioid dose, mg oral morphine equivalents	45.4 (28.4–71.5)	42.0 (30.6–70.4)	0.076
Propofol dose, mg	200 (150–230)	200 (160–250)	0.136
Non-depolarising neuromuscular blocking agents, ED95	2.1 (1.0–3.3)	2.7 (1.6–4.1)	0.302
Crystalloid and colloid infusion, ml	1200 (800–1950)	1300 (1000–2000)	0.089
Packed red blood cells units	0 (0–0)	0 (0–0)	0.037
Vasopressors, mg norepinephrine equivalents	0.0 (0.0–0.3)	0.0 (0.0–0.1)	0.120
Hypotension, MAP below 55 mmHg >5 min	5762 (11.3%)	4283 (10.2%)	0.036
Hypoxaemia, SpO2 below 90% >2 min	8020 (15.8%)	4833 (11.5%)	0.124
Antiemetics	41,228 (81.1%)	39,503 (94.1%)	0.403

Data are presented as median (interquartile range) for continuous measures, and n (%) for categorical measures.

ASA, American Society of Anesthesiologists; Abs. Std. Diff, Absolute standardised difference; MAP, mean arterial pressure; SpO₂, peripheral capillary oxygen saturation.

^aFrailty was categorised following the risk analysis index based on diagnostic codes: robust <27 points, normal 27–35 points, frail >35–45 points, very frail >45–81 points.

2575 (2.8%) patients developed postoperative delirium, 580 (1.4%) among patients receiving dexamethasone and 1995 (3.9%) among patients not receiving dexamethasone. Dexamethasone administration decreased, while delirium incidence increased with higher patient age (Supplemental digital content 1, Fig. S4). In adjusted analyses, intraoperative administration of dexamethasone was associated with a lower risk of postoperative delirium (aOR 0.63, 95% CI 0.56–0.70; p < 0.001; adjusted absolute risk difference −1.1%, 95% CI −1.3 to −0.8). Both low and high doses of dexamethasone were associated with lower risk of delirium (≤0.09 mg/kg n = 20,155; aOR 0.64, 95% CI 0.56–0.73; p < 0.001; >0.09 mg/kg n = 21,828; aOR 0.61, 95% CI 0.53–0.70; p < 0.001). The association was consistent when treating dexamethasone as a continuous variable (Fig. 2 and Supplemental digital content 1, Fig. S5).

Fig. 2.

Dexamethasone dose-dependent delirium risk. Absolute predicted risk (mean with 95% confidence interval) of delirium with dexamethasone in mg per kg bodyweight (BW; red, right y-axis), modelled with a 1st degree fractional polynomial regression analysis. Dashed reference line indicating the predicted delirium risk without dexamethasone (right y-axis). Histogram displaying the amount of people receiving dexamethasone in the respective dose (grey, left y-axis). No statistically significant differences were observed among various dexamethasone doses compared to no dexamethasone.

19,484 (21.0%) patients had a documented diagnosis of diabetes mellitus, 2141 (11.0%) of them with type 1 and 17,343 (89.0%) with type 2 diabetes. Compared to 37,047 (50.5%) patients without, dexamethasone was administered to 4936 (25.3%) patients with a history of diabetes mellitus. The incidence of delirium was lower in patients without (2.3%, n = 1696) compared to patients with diabetes (4.5%, n = 879). The association of dexamethasone with delirium was magnified in non-diabetic and attenuated among diabetic patients (p_interaction = 0.006; no diabetes aOR 0.58, 95% CI 0.51–0.66; p < 0.001; diabetes aOR 0.77, 95% CI 0.63–0.94; p = 0.011; Fig. 3a and b).

Fig. 3.

Panel a and b: Delirium risk by glucose level and state of diabetes mellitus. Absolute predicted probability of 7-day postoperative delirium with (red) and without (blue) intraoperative dexamethasone administration (mean margin in groups of five blood glucose values, with 95% confidence intervals; right y-axis). Bar chart (black, left y-axis) displaying the absolute number of patients per blood glucose level (groups of 5 mg/dL bins). Blood glucose values were defined as the peak value within the first 24-h postoperative period. Fig. 3a: Patients without a diagnosis of diabetes mellitus one year prior to surgery, Fig. 3b: Patients with diabetes mellitus.

Postoperative glucose measurements as part of clinical routine were available in 63,046 (67.9%) of 92,832 patients. Of these patients, 8233 (13.1%) had a documentation of hyperglycaemia within 24 h after surgery. Postoperative hyperglycaemia itself was associated with higher odds of 7-day delirium (aOR 1.47, 95% CI 1.31–1.64; p < 0.001). In adjusted logistic regression analysis, the odds of developing hyperglycaemia after dexamethasone administration were 1.55 (95% CI 1.46–1.65; p < 0.001), which was more pronounced when receiving high doses of dexamethasone (>0.09 mg/kg aOR 1.72, 95% CI 1.59–1.86; p < 0.001) compared to lower doses (≤0.09 mg/kg aOR 1.44, 95% CI 1.33–1.55; p < 0.001). When hyperglycaemia occurred, dexamethasone was no longer associated with lower odds of delirium (p_interaction < 0.001; no hyperglycaemia aOR 0.59, 95% CI 0.51–0.67; p < 0.001; hyperglycaemia aOR 0.85, 95% CI 0.68–1.07; p = 0.17; Fig. 3a and b).

Post-hoc four-way effect decomposition analysis confirmed the association of dexamethasone with a lower risk of delirium (excess relative risk −0.42, 95% CI −0.49 to −0.36; p < 0.001). This association was stronger when hyperglycaemia did not occur (excess relative risk −0.47, 95% CI −0.53 to −0.40; p < 0.001). Across all patients, the occurrence of hyperglycaemia reduced dexamethasone's effect on delirium by 10.4% (overall proportion eliminated; 95% CI 16.4%–4.4%; p = 0.001). A graphical depiction of the four-way effect decomposition is presented in Fig. 4.

Fig. 4.

Four-way effect decomposition. In four-way effect decomposition analysis, intraoperative dexamethasone administration was associated with a relative risk reduction of 7-day postoperative delirium (“clinical effect”, excess relative risk [ERR]). The occurrence of 24-h postoperative hyperglycaemia (peak blood glucose >180 mg/dL) was associated with a higher risk of delirium. By hypothetically intervening and preventing both dexamethasone-associated hyperglycaemia (mediation) and hyperglycaemia due to other reasons (interaction) from occurring, the association between dexamethasone and delirium would be 10.4% larger (overall proportion eliminated). OR_adj, adjusted odds ratio.

Dexamethasone was not associated with higher risks of 30-day postoperative surgical site infections (aOR 0.77, 95% CI 0.71–0.84; p < 0.001), even when hyperglycaemia occurred (aOR 0.88, 95% CI 0.69–1.13; p = 0.32).

The primary findings were consistent among patients with preoperative (aOR 0.62, 95% CI 0.45–0.85; p = 0.003) and postoperative (aOR 0.61, 95% CI 0.53–0.71; p < 0.001) leucocytosis, and in patients undergoing procedures with high operative stress scores (aOR 0.68, 95% CI 0.50–0.93; p = 0.015). Dexamethasone was associated with lower risks of PACU discharge delays due to PONV (aOR 0.82, 95% CI 0.78–0.86; p < 0.001) and postoperative pain (aOR 0.80, 95% CI 0.74–0.86; p < 0.001). Neither PONV nor pain discharge delays mediated the primary association (Supplemental digital content 1, Sections 4.3 and 4.4).

The primary results were robust across all sensitivity analyses. The association between dexamethasone and delirium was not modified by the patient's age (p_interaction = 0.26; Supplemental digital content 1, Fig. S6) or sex (p_interaction = 0.89) and was consistent across various statistical approaches and in emergencies, in high-risk procedures, or same-day versus inpatient admission. The primary association was magnified in older patients, and in patients with dementia or frailty (Fig. 5 and Supplemental digital content 1, Section 5). Using the stress-hyperglycaemia ratio yielded comparable results to those with absolute hyperglycaemia (Supplemental digital content 1, Section 5.9).

Fig. 5.

Forest plot of sensitivity analyses. Forest plot summarising results (adjusted absolute risk differences with 95% confidence intervals) of the primary multivariable logistic regression analysis of intraoperative dexamethasone administration and 7-day postoperative delirium, and sensitivity analyses. The adjusted absolute risk difference refers to the delirium risk without dexamethasone. More negative values indicate stronger effect size of the association between dexamethasone administration and the development of delirium. ICU, Intensive care unit; ATE, average treatment effect.

Discussion

Among 92,832 adult hospitalised patients undergoing general anaesthesia for surgery, intraoperative administration of dexamethasone was associated with a lower risk of delirium within seven days after surgery. This association was pronounced among non-diabetic patients. Importantly, dexamethasone exerted competing effects with regards to postoperative delirium, and the occurrence of hyperglycaemia compromised the protective effects of dexamethasone.

Previous studies in patients undergoing hip surgery reported beneficial effects on delirium with 10–20 mg dexamethasone, with Kluger et al. (20 mg) describing reduced delirium severity but not incidence.^7,9 In cardiac surgery, Mardani et al. showed lower delirium rates after administration of 8 mg dexamethasone initially, followed by 24 mg/day for 3 days.⁸ The multicentre Dexamethasone in Cardiac Surgery (DECS) RCT reported a reduced risk of delirium after 1 mg/kg dexamethasone, however, this was as secondary analysis and delirium was defined as the requirement for neuroleptic medication.¹⁰ In a single-centre substudy of the DECS trial, Sauër et al. prospectively screened 737 cardiac surgical patients for delirium using structured assessment tools, and did not find a difference in delirium between the dexamethasone (1 mg/kg) and placebo group.¹¹ Our findings help to explain these seemingly contradictory results: In the studies by Mardani et al. or Huang et al. who reported beneficial effects,^7,8 dexamethasone was perioperatively administered at a dose of 8–10 mg, which is in line with what most patients received in our cohort. By contrast, high doses of dexamethasone did not reduce delirium incidence after cardiac surgery, when delirium was assessed with structured tools.¹¹ Interestingly, in a subgroup of critically ill patients analysed retrospectively by Wu et al. who received 10 mg or less dexamethasone, delirium incidence was also reduced.²⁶ Thus, our findings, in context with previous studies, emphasise the importance of appropriate dosing when administering dexamethasone to reduce the risk of delirium in patients. Since low doses of dexamethasone are equally effective for PONV prophylaxis and associated with fewer side effects,^3,5 they might be preferred in the perioperative setting.

Despite its single-centre design, this study comprises one of the largest and most diverse surgical populations on this topic. Demographic and clinical characteristics in our cohort, including the near-even sex distribution (55.4% female), the median age of 60 years, and comorbidity burden closely align with national-level epidemiologic data from large United States perioperative cohorts.²⁷ In a sensitivity analysis, we found no difference with regard to the effect of dexamethasone between female and male patients. While 71.9% (n = 66,785/92,832) patients reported their race as White, 9425 (10.2%) patients identified themselves as Black, 4663 (5.0%) as Hispanic, and 3422 (3.7%) patients as Asian, which reflects the surgical demographics for Boston-area academic hospitals, which may, however, not be representative of non-academic or rural populations.²⁸ In our cohort, 45.2% of patients received dexamethasone, which is consistent with a large United States-based quality improvement study reporting a rate of 44.0%.²⁹ Among these patients, younger and less sick individuals were more likely to receive dexamethasone, which aligns with recommendations to use it as a first-line drug to prevent PONV, a condition more commonly observed among younger female patients.³ Appropriateness of dexamethasone administration is determined by the patient's risk of PONV based on the Apfel score,¹⁶ which we included into our model. Its individual components female sex, history of PONV or motion sickness, nonsmoking status, and perioperative opioid use are independent of age or comorbidity.¹⁶ Nevertheless, given that clinical guidelines provide recommendations rather than mandates, and acknowledging that this study reflects real-world clinical practice with inherent variability, we strongly advocate for prospective, randomised controlled trials incorporating diverse, representative patient populations to definitively establish dexamethasone's efficacy in preventing postoperative delirium across different demographic and clinical subgroups. Based on our data, targeting older patients (e.g., age ≥80 years) might be an effective enrichment strategy and particularly suitable to investigate relevant effects of dexamethasone.

We observed that the association between intraoperative dexamethasone administration and postoperative delirium was modified by a prior diagnosis of diabetes mellitus, and a higher incidence of delirium was observed in patients with diabetes compared to those without. In clinical practice, physicians are often concerned about dexamethasone-induced hyperglycaemia which might translate into a higher risk of surgical site infection and further complications.^5,6 This was in part reflected in our study, where diabetic patients exhibited higher rates of postoperative hyperglycaemia, but did not show a higher risk of surgical site infection. While diabetes mellitus is a risk factor for postoperative delirium,¹ the role of elevated blood glucose levels, induced through dexamethasone, has so far been unclear in patients both with and without diabetes mellitus. We found that dexamethasone was dose-dependently associated with hyperglycaemia within 24 h after surgery, and that the occurrence of hyperglycaemia may abolish protective effects of dexamethasone on delirium. This, in turn, may explain the lower effectiveness among diabetic patients. Notably, we did not identify a dose or scenario where dexamethasone exacerbated the risk of postoperative delirium, even in the presence of hyperglycaemia. In general, four-way effect decomposition analysis suggested that perioperative strategies aimed at preventing hyperglycaemia might increase the protective effect of dexamethasone on delirium by 10.4%. While dexamethasone was associated with the occurrence of hyperglycaemia, it is important to note that this association was modest compared to the much stronger effects on delirium. Based on these data, the risk of hyperglycaemia alone should not justify withholding dexamethasone, particularly since hyperglycaemia did not modify dexamethasone-associated risk of surgical site infections. However, further research is needed to specifically investigate the impact of intraoperative dexamethasone administration on postoperative delirium among patients at very high risk of hyperglycaemia. We further acknowledge that perioperative blood glucose control is a subject of ongoing debate, since overly strict control and relative hypoglycaemia have been linked to delirium.^30,31 Corroborating evidence from prospective trials is needed to demonstrate that consistent glycaemic control following dexamethasone administration restores the delirium-protective effect. Our study, together with others, therefore encourages the implementation of strategies to ensure appropriate and balanced perioperative blood glucose control while supporting the administration of low to moderate doses of dexamethasone for delirium prevention.

This study is subject to the inherent limitations of retrospective analyses using electronic hospital registry data. A key concern is the under-reporting and under-detection of postoperative delirium in routine clinical documentation.³² Notably, even randomised trials report heterogeneous and frequently low delirium rates, which are consistent with those we observed in this broad clinical dataset. A recent randomised trial on the effect of anaesthetic approach on postoperative delirium risk in older patients undergoing hip fracture repair reported a delirium incidence of 5.6%, which is comparable to the 6.3% in our data when applying the same inclusion and exclusion criteria.³³ Among patients with structured CAM-ICU assessment, the 7-day incidence (21.1%) aligns with previous, prospective data (e.g., 17.9%)³⁴ and CAM-based estimates from elective surgery cohorts (23–24%),³⁵ while effect estimates remained similar. The addition of structured assessments and chart review increased case capture relative to diagnostic codes alone, yet effect estimates remained highly consistent across all strategies, indicating that residual misclassification is unlikely to have materially affected our findings. Continued improvements in clinical documentation, prospective validation, and improved algorithmic tools are needed to enhance delirium detection in future research. Our study further illustrates the need for standardised delirium screening to become part of routine care and to be consistently documented to better support future studies. In addition, prospective trials are typically limited in sample size due to their labour and cost intensity. Thus, certain interventions such as mediation analysis and other means of isolating associations can often not be performed to the extend done in retrospective analyses based on real-world data. Therefore, delirium detection by the bundle of ICD-9/10-CM codes, nursing assessments, and keyword-triggered chart review as previously published by our group^14,36 is one of the most elegant solutions for retrospective research. While machine learning approaches show promise for improving detection accuracy, current models are still limited by the quality of available labels and, nonetheless, under-recognition of delirium reflects a system-level rather than study-specific limitation.³² Our results may reflect selection bias, although a diverse surgical cohort was selected. Furthermore, confounder selection was limited by data availability in the medical record, and unmeasured confounding can never be completely ruled out. However, we are confident that unmeasured confounding did not bias our results, as the E-value (2.58 for the point estimate, Supplemental digital content 1, Section 5.5) suggests that substantial unmeasured confounding would be needed to invalidate the associations. Nevertheless, caution is warranted when interpreting the four-way effect decomposition causally, as it inherently relies on the absence of unmeasured confounding.²⁷ During surgery, dexamethasone is usually administered in standardised doses between 4 mg and 10 mg.³ As this study reflects real-world clinical practice, we were unable to analyse the effect of high doses (>10 mg) of dexamethasone, which should be explored in future studies.

In conclusion, the intraoperative administration of dexamethasone is associated with a lower risk of delirium within seven days of surgery. This association is magnified among non-diabetic patients and abolished in patients experiencing hyperglycaemia. These findings provide a rationale for prospective studies investigating the effect of dexamethasone on delirium and for determining the role of balanced glucose control in preventing delirium.

Contributors

SR: conceptualisation, data curation, investigation, formal analysis, visualisation, writing—original draft.

TT: conceptualisation, data curation, formal analysis, visualisation, writing—original draft.

EA: conceptualisation, methodology, visualisation, writing—original draft.

LJW: conceptualisation, data curation, formal analysis, writing—original draft.

BSP: data curation, formal analysis, writing—review & editing.

DS: methodology, writing—review & editing.

DvW: methodology, writing—review & editing.

KL: data curation, visualisation, writing—review & editing.

JPS: data curation, formal analysis, writing—review & editing.

SCF: conceptualisation, methodology, writing—review & editing.

LK: data curation, writing—review & editing: methodology, software, writing—review & editing.

AK: conceptualisation, writing—review & editing.

LV: methodology, supervision, writing—review & editing.

VN: conceptualisation, methodology, software, writing—review & editing.

TSK: conceptualisation, supervision, writing—review & editing.

HM: conceptualisation, supervision, writing—review & editing.

MSS: conceptualisation, project administration, funding acquisition, resources, supervision, writing—original draft.

SR, TT, and MSS have accessed and verified the data. SR and MSS were responsible for the decision to submit the manuscript. All authors read and approved the final version of the manuscript.

Data sharing statement

Due to the sensitive nature of the data collected for this study, requests to access the dataset from qualified researchers trained in human subject research and confidentiality may be sent to Maximilian S. Schaefer at msschaef@bidmc.harvard.edu.

Declaration of interests

TT received a grant unrelated to this work by the German Research Foundation (Walter Benjamin Fellowship, project number: 522518834). TT received travel support to attend conferences through a DAAD (German Academic Exchange Service, Deutscher Akademischer Austauschdienst) congress stipend and a HeRa (Heine Research Academies) congress travel grant. EA is an associate editor for BMC Anesthesiology. LJW is an associate editor for BMC Anesthesiology and received funding for an investigator-initiated study from Merck & Co., which does not pertain to this manuscript. DS is an associate editor for BMC Anesthesiology. JPS received an unrelated grant by the German Research Foundation (Walter Benjamin Fellowship, project number: 567508647). XX received funding from the National Natural Science Foundation of China (project number: 72304281). LV received funding from the National Institute of Environmental Health Sciences and the National Institute on Ageing. She received payment for short course instructions at Statistical Horizons, the Erasmus Universiteit Rotterdam, the University of Michigan, and Harvard University. LV was reimbursed for travel and accommodation for an invited talk at the Université de Bordeaux. She serves as statistician in the Data Safety Monitoring Board at New York University.

MSS received an unrestricted philanthropic grant from Jeffrey and Judith Buzen. MSS is an associate editor for BMC Anesthesiology. He received honoraria for lectures from Fisher & Paykel Healthcare and Mindray Medical International Limited. MSS received funding for investigator-initiated studies from Merck & Co., which do not pertain to this manuscript. EA, LJW, and MSS have received a Mentoring Grant from the ASA Committee on Professional Diversity Mentoring. All funders had no role in the study design; in the collection, analysis, and interpretation of data; in the writing of the report; and in the decision to submit the article for publication.

SR, BSP, DvW, KL, SCF, LK, AK, VN, TSK, and HM have no support from any organisation for the submitted work, no financial relationships with any organisations that might have an interest in the submitted work in the previous three years, and no other relationships or activities that could appear to have influenced the submitted work.