This meta-analysis is based on previously conducted studies and does not involve any new studies of human or animal subjects performed by any of the authors.
Bibliographic Search and Analysis
We conducted this meta-analysis according to the guidelines of the Cochrane Handbook for Systematic Reviews [
8] of intervention and the PRISMA statements [
9]. Literature databases included PubMed, EMBASE, Cochrane central register of controlled trials, clinical trials register, and open-access journals not indexed in major databases (Directory of Open Access Journals, Open Journal of Anesthesiology, Anesthesiology Research and Practice, Journal of Anesthesia and Clinical Research, Journal of Anesthesiology and Clinical Science, Journal of Anesthesiology and Critical Care Medicine). The following queries were used to discard irrelevant results related to postoperative Dex use: “Dexmedetomidine” and “children or child or infant or infants”. No language restriction was applied for searches. In addition, a manual search of the references found in all selected articles was performed, including reviews and meta-analyses. Identified articles were independently assessed by four anesthesiologists (Myriam Bellon, Alix Le bot, Daphnée Michelet, Julie Hilly) and only those which fulfilled the following criteria were included: randomized-controlled, double-blind studies, patients with neurological and/or psychiatric diseases excluded, standardized protocols for anesthesia, analgesia and rescue analgesics, presence of a control group (placebo with no active anesthetic or analgesic agent) and of at least one outcome in relation to: postoperative analgesia or opioid consumption. Given the potential impact of neurosurgery on postoperative neurological function and the preoperative alteration of those functions in congenital heart diseases, both cardiac surgery and neurosurgery were excluded from the area of this meta-analysis. Abstracts presented at meetings were not included. The most recent search was performed in December 2015.
Each reader evaluated the potential presence of bias and study quality based on the following criteria: randomization and allocation concealment (clear, sufficiently detailed description of methodology demonstrating whether intervention allocations could have been foreseen before or during enrolment), double blinding, incomplete data report statements (concerning excluded patients and data) and selective reporting (presence of studied outcomes report verified). For the studies meeting these criteria, data were then independently collated by two anesthesiologists and included: patient American Anesthesiologists Association (ASA) physical status and age, type of surgery, sedative anesthetic premedication (dose, timing, and route of administration), Dex administration characteristics (doses and timing, bolus and infusion), other hypnotic agents used, intraoperative analgesia administration (both systemic or regional analgesia), postoperative analgesic administered and endpoints of each study. The primary endpoint of the study was the opioid-sparing effect of intraoperative Dex (either expressed as continuous data or as percentage of patients receiving opioids). Secondary endpoints were: the quality of postoperative [either the intensity of postoperative pain or the presence of a significant pain defined as: FLACC (Face, Legs, Activity, Cry, Consolability) >3, visual or numerical pain scale >3, facial pain scale >3, and Objective Pain Scale (OPS) >3] [
10‐
13] and the occurrence of PONV (either both or the presence of vomiting). Other outcomes such as emergence agitation and hemodynamic effects of Dex were not analyzed because of the restrictive search on studies with postoperative analgesia outcomes. When conflicting results were found, the article was rechecked twice by the two anesthesiologists until a consensus was found.
Statistical Analysis
Statistical analyses were performed using Review Manager 5 software (RevMan 5.3, The Cochrane Collaboration, Oxford, UK) and the Trial Sequential Analysis Software (Copenhagen Trial Unit’s Trial Sequential Analysis Software, hereafter: TSA Software, Copenhagen, Sweden). Where original data were expressed as continuous variables, meta-analyses were performed using the mean difference (MD) or standardized mean difference (SMD). SMD is calculated using the formula: difference in mean outcome between studies/standard deviation (SD) of outcome among participants. This method allows aggregation of outcomes measured using different scales (opioid consumption when combining different opioid agents, times when combining hours and minutes, score rating when using five-point or ten-point scales, etc.). In all other cases, outcome incidence analysis was performed using the risk ratios (RR). In order to include a maximum number of appropriate studies and avoid publication bias, incomplete data were obtained by contacting the corresponding author or estimation of the mean and the SD on the basis of the sample size, median, and range according to the method described by Hozo and collaborators [
14]. Where no validated method was identified to convert median and interquartile ranges to means and SD, data were discarded. In articles where outcomes were expressed as continuous variables, a partial standardized mean ratio was initially computed for each study, than transformed into partial odds ratio (OR) using Chinn’s formula [
15]: LnOR = 1.814 × SMD (Ln: natural logarithm). The data were then included as Ln(OR) and SD(LnOR) in the software (Review Manager 5 software). Overall SDM or RR (and 95% confidence intervals) were then calculated using the inverse variance method [
8]. Regarding common cut-off values for SMD, the Dex effect was considered small when the SMD was greater than −0.4, moderate when it was lying between −0.4 and −0.7, and large when it was smaller than −0.7 [
8].
Heterogeneity was assessed using
I
2 statistics. This approach describes the percentage of the variability in effect estimates (OR, MD, or SMD) that is due to heterogeneity rather than sampling error. According to the Cochrane Review guidelines [
8], the threshold for heterogeneity is an
I
2 > 40% and a
p < 0.1 and indicated the use of a random effect in OR and SMD computation rather than a fixed-effects model. The random-effects model assumes that the observed effects are estimating different intervention effects while a fixed-effects model estimates the same “true” intervention effect. Based on this principle, studies were weighted. In the random-effects model, all studies are equally weighted while in the fixed-effects model, each study is weighted according to the number of included patients. In addition, because of the potential effect of some confounding factors on results, subgroup analyses for Dex effect were performed (when at least two studies included the considered outcome for the considered subgroup) according to: the type of procedure, the mode of administration (bolus alone, or infusion with or without bolus), and the dose of bolus administered (threshold for defining low and high boluses was considered as the mean for number of included studies >30 or the median if the number of included studies <30). Finally, overall results were also computed in studies displaying low-risk bias for all checked items.
In order to confirm results of our meta-analysis on the primary outcome, a second set of analyses were performed using the trial sequential method [
16,
17]. This statistical method allows combining effects of studies and previous meta-analysis performed on the same subject to correct results (the adjustment of alpha-risk related to multiple comparison in previous meta-analyses), predict the possibility of a significant result in case of low power of the actual analysis and estimates the effect-size to be included in a meta-analysis (termed the information size for meta-analysis) to find a significant result. This analysis was performed on the freeware Copenhagen Trial Unit’s Trial Sequential Analysis Software, hereafter: TSA Software, Copenhagen, Sweden.
In studies with more than one intervention arm, in order to take into account all data, each arm was considered as a study and compared to the control group. However, given the weight taken by those studies in overall results, a sensitivity analysis was performed by removing one arm and another in order to assess the effect of these studies on outcome. Finally, to avoid calculation failure related to zero values in RevMan and TSA, a 1 or 0.001 was added to all groups when the number of events was equal to 0 in one group (for RevMan and TSA, respectively).
Statistical methods are available to assess the effects of unpublished studies on meta-analysis results (publication bias). Publication bias is assessed by studying the distribution of results on a funnel plot, which is a scatter plot of the intervention effect (RR, MD, or SMD) estimates from individual studies against some measure of each study’s size or precision (standard error of the intervention effect). Funnel plot asymmetry may indicate that some studies went unpublished [
18,
19]. This asymmetry can also indicate result heterogeneity or poor methodology in included studies [
18,
19]. According to the Cochrane collaborative guideline [
8], it is suitable to assess publication bias when analysis aggregates at least ten studies.
Results are expressed as RR, MD, or SMD (95% confidence interval), I
2, p value for I
2 statistics.