Dexamethasone and haemorrhage risk in paediatric tonsillectomy: a systematic review and meta-analysis

Transcription

1 British Journal of Anaesthesia 3 (): 3 4 (4) doi:.93/bja/aeu5 Dexamethasone and haemorrhage risk in paediatric tonsillectomy: a systematic review and meta-analysis J. R. Bellis *, M. Pirmohamed 4, A. J. Nunn 3,Y.K.Loke 5,S.De, S. Golder 6 and J. J. Kirkham 7 Research and Development, Department of Paediatric Otolaryngology, Head and Neck Surgeryand 3 Department of Women s and Children s Health, Institute of Translational Medicine (Child Health), University of Liverpool, Alder Hey Children s NHS Foundation Trust, Eaton Road, Liverpool L AP, UK 4 Department of Molecular and Clinical Pharmacology, University of Liverpool, Ashton Street, Liverpool L69 3GE, UK 5 Norwich Medical School, Faculty of Medicine and Health Sciences, University of East Anglia, Norwich NR4 7TJ, UK 6 Centre for Reviews and Dissemination, University of York, York YO 5DD, UK 7 Department of Biostatistics, University of Liverpool, Shelley s Cottage, Brownlow Street, Liverpool L69 3GS, UK * Corresponding author. jennifer.bellis@alderhey.nhs.uk Editor s key points The authors performed a systematic review to examine the effect of dexamethasone and NSAIDs on posttonsillectomy haemorrhage. They found no influence of these drugs on posttonsillectomy haemorrhage, but noted that further prospective research is needed. Summary. In children undergoing tonsillectomy, dexamethasone is recommended to reduce the risk of postoperative nausea and vomiting while non-steroidal anti-inflammatory drugs (NSAIDs) are used for pain relief. We aimed to determine whether children who receive dexamethasone or dexamethasone with NSAID are more likely to experience haemorrhage post-tonsillectomy. Randomized and non-randomized studies in which children undergoing tonsillectomy received dexamethasone or dexamethasone and NSAID were sought within bibliographic databases and selected tertiary sources. The risk of bias assessment and evaluation of haemorrhage rate data collection and reporting were assessed using the Cochrane Risk of Bias Tool and McHarm tool. Synthesis methods comprised pooled estimate of the effect of dexamethasone on the risk of haemorrhage rate using the Peto odds ratio (OR) method. The pooled estimate for haemorrhage rate in children who received dexamethasone was 6.%, OR.4 (95% confidence interval.89., P¼.5). There was risk of bias and inconsistent data collection and reporting rates of haemorrhage in many of the included studies. Clinical heterogeneity was observed between studies. The pooled analysis did not demonstrate a statistically significant increase in the risk of post-tonsillectomy haemorrhage with dexamethasone with/without NSAID use in children. However, the majority of the included studies were not designed to investigate this endpoint, and thus large studies which are specifically designed to collect data on haemorrhage rate are needed. Keywords: dexamethasone; paediatrics; tonsillectomy Children who undergo tonsillectomy or adenotonsillectomy are at risk of experiencing complications. These include postoperative nausea and vomiting (PONV) and postoperative haemorrhage. In studies where intraoperative anti-emetics were not administered, post-tonsillectomy nausea and vomiting rates as large as 7% have been reported. Post-tonsillectomy haemorrhage rates range from.% to 8.%. The Association of Paediatric Anaesthetists of Great Britain and Ireland (APA) guideline on the prevention of postoperative vomiting (POV) recommends the use of dexamethasone (.5 mg kg in combination with ondansetron.5 mg kg )to minimize the risk of POV in children undergoing tonsillectomy. 3 A Cochrane review of dexamethasone use in tonsillectomy supports this recommendation; children receiving a single intraoperative dose of dexamethasone at a dose between.5 and mg kg were half as likely to vomit within 4 h of their operation. 4 Similarly, a systematic review and metaanalysis of anti-emetic use in paediatric tonsillectomy determined that the most effective agents for the prevention of POV in the first 4 h after surgery were dexamethasone and the 5HT 3 antagonists including ondansetron. The dose range for dexamethasone was.5 mg kg and the combined odds ratio (OR) was.3 [95% confidence interval (CI).6.33]; the dose range for ondansetron was..3 mg kg and the combined OR was.36 (95% CI.6.46). 5 A study of five different doses of dexamethasone in paediatric tonsillectomy considered the following outcomes: vomiting, analgesic requirements, time to first liquid, and change in voice. The study found no difference between any of these outcomes at any of the five doses (.6 mg kg ) studied. 6 Although there is evidence of the efficacy of dexamethasone in the context of tonsillectomy, evidence of its safety is less well established. Czarnetzki and colleagues 7 showed that postoperative haemorrhage rates were increased by 6.5-fold (95% CI ) with intraoperative dexamethasone use in children undergoing tonsillectomy (a single dose of.5,.5, or.5 mg kg ). Three recent meta-analyses & The Author [4]. Published by Oxford University Press on behalf of the British Journal of Anaesthesia. All rights reserved. For Permissions, please journals.permissions@oup.com

2 BJA Bellis et al. have also addressed this issue: (i) using 4 randomized controlled trials (RCTs) of adults and children who underwent tonsillectomy, no difference in bleeding rates was demonstrated between dexamethasone and comparator arms (RR., 95% CI.65.6, P¼.9); 8 (ii) using out of 3 studies thatreported data on haemorrhage rates, 9 again no significant difference in postoperative bleeding was identified in children receiving single-dose dexamethasone vs placebo (OR.7, 95% CI.58.98, P¼.8); and (iii) using data from 9 RCTs of adult and paediatric tonsillectomy patients, a pooled effects estimate again revealed no significant difference in postoperative haemorrhage in patients who had received dexamethasone vs those who had not (OR.96, 95% CI.66.4, P¼.83). This review aims to determine whether the use of dexamethasone or dexamethasone with NSAIDs in paediatric tonsillectomy affects the rate of post-tonsillectomy haemorrhage in children. There are limitations to the use of standard systematic review methodology in the evaluation of adverse event outcomes. This is particularly true of post-tonsillectomy haemorrhage because it is a rare adverse event and therefore haemorrhage rate data derived from small RCTs of dexamethasone may not be generalizable. It has been recommended that systematic reviews of rare adverse effects should include nonrandomized studies (NRS) which may cover a broader population than RCTs and in which the adverse event may be the primary outcome. It is possible that haemorrhage rate data for dexamethasone used in this context may be unpublished (publication bias) or that it may have been selectively unreported due to undesirable outcome results (outcome reporting bias). 3 In order to address some of the limitations of standard systematic review methodology in the evaluation of a rare adverse event outcome, we included both RCTs and NRS and furthermore assessed the methodological quality of haemorrhage rate recording and reporting. Methods Search strategy Search strategies were developed specifically for each database (the search strategy for MEDLINE is presented in Supplementary Table S; details of similar strategies for the other databases can be provided from the authors upon request). The databases and tertiary sources used in this review are listed in Supplementary Table S. Searches were carried out in November. The reference lists of previous systematic reviews, identified during the search, were also examined for additional references. After the selection of studies via this process, forward and backward citation tracking was undertaken for each study if it was indexed in the Scopus database. Contact was made with experts to identify other potentially relevant published and unpublished studies. Study eligibility RCTs and quasi-rcts (q-rcts) that considered dexamethasone (except by peri-tonsillar infiltration) or, dexamethasone in combination with NSAIDs, in the context of paediatric tonsillectomy or adenotonsillectomy in the immediate perioperative period were included (note: for the remainder of this report, the term tonsillectomy will be used to encompass both tonsillectomy and adenotonsillectomy). The immediate perioperative period was defined as: within the 4 h before the procedure, during the procedure, or in the 4 h which followed the procedure. This review considered only children up to the age of 8 yr; studies that included both adults and children were also considered, and, if possible, only the data for children were used. Studies were only considered for inclusion if they were published in English. Types of outcome measure The main outcome of interest extracted from each study was haemorrhage rate. The definition of haemorrhage included any bleeding which required a change in postoperative management, for example, re-operation, blood transfusion, prolonged hospital stay, re-admission, or contact with a healthcare provider, for example, an accident and emergency department or a general practitioner. In some studies where participants may have experienced more than one haemorrhage, we recorded the number of haemorrhages rather than the number of patients who experienced a haemorrhage. This is because we considered recurrent bleeding to be a clinically important outcome in postoperative patients. Study selection Inspection of citations After duplicate citations were removed, all titles and abstracts were independently reviewed by two reviewers (J.R.B., J.J.K.) with reference to the inclusion/exclusion criteria (Supplementary Table S3), and a decision was made whether to retrieve the full report of the study. The number of titles/abstracts identified, accepted, and rejected was recorded. Inspection of retrieved reports Once the full reports were retrieved, they were inspected for relevance to the review and the inclusion and exclusion criteria applied. Studies not meeting the predetermined criteria were excluded. If there was any disagreement about whether to include any of the studies, a third reviewer (A.J.N.) assessed them and, together with the other reviewers, made a consensus decision about whether to include or exclude. A record was made of the number of papers retrieved and the number of papers excluded. For quality assurance purposes, 5% of studies excluded at title and abstract stage were re-reviewed (J.R.B., J.J.K.) for inclusion and five studies excluded at the full article stage were re-reviewed by M.P. or A.J.N. Data collection A formal data extraction form was designed, piloted independently on a small selection of studies, and adjusted as necessary. For each study, information regarding methods, participants, comparison groups, interventions, and outcomes was 4

3 Dexamethasone and haemorrhage risk in tonsillectomy BJA tabulated. Where they were recorded or provided by the author, the following data were extracted for each randomized study: (i) Study characteristics: number of participants, number of participants in each intervention group, year completed, setting, inclusion criteria, definition of postoperative haemorrhage, and length of follow-up. (ii) Participant characteristics: age, sex, underlying disease, indication for surgery. (iii) Interventions: number of intervention groups, intervention details: perioperative medicines inclusive of dose, surgical technique. (iv) Outcomes reported: postoperative haemorrhage rate. (v) Additional data relating to haemorrhage: severity, timing (i.e. need for intervention, primary or secondary). Where it was recorded or provided by the author, the following data were extracted for each non-randomized study: (i) Study characteristics: number of participants, number of participants in each intervention group, year completed, setting, inclusion criteria, definition of postoperative haemorrhage, and length of follow-up. (ii) Participant characteristics: age, sex, underlying disease, indication for surgery. (iii) Characteristics of surgery: surgical technique, perioperative medicines inclusive of dose. (iv) Characteristics of perioperative care: medicines. (v) Data relating to haemorrhage: number of haemorrhages, severity, timing (i.e. need for intervention, primary or secondary), risk factors identified. Quality assessment For RCTs that compared dexamethasone with another intervention and reported haemorrhage rate or for which haemorrhage rate data were obtained from the author(s), the methodological quality was assessed using the Cochrane Collaboration s domain-based evaluation tool for assessing risk of bias. 4 The methodological quality of haemorrhage rate recording and reporting was assessed for both randomized and NRS using selected elements of the McMaster Quality Assessment Scale of Harms for primary studies (the McHarm Scale), The elements used were selected based on an evaluation of their relevance to our research question and they aimed to evaluate: the quality and appropriateness of study design and reporting, the applicability of the study findings to the population, and measures taken to reduce bias (Supplementary Table S4). 5 Both data collection and quality assessment of studies were undertaken byone reviewer (J.R.B.), with three randomized and three NRS assessed by a second reviewer (J.J.K.) to check for consistency. Statistical analysis and synthesis Statistical analyses were performed using RevMan (version 5. software). As haemorrhage rate data are dichotomous, the data were analysed by calculating the Peto OR for each randomized study and for NRS, ORs were calculated with the corresponding 95% CIs. For each study, we only included data for participants who were not excluded after randomization and for whom follow-up was complete. Meta-analysis We aimed to conduct two meta-analyses. For RCTs, dexamethasone alone was compared with any other intervention used in paediatric tonsillectomy. For NRS, dexamethasone alone was compared with anyother intervention used in paediatric tonsillectomy. Heterogeneity and subgroups Ifclinical heterogeneitywastoo great, studies were notpooledina meta-analysis and narrative synthesis was used to compare studies. A x test for statistical heterogeneity was undertaken, and the I statistic was calculated. Where the necessary data were available, the following subgroup analysis was also planned: a comparison of primary and secondary haemorrhage rates; a comparison of studies in which some participants received NSAIDs in addition to dexamethasone with those in which no participants received NSAIDs. Studies in which all participants received dexamethasone For RCTs: report the haemorrhage rate for each arm of the trial. For NRS: report the haemorrhage rate for each study. Publication bias Publication bias for the randomized trials included in the meta-analysis was assessed by visual inspection of a funnel plot. All study authors were contacted where possible for missing outcome data. Results Search results The database searches undertaken in November identified 349 abstracts for screening after duplicate records were removed. After review of abstracts, 39 full-text articles were reviewed. Of these, 5 unique studies (37 RCTs and 5 NRS) fulfilled the inclusion criteria (Fig. ). Forward and backward citation tracking for all the eligible articles in the database search, plus the reference sections of 5 review articles identified in our database search, identified 96 potentially relevant citations not identified in our initial searches. After review of abstracts, 88 articles were excluded, leaving 8 full articles to be reviewed. Of these, additional articles (three RCTs and seven NRS) met the inclusion criteria (Fig. ). One article reported an RCT already identified in another article picked up by our database search. In total, 6 studies were included in this review (39 RCTs and NRS).

4 BJA Bellis et al. Included Eligibility Screening Identification Records identified through database searching (n=433) Records after duplicates removed (n=349) Records screened (n=349), six required third reviewer Full-text articles assessed for eligibility (n=39) Eligible articles (n=64 representing 59 studies) Included articles (n=57 representing 5 studies, 37 RCT and 5 NRS) Additional records identified through other sources (n=95 Medscape) Records excluded (n=38) Full-text articles excluded, with reasons (n=73) Full-text articles could not be obtained (n=) Integrity of author questioned (n=) Combined data on children and adults (n=4) 88 9 Ongoing trial (n=) 8 83 Fig Database search flow diagram. Included studies Of the 39 RCTs, 3 compared dexamethasone with another intervention; of these, 7 studies involving 973 participants did not report haemorrhage rate. In the remaining seven randomized studies, all participants received dexamethasone, and of these, four did not report haemorrhage rate (588 participants)

5 Dexamethasone and haemorrhage risk in tonsillectomy BJA Screening Identification Records from backward citation tracking for 6/64 titles (n=99) Records after duplicates removed (n=487) Records not found in our search (n=38) Records from forward citation tracking for 54/64 titles available in Scopus (n=3) Records after duplicates removed (n=49) Records not found in our search (n=33) Duplicates removed and titles and abstracts screened (n=96 ) Records from backward citation tracking for the 5 reviews identified (n=684) Records after duplicates removed (n=485) Records not found in our search (n=369) Records excluded (n=88) Records excluded (n=67) Eligibility Full text assessed for eligibility (n=8) Could not obtain (n=) 9 Full text non-english (n=) 93 Incorrectly referenced (n=) Eligible articles (n= representing studies) Combined data on children and adults (n=) Included Eligible articles (n= representing studies, 3 RCT and 7 NRS) Fig Citation tracking flow diagram. 7

6 BJA Bellis et al. Table Randomized studies which compared dexamethasone with another intervention and reported haemorrhage rate. *Count of bleeds, there were patients affected by 6 bleeds. Category : history of bleeding leading to readmission but without evidence of bleeding at re-admission. Category : readmission due to bleeding with evidence if bleeding at medical examination but no need for reoperation. Category 3: emergency reoperation due to bleeding. Level I: all children who reported to have any history of postoperative haemorrhage, whether or not there was clinical evidence. Level II: all children who required inpatient admission for postoperative haemorrhage regardless of the need for operative intervention. This level excludes children undergoing evaluation in the emergency department for reported postoperative haemorrhage who had no evidence of clot formation or haemorrhage and were deemed safe discharge. Level III: all children who required return to the operating department forcontrol of postoperative bleeding. Cat, category; dex, dexamethasone; ED, emergency department; NSAID, non-steroidal anti-inflammatory drug; Obs, observations; postop, postoperative; temp, temperature Study No. of participants No. of participants in analysis Length of follow-up Primary outcome(s) Interventions Participants per intervention group Dissection technique Haemostasis technique Alajmi and days Postop pain, nausea, colleagues 6 vomiting, and oedema April and h Postop oral intake, colleagues 9 pain, vomiting, temp, and complications Dex Other Dex vs placebo 4 38 Sharp dissection Packs or sutures, electrocautery if persistent bleed Dex vs placebo 4 39 Electrodissection Suction cautery Catlin and Grimes 9 7 days Length of stay, i.v. fluid requirement, pain, nausea, emesis, fever, postop analgesia, complications, appetite Czarnetzki and 7 days Prevention of postop colleagues 7 nausea and vomiting at 4 h Gallagher and days Rate and severity of colleagues post-tonsillectomy haemorrhage Dex vs placebo 5 Adenoidectomy by curette, excision of tonsils by sharp and blunt dissection and snare Dex vs placebo Cold steel, electrocautery Dex vs placebo 54 5 Monopolar electrocautery and a spatula tip Electrocautery Gauze compression, electrocautery Suction cautery Giannoni and 5 5 days Postop pain assessment Dex vs placebo Electrocautery Not reported colleagues 6 Hanasono and 3 days Oral intake, pain scores, colleagues 8 vomiting Holt and 3 6 days Postop nausea and colleagues 9 vomiting Dex vs placebo 6 3 Electrocautery or sharp wire snare transection Dex+tropisetron vs tropisetron Electrocautery or directed cautery Sharp dissection Suture ligation Kaan and colleagues h Early oral intake, pain and vomiting Malde and days Postop pain and postop colleagues 45 nausea and vomiting Mohammad and h Vomiting, trismus, pain, colleagues 4 fever, time to first solid intake, primary haemorrhage Nawasreh and 4 h Temp, vomiting, oral colleagues 4 intake Dex vs placebo 3 3 Sharp dissection Suture ligation Dex vs placebo Sharp dissection snare technique Dex vs placebo Sharp dissection snare technique Dex vs placebo 6 58 Electrocautery dissection, enlarged adenoid removed by shaving+currette Ohlms and colleagues days Pain scores Dex vs placebo Sharp dissection snare technique, adenoid removed using curettes if indicated Ligation using ties, packs, or sutured Electrocautery or ligation with silk Not reported Using packs, electrocautery if persistent bleeding Pappas and h from colleagues 46 discharge Postop nausea and vomiting Volk and colleagues days Fever, mouth odour, oral intake, pain, activity, weight loss, trismus, and analgesic usage Dex vs placebo Electrodissection Not reported Dex vs placebo 4 Combination of blunt and sharp dissection Suction cautery 8

7 Dexamethasone and haemorrhage risk in tonsillectomy BJA Table Continued Dexamethasone dose Perioperative NSAID Postoperative haemorrhage rate Additional information about haemorrhages Contact with Primary Secondary Total Definition of haemorrhage Severity author Dex Other Dex Other Dex Other Dex Other mgkg mgkg (max 6 mg) 3 3 None None All re-admitted None profinal (ibuprofen) 5 mg kg p.o. if required None Primary,4 h, secondary 4 days Treated with suction and silver nitrate but not admitted 8 mgm None None day 5, neither required treatment Day 7, admitted and required Obs and i.v. fluids day 6, did not require treatment Author provided data on NSAID use None.5.5 mg kg Yes some patients 5 9 4* Cat 3 7 cat ; 8 cat ; 9 cat 3 cat ; cat None level I;.5 mg kg None Severity levels I (max mg) III 3 level II; 3 level III mgkg (max Single preop 6 mg) dose ibuprofen 5 mg kg mgkg None None Day 3 requiring re-admission.5 mg kg (max 8 mg).5 mg kg (max 6 mg) 7 level I; 5 level II; level III None None None Day requiring re-admission None None None None Author could not provide data on timing of bleeds by intervention None None Author provided data on bleeds.5 mg kg None None Day 4 None None None mgkg (max. mg) Diclofenac i.v. if required None None None mgkg (max 6 mg) Not reported Primary,4 h; secondary postoperative day 4 None None Author provided data on timing of bleeds.5 mg kg (max mg) mgkg (max mg) None 3 3 None Day 3: cauterized in ED; day : observed for 4 h; day : cauterized in theatre None None None Author provided data on bleeds mg None None Minor: controlled in the operating theatre Minor: controlled in the operating theatre Author confirmed that these were delayed bleeds 9

8 BJA Bellis et al. Nine of the NRS compared dexamethasone with another intervention Ofthese,sixstudiesinvolving688participants did not report haemorrhage rate In the remaining 3, all participants received dexamethasone 6 74 and of these, 64 7 two did not report haemorrhage rate (8 participants). Randomized studies which compared dexamethasone with another intervention and reported haemorrhage rate These5studiesinvolving693participants are summarized in Table. They were all published between 99 and, the length of follow-up ranged from 4 h to 6 days, and all of the studies compared dexamethasone with placebo. Seven of the studies used sharp dissection to remove the tonsils, three used more than one method, while the remainder used electrodissection The dose of perioperative dexamethasone ranged from.5to mg kg ; in one study, all childrenreceived mg rather than a weight-based dose. 43 At least some of the participants of four of the studies 7664 received NSAIDs. The primary outcome in the majority of studies was PONV, postoperative pain, or both. Haemorrhage rate and severity was the primary outcome in one study. Four studies predefined haemorrhage 74 and seven reported additional information about the haemorrhages they detected The risk of bias assessment using the Cochrane risk of bias tool was completed for these studies (Table ): had a high overall risk of bias, and the remainder had an unclear overall risk of bias The results of the McHarm scale assessment were as follows: only one of these studies predefined haemorrhage, four actively collected data on haemorrhage rate, two passively collected data on haemorrhage rate, 7 and one used a standard checklist for haemorrhage rate data collection. It was unclear whether there was a possibility of selective outcome reporting bias for five of these studies. The overall haemorrhage rate for participants who received placebo ranged from % to 8.6% and the overall haemorrhage rate for participants who received dexamethasone ranged from % to 5.6%. The pooled estimate of haemorrhage rate for children who received dexamethasone was 6.%. Of the 5 studies that reported data on haemorrhage rate, there was a non-significant increase in the risk of haemorrhage for the dexamethasone intervention group (Peto OR.4, 95% CI.89., P¼.5) (Fig. 3). Fourteen out of 5 studies reporting data on haemorrhage separated the data into primary and secondary haemorrhage rates. For primary haemorrhage, only seven events were observed in the dexamethasone group and three on placebo; the pooled estimate demonstrated a non-significant increase in haemorrhage rate (Peto OR.4, 95% CI , P¼.6; Fig. 4). For secondary haemorrhage, the pooled estimate again suggested that there was a non-significant increase in the risk of haemorrhage for the dexamethasone intervention group (Peto OR.4, 95% CI.86.35, P¼.7; Fig. 4). In the four studies in which some children also received NSAIDs, the pooled estimate indicated that there was a nonsignificant increase in the risk of haemorrhage for the dexamethasone intervention group (Peto OR.56, 95% CI , P¼.8). For the eight studies in which no children received NSAIDs, again there was a non-significant increase in the risk of haemorrhage in the dexamethasone group (Peto OR.3, 95% CI.73.37, P¼.36) (Fig. 5). A funnel plot of the studies included in our meta-analysis shows no evidence of publication bias (Fig. 6). Randomized studies in which all participants received dexamethasone and reported haemorrhage rate These three studies involved 445 participants and were published between 3 and. Rawlinson and colleagues 5 followed children for 4 days, Derkay and colleagues 47 for 4 h, and O Flaherty and Liu 49 for 4 h. The primary outcome in all three studies was postoperative pain, PONV, or both. In Rawlinson and colleagues study, 5 all children received dexamethasone.4 mg kg and the electrocautery and microdebrider dissection techniques were compared; no children received NSAIDs. Derkay and colleagues administered dexamethasone. mg kg to all participants and tonsillectomy was undertaken by cold steel dissection. In their study which compared magnesium and ketamine use, O Flahertyand colleagues administered. mg kg dexamethasone to all participants, but did not report the tonsillectomy technique(s) used and their reporting of early postoperative haemorrhages was ambiguous (Table 3). One study predefined haemorrhage, 47 and another actively collected haemorrhage rate data and specifying the timing and frequency of haemorrhage rate data collection. 49 The haemorrhage rates in two of these studies ranged from % to %. The rate was unclear in one of these studies. 49 NRS which compared dexamethasone with another intervention and reported haemorrhage rate There were 88 participants in these three studies published between 999 and. They were all retrospective case note reviews and two of them retrieved week follow-up data. In one of the studies, a single method of tonsil dissection was used, 6 while in the other studies, there were several methods recorded In the retrospective chart review conducted by Conley and Ellison, 54 children who received dexamethasone were operated on using a standard surgical technique cold-knife dissection and snare and haemostasis was achieved using tonsillar packs dipped in bismuth subgallate phenylephrine hydrochloride mixture followed by suction electrocautery and a 3 min observation period. However, children who did not receive dexamethasone were operated on using either cold-knife dissection, snare, and suction electrocautery or electrocautery dissection; tonsillar packs were not used in any of these children. The dexamethasone doses administered in these three studies ranged from.4 to.6 mg kg. The report of one study specifies that dexamethasone was administered or withheld according to the anaesthetist s preference. 6 NSAIDs were administered 3

9 3 Table Risk of bias for randomized studies which compare dexamethasone with another intervention and reported haemorrhage rate. Dex, dexamethasone; ED, emergency department; NSAID, non-steroidal anti-inflammatory drug; POD, postoperative day; PONV, postoperative nausea and vomiting; postop, postoperative Study Alajmi and colleagues 6 How was allocation sequence generated? Unclear first two patients were given dex and the rest were given saline on the next operation theatre this was reversed How was allocation sequence concealed? What measures were taken to blind participants and personnel? Unclear states that participants were blinded What measures were taken to blind outcome assessors? Unclear we know some personnel knew who had which intervention, do not know who did observations Is the outcome data complete? Did the authors report exclusion and attrition and give reasons for these? Low outcomes reported for all participants Catlin and Grimes Unclear Unclear Unclear Unclear 9 patients recruited but four lost to follow up, unclear which intervention group these four were in Czarnetzki and colleagues 7 Gallagher and colleagues Giannoni and colleagues 6 Unclear randomization was done in blocks of 4 children ( per group) study medications were produced and randomized Low random number generator Low random number generator (Excel) Unclear see previous comment Low carried out by hospital pharmacy Low study drug was supplied as syringes of a liquid, identical in colour and volume but designated by a letter, to one of the study groups Low indistinguishable syringes Low identical packaging of dex and placebo Low see previous comment Low anaesthetist did not know what had been given so could not tell nurses/surgeons/ parents what had been given Low anaesthetist, surgeon, patients, guardians, data collectors were blinded Low see previous comment, states that all physicians, nurses, patients, parents, and others caring for the subjects were blinded to the Low patients lost to follow-up or excluded because did not meet inclusion criteria are Low 9 excluded, 3 received additional postop steroid, 6 lost to follow-up but clear which intervention groups these 9 were in 3 patients, all from the dex group had data collection on the day of surgery but did not complete the evaluation period required steroid injection for asthma exacerbation on Is there a possibility of selective outcome reporting? outcomes not reported for each follow-up visit, haemorrhage only reported in the context of readmission authors state that no re-admission signified no complication but do not report what was recorded at POD 7, POD, and POD 6 follow-ups reporting of pain and nausea and vomiting rates is incomplete Low all endpoints are reported on albeit with limited detail for some data on secondary outcomes not fully reported Low all endpoints are reported on Are there any other potential sources of bias? Low none Low none questionnaire at home was not completed for 3 children so do not know if had NSAIDs, minor bleed, or PONV that has not been recorded. There could be partial completion of some of the returned questionnaires. Early termination of this trial may have exaggerated harm Unclear strict instructions to return with bleeding but might have gone elsewhere or not attended if only minor Low none Summary assessment of risk of bias Continued Dexamethasone and haemorrhage risk in tonsillectomy BJA

10 3 Table Continued Study Hanasono and colleagues 8 Holt and colleagues 9 Malde and colleagues 45 Mohammad and colleagues 4 How was allocation sequence generated? Unclear random number list does not describe how it was generated Low random number generation tables Low computer-generated random number table How was allocation sequence concealed? Unclear study medication was supplied in a blinded manner and medication records were maintained in pharmacy until the end of the study Low packed in pharmacy and numbered according to randomization What measures were taken to blind participants and personnel? Unclear see previous comment Low two ampoules per study arm (saline instead of dex in one group) What measures were taken to blind outcome assessors? assignment until the conclusion of the study Unclear see previous comment Low anaesthetist took no part in outcome assessment, this was undertaken by nursing staff who did not know which patients had received which intervention Low undertaken by second anaesthetist who was unaware about the drug administered Is the outcome data complete? Did the authors report exclusion and attrition and give reasons for these? day and could not be contacted after day so data for these 3 patients were excluded from analysis fromdaysto outcome measures for pain, emesis, and oral intake only reported for 73 of 9 participants Low 3 enrolled 7 excluded: 3 received propofol, 4 tonsillectomy cancelled follow-up details were not available for six patients of each group does not specify from which point, assume means post-discharge follow-up Unclear method of scoring pain is not well Is there a possibility of selective outcome reporting? Low all outcomes are reported on followed up on day 6 no report of haemorrhage rate despite this being reported for patients in the pre-discharge period Low all outcomes reported on patients observed for 4 h and discharged the day aftersurgery butstudy reports outcome data for secondary haemorrhage, oral intake at 36 and 7 h, note the statement about contacting Are there any other potential sources of bias? only followed up for 3 days so would not observe haemorrhage occurring after that. Notice that patients were only asked about postop problems leading to unplanned office or ED visit on POD 3 minor bleeds might not have been reported 5 patients could not be contacted so no 6 day follow-up data available for them 3 tropisetron, tropisetron+dex Low none Low none Summary assessment of risk of bias BJA Bellis et al.

11 33 Nawasreh and colleagues 4 Pappas and colleagues 46 Unclear patients were divided into two groups Low computer-generated table first group received dex, second group received placebo Volk and colleagues 43 April and colleagues 9 Kaan and colleagues 3 Low table of random numbers distributed in blocks of six according to diagnosis Low was used Low patients weight and diagnosis list was sent to pharmacy and a syringe was prepared based on the random number Ohlms and colleagues 44 see previous comment states administered in a randomized double-blind fashion study drugs were prepared by pharmacy but do not know if they looked identical Unclear cannot tellif both study drugs looked the same Low numbered otherwise unmarked syringe containing colourless dex or saline Unclear all procedures were performed in a double-blind fashion no details of how Low medication prepared in pharmacy and administered in a double-blind fashion Low parents undertook observations they did not know which intervention the child had received Unclear independent observer no details of how or whether they were blinded Low see previous comment Low outcomes reported for all participants Low had to be excluded both were from the dex group had problems with compliance with the postoperative questionnaire so only got complete follow-up for 9/ dex and 6/4 placebo Low outcomes reported for all participants Low outcomes reported for all participants Low outcomes reported for all participants people by phone only if they had the facility Unclear reporting incomplete about pain medication requirements do not report compliance with analgesic regime at home Low all outcomes reported for the patients followed up Unclear reporting incomplete about pain medication requirements and pain rating Low all outcomes are reported on Low all outcomes reported for all patients Unclear follow up beyond discharge for adverse events seems to have only been if patients presented with a problem Unclear relied on all parents completing diary adequately and reporting accurately Low none Low none Unclear instructed to return if bleeding, + vomiting or inadequate oral intake might have decided not to return so would not have picked up these adverse events Unclear although haemorrhage rate reported, looks like only detected if patients presented Unclear Unclear Unclear Dexamethasone and haemorrhage risk in tonsillectomy BJA

12 BJA Bellis et al. Study or subgroup Catlin (99) Volk (993) Ohlms (995) April (996) Nawasreh () Holt () Giannoni () Hanasono (4) Malde (5) Mohammad (6) Czarnetzki (8) Alajmi (8) Gallagher () Dexamethasone Control Peto odds ratio Events Total Events 3 3 Total Weight 3.7% 4.% 4.%.8% 5.5% 4.%.8%.8%.4%.4% 4.6% 4.% 38.4% Peto, Fixed, 95% CI Year 3.36 [.3, 37.5] 99.9 [.9, 9.4] [.8, 8.5] [.6, 5.48] [.3, 6.8].45 [.5, 4.44]. [.6, 6.45].9 [.7, 7.67] 4.4 [., 6.8] 5.4 [., 6.8] 7.9 [.4, 7.46] 8. [.,.4] 8.3 [.6,.79] Peto odds ratio Peto, Fixed, 95% CI Total (95% CI) 783 Total events 54 9 Heterogeneity: c =3.7, df= (P=.3); I =3% Test for overall effect: Z=.45 (P=.5) 677.%.4 [.89,.].. Control increased risk Dex increased risk The following reported zero haemorrhages in both intervention groups: Pappas (998) dexamethasone (/63) control (/65) Kaan (6) dexamethasone (/3) control (/3) Fig 3 Haemorrhage rates for randomized studies which compared dexamethasone with another intervention. to some of the participants in one study. 58 The primary outcome was haemorrhage rate for all of these studies. Two of these studies predefined haemorrhage 54 6 and all reported their haemorrhages in detail, providing information on the need for and types of re-intervention (Table 4). All of these studies predefined haemorrhage, but none actively collected data on haemorrhage rate. None of these studies used a standard checklist for haemorrhage rate data collection and for all of them, it was unclear whether there was a possibilityof selective outcome reporting bias. The haemorrhage rates for participants who did and did not receive dexamethasone ranged from.% to 8.3% and 3.8% to 9.7%, respectively (Fig. 7). NRS in which all participants received dexamethasone and reported haemorrhage rate These studies included 6 participants, and were published between and. They comprise five retrospective chart reviews, three prospective observational studies, two audits, 7 7 and a retrospective analysis of data from an RCT. 6 The period of follow-up in these studies ranged from 6 h to 3 days and seven reported the technique used to remove the tonsils The primary outcome was haemorrhage rate in two of these studies, 6 65 while most of the remaining studies examined postoperative pain and vomiting. Dexamethasone doses ranged from.6 to.5 mg kg and NSAID use was reported in four of the studies (Table 5). Two of the studies predefined haemorrhage, one undertook passive haemorrhage rate data collection, 65 the timing and frequency of haemorrhage rate data collection was specified by two authors, 67 7 and there was a possibility of selective outcome reporting in two studies The haemorrhage rate in these studies ranged from.4% to 5.7%. Discussion Summary of findings Data from the 5 RCTs included in our meta-analysis and the three NRS in which dexamethasone was compared with another intervention indicate that the overall risk of posttonsillectomy haemorrhage in children is not significantly increased by the perioperative use of dexamethasone. Limitations of included studies The overall risk of bias was high or unclear for all of the included RCTs. Among both the randomized studies and NRS, there was clinical heterogeneity. Some of the elements of study design that have an impact on haemorrhage risk are: dissection technique, haemostasis technique, 75 patient age, gender, and the indication for surgery and the perioperative use of NSAIDs. 77 Our evaluation of haemorrhage rate data collection and reporting indicates that this was inadequately reported in the majority of studies. Many studies did not follow children beyond the point at which primary outcomes would be measured. Studies which relied on re-admission as a method of haemorrhage detection may have missed minor self-limiting bleeds or bleeds attended to at other healthcare facilities. Studies which used patient and parent questionnaires to detect bleeding episodes may have missed haemorrhages due to questionnaires being incomplete or unreturned. Very few studies active surveillance for adverse outcomes; for example, a telephone call or face-to-face contact 34

13 Dexamethasone and haemorrhage risk in tonsillectomy BJA Dexamethasone Control Peto odds ratio Study or subgroup Events Total Events Total Weight Peto, Fixed, 95% CI Number of primary haemorrhages Mohammad (6) Czarnetzki (8) Gallagher () Subtotal (95% CI) Total events Heterogeneity: c =.49, df= (P=.9); I =% Test for overall effect: Z=.5 (P=.6) % 5.3% 5.6%.4% Year.4 [., 6.8] [.5, 9.93] 8.98 [.4, 7.3].4 [.38, 5.36] Peto odds ratio Peto, Fixed, 95% CI Number of secondary haemorrhages Catlin (99) Volk (993) Ohlms (995) April (996) Nawasreh () Giannoni () Hanasono (4) Malde (5) Mohammad (6) Czarnetzki (8) Alajmi (8) Gallagher () Subtotal (95% CI) Total events Heterogeneity: c =.7, df= (P=.39); I =6% Test for overall effect: Z=.38 (P=.7) % 4.% 4.%.8% 5.6%.8%.8%.4%.4%.6% 4.% 34.% 87.6% 3.36 [.3, 37.5] 99.9 [.9, 9.4] [.8, 8.5] [.6, 5.48] [.3, 6.8]. [.6, 6.45].9 [.7, 7.67] 4.4 [., 6.8] 5.4 [., 6.8] 7.55 [.9, 7.4] 8. [.,.4] 8.37 [.6, 3.5].4 [.86,.35] Total (95% CI) % Total events 53 8 Heterogeneity: c =4.9, df=4 (P=.44); I =% Test for overall effect: Z=.48 (P=.4) Test for subgroup differences: c =., df= (P=.), I =%.4 [.89,.7].. Control increased risk Dex increased risk Holt () reported only overall haemorrhage rate. The following reported zero primary haemorrhages in both intervention groups: Volk (993), Ohlms (995), April (996), Pappas (998), Nawasreh (), Giannoni (), Hanasono (4), Malde (5), Kaan (6), Catlin (6), Alajmi (8) The following reported zero secondary haemorrhages in both intervention groups: Pappas (998), Kaan (6) Fig 4 Subgroup analysis: primary and secondary haemorrhage rates. whereby participants were specifically asked about any haemorrhages experienced. Comparison with results of previous studies In our meta-analysis of 5 RCTs, the pooled estimate risk ratio for haemorrhage in children who received dexamethasone was.4 (95% CI.89., P¼.5; I ¼3%). The interpretation of our finding conforms to that of three previous meta-analyses of randomized studies, although the difference in the summary statistic reflects differences in study identification and inclusion. Geva and Brigger 8 calculated a relative risk of haemorrhage of. (95% CI.65.6, P¼.9) for adult and paediatric patients in their meta-analysis who received dexamethasone. The haemorrhage rate among those who received dexamethasone was 5.9%. Their analysis included 4 studies, nine of which were also included in our analysis; of the five that were not included in our analysis, three included adult participants only, one was in Hebrew, and for one, we were not able to obtain sufficient detail from the author about the participants who experienced a haemorrhage. A meta-analysis of studies involving only children calculated an OR of haemorrhage for children who had received dexamethasone compared with those who did not (OR¼.7; 95% CI.58.98, P¼.8). 9 Their analysis included studies; of which, were the same as those in our review. The two studies we did not include were those where we could not obtain a haemorrhage rate from the authors. The haemorrhage rate for children who received dexamethasone in this study was 6.%. Finally, the most recent meta-analysis calculated an OR of.96 (95% CI.66.4, I ¼%). This review included 9 studies of systemic steroid use in tonsillectomy and haemorrhage rate in adults and children. The haemorrhage rate for patients who received systemic steroids was 4.6%. We included 3 of these 9 plus an additional study 35

14 BJA Bellis et al. Dexamethasone Control Peto odds ratio Study or subgroup Events Total Events Total Weight Peto, Fixed, 95% CI Year Number of haemorrhages in studies in which some participants received NSAIDs Giannoni () Mohammad (6) Alajmi (8) Czarnetzki (8) Subtotal (95% CI) Total events Heterogeneity: c =8.4, df=3 (P=.4); I =64% Test for overall effect: Z=.7 (P=.8) %.5% 4.3%.7% 34.4%. [.6, 6.45].4 [., 6.8]. [.,.4].9 [.4, 7.46].56 [.69, 3.5] Peto odds ratio Peto, Fixed, 95% CI Number of haemorrhages in studies in which none of the participants received NSAIDs Catlin (99) Volk (993) Ohlms (995) April (996) Holt () Hanasono (4) Malde (5) Gallagher () Subtotal (95% CI) Total events Heterogeneity: c =6.4, df=7 (P=.49); I =% Test for overall effect: Z=.9 (P=.36) % 4.% 4.3%.9% 5.7%.9%.5% 4.% 65.6% 3.36 [.3, 37.5].9 [.9, 9.4] 8.9 [.8, 8.5].95 [.6, 5.48].3 [.4,.33].9 [.7, 7.67].4 [., 6.8].3 [.6,.79].3 [.73,.37] Total (95% CI) 7 69.% Total events 5 8 Heterogeneity: c =4.75, df= (P=.9); I =% Test for overall effect: Z=.37 (P=.7) Test for subgroup differences: c =., df= (P=.74), I =%.39 [.87,.].. Control increased risk Dex increased risk The following reported zero haemorrhages in both intervention groups: Pappas (998), Kaan (6) For the following it was unclear whether NSAIDs had been administered: Nawasreh () Fig 5 Subgroup analysis: haemorrhage rates with NSAID use..5.5 SE(log[Peto OR]) Peto OR.. Fig 6 Funnel plot. they did not identify 46 and another where the results were published more recently. Of the 6 studies, we did not include: five were not in English, six involved adults only, for three we could not obtain data on haemorrhage rate, one reported combined data on adults and children, and in one the intervention was prednisolone rather than dexamethasone. Data not included in previous reviews We included two RCTs not included by previous systematic reviews. One of these studies was identified in both the EMBASE and MEDLINE. Although it did not report haemorrhage rate, we obtained the necessary data from the author. 46 The results of the second randomized study by Gallagher and colleagues were published after the previous reviews. This was the only randomized study included here which had the rate and severity of post-tonsillectomy haemorrhage as its primary outcome measure. The studywas a non-inferioritystudy(onesided test). Non-inferiority was tested for haemorrhage events in each of three groups (levels I III, see footnote to Table for definitions) rather than grouping all haemorrhage events together. The hypothesis was that dexamethasone would not increase haemorrhage rate by more than 5% (rather than that there would be no difference between the two groups). Noninferiority was not shown for level I bleeding events, but it was demonstrated for both level II and level II bleeding events. Unsurprisingly, the assessments of risk of bias and methodological quality show that the risk of bias for this study is generally low, and overall, the methodology for haemorrhage rate detection and reporting was the most robust. Consequently, the haemorrhage rate in this study is at the upper end of the range for our 36

15 Dexamethasone and haemorrhage risk in tonsillectomy BJA Table 3 Randomized studies in which all participants received dexamethasone and in which haemorrhage rate was reported. Dex, dexamethasone; NSAID, non-steroidal anti-inflammatory drug; PONV, postoperative nausea and vomiting; postop, postoperative Contact with authors Postoperative haemorrhage rate Dex dose Perioperative NSAID Haemostasis technique Interventions Dissection technique Primary outcome(s) Length of follow-up Study No. of participants Primary Secondary Total 3 None Not reported Not reported Not reported.4 mg kg (max mg) Derkay and 5 4 days Postop pain All had dex Electrocautery Suction colleagues 47 electrocautery 5 patient had Microdebrider Not Not reported reported 4 h Postop pain All had dex Not reported Not reported. mg Unclear None kg ibuprofen none Unclear O Flaherty and, ketamine/ colleagues 49 placebo, magnesium/ placebo patient had Unclear ibuprofen None Unclear 9, ketamine/ magnesium 7, placebo/ placebo. mg Ibuprofen 5 mg None kg kg Cold steel Bipolar diathermy All had dex+ondansetron Rawlinson and 3 4 h PONV and colleagues 5 pain 3 Diclofenac mg kg i.v. randomized studies (9.6%). We calculated the overall relative risk of haemorrhage with dexamethasone administration in this study and found a non-significant increase in risk:.3 (95% CI , P¼.5653). The categorization of haemorrhages by severity level was one strength of this study; this was undertaken by only six of our included randomized studies. When haemorrhages are categorized in such a way, the results become useful to clinicians and policy-makers. In Gallagher and colleagues study, although there were a total of 3 haemorrhages, 8 did not require any change in postoperative care and only four required re-operation. The evaluation of haemorrhage rate data from NRS was not conducted in previous reviews. Our evaluation of three studies provides conflicting results. The data for two of these studies indicated that children in the control group were more likely to experience a haemorrhage than those who received dexamethasone In one of these studies, there were too many differences in the way the two intervention groups were managed to draw any meaningful conclusions about the effects of dexamethasone administration on haemorrhage risk. 54 In a third study, there was no difference between the intervention groups, but dexamethasone had been administered according to anaesthetist preference (no further details available), and it is therefore possible that children with an increased risk of haemorrhage were not administered dexamethasone. 6 All three studies were retrospective chart reviews using hospital records. Although they report haemorrhage rates which correspond to those quoted in the literature, there is some evidence that retrospective chart review underestimates haemorrhage rate. 78 Conclusions We did not find any evidence that dexamethasone statistically or clinically significantly increases the risk of post-tonsillectomy haemorrhage. However, the degree of imprecision in our pooled estimate of the OR (upper bounds of the 95% CI were.) prevents us from ruling out a clinically relevant doubling in risk. There were insufficient data to determine any additional impact of NSAID use, an issue which is likely to gain greater prominence, given the recent moves away from the use of codeine as an analgesic in children after tonsillectomy. In the studies included here, we detected inadequacies in haemorrhage rate detection and reporting. Further large studies (both randomized and observational) are needed to provide evidence about the safety of dexamethasone+nsaids in paediatric tonsillectomy. For instance, a factorial trial would be an elegant method of addressing both dexamethasone and NSAID use, individually and in combination. The findings of four ongoing trials will provide additional data on outcomes for the use of dexamethasone and ibuprofen in paediatric tonsillectomy. Future research studies need to have haemorrhage rate as their primary outcome with a consensus group of clinicians, caregivers, and patients agreeing beforehand on predefined levels of severity that can be used in research worldwide. 37