Safety of Perioperative Glucocorticoids in Elective Noncardiac Surgery: A Systematic Review and Meta-analysis

WE are witnessing an exponential growth in the perioperative use of glucocorticoids for an expanding number of indications. When used in cardiac surgery patients, they improve pulmonary function, decrease rates of atrial fibrillation, and reduce the incidence of infection.^1,2  In hip and knee surgery, they are recommended to confer analgesic benefit,³  and they decrease sore throat in intubated patients^4,5  and swelling in maxillofacial surgery.⁶  The principal anesthesia indication is in the prevention of postoperative nausea and vomiting.⁷ 

Glucocorticoids are not, however, without harm.⁸  They are associated with atypical or opportunistic infections in either short-term high-dose or chronic long-term (more than 21 days) administration. Recognized complications include hyperglycemia, psychosis, peptic ulceration, and poor bone healing.^9,10  Although high-quality studies have shown glucocorticoid use in cardiac surgery to be both beneficial and safe,¹  there are three major safety concerns regarding the perioperative use of glucocorticoids in noncardiac surgical patients, including a possible increase in the risk of (1) hyperglycemia, (2) infection, and (3) perioperative bleeding.^9–11  None of the published trials in this field are of sufficient size to have adequate power to detect a meaningful effect upon these outcomes,¹²  and there is a need to examine these questions with a meta-analysis. We, therefore, hypothesized that important adverse outcomes are related to the use of perioperative glucocorticoids, principally the occurrence of postoperative wound infection and hyperglycemia, and these relationships are dose-dependent.

This meta-analysis was registered with the International Prospective Register of Systematic Reviews (PROSPERO 2016: CRD42016038280, http://www.crd.york.ac.uk/PROSPERO). We conducted a systematic literature search of MEDLINE and EMBASE via Ovid, PubMed, CINAHL, the Cochrane controlled trial register, and Web of Science from database inception to April 2, 2016. Each database was searched separately to improve functionality and to allow mapping to relevant subject headings. The strategy used validated methods of the Cochrane Collaboration and the Preferred Reporting Items for Systematic Reviews and Meta-analyses statement.¹³  Search terms included combinations of the Medical Subject Headings: dexamethasone, methylprednisolone, glucocorticoid, steroid, perioperative period, perioperative care, and combinations with the key words: surgery, wound infection, surgical-site infection, pneumonia, infection, sepsis, sept*, pyrexia, febrile, C-reactive protein (CRP), wound dehiscence, anastomotic leak, blood glucose, hyperglycemia, malignancy, cancer, neoplasm, bleeding, hemorrhage, mortality, and death. Subject headings were exploded to include all relevant subheadings. Search limits included randomized trials, systematic reviews, and meta-analyses in adults (18 yr and older). There were no language restrictions. Two researchers (A.J.T. and V.G.) independently screened the articles by their titles and abstracts to identify studies specifically reporting on the primary outcomes of any wound infection or blood glucose concentration, as well as secondary outcomes relevant to an assessment of safety. The full text of these articles and the gray literature were then screened according to fixed eligibility criteria.

Published and unpublished studies that met all of the following criteria were eligible for inclusion: (1) randomized controlled trial; (2) evaluation of single or repeated doses of intravenous glucocorticoid against placebo, with the first dose commencing preoperatively or intraoperatively; (3) study population underwent elective noncardiac surgery; (4) the studies quantitatively defined at least one predetermined safety outcome as an endpoint and reported this in Results; and (5) an English translation of foreign language studies was available through our institutional library. The predetermined primary outcomes were any wound infection and blood glucose concentrations, and the secondary outcomes included wound healing, anastomotic leak, inflammatory response (as determined by CRP concentrations), length of stay, bleeding, and any other infection. Studies that reported no adverse effects or complications in either group were excluded unless systematic postoperative screening had taken place. Statements such as “no adverse events occurred” were insufficient in isolation to warrant inclusion. Reference lists of included articles were hand searched, and relevant trials were subjected to the same criteria.

Two researchers (A.J.T. and V.G.) independently examined and recorded trial characteristics and outcomes, using a predesigned data abstraction form. This abstraction form was used to record information regarding the quality of the trial such as allocation concealment, randomization, blinding, exclusion criteria, and systematic outcome measurement. The grading of allocation concealment was based on the Cochrane approach, that is, adequate, uncertain, or clearly inadequate. Quality assessment was performed using the Jadad scale.¹⁴  Authors of the primary studies were contacted, when possible, if information was missing or unclear. There were no disagreements between the reviewers in relation to any of the data extracted.

Studies reporting categorical adverse outcomes were grouped based on the following definitions: “any wound infection”—Centers for Disease Control and Prevention surgical-site infection criteria or no criteria provided; “deep wound infection”—Centers for Disease Control and Prevention deep/organ space surgical-site infection criteria or abdominal abscess at the site of surgery; “any infection”—specifically reported in results table (not amalgamated from individual complications, thereby avoiding double counting); “impaired wound healing” or “anastomotic leak” or “postoperative hemorrhage” or “malignancy recurrence”—specifically reported in results table. Studies were also grouped if the following continuous outcomes were measured perioperatively: blood glucose (mg/dl), CRP (mg/l), intraoperative blood loss (ml), and hospital length of stay measured in days.

The impact of glucocorticoid dose was assessed using dexamethasone as the reference drug; each outcome group was stratified according to dexamethasone dose equivalents in the first 24 h perioperatively into low (less than 8 mg), medium (8 to 16 mg), and high (more than 16 mg), using an online steroid equivalence converter (http://www.medcalc.com). These thresholds were selected to correspond to antiemetic doses,¹⁵  doses to control swelling in maxillofacial surgery,⁶  and doses aggressively targeting the systemic inflammatory response.²  Studies assessing more than one dose of glucocorticoid against a single control group were considered as a separate study for each dose. Studies reporting outcomes in diabetic and nondiabetic subgroups were also considered as separate studies where relevant.

Statistical analyses were conducted using MetaXL version 3.1 (EpiGear International Pty Ltd, Australia). The presence and extent of heterogeneity between studies were assessed with Cochran Q and I² statistics. Pooled categorical outcomes incorporating studies with sparse event data were meta-analyzed using Peto odds ratio (OR) method with a 0.5 continuity correction where zero events occurred in both groups. Sensitivity analyses were subsequently performed using high-quality trials only (Jadad scale score, 4 to 5). Continuous outcomes were analyzed with the quality effects model incorporating the Jadad scale and expressed as weighted mean difference. Where median (range) data only were available, conversion to mean (SD) followed established guidance.¹⁶ 

For the primary outcome of blood glucose, the time point within each relevant study from the intraoperative or early postoperative (up to 12 h after surgery) periods with the highest mean values across both arms was used to compare peak perioperative glucose concentrations in the presence or absence of glucocorticoid. A similar approach was employed to assess peak perioperative CRP concentrations, using the time point within 3 days of surgery with the highest mean values across both arms. A sensitivity analysis with respect to blood glucose was conducted for trials with explicit exclusion criteria for diabetes. For the primary outcome of any wound infection, sensitivity analyses across all trials were performed using the Mantel–Haenszel, inverse variance, and quality effects models. Further sensitivity analyses were conducted for trials employing wound infection surveillance against Centers for Disease Control and Prevention criteria and for trials using dexamethasone as the glucocorticoid intervention.

Mixed meta-regression (methods of moment) was used to assess any potential interaction between dexamethasone dose (or equivalent) in the first 24 h and the risk of any wound infection and any interaction between the initial dexamethasone dose (or equivalent) and peak perioperative blood glucose concentrations (Comprehensive Meta-analysis version 2.2.034; Biostat, USA). P < 0.05 was used to denote statistical significance as this corresponded to an acceptable risk of false-positive results (type 1 error) for the primary outcomes within our specific meta-analysis design.¹⁷ 

The Preferred Reporting Items for Systematic Reviews and Meta-analyses flow diagram outlining the study selection process is detailed in figure 1. Fifty-six randomized controlled trials employing perioperative glucocorticoids including 5,607 patients from 18 countries were identified and subject to meta-analysis (median sample size, 58; interquartile range, 36 to 106).^18–73  The median Jadad score was 4 (interquartile range, 3 to 5); 31 trials were ascribed a score of 4 or 5. The characteristics of the studies and their interventions are summarized in table 1. The number of studies and patients in each outcome group are summarized in table 2. Twenty-nine studies used dexamethasone, 24 used methylprednisolone, 2 used hydrocortisone, and 1 used betamethasone. The drug was administered intravenously in all trials, predominantly as a single preoperative or early intraoperative dose. Repeated doses were given in two studies using dexamethasone, two studies using methylprednisolone, and two studies using hydrocortisone. Eighteen trials reported on any wound infection as an outcome (table 3). Three used Centers for Disease Control and Prevention criteria, and three used clinical signs together with a positive culture and/or antibiotic therapy for diagnosis of wound infection. In only one trial was wound infection the primary outcome, and in one trial, wound infection was a component of a composite primary outcome. In 12 trials, the criteria for reaching a diagnosis of wound infection were either unclear or not stated.

There was no difference in the incidence of any wound infection in patients randomized to perioperative glucocorticoid (OR, 0.8; 95% CI, 0.6 to 1.2; P = 0.80; fig. 2A). This was unchanged after excluding low-quality trials (OR, 1.0; CI, 0.7 to 1.4) or including only trials that employed surveillance for wound infection against Centers for Disease Control and Prevention criteria (OR, 1.1; CI, 0.8 to 1.6; fig. 2B) or trials using dexamethasone (OR, 0.9; CI, 0.6 to 1.3). Across all trials, findings were unchanged with the Mantel–Haenszel model (OR, 0.9; CI, 0.6 to 1.2), the inverse variance model (OR, 0.9; CI, 0.7 to 1.3), and the quality effects model (OR, 0.9; CI, 0.7 to 1.3). Patients randomized to perioperative glucocorticoid had similar rates of deep wound infection (OR, 1.2; CI, 0.8 to 1.8; fig. 2C), impaired wound healing (OR, 1.0; CI, 0.5 to 2.1), any infection (OR, 0.9; CI, 0.6 to 1.3), anastomotic leak (OR, 1.0; CI, 0.5 to 2.2), and postoperative hemorrhage (OR, 1.4; CI, 0.7 to 2.7). These findings were unchanged restricting analyses to high-quality trials only or those using Centers for Disease Control and Prevention criteria (fig. 2D).

The highest glucose value measured at a specific time intraoperatively or within the first 12 h after surgery was greater in patients receiving glucocorticoids (weighted mean difference, 20.0 mg/dl; CI, 11.4 to 28.6; P < 0.001; fig. 3A). When studies excluding diabetic patients only were analyzed, a smaller but nonetheless statistically significant rise in blood glucose was observed (weighted mean difference, 14.0 mg/dl; CI, 2.0 to 25.9; P = 0.02). Hyperglycemia in studies with no diabetic exclusion criteria appeared more marked (weighted mean difference, 30.1 mg/dl; CI, 22.1 to 38.0; P < 0.001). The highest CRP value measured at a specific time within the first 3 postoperative days was lower in patients receiving glucocorticoids (weighted mean difference, −22.1 mg/l; CI, −31.7 to −12.5; P < 0.001; fig. 3B). The length of hospital stay was no different with perioperative glucocorticoids (weighted mean difference, −0.3 days; CI, −1.4 to 0.8; P = 0.65; fig. 3C). Intraoperative bleeding was unchanged after glucocorticoid administration (weighted mean difference, −6.4 ml; CI, −28.2 to 15.4; P = 0.58). Only a single study reported on malignancy recurrence, using follow-up data from a randomized trial of perioperative dexamethasone assessing the primary outcomes of peritoneal cytokines and fatigue.⁴⁹  Malignancy recurrence was, therefore, not subjected to meta-analysis. Stratification according to low-, medium-, or high-dose dexamethasone equivalents did not change the statistical significance of any of the categorical or continuous outcomes.

In a meta-regression analysis, there was no association between glucocorticoid dose and the subsequent risk of any wound infection (slope of regression line, −0.004; 95% CI, −0.001 to 0.003; P = 0.30; fig. 4A). There was also no association between the doses of dexamethasone and the difference in maximum glucose concentrations between the included studies (slope of regression line, 0.02; 95% CI, −0.01 to 0.05; P = 0.21; fig. 4B).

Funnel and Doi plot analyses suggest that less precise studies report greater reductions in any wound infection incidence after perioperative glucocorticoid administration (fig. 5, A and B). Blood glucose trials did not exhibit significant bias (fig. 5, C and D).

The most important finding of this meta-analysis is that perioperative glucocorticoid administration does not result in an increased incidence of any wound infection, deep wound infection, anastomotic leak, impaired wound healing, or bleeding in patients undergoing noncardiac surgery. Glucocorticoids did not influence the length of stay but were associated with a clinically unimportant increase in blood glucose concentrations and a decrease in CRP concentrations. Although these findings appear to affirm the safety of short-term administration of glucocorticoids, the literature regarding safety outcomes is not particularly robust. Given the expanding use of these agents and the implications of harm for a large number of patients, this requires further discussion.

Postoperative infections, particularly surgical-site infections, are important as they prolong hospital stay, increase costs, and have an impact on postoperative mortality that extends to at least to 30 days.⁷⁵  Among the glucocorticoids, dexamethasone is the most commonly administered agent in the perioperative period being a potent, cheap, and effective antiemetic,^7,15  with analgesic properties and the capacity to improve the quality of recovery, facilitating early hospital discharge.^76–78  While several meta-analyses and practice guidelines have asserted the apparent safety of perioperative glucocorticoids in general and dexamethasone in particular in terms of infection risk,^3,15,79  the lack of definitions of adverse outcomes or use of postoperative surveillance has provoked much discussion.^11,80  Two small retrospective studies have produced conflicting results, with one suggesting that dexamethasone increases infection risk,⁸¹  while a cohort study did not confirm these findings.⁸²  A large randomized controlled trial in cardiac surgery has reported a 4.9% absolute reduction in the risk of postoperative infection in patients receiving 1 mg/kg dexamethasone, principally related to the occurrence of pneumonia.¹  Our results do not suggest an effect of glucocorticoids on surgical-site infection, wound healing, and anastomotic leak nor a dose-related trend. When our analyses were restricted to high-quality trials that used objective criteria and postoperative surveillance to day 30, the results did not change. Hence, these results are likely a robust finding. The Doi plot reveals that there is clear asymmetry, favoring the publication of small trials with lower rates of infection.

The hyperglycemic effect of perioperative glucocorticoids are real but small. Acute hyperglycemia may impair leukocyte functions⁸³  and wound healing,⁸⁴  these being worse with surges of blood glucose rather than sustained hyperglycemia. While the clinical importance of perioperative hyperglycemia is as yet not fully evaluated, particularly in terms of causality, there is a growing body of observational data to support a strong relationship, especially in nondiabetic patients.^85,86  Studies have shown infectious and noninfectious complications to be directly related to the degree of hyperglycemia, with even a single elevation of blood glucose in the perioperative period being harmful.^87,88  Our results show that perioperative glucocorticoids increased blood glucose concentrations, but with the limited doses of dexamethasone used in the included trials in this meta-analysis (almost all less than or equal to 8 mg dexamethasone), the absolute difference in blood glucose (20.0 mg/dl) was modest and was only marginally less than that observed with a much larger dose in cardiac surgery patients.¹  We chose the maximal blood glucose concentration measured intraoperatively or postoperatively up to 12 h as this enabled the inclusion of the maximum number of trials. Whether this is the period in which a peak effect of glucocorticoid on blood glucose concentration is evident or not is unknown. Hyperglycemia after glucocorticoid was significant whether trials excluded diabetic patients or not, but the effect size appeared more marked in studies with no diabetic exclusion criteria. This observation was strongly influenced by a single, well-conducted trial reporting no significant hyperglycemia with dexamethasone after gynecologic surgery in nondiabetics.³⁸  In contrast, a separate analysis of the data from one of the included randomized trials¹⁸  has been published, and it suggests that the hyperglycemic effect is more marked in nondiabetic patients, even when controlling for insulin treatment in the diabetes group.⁷⁴  Overall, the importance of small blood glucose increases is unknown¹⁰  and remains to be established in a large, properly conducted trial. Beyond infection and hyperglycemia, the effect of glucocorticoids on postoperative CRP concentrations is consistent with their well-documented antiinflammatory actions⁸⁹  and is not surprising.

Some previous meta-analyses in noncardiac surgery have reported reductions in specific or composite postoperative complications and length of stay after glucocorticoid administration.^79,89  The limitations of these studies include summation of individual complications and an absence of adjustment for study quality or glucocorticoid dose.⁹⁰  Other systematic reviews have reported no impact on adverse events but were based on searches restricted to specific outcomes such as pain and were not suitable for a safety assessment.^3,78  Overall assertions of safety have been criticized as few trials systematically defined or sought complications through postoperative surveillance. In this work, three recent high-quality trials with appropriate surveillance were included (and considered in isolation), and no impact on discrete complications was observed. Furthermore, trial quality and glucocorticoid dose were adequately adjusted for, lending further validity to the results.

Finally, we would like to acknowledge the limitations of this study, which are common to many meta-analyses⁹¹  and which in our case are compounded by the poor quality of the definitions of the endpoints of interest. The qualitative outcomes (infection, anastomotic leak, and wound healing) lacked uniformity of approach and definition. In only one of the 18 trials reporting on infection rates was this a primary outcome, and in only six were any diagnostic definitions provided, a problem common to other fields.⁹²  The total number of patients included in this study was limited because we sought to specifically examine the effect of glucocorticoids in noncardiac surgery and nonobstetric patients as these patients represent the majority of patients undergoing general anesthesia globally every day. Combining the results of independent studies, with variable amounts of sampling error due to differing conditions, surgical populations, glucocorticoid type and dose, and sample size, is fraught with challenges.⁹⁰  The I² statistic (which was low) for infective outcomes may be underpowered to detect heterogeneity when the number of studies included in the meta-analysis is small (k < 20) and/or the average sample size of the studies is less than 80 (both conditions pertain to this meta-analysis).⁹³  To address this, we employed both quality effects and inverse variance heterogeneity models, which provide more reliable estimates than the random effects model.^94,95  Smaller, lower quality, and less precise trials tended to report larger effect sizes after glucocorticoid, but their influence was removed or attenuated by sensitivity analysis or the quality effects model, respectively. Our meta-analysis provides data on any wound infection rates in 2,138 patients and on deep wound infection rates in 1,196 patients. Based on these data, with any and deep wound infection rates in the control groups of 8.9% and 7.0 %, the detection of a small but clinically meaningful difference in infection rates of 2% between glucocorticoid and control groups at 90% power would require a trial with 3,543 patients in each arm.

In conclusion, our meta-analysis confirms many of the findings that have been asserted in previous meta-analyses and guideline documents. The administration of perioperative glucocorticoids to patients undergoing noncardiac and nonobstetric surgery appears to be safe in terms of postoperative infection risk, anastomotic leak, wound healing, and bleeding risk. They have a clear antiinflammatory effect without reducing the length of stay after surgery. Blood glucose concentrations do increase in the perioperative period in patients receiving glucocorticoids, but the magnitude of changes is of questionable clinical importance especially since an increased risk of infection is not observed. An 8,800-patient trial of dexamethasone and surgical-site infection is in progress (Perioperative ADministration of Dexamethasone and Infection trial; ACTRN12614001226695). Pending those results, expected in 2019, available data suggest that single-dose perioperative dexamethasone does not provoke substantive complications. We can, therefore, be assured that the current literature does not raise any safety concerns that should rule out using low to moderate doses of glucocorticoids in the elective noncardiac surgical patient.

Royal Perth Hospital Department of Anaesthesia and Pain Medicine (Perth, Western Australia, Australia) funded academic time for database searches, data analysis, and manuscript preparation.

Drs. Ho and Corcoran are funded by WA Health and Raine Medical Research Foundation (Perth, Western Australia, Australia) through the Raine Clinical Research Fellowship. Drs. Chan and Ho are steering committee members for Perioperative ADministration of Dexamethasone and Infection (PADDI). Dr. Corcoran is the principal investigator for the PADDI trial. The other authors declare no competing interests.