Adenoidectomy with or without grommets for children with otitis media: an individual patient data meta-analysis

Article history

The research reported in this issue of the journal was funded by the HTA programme as project number 10/124/01. The contractual start date was in December 2011. The draft report began editorial review in February 2013 and was accepted for publication in June 2013. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The HTA editors and publisher have tried to ensure the accuracy of the authors’ report and would like to thank the reviewers for their constructive comments on the draft document. However, they do not accept liability for damages or losses arising from material published in this report.

Copyright statement

© Queen’s Printer and Controller of HMSO 2014. This work was produced by Boonacker et al. under the terms of a commissioning contract issued by the Secretary of State for Health. This issue may be freely reproduced for the purposes of private research and study and extracts (or indeed, the full report) may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising. Applications for commercial reproduction should be addressed to: NIHR Journals Library, National Institute for Health Research, Evaluation, Trials and Studies Coordinating Centre, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.

Aims and objectives

In this IPD meta-analysis we therefore (1) developed a model to predict the risk of children referred for adenoidectomy having a prolonged duration of their otitis media. Then, (2a) having evaluated the overall effect of adenoidectomy, with or without grommets, on OM using these IPD, we (2b) identified those subgroups of children who benefit most, or who are most likely to benefit, from adenoidectomy, with or without grommets.

Protocol and registration

The original protocol for this IPD meta-analysis was published on the website of the Health Technology Assessment programme of the National Institute for Health Research (see www.hta.ac.uk/2576). The protocol is included in this report (see Appendix 1). The study was also registered on 13 September 2011 in the PROSPERO register with the number CRD42011001549 (see www.crd.york.ac.uk/PROSPERO/display_record.asp?ID=CRD42011001549).

As in any IPD meta-analysis, small but potentially important changes to intentions have to be made on the basis of the specific data that are made available to the research team. In this regard, discussion with the principal investigators and/or study representatives included in the IPD meta-analysis took place at two meetings (Amsterdam, the Netherlands, 21 May 2012, and Oxford, UK, 3 September 2012). This led to the decision to promote to the primary outcome a composite outcome measure that was already included in the protocol as a secondary outcome, because the primary outcome originally proposed could not be determined in a significant proportion of studies, severely limiting the value of any ensuing meta-analysis. Specifically, number of OM episodes was replaced by failure at 12 months. By using such a composite end point we were able to aggregate our data and generalise results. This was critical to accomplish our aim of establishing which children benefit more or less from adenoidectomy.

An overview of the adaptations to the study protocol and the reasons for them can be found in Appendix 2.

Selection of the trials and quality assessment

For this IPD meta-analysis the same search strategy was used as in our Cochrane review. 28 We searched the following databases from their inception: the Cochrane Ear, Nose and Throat Disorders Group Trials Register, the Cochrane Central Register of Controlled Trials (CENTRAL, The Cochrane Library Issue 1, 2009), PubMed, EMBASE, the Cumulative Index to Nursing and Allied Health Literature (CINAHL), Latin American and Caribbean Health Sciences Literature (LILACS), KoreaMed, IndMED, PakMediNet, CAB Abstracts, Web of Science, BIOSIS Previews, China National Knowledge Infrastructure (CNKI), the metaRegister of Current Controlled Trials (mRCT), ClinicalTrials.gov, the International Clinical Trials Registry Platform (ICTRP), ClinicalStudyResults.org and Google. We used the following keywords with their synonyms: ‘adenoidectomy’ and ‘otitis’ (see Appendix 3 for the complete search strategy). We checked the bibliography of all relevant studies and reviews to identify supplemental studies. Unpublished studies were also eligible for inclusion. We imposed no language restriction on the searches. The date of the last search for the Cochrane review was 30 March 2009. Thereafter, we received weekly updates from PubMed and performed a full updated search on 7 June 2012 (including CENTRAL, The Cochrane Library Issue 5, 2012).

The final selection of eligible studies and the quality assessment were carried out by two of the authors of the Cochrane review (CB and MvdA) and disagreement was resolved by discussion.

For the quality assessment we used The Cochrane Collaboration quality assessment (risk of bias) tool,29 which requires a judgement on sequence generation, allocation concealment (whether or not assignment to the intervention or control group could be foreseen by the participants or the investigators), blinding, incomplete outcome data, selective outcome reporting and evaluation of other possible biases. The judgement on the risk of bias for this IPD meta-analysis may differ from the judgement in the Cochrane review because for this IPD meta-analysis we obtained the original raw data sets from the primary investigators of all of the included trials and we did not have to rely on the published information alone.

We checked and reanalysed all data sets to see if it was possible to recalculate the published results. When needed, we contacted the principal investigators to resolve any questions. The two data sets that were provided in Finnish16,17 were translated into English by one of the project members (CB). The dictionary developed during the translation of the first data set16 was approved by the principal investigator and was then used to translate the second dataset. 17 After checking the quality of the data sets we developed one overall data set by recoding the variables from the individual data sets to the set of variables used for the IPD meta-analysis. For season, we used country-specific seasons, that is, June could be either spring or summer, depending on the country in which the study was performed. We used date of surgery to calculate the season and winter/spring was used in the analyses as the measure for seasonality. For children in the non-surgical group we used the date of enrolment.

Types of studies and patients

Studies were eligible for inclusion in this IPD meta-analysis if they were a randomised controlled trial (RCT) in children up to 12 years of age diagnosed with OM (being recurrent AOM and/or persistent OME) in which adenoidectomy (with or without grommets) was compared with non-surgical treatment or grommets. Trials in which all children underwent adenoidectomy were therefore not eligible. We included trials in which the method of randomisation was not specified in detail, but we excluded quasi-randomised trials (e.g. allocation by date of birth or record number). Studies had to have a follow-up period of at least 1 year. Desirable time points for outcome assessment were 6, 12, 18 and 24 months.

Types of interventions

Comparator

The comparators evaluated were:

1. unilateral or bilateral grommets

2. non-surgical treatment or myringotomy alone.

To evaluate the effects of adenoidectomy we compared the following interventions:

• main comparison (Figure 1a):

  • adenoidectomy with or without grommets compared with non-surgical treatment or grommets only (this comparison was selected on the basis of prior knowledge of the available data sets included in our conventional meta-analysis and was chosen to maximise the statistical power to identify subgroups).

• other comparisons (Figure 1b):

  • adenoidectomy with unilateral or bilateral grommets compared with non-surgical treatment

  • adenoidectomy with unilateral or bilateral grommets compared with unilateral or bilateral grommets

  • adenoidectomy without grommets compared with non-surgical treatment

  • adenoidectomy without grommets compared with unilateral or bilateral grommets.

FIGURE 1.

(a) Main comparison: adenoidectomy with or without grommets compared with non-surgical treatment or grommets only; and (b) other comparisons.

We did not compare the following interventions as these analyses do not fulfil the criteria for inclusion in this IPD meta-analysis:

• adenoidectomy with unilateral or bilateral grommets compared with adenoidectomy without grommets

• unilateral or bilateral grommets compared with non-surgical treatment.

Statistical analysis

Medical ethical approval

The study team discussed the project with the Medical Ethical Committee of the University Medical Center Utrecht. The committee confirmed that ethical approval was not required as the study uses only anonymous data from previously performed studies for which both informed consent and ethical approval had already been obtained.

Searches

Our searches (2009 and 2012) retrieved a total of 503 articles (Figure 2). We first screened the articles by title/abstract and excluded five trials in which all children underwent adenoidectomy. 31–35 An additional 467 studies were excluded for other reasons, for example follow-up for <12 months. In total, 31 articles were eligible for further assessment. One study by Sagnelli et al. 36 could not be retrieved. We excluded three articles,37–39 describing two trials, because of inadequate randomisation or concealment of allocation. One article40 was excluded because it was found to report a non-RCT. Furthermore, we identified 11 of the publications9,41–50 as providing the same data as more recent articles included in this meta-analysis. Publications focused on adenotonsillectomy were also excluded except for those in which a separate analysis for adenoidectomy was undertaken. 19,22 No additional trials were identified from checking the bibliographies of the selected trials and reviews, nor by contacting the first or corresponding author of the eligible trials.

FIGURE 2.

Flow chart for adenoidectomy studies. n, number of publications (there could be more than one publication per study); k, number of studies.

In total, 15 trials10–25,51,52 were eligible for inclusion in this IPD meta-analysis. We contacted the principal investigators of these trials to ask if they were willing and able to provide their raw data. We were unable to contact the principal investigator of one trial. 24 The data for four older trials (1978–99) were no longer available. 14,22,23,25 For one trial, additional follow-up data were available. 15,51 For another study, additional unpublished data were also provided. 20 Three older studies (1986–90) used a unilateral grommet,10,12,52 whereas the others used no grommets16 or bilateral grommets. 11,15,17,18,20,21

Characteristics of included studies

The characteristics of the included studies are shown in Table 1. The inclusion and exclusion criteria are shown in Appendix 5, the indication for surgery and the age ranges within studies are shown in Appendix 6 and the definitions of recurrent AOM and persistent OME used in the included studies are shown in Appendix 7.

We included raw data from 10 trials,10–12,15–21,51,52 including 1761 children. The trials were published between 1990 and 2012. Eight trials were undertaken in Europe (four in Finland15–18 and four in the UK10,12,20,52), one in the USA11 and one in Canada. 21

Specific differences between included studies

The 10 trials differed in a number of ways but the most important were in terms of indication for surgery, interventions studied and frequency of outcome assessment.

Methodological quality of the included studies

The results of the risk of bias assessment are presented in Table 2. We classified the risk of selection bias because of inadequate sequence generation as low for eight10–12,15–17,20,52 out of 10 studies and unclear in two studies. 18,21 We classified the risk of selection bias because of inadequate allocation concealment as unclear in three studies11,18,21 and as low in the remaining seven studies. 10,12,15–17,20,52 The risk of performance bias (blinding of participants and personnel) and detection bias (blinding of outcome assessment) was unclear in seven studies10,11,15–18,21 and six studies11,15–18,21 respectively. Incomplete outcome data may have caused attrition bias in five studies10,12,16,20,52 and is likely to have caused bias in two studies. 18,21 As the full data sets were provided, the risk of reporting bias because of selective reporting is low in all studies. We classified one study21 as having a high risk of other bias. In this study the randomisation did not result in two similar groups. Age and degree of adenoid hypertrophy differed between the two groups. Furthermore, the follow-up data were obtained by questionnaires completed retrospectively at 6-monthly intervals. This could have led to recall bias. The risk of other bias was classified as unclear for five studies10–12,18,52,53 and low for the remaining four studies. 15–17,20

Overall we concluded that eight studies10–12,15–17,20,52 were at a low risk of bias and two18,21 were at a moderate risk of bias.

Note that the judgement of the risk of bias for selective reporting differs from the judgement in the Cochrane review28 because in this IPD meta-analysis the complete data sets were obtained and we did not have to rely on the publications alone, resulting in low risk for selective reporting in all studies. Also, the judgement on blinding is now divided into blinding of participants and personnel, and blinding of outcome assessment.

Characteristics of participants at baseline

1. Development of a model to predict the risk of children referred for adenoidectomy having a prolonged duration of their otitis media

To determine the predictive factors for failure at 12 months in children referred for adenoidectomy but randomised to a non-surgical treatment we studied the following factors: age, sex, indication, season, day-care attendance, breastfeeding, passive smoking, atopy or allergy, number of previous AOM episodes (less than four vs. four or more), the occurrence of one or more episodes of AOM in the previous year, age at first occurrence of AOM, baseline hearing level, duration of symptoms, and OME at study entry. We could not include pneumococcal vaccination status as this was not reported in any of the studies included in this analysis.

2a. Evaluation of the overall effect of adenoidectomy, with or without grommets on otitis media using individual patient data

Main comparison: adenoidectomy with or without grommets compared with non-surgical treatment or grommets only

Primary outcome

Figure 3 provides an overview of the overall effect within each individual study.

FIGURE 3.

Overall failure at 12 months: effect of adenoidectomy with or without grommets compared with non-surgical treatment or grommets only by study. (a) RD; and (b) RR. Note that the overall pooled effect size is not adjusted for the study. Table 6 provides the calculated RDs, RRs and adjusted RRs which differ slightly from those in the forest plots as the underlying models differ.

In total, the proportion of children who failed at 12 months in the adenoidectomy group (adenoidectomy with or without grommets) was 31.9% whereas the proportion of children who failed at 12 months in the no adenoidectomy (non-surgical or grommets alone) group was 44.7%. The RD for failure at 12 months was −12.7% (95% CI −17.3% to −8.2%), resulting in a NNT of eight children to prevent one failure. The RR was 0.72 (95% CI 0.63 to 0.81) and the adjusted RR (using a dummy for study) was 0.76 (95% CI 0.69 to 0.85) (Table 6).

TABLE 6 - Overall effect of adenoidectomy with or without grommets compared with non-surgical treatment or grommets only (n  = 1761)

k, number of studies; n, number of children; SD, standard deviation.

Bold text indicates values that are statistically significant.

Outcome	Adenoidectomy with or without unilateral or bilateral grommets (n = 767, k = 10)	Non-surgical treatment or unilateral or bilateral grommets only (n = 994, k = 10)	RD (95% CI)a	NNTa	RR or mean difference (95% CI)a	Adjusted RR or mean difference (95% CI)a
Primary outcome, n/N (%)
Failure at 12 months	245/767 (31.9)	444/994 (44.7)	−12.7% (−17.2% to −8.2%)	8	0.72 (0.63 to 0.81)	0.76 (0.69 to 0.85)
Secondary outcome: individual items of failure at 12 months, n/N (%)
Four or more AOM episodes per year	52/508 (10.2)	111/737 (15.1)	−4.8% (−8.5% to −1.1%)	21	0.65 (0.50 to 0.93)	0.78 (0.58 to 1.04)
Presence of effusion for ≥ 50% of the time	153/475 (32.2)	254/652 (39.0)	−6.7% (−12.4% to −1.1%)	15	0.83 (0.70 to 0.97)	0.78 (0.68 to 0.89)
Surgery during follow-up	38/764 (5.0)	172/994 (17.3)	−12.3% (−15.1% to −9.5%)	9	0.29 (0.21 to 0.40)	0.30 (0.22 to 0.42)
Improvement in hearing < 10 dB	55/197 (27.9)	126/301 (41.9)	−13.9% (−22.3% to −5.6%)	8	0.67 (0.51 to 0.87)	0.67 (0.52 to 0.87)
Secondary outcome: number of AOM episodes during follow-up, mean (SD), range
In the first 6 months	0.8 (1.3), 0–8.0	1.1 (1.6), 0–10.5	–	–	−0.3 (−0.5 to −0.2)	−0.2 (−0.4 to −0.1)
In the first 12 months	1.2 (1.7), 0–9.5	1.6 (2.1), 0–13.0	–	–	−0.4 (−0.7 to −0.2)	−0.4 (−0.6 to −0.1)
In the first 18 months	1.6 (2.2), 0–10.0	2.4 (2.9), 0–15.0	–	–	−0.9 (−1.3 to −0.4)	−0.5 (−0.9 to −0.0)
In the first 24 months	2.3 (2.6), 0–15.0	2.9 (3.2), 0–16.0	–	–	−0.6 (−1.0 to −0.2)	−0.5 (−0.7 to 0.1)
Secondary outcome: time with effusion during follow-up (weeks), mean (SD), range
In the first 6 months	10.0 (8.7), 0.0–26.0	12.6 (10.2), 0.0–26.0	–	–	−2.6 (−3.8 to −1.4)	−2.6 (−3.4 to −1.8)
In the first 12 months	16.4 (16.2), 0.0–52.0	20.1 (19.7), 0.0–52.0	–	–	−3.8 (−5.9 to −1.6)	−4.5 (−6.0 to −2.9)
In the first 18 months	25.3 (23.2), 0.0–78.0	33.2 (28.6), 0.0–78.0	–	–	−7.9 (−11.4 to −4.4)	−7.9 (−10.6 to −5.3)
In the first 24 months	31.9 (30.2), 0.0–104.0	43.0 (36.8), 0.0–104.0	–	–	−11.1 (−15.7 to −6.6)	−11.2 (−14.7 to −7.6)
Secondary outcome: additional surgery during follow-up (yes), n/N (%)
In the first 6 months	18/764 (2.4)	93/994 (9.4)	−7.0% (−9.1% to −4.9%)	15	0.25 (0.15 to 0.41)	0.32 (0.20 to 0.52)
In the first 18 months	50/480 (10.4)	179/631 (28.4)	−18.0% (−22.4% to −13.5%)	6	0.37 (0.28 to 0.49)	0.36 (0.27 to 0.47)
In the first 24 months	64/480 (13.3)	211/631 (33.4)	−20.1% (−24.9% to −15.3%)	5	0.40 (0.31 to 0.51)	0.38 (0.30 to 0.48)
Secondary outcome: hearing loss (dB), mean (SD), range
After 6 months’ follow-up	15.9 (8.3), 3.2–54.4	20.3 (9.3), −1.9–53.8	–	–	−4.4 (−5.9 to −2.8)	−4.4 (−5.9 to −2.9)
After 12 months’ follow-up	16.8 (8.5), 1.9–48.8	20.4 (9.6), 1.9–49.4	–	–	−3.6 (−5.2 to −2.0)	−3.5 (−5.1 to −1.9)
After 18 months’ follow-up	15.5 (7.0), 1.9–39.4	20.5 (10.0), 3.1–55.0	–	–	−5.0 (−6.8 to −3.3)	−5.0 (−6.6 to −3.4)
After 24 months’ follow-up	15.4 (7.8), 1.3–39.4	19.2 (8.8), 4.4–43.8	–	–	−3.9 (−5.4 to −2.2)	−4.1 (−5.7 to −2.6)
Secondary outcome: improvement in hearing level, n/N (%)
< 10 dB after 6 months’ follow-up	47/204 (23.0)	115/305 (37.7)	−14.7% (−22.6% to −6.7%)	7	0.61 (0.46 to 0.82)	0.61 (0.46 to 0.81)
< 10 dB after 18 months’ follow-up	37/160 (23.1)	103/260 (39.6)	−16.5% (−25.3% to −7.7%)	6	0.58 (0.42 to 0.80)	0.57 (0.42 to 0.80)
< 10 dB after 24 months’ follow-up	39/165 (23.6)	96/275 (34.9)	−11.3% (−19.9% to −2.7%)	9	0.68 (0.49 to 0.93)	0.64 (0.47 to 0.89)

In this analysis the proportion of children without grommets, with a unilateral grommet and with bilateral grommets differed between the groups. In the adenoidectomy group 12% did not have grommets, 25% had a unilateral grommet and 62% had bilateral grommets. In the no adenoidectomy group these percentages were 35%, 20% and 46% respectively (see Table 3). To study whether this imbalance influenced our results we performed sensitivity analyses by omitting one group at a time (i.e. no grommets, unilateral grommets or bilateral grommets). The results of the sensitivity analyses did not differ and therefore the imbalance did not affect the outcome.

Secondary outcomes

Data on all available secondary outcomes are also presented in Table 6.

For all secondary outcomes, results in children in the adenoidectomy group were statistically significantly better than results in those in the no adenoidectomy group. During follow-up, children in the adenoidectomy group had a lower number of AOM episodes, less time with effusion, less additional surgery, less hearing loss and improved hearing levels at 6, 12, 18 and 24 months’ follow-up. The clinical importance of these findings is not clear. The reductions in both time with effusion and number of AOM episodes are very modest. However, about one in five children avoid the need for additional surgery. The clinical importance of an average 4-dB improvement in hearing loss has been debated and we consider it further below.

Secondary comparisons

Adenoidectomy with unilateral or bilateral grommets compared with non-surgical treatment

Primary outcome

Figure 4 provides an overview of the overall effect within each individual study.

FIGURE 4.

Overall failure at 12 months: effect of adenoidectomy with unilateral or bilateral grommets compared with non-surgical treatment by study. (a) RD; and (b) RR. Note that the overall pooled effect size is not adjusted for the study. Table 7 provides the calculated RDs, RRs, and adjusted RRs which differ slightly from those in the forest plots as the underlying models differ and some indirect comparisons could be added to the regression analyses.

The proportion of children who failed at 12 months was 32.4% in the group in which all children had adenoidectomy with a grommet in one ear or both and 56.3% in the non-surgical group. The RD for failure at 12 months was −23.8% (95% CI −30.2% to −17.5%), resulting in a NNT of five children to prevent one failure. The RR was 0.58 (0.50 to 0.67) and the adjusted RR (using a dummy for study) was 0.48 (0.40 to 0.59) (Table 7).

TABLE 7 - Overall effect of adenoidectomy with unilateral or bilateral grommets compared with non-surgical treatment (n  = 1015)

k, number of studies; n, number of children; SD, standard deviation.

Black et al. ,52 Casselbrant et al. ,11 Dempster et al. ,12 Hammarén-Malmi et al. ,15 Kujala et al. ,17 Mattila et al. ,51 Maw and Bawden,10 Maw and Herod,19 MRC Multicentre Otitis Media Study Group20 and Nguyen et al. 21

MRC Multicentre Otitis Media Study Group,20 Koivunen et al. 16 and Kujala et al. 17

Bold text indicates values that are statistically significant.

Outcome	Adenoidectomy with unilateral or bilateral grommets (n = 672, k = 9)a	Non-surgical treatment (n = 343, k = 3)b	RD (95% CI)c	NNTc	RR or mean difference (95% CI)c	Adjusted RR or mean difference (95% CI)c
Primary outcome, n/N (%)
Failure at 12 months	218/672 (32.4)	193/343 (56.3)	−23.8% (−30.2% to −17.5%)	5	0.58 (0.50 to 0.67)	0.48 (0.40 to 0.59)
Secondary outcome: individual items of failure at 12 months, n/N (%)
Four or more AOM episodes per year	35/413 (8.5)	77/333 (23.1)	−14.6% (−19.9% to −9.4%)	7	0.37 (0.25 to 0.53)	0.38 (0.21 to 0.68)
Presence of effusion for ≥ 50% of the time	153/386 (39.6)	83/253 (32.8)	6.8% (−0.1% to 14.4%)	–	1.21 (0.97 to 1.50)	0.45 (0.34 to 0.60)
Surgery during follow-up	27/669 (4.0)	96/343 (28.0)	−24.0% (−28.9% to −19.0%)	5	0.14 (0.10 to 0.22)	0.12 (0.06 to 0.24)
Improvement in hearing < 10 dB	55/197 (27.9)	40/100 (40.0)	−12.1% (−23.5% to −0.6%)	9	0.70 (0.50 to 0.97)	0.68 (0.46 to 0.99)
Secondary outcome: number of AOM episodes during follow-up, mean (SD), range
In the first 6 months	0.7 (1.2), 0–7.0	1.6 (1.9), 0–10.5	–	–	−0.9 (−1.1 to −0.7)	−0.7 (−1.0 to −0.4)
In the first 12 months	1.1 (1.6), 0–9.5	2.2 (2.5), 0–13.0	–	–	−1.0 (−1.3 to −0.8)	−0.7 (−1.1 to −0.7)
In the first 18 months	1.2 (2.0), 0–9.5	3.1 (3.2), 0–15.0	–	–	−1.8 (−2.5 to −1.2)	−1.8 (−1.6 to −0.1)
In the first 24 months	2.4 (2.6), 0–15.0	3.6 (3.6), 0–16.0	–	–	−1.3 (−1.9 to −0.6)	−1.2 (−2.2 to −0.2)
Secondary outcome: time with effusion during follow-up (weeks), mean (SD), range
In the first 6 months	12.3 (8.3), 0.0–26.0	10.8 (11.4), 0.0–26.0	–	–	1.5 (−0.1 to 3.1)	−10.1 (11.7 to −8.5)
In the first 12 months	19.7 (16.2), 0.0–52.0	17.0 (20.4), 0.0–52.0	–	–	2.7 (−0.2 to 5.5)	−12.4 (−15.5 to −9.2)
In the first 18 months	31.6 (22.4), 0.0–78.0	27.8 (29.5), 0.0–78.0	–	–	5.9 (1.4 to 10.3)	−17.6 (−23.4 to −11.9)
In the first 24 months	39.5 (29.5), 0.0–104.0	33.0 (38.2), 0.0–104.0	–	–	6.6 (0.8 to 12.4)	−21.5 (−29.4 to −13.7)
Secondary outcome: additional surgery during follow-up (yes), n/N (%)
In the first 6 months	12/669 (1.8)	66/343 (19.2)	−17.4% (−21.7% to −13.2%)	6	0.09 (0.05 to 0.17)	0.15 (0.07 to 0.32)
In the first 18 months	35/385 (9.1)	97/242 (40.1)	−31.0% (−37.8% to −24.2%)	4	0.23 (0.16 to 0.32)	0.03 (0.08 to 0.13)
In the first 24 months	48/385 (12.5)	102/242 (42.1)	−29.7% (−36.7% to −22.6%)	4	0.30 (0.22 to 0.40)	0.05 (0.02 to 0.15)
Secondary outcome: hearing loss (dB), mean (SD), range
After 6 months’ follow-up	15.9 (8.3), 3.1–54.4	23.1 (10.1), −1.9–53.8	–	–	−7.2 (−9.3 to −5.1)	−7.8 (−10.2 to −5.3)
After 12 months’ follow-up	16.8 (8.5), 1.9–48.8	20.5 (10.1), 4.4–45.0	–	–	−3.6 (−5.8 to −1.4)	−3.4 (−6.0 to −0.8)
After 18 months’ follow-up	15.5 (7.0), 1.9–39.4	19.7 (10.4), 4.4–55.0	–	–	−4.2 (−6.3 to −2.1)	−4.0 (−6.5 to −1.5)
After 24 months’ follow-up	15.4 (7.7), 1.3–39.3	18.2 (8.1), 5.6–41.3	–	–	−2.9 (−4.8 to −9.3)	−3.5 (−5.6 to −1.4)
Secondary outcome: improvement in hearing level, n/N (%)
< 10 dB after 6 months’ follow-up	47/204 (23.0)	49/105 (46.7)	−23.6% (−34.8% to −12.5%)	5	0.49 (0.36 to 0.68)	0.46 (0.31 to 0.69)
< 10 dB after 18 months’ follow-up	37/160 (23.1)	30/98 (30.6)	−7.5% (−18.7% to 3.7%)	14	0.76 (0.50 to 1.14)	0.83 (0.53 to 1.29)
< 10 dB after 24 months’ follow-up	39/165 (23.6)	27/102 (26.5)	−2.8% (−13.6% to 7.9%)	36	0.89 (0.58 to 1.37)	0.80 (0.49 to 1.30)

Secondary outcomes

Data on all available secondary outcomes are also presented in Table 7.

Adenoidectomy with unilateral or bilateral grommets compared with unilateral or bilateral grommets

Primary outcome

Figure 5 provides an overview of the overall effect within each individual study. The proportion of children who failed at 12 months was 32.4% in the group in which all children had adenoidectomy with a grommet in one or both ears and 38.6% in the group of children who had grommets only in one or both ears. The RD for failure at 12 months was −6.2% (95% CI −11.3% to −1.0%), resulting in a NNT of 17 children to prevent one failure. The RR was 0.84 (95% CI 0.73 to 0.97) and the adjusted RR (using a dummy for study) was 0.86 (95% CI 0.76 to 0.97) (Table 8).

FIGURE 5.

Overall failure at 12 months: effect of adenoidectomy with unilateral or bilateral grommets compared with unilateral or bilateral grommets by study. (a) RD; and (b) RR. Note that the overall pooled effect size is not adjusted for the study. Table 8 provides the calculated RDs, RRs, and adjusted RRs which differ slightly from those in the forest plots as the underlying models differ and some indirect comparisons could be added to the regression analyses.

TABLE 8 - Overall effect of adenoidectomy with unilateral or bilateral grommets compared with unilateral or bilateral grommets (n  = 1323)

k, number of studies; n, number of children; SD, standard deviation.

Black et al. ,52 Casselbrant et al. ,11 Dempster et al. ,12 Hammarén-Malmi et al. ,15 Kujala et al. ,17 Mattila et al. ,51 Maw and Bawden,10 Maw and Herod,19 MRC Multicentre Otitis Media Study Group20 and Nguyen et al. 21

Black,9 Casselbrant et al. ,11 Dempster et al. ,12 Hammarén-Malmi et al. ,15 Kujala et al. ,17 Mattila et al. ,51 Maw and Bawden,10 Maw and Herod,19 MRC Multicentre Otitis Media Study Group20 and Nguyen et al. 21

Bold text indicates values that are statistically significant.

	Adenoidectomy with unilateral or bilateral grommets (n = 672, k = 9)a	Unilateral or bilateral grommets (n = 651, k = 9)b	RD (95% CI)c	NNTc	RR or mean difference (95% CI)c	Adjusted RR or mean difference (95% CI)c
Primary outcome, n/N (%)
Failure at 12 months	218/672 (32.4)	251/651 (38.6)	−6.1% (−11.3% to −1.0%)	17	0.84 (0.73 to 0.97)	0.86 (0.76 to 0.97)
Secondary outcome: individual items of failure at 12 months, n/N (%)
Four or more AOM episodes per year	35/413 (8.5)	34/404 (8.4)	0.1% (−3.8% to 3.9%)	–	1.01 (0.64 to 1.58)	0.97 (0.62 to 1.52)
Presence of effusion for ≥ 50% of the time	153/386 (39.6)	171/399 (42.9)	−3.2% (−10.1% to 3.7%)	32	0.93 (0.78 to 1.09)	0.88 (0.77 to 1.02)
Surgery during follow-up	27/669 (4.0)	76/651 (11.7)	−7.6% (−10.5% to −4.8%)	13	0.35 (0.23 to 0.53)	0.35 (0.23 to 0.53)
Improvement in hearing < 10 dB	55/197 (27.9)	86/201 (42.8)	−14.9% (−24.1% to −5.6%)	7	0.65 (0.50 to 0.86)	0.66 (0.50 to 0.87)
Secondary outcome: number of AOM episodes during follow-up, mean (SD), range
In the first 6 months	0.7 (1.2), 0–7.0	0.7 (1.1), 0–6.0	–	–	0.0 (−0.1 to 0.2)	0.0 (−0.1 to 0.2)
In the first 12 months	1.1 (1.6), 0–9.5	1.2 (1.6), 0–10.0	–	–	−0.1 (−0.3 to 0.2)	−0.1 (−0.3 to 0.1)
In the first 18 months	1.2 (2.0), 0–9.5	1.4 (1.9), 0–8.5	–	–	−0.2 (−0.6 to 0.3)	−0.2 (−0.6 to 0.3)
In the first 24 months	2.4 (2.6), 0–15.0	2.2 (2.6), 0–12.0	–	–	0.2 (−0.3 to 0.7)	0.1 (−0.4 to 0.6)
Secondary outcome: time with effusion during follow-up (weeks), mean (SD), range
In the first 6 months	12.3 (8.3), 0.0 to 26.0	13.8 (9.2), 0.0 to 26.0	–	–	−1.5 (−2.8 to −0.2)	−1.7 (−2.7 to −0.7)
In the first 12 months	19.7 (16.2), 0.0 to 52.0	22.1 (18.9), 0.0 to 52.0	–	–	−2.4 (−4.9 to 0.0)	−3.3 (−5.2 to −1.4)
In the first 18 months	31.6 (22.4), 0.0 to 78.0	39.1 (26.4), 0.0 to 78.0	–	–	−7.4 (−11.4 to −3.5)	−7.5 (−11.1 to −4.0)
In the first 24 months	39.5 (29.5), 0.0 to 104.0	51.0 (33.6), 0.0 to 104.0	–	–	−11.5 (−16.6 to −6.4)	−11.7 (−16.4 to −7.0)
Secondary outcome: additional surgery during follow-up (yes), n/N (%)
In the first 6 months	12/669 (1.8)	27/651 (4.1)	−2.4% (−4.2% to −0.5%)	42	0.43 (0.22 to 0.85)	0.44 (0.23 to 0.84)
In the first 18 months	35/385 (9.1)	82/389 (21.1)	−12.0% (−17.0% to −7.0%)	9	0.43 (0.30 to 0.62)	0.44 (0.31 to 0.62)
In the first 24 months	48/385 (12.5)	109/389 (28.0)	−15.6% (−21.1% to −10.0%)	7	0.45 (0.33 to 0.61)	0.45 (0.34 to 0.60)
Secondary outcome: hearing loss (dB), mean (SD), range
After 6 months’ follow-up	15.9 (8.3), 3.1–54.4	18.9 (8.6), 5.0–47.5	–	–	−2.9 (−4.6 to −1.3)	−2.9 (−4.5 to −1.3)
After 12 months’ follow-up	16.8 (8.5), 1.9–48.8	20.4 (9.3), 1.9–49.4	–	–	−3.6 (−5.3 to −1.9)	−3.6 (−5.3 to −1.9)
After 18 months’ follow-up	15.5 (7.0), 1.9–39.4	21.0 (9.8), 3.1–52.0	–	–	−5.5 (−7.3 to –3.7)	−5.5 (−7.3 to −3.7)
After 24 months’ follow-up	15.4 (7.7), 1.3–39.3	19.8 (9.1), 4.4–43.8	–	–	−4.4 (−6.2 to −2.6)	−4.3 (−6.1 to −2.6)
Secondary outcome: improvement in hearing level, n/N (%)
< 10 dB after 6 months’ follow-up	47/204 (23.0)	66/200 (33.0)	−10.0% (−18.7% to −1.3%)	10	0.70 (0.51 to 0.96)	0.71 (0.52 to 0.97)
< 10 dB after 18 months’ follow-up	37/160 (23.1)	73/162 (45.1)	−21.9% (−32.0% to −11.9%)	5	0.51 (0.37 to 0.71)	0.51 (0.37 to 0.71)
< 10 dB after 24 months’ follow-up	39/165 (23.6)	69/173 (39.9)	−16.2% (−26.0% to −6.5%)	7	0.59 (0.43 to 0.82)	0.60 (0.43 to 0.83)

The proportions of unilateral and bilateral tubes were similar in both groups. In the adenoidectomy with grommets group 29% had a unilateral grommet and 71% had bilateral grommets whereas in the grommets only group the proportions were 30% and 70% respectively. It is therefore unlikely that the results are influenced by the proportion of unilateral or bilateral tubes within each group.

Secondary outcomes

Data on all available secondary outcomes are also presented in Table 8.

Adenoidectomy without grommets compared with non-surgical treatment

Primary outcome

Figure 6 provides an overview of the overall effect within each individual study. The proportion of children who failed at 12 months was 28.4% in the group in which all children had adenoidectomy alone and 56.3% in the group of children who had non-surgical treatment. The RD for failure at 12 months was −27.8% (95% CI −38.3% to −17.4%), resulting in a NNT of four children to prevent one failure. The RR was 0.51 (95% CI 0.36 to 0.70) and the adjusted RR (using a dummy for study) was 0.74 (95% CI 0.49 to 1.12) (Table 9).

FIGURE 6.

Overall failure at 12 months: effect of adenoidectomy without grommets compared with non-surgical treatment by study. (a) RD; and (b) RR. Note that the overall pooled effect size is not adjusted for the study. Table 9 provides the calculated RDs, RRs, and adjusted RRs which differ slightly from those in the forest plots as the underlying models differ and some indirect comparisons could be added to the regression analyses.

TABLE 9 - Overall effects of adenoidectomy alone compared with non-surgical treatment (n  =  438)

k, number of studies; n, number of children; NA, not applicable; SD, standard deviation.

Casselbrant et al. 11 and Koivunen et al. 16

MRC Multicentre Otitis Media Study Group,20 Koivunen et al. 16 and Kujala et al. 17

Bold text indicates values that are statistically significant.

	Adenoidectomy with or without myringotomy (n = 95, k = 2)a	Non-surgical treatment (n = 343, k = 3)b	RD (95% CI)c	NNTc	RR or mean difference (95% CI)c	Adjusted RR or mean difference (95% CI)c
Primary outcome, n/N (%)
Failure at 12 months	27/95 (28.4)	193/343 (56.3)	−27.8% (−38.3% to −17.4%)	4	0.51 (0.36 to 0.70)	0.74 (0.49 to 1.12)
Secondary outcome: individual items of failure at 12 months, n/N (%)
Four or more AOM episodes per year	17/95 (17.9)	77/333 (23.1)	−5.2% (−14.2% to 3.7%)	20	0.77 (0.48 to 1.24)	0.78 (0.48 to 1.27)
Presence of effusion for ≥ 50% of the time	0/89 (0.0)	83/253 (32.8)	−32.8% (−38.3% to −27.0%)	3	NA	NA
Surgery during follow-up	11/95 (11.6)	96/343 (28.0)	−16.4% (−24.4% to −8.4%)	6	0.41 (0.23 to 0.74)	0.33 (0.14 to 0.82)
Secondary outcome: number of AOM episodes during follow-up, mean (SD), range
In the first 6 months	1.1 (1.5), 0–8.0	1.6 (1.9), 0–10.5	–	–	−0.6 (−1.0 to −0.2)	−0.6 (−1.2 to −0.1)
In the first 12 months	1.5 (2.0), 0–9.0	2.2 (2.5), 0–13.0	–	–	−0.7 (−1.2 to −0.2)	−0.8 (−1.6 to −0.1)
In the first 18 months	2.0 (2.4), 0–10.0	3.1 (3.2), 0 –15.0	–	–	−1.1 (−1.8 to −0.4)	−0.7 (−1.5 to 0.2)
In the first 24 months	2.1 (2.6), 0–10.0	3.6 (3.6), 0–16.0	–	–	−1.5 (−2.3 to −0.7)	−0.8 (−1.7 to 0.2)
Secondary outcome: time with effusion during follow-up (weeks), mean (SD), range
In the first 6 months	1.2 (2.8), 0.0–17.5	10.8 (11.4), 0.0–26.0	–	–	−9.6 (−12.0 to −7.2)	−0.1 (−1.1 to 0.8)
In the first 12 months	2.0 (3.5), 0.0–17.5	17.0 (20.4), 0.0–52.0	–	–	−15.0 (−19.3 to −10.8)	−0.4 (−1.4 to 0.7)
In the first 18 months	2.6 (4.6), 0.0–26.0	27.8 (29.5), 0.0–78.0	–	–	−23.1 (−29.6 to −16.7)	−0.3 (−1.8 to 1.2)
In the first 24 months	3.3 (5.2), 0.0–26.0	33.0 (38.2), 0.0–104.0	–	–	−29.7 (−38.2 to −21.2)	−0.8 (−2.5 to 0.9)
Secondary outcome: additional surgery during follow-up (yes), n/N (%)
In the first 6 months	6/95 (6.3)	66/343 (19.2)	−12.9% (−19.4% to −6.5%)	8	0.33 (0.15 to 0.73)	0.42 (0.15 to 1.18)
In the first 18 months	15/95 (15.8)	97/242 (40.1)	−24.3% (−33.9% to −14.7%)	5	0.39 (0.24 to 0.64)	0.36 (0.17 to 0.75)
In the first 24 months	16/95 (16.8)	102/242 (42.1)	−25.3% (−35.1% to −15.5%)	4	0.40 (0.25 to 0.64)	0.38 (0.19 to 0.76)

Secondary outcomes

Data on all available secondary outcomes are also presented in Table 9.

A large difference was observed between groups during follow-up in the proportion of children with the presence of effusion for ≥ 50% of the time. However, this may be due to a measurement artefact because, as some studies had zero events, it was not possible to calculate an adjusted RR. This adjustment is particular important for these analyses as, in the studies contributing participants to the adenoidectomy only group,11,16 children were examined more frequently and more observations were made than in those studies contributing participants to the non-surgical group. 16,17,20

Adenoidectomy without grommets compared with unilateral or bilateral grommets

Primary outcome

Figure 7 provides an overview of the overall effect within each individual study. The proportion of children who failed at 12 months was 28.4% in the group in which all children had adenoidectomy alone and 38.6% in the group who had grommets. The RD for failure at 12 months was −10.4% (95% CI −20.0% to −0.3%), resulting in a NNT of 10 children to prevent one failure. The RR was 0.74 (95% CI 0.53 to 1.03) and the adjusted RR (using a dummy for study) was 2.07 (95% CI 0.58 to 7.31) (Table 10).

FIGURE 7.

Overall failure at 12 months: effect of adenoidectomy without grommets compared with unilateral or bilateral grommets by study. (a) RD; and (b) RR. Note that the overall pooled effect size is not adjusted for the study. Table 10 provides the calculated RDs, RRs, and adjusted RRs which differ slightly from those in the forest plots as the underlying models differ and some indirect comparisons could be added to the regression analyses.

TABLE 10 - Overall effects for adenoidectomy without grommets compared with unilateral or bilateral grommets (n = 746)

k, number of studies; n, number of children; NA, not applicable; SD, standard deviation.

Casselbrant et al. 11and Koivunen et al. 16

Black et al. ,52 Casselbrant et al. ,11 Dempster et al. ,12 Hammarén-Malmi et al. ,15 Kujala et al. ,17 Mattila et al. ,51 Maw and Bawden,10 Maw and Herod,19 MRC Multicentre Otitis Media Study Group20 and Nguyen et al. 21

Bold text indicates values that are statistically significant.

	Adenoidectomy with or without myringotomy (n = 95, k = 2)a	Unilateral or bilateral grommets (n = 651, k = 9)b	RD (95% CI)c	NNTc	RR or mean difference (95% CI)c	Adjusted RR or mean difference (95% CI)c
Primary outcome, n/N (%)
Failure at 12 months	27/95 (28.4)	251/651 (38.6)	−10.1% (−20.0% to −0.3%)	10	0.74 (0.53 to 1.03)	2.07 (0.58 to 7.31)
Secondary outcome: individual items of failure at 12 months, n/N (%)
Four or more AOM episodes per year	17/95 (17.9)	34/404 (8.4)	9.5% (1.3% to 17.7%)	–	2.13 (1.24 to 3.6)	NA
Presence of effusion for ≥ 50% of the time	0/89 (0.0)	171/399 (42.9)	−42.9% (−47.7% to −38.0%)	3	NA	NA
Surgery during follow-up	11/95 (11.6)	76/651 (11.7)	−0.1% (−7.0% to 6.8%)	–	0.99 (0.55 to 1.80)	1.77 (0.48 to 6.49)
Secondary outcome: number of AOM episodes during follow-up, mean (SD), range
In the first 6 months	1.1 (1.5), 0–8.0	0.7 (1.1), 0–6.0	–	–	0.3 (0.1 to 0.6)	0.2 (−0.1 to 0.5)
In the first 12 months	1.5 (2.0), 0–9.0	1.2 (1.6), 0–10.0	–	–	0.3 (−0.1 to 0.7)	0.3 (−0.1 to 0.7)
In the first 18 months	2.0 (2.4), 0–10.0	1.4 (1.9), 0–8.5	–	–	0.6 (0.0 to 1.1)	0.2 (−0.3 to 0.7)
In the first 24 months	2.1 (2.6), 0–10.0	2.2 (2.6), 0–12.0	–	–	−0.1 (−0.7 to 0.5)	0.1 (−0.5 to 0.6)
Secondary outcome: time with effusion during follow-up (weeks), mean (SD), range
In the first 6 months	1.2 (2.8), 0.0–17.5	13.8 (9.2), 0.0–26.0	–	–	−12.6 (−14.5 to −10.7)	0.7 (0.1 to 1.2)
In the first 12 months	2.0 (3.5), 0.0–17.5	22.1 (18.9), 0.0–52.0	–	–	−20.2 (−24.1 to −16.2)	1.5 (0.2 to 2.7)
In the first 18 months	2.6 (4.6), 0.0–26.0	39.1 (26.4), 0.0–78.0	–	–	−36.4 (−42.2 to −30.7)	1.8 (0.2 to 3.4)
In the first 24 months	3.3 (5.2), 0.0–26.0	51.0 (33.6), 0.0–104.0	–	–	−47.8 (−55.2 to −40.3)	1.4 (−1.0 to 3.8)
Secondary outcome: additional surgery during follow-up (yes), n/N (%)
In the first 6 months	6/95 (6.3)	27/651 (4.1)	2.2% (−3.0% to 7.3%)	–	1.52 (0.65 to 3.59)	NA
In the first 18 months	15/95 (15.8)	82/389 (21.1)	−5.3% (−13.7% to 3.1%)	19	0.75 (0.45 to 1.24)	1.42 (0.52 to 3.88)
In the first 24 months	16/95 (16.8)	109/389 (28.0)	−11.2% (−19.9% to −2.4%)	9	0.60 (0.37 to 0.97)	1.18 (0.46 to 3.03)

Secondary outcomes

Data on all available secondary outcomes are also presented in Table 10. The same large difference between groups during follow-up in the proportion of children with effusion for ≥ 50% of the time was noted here, which again may be due to the measurement artefact described in the previous section.

2b. Identification of those subgroups of children who benefit most, or who are most likely to benefit, from adenoidectomy with or without grommets

The prognostic analyses revealed indication (recurrent AOM or persistent OME) as one potential subgrouping variable and, in addition, we had decided a priori to include age. We studied these factors individually and in combination.

All significant interaction effects (for both primary and secondary outcomes) for all comparisons are presented in Appendices 11–15. In the text and figures that follow we report only significant interaction effects for the analyses stratified on indication.

Adverse effects or events

As adverse effects and events are important secondary outcomes we report the available data here. We sought evidence of the specific complications of adenoidectomy: primary and secondary haemorrhage, hypernasal speech and the complications of general anaesthesia. Two studies11,20 reported data on postoperative complications and adverse effects, three12,16,17 reported that no surgery-related complications had occurred and five15,18,19,21,52 did not report on complications or adverse effects.

The adenoidectomy complication reported by Casselbrant et al. 11 was a single child in the adenoidectomy with grommets group (1/32, 3.1%) in whom difficulty during anaesthesia led to the child subsequently being treated with grommets only. In the MRC study20 one child (1/165, 0.6%) had to return to theatre because of postoperative haemorrhage.

Main findings

This IPD meta-analysis includes 10 trials involving 1761 children with OM (recurrent AOM or persistent OME). Nine out of the 10 studies made a comparison between adenoidectomy and grommets and grommets alone, that is, the direct comparisons from these nine studies provide the most robust evidence. Most of the included studies had a low risk of bias. Twelve months after surgery, children who have had their adenoid removed have a greater chance of clinical improvement. The size of that effect is, in general, small but persists for at least 2 years after surgery. The RD of 13% corresponds to a NNT of eight to prevent one failure; the adjusted RR was 0.76 (95% CI 0.69 to 0.85). Improvement was seen in outcomes that were not found to be significant in a previous non-IPD meta-analysis,28 for example lower number of AOM episodes, less time with effusion, less additional surgery, less hearing loss and improvement in hearing levels up to 2 years after surgery. Effect sizes are small but consistent, with NNTs ranging from 5 to 21.

To predict which children are likely to have a prolonged duration of their OM we looked specifically at those children who did not receive any surgery at all (adenoidectomy or grommets) at study entry and who are likely to best reflect the natural history of untreated disease (‘non-surgical children’). Of these children 56% failed to improve at 12 months. The only factor that predicted which children had a prolonged duration of their OM was the indication for surgery (recurrent AOM or persistent OME). In children with recurrent AOM as the indication for intervention 38% failed to improve, whereas in children with persistent OME as the indication for intervention 89% failed to improve.

In the subgroup analyses we found the anticipated differences in effects based on the indication for surgery (recurrent AOM compared with persistent OME). Two subgroups of children are most likely to benefit from adenoidectomy. These are (1) children with recurrent AOM aged < 2 years and (2) older children aged ≥ 4 years with persistent OME.

The proportion of children with recurrent AOM aged < 2 years who failed at 12 months was 16% in the adenoidectomy group and 27% in the group who did not have adenoidectomy. This RD of 12% corresponds to a NNT of nine to prevent one failure. The adjusted RR was 0.63 (95% CI 0.47 to 0.85). In contrast, in children aged ≥ 2 years, no benefit of adenoidectomy was seen.

The proportion of children with persistent OME aged ≥ 4 years who failed at 12 months was 51% in the adenoidectomy group and 70% in the group who did not have adenoidectomy. This RD of 19% corresponds to a NNT of six to prevent one failure. The adjusted RR was 0.77 (95% CI 0.68 to 0.86). In children aged < 4 years no significant benefit of adenoidectomy was seen.

These findings relate to the main comparison, that is, adenoidectomy or adjuvant adenoidectomy (adenoidectomy with grommets) compared with no surgery or grommets alone. When the subgroup analyses were repeated with our secondary comparisons, the statistical power to detect significant interactions was necessarily reduced. However, we still found a difference in subgroup effects based on indication for surgery and age when adenoidectomy with grommets was compared with grommets alone (the ‘adding adenoidectomy’ comparison, assessing the effect of so-called ‘adjuvant adenoidectomy’). Age was found to be of influence only in the recurrent AOM group.

In the development of the predictive model age was not identified as a factor predicting those children with a prolonged (12-month) duration of their OM. We had anticipated that it might be a factor and that there might be a detectable interaction between age and indication. Yet in the subgroup analysis, in children with recurrent AOM, age was a factor associated with failure at 12 months in young children aged < 2 years compared with older children. The reason for this is that almost all of the children in the trials that evaluate interventions for recurrent AOM are young – aged < 2 years. As one moves from analysing the whole group to subgroups, the recurrent AOM subgroup and the younger children (< 2 years) subgroup contain the same children. Identifying an interaction between indication and age was therefore impossible: there were insufficient data to detect an interaction even if one exists because there are not enough older children (aged ≥ 4 years) in the recurrent AOM group and also not enough younger children (aged < 2 years) in the persistent OME group.

These findings support the general understanding of at least two clinical entities within a spectrum of OM across the age range from birth to 12 years: recurrent AOM in younger children (aged < 2 years) and persistent OME in older children (aged ≥ 4 years). Children aged ≥ 4 years with recurrent AOM are rare within our IPD data set either because the trial entry criteria specifically excluded them (we know that this is true for some of our included trials – all of the trials that enrolled children with recurrent AOM only specifically excluded children aged > 2 years) or because this combination is a rare occurrence. We did in fact include trials in which children aged > 2 years with recurrent AOM could have been included, but they are present in very low numbers.

Strengths and limitations

The main strength of our study is that by reanalysing the original data from 10 trials of adenoidectomy we could include 1761 children in our IPD meta-analysis, which gave us the power to identify subgroups that could benefit most from adenoidectomy.

Some of our findings deserve further discussion:

1. We contacted the principal investigators of all 15 eligible RCTs to ask if they were willing and able to provide their raw data. We were unable to contact the principal investigator of one trial24 and the data for four older trials (1978–99)14,22,23,25 were no longer available. The main characteristics of these five trials are, however, much the same as those of the 10 included trials. Moreover, the overall results of our subset of 10 trials are very similar to the overall results reported by the Cochrane review28 that did include all of these trials. A funnel plot of the included studies (not shown) also indicates that publication bias is unlikely in this IPD meta-analysis.

2. Not all variables were measured in all eligible trials so we chose to use a composite outcome, failure at 12 months, defined as failing on at least one out of four components. Only one of the 10 included trials measured all four components; six trials measured three of the four components and the other three trials measured two of the four components (see also Appendix 10). Furthermore, we could not study all potential effect modifiers. The total number of baseline variables measured in the included trials varied between 9 and 39 per trial. Of the variables of interest, only three were measured in all 10 trials.

3. In our prognostic analyses we included only those children allocated to non-surgical treatment. This is appropriate because adenoidectomy with or without grommets might influence the course of the disease and might result in an invalid natural history model.

4. We did not study all potential subgroups. We selected previously established predictors of a prolonged course and some clinically relevant variables and performed stratified analyses only for those variables that showed a significant p-value (< 0.1) for interaction in the regression model. Did we therefore ‘miss’ a relevant subgroup? Our approach follows the recommendations for the study of subgroups. The strength of this approach is that our prognostic analyses revealed only a few relevant subgroups. By limiting the number of subgroup analyses we minimise the likelihood of false-positive findings (type I error) that could be caused by multiple testing. We were unable to study some clinically relevant subgroups that might benefit more from adenoidectomy, such as children with Down syndrome, because such children were excluded from the individual trials. The experience of many clinicians – that these and other subgroups of children benefit more from adenoidectomy than others – has not yet been evidenced in RCTs. Neither could we study the effect of adenoidectomy in children with differing baseline hearing levels as hearing was not assessed in all included studies. In the studies that did assess hearing at baseline, the effects of adenoidectomy did not differ between groups with different baseline hearing levels.

5. Although an overall measure of severity including both persistent AOM and recurrent OME is not available, frequency of AOM, duration of symptoms and degree of hearing loss are potential measures of severity. They were included in the prognostic analyses and were not found to be of influence when tested in the multivariate analyses.

6. Because of relatively wide variation in the specific surgical interventions across the 10 included trials, we had to make a fairly large number of comparisons, which renders interpretation of the results more difficult and introduces a degree of obfuscation. Although we focused on our main comparison, adenoidectomy with or without grommets compared with non-surgical treatment or grommets only, we also compared adenoidectomy with unilateral or bilateral grommets with non-surgical treatment; adenoidectomy with unilateral or bilateral grommets with unilateral or bilateral grommets; adenoidectomy without grommets with non-surgical treatment; and adenoidectomy without grommets with unilateral or bilateral grommets. Because of the smaller numbers in some of these secondary comparisons, the subgroup effects are inevitably less obvious. However, there were no inconsistencies in effect directions.

7. In all of our studies short-term tubes were used; therefore, ‘tube life’ is unlikely to have influenced the results.

8. In our pooled analysis of failure at 12 months we combined several outcomes, which were measured slightly differently across the 10 included trials. Sensitivity analysis using an alternative set of definitions of ‘failure’ showed similar results. Furthermore, the time points for follow-up assessments varied across the included trials (e.g. some took measurements every month, others every 3 or 6 months). As a result, a measurement artefact cannot be precluded. To evaluate this further we undertook a sensitivity analysis with the trials that assessed the participants most frequently and found that the results were consistent with the overall results.

9. We included trials that studied the effects of surgery in children with persistent OME as well as those studying the effects of surgery in children with recurrent AOM or combinations of persistent OME and recurrent AOM. We believe that these two clinical conditions, although distinctly defined, are in fact closely related as part of a spectrum of disease. Our analyses showed different effects based on the indication for surgery for both the main effect and the subgroup effects. This confirms that these two indications should be distinguished in both clinical and research settings, with implications for clinical practice and future research.

10. We chose a composite outcome as the included trials each used slightly different outcomes. It is known that combined outcomes can make a treatment seem more or less effective than it really is,54 that is, higher event rates and larger treatment effects associated with less important components may result in misleading impressions of the impact of treatment. We therefore performed sensitivity analyses using different composite outcomes, which all showed similar results. By using such a composite end point we were able to aggregate our data and generalise results. This was critical to accomplish our aim of establishing which children benefit more or less from adenoidectomy.

11. Some studies randomised ears rather than children, and it is self-evident that this applies only to the insertion of grommets (i.e. in some studies a grommet was inserted unilaterally). The focus of our IPD meta-analysis was the effect of adenoidectomy, and this intervention was always randomised to children. We compared children and those randomised to adenoidectomy with/without grommets with those randomised to non-surgical treatment or grommets only. Other comparisons included adenoidectomy with unilateral or bilateral grommets compared with adenoidectomy without grommets, and unilateral or bilateral grommets compared with non-surgical treatment. We pooled the studies that randomised ears with those that randomised children. Sensitivity analysis in which we excluded the (older) trials that randomised ears rather than children did not change our results.

12. We were unable to calculate adjusted RDs because of the calculation artefact that occurs when there are zero events in a study. 21 We decided to also present the unadjusted RDs as these are the easiest to understand and interpret. In most cases the adjusted RR and the raw RR did not differ much, suggesting that confounding is not a big issue.

13. Although informative, we did not compare (1) adenoidectomy plus unilateral or bilateral grommets with adenoidectomy without grommets or (2) unilateral or bilateral grommets with non-surgical treatment as these analyses do not fulfil the criteria for inclusion in this IPD meta-analysis.

14. Most data on recurrent AOM come from Finland and this may have influenced our results. Personal communication with the authors of these Finnish trials led us to realise that early surgery for recurrent AOM might be a cultural issue. Reasons for performing these trials included the high adenoidectomy rate in Finland compared with that in other Scandinavian countries, for example Norway. The authors also confirmed that the number of children undergoing adenoidectomy has decreased since the results of these trials were published. We are not aware of any societal or biological factors varying across Western countries that may cause the effects of treatment in Finland to differ from the effects of treatment in other countries.

15. In surgical trials such as those included in our IPD meta-analysis, only the children who are assigned to the control arm, who do not undergo the operation under study, can transfer, that is, move to a treatment arm because of persisting problems. Per-protocol analyses that exclude children who transfer to the other arm of the trial will therefore underestimate the effect of treatment. Conversely, analysing children on the basis of time spent in a treatment arm might overestimate or underestimate this effect.

16. Finally, nine out of the 10 included studies made a comparison between adenoidectomy with grommets and grommets alone. The comparisons from these nine studies therefore seem to provide the most robust evidence.

Clinical implications

The effect of adenoidectomy seems to vary with age both in children with recurrent AOM and in children with persistent OME. However, it may not be age per se that is relevant but rather the presence of differing pathophysiological mechanisms in different age groups that serves as the modifier, with age simply being a proxy for these. OM is known to result from an interplay between microbial load and the immune response. The Eustachian tube is the port of entry for middle ear pathogens from the nasopharynx, but it also plays an important part in clearing middle ear secretions. One responsible mechanism may relate to the relative immaturity of the immune response in younger children, particularly those children aged < 2 years. As the immune response improves with age, problems with infection may recede and those associated with a middle ear effusion, that is, hearing loss, become relatively more apparent. This coincides with a period of social and behavioural change in the child’s life. Starting school and becoming part of a peer group may be the factors that initiate concerns about performance and increase awareness of the child’s hearing. More insight into the pathophysiology of OM is needed to understand better the causal mechanisms of the subgroup effects. In the meantime, clinicians can still use the information about age and indication as part of the shared decision-making process when deciding with parents whether or not to operate on their child.

Like all health-care decisions, the decision whether or not to undertake adenoidectomy will be influenced by existing local practice. It is extremely unusual for adenoidectomy to be undertaken in the UK in children aged < 2 years for any indication other than obstructive sleep apnoea, when it may be performed with tonsillectomy. In other northern European countries adenoidectomy is common in young children, as the presence of such children in some of the included studies indicates.

In our analyses we found no evidence to support the suggestion by NICE that possible benefits of adenoidectomy should be considered in children selected for grommets for OME who also suffer from respiratory symptoms. The number of upper respiratory tract infections was not a predictor of failure at 12 months. However, this might be because upper respiratory tract infections were measured in only one trial, resulting in a relatively low number of children in whom upper respiratory tract infections could be studied as a predictor.

As with all interventions (and particularly in the case of surgical procedures), consideration must be given to the ratio between benefits and harms. In the children aged < 4 years a modest benefit may easily be outweighed by the potential risks and harms. Those same risks and harms are the reason why the overall small beneficial effect found for children of all ages should not be used as the rationale for treating all children.

The trials included in this IPD meta-analysis provide very limited information on the adverse effects of adenoidectomy, an understanding of which is clearly critical when undertaking a risk–benefit analysis before making a treatment decision.

Contribution of authors

Disclaimers

This report presents independent research funded by the National Institute for Health Research (NIHR). The views and opinions expressed by authors in this publication are those of the authors and do not necessarily reflect those of the NHS, the NIHR, NETSCC, the HTA programme or the Department of Health. If there are verbatim quotations included in this publication the views and opinions expressed by the interviewees are those of the interviewees and do not necessarily reflect those of the authors, those of the NHS, the NIHR, NETSCC, the HTA programme or the Department of Health.

Which children with otitis media with effusion benefit most from adenoidectomy with or without grommets (ventilation tubes)? An individual patient data meta-analysis

Authors: Maroeska Rovers, Martin Burton, Arno Hoes, Anne Schilder

Summary

Current evidence suggests that overall effects of adenoidectomy are limited. Nevertheless, in day-to-day practice, both general practitioners and ENT-surgeons believe that certain subgroups of patients may benefit from adenoidectomy. So far, identifying these subgroups has been problematic, as the individual trials have been too small to allow valid and reliable subgroup analyses to be performed. Consequently it is likely that both over- and under-treatment occurs. The proposed Individual Patient Data (IPD) meta-analysis will include the individual data of about 2000 children aged 10 months to 12 years who participated in 9 to 12 randomised controlled trials into the effectiveness of adenoidectomy.

We aim to:

1. predict which children referred for adenoidectomy have a higher risk for a prolonged course of their upper respiratory tract infections, and

2. identify subgroups of children that most likely benefit from adenoidectomy.

The current proposal will lead to a better diagnostic and treatment protocol for children with otitis media for whom the general practitioner considers referral for adenoidectomy.

Background

Adenoidectomy is one of the most frequently performed surgical procedures in children in high-income countries. Annual adenoidectomy rates differ, however, between countries. When considering rates per 10,000 children per year, they vary, being 127 in Belgium, 101 in the Netherlands, 39 in England and 24 and 17 in the United States and Canada, respectively. 1

Indications for adenoidectomy include recurrent episodes of acute otitis media (AOM) and persistent otitis media with effusion (OME). 2 The operation involves removing the adenoid – a nasopharyngeal reservoir of potential respiratory pathogens and a potential cause of obstruction of the nasal airway. Removal is thought to improve Eustachian tube function.

We recently performed a Cochrane review, which showed a significant benefit of adenoidectomy as far as the resolution of middle ear effusion in children with OME is concerned. 3 However, the benefit to hearing is small and the effects on changes in the tympanic membrane are unknown. The risks of operating should therefore be weighed against these potential benefits. Furthermore, no benefit of adenoidectomy was found on AOM. However, in day-to-day general practice about one in ten children with otitis media is referred to an ENT-surgeon and in some countries the vast majority of these children are selected for surgery. This confirms that both general practitioners and ENT-surgeons believe that certain subgroups of patients may benefit from adenoidectomy. The identification of these subgroups has, however, been problematic, as the individual trials performed so far were too small to perform valid and reliable subgroup analyses. Consequently it is likely that both over- and under-treatment occurs. An IPD meta-analysis, i.e. a meta-analysis on the individual original data of previously performed trials, offers the unique opportunity to identify subgroups more likely to benefit from an intervention. 4,5 Since IPD meta-analyses include more detailed data, they have much more statistical power to carry out informative subgroup analyses. Furthermore, by using individual patient data the flexibility of subgroup analyses is enhanced. Consequently, the estimated subgroup effects may be less influenced by misclassification and (ecological) bias. IPD meta-analysis therefore allows a more thorough assessment as to whether differences in treatment effects across subgroups are spurious or not. 6 An earlier IPD meta-analysis from our research group in children with acute otitis media showed that such an approach is capable of identifying subgroups of children that are most likely to benefit from antibiotic therapy. 7,8

Objective

In this IPD meta-analysis with the individual data of at least nine (but probably 12) randomized controlled trials into the effectiveness of adenoidectomy including more than 2000 children aged 10 months to 12 years, we aim to:

1. predict which children referred for adenoidectomy have a higher risk for a prolonged course of their upper respiratory tract infection

2. identify subgroups of children that benefit most or are most likely to benefit from adenoidectomy.

Methods

Time schedule

The research will start in 2011. A time line is provided below.

The proposal will be undertaken in Utrecht by a PhD student (Chantal Boonacker) under primary supervision of Professor Maroeska Rovers (clinical-epidemiologist, Utrecht) and Dr. M. Burton (ENT-surgeon, Oxford) and supported by a systematic reviewer based in Oxford. Professor Arno Hoes (clinical-epidemiologist, Utrecht), Professor Anne Schilder (Paediatric ENT-surgeon and clinical trialist, Utrecht and London) and Professor George Browning (ENT-surgeon with special expertise in otitis media trials) will be available for further support. Project management and administrative support will be provided by staff from the Cochrane ENT Disorders Group in Oxford.

The PIs of the original trials will also be involved by means of regularly updates. The commitment and enthusiasm of several has been demonstrated by their already having provided their raw data. A formal meeting will be held in Oxford (September 2012) to discuss the results and funding for the senior PIs to travel to this meeting (several from the USA) has been included in the budget. It seems likely that new research questions will emerge from this work and an important element of this project is to continue a process of international engagement and dialogue with the world’s most experienced otitis media triallists to ensure that the results of this review are used to best effect.

If statistical questions or problems arise, which Maroeska Rovers cannot solve, she will consult her colleagues from the Cochrane IPD meta-analysis methods group, i.e. Mike Clarke, Jayne Thierney or Lesley Clarke. For more details see the table at the following page.

References

Methods

Outcomes

The most notable adaptation was the decision to change the primary outcome from ‘number of otitis media episodes’ to ‘failure at 12 months’ – a composite endpoint which was already included in the protocol as a secondary outcome. When our study was underway and we were analysing the available datasets, we realised that data on the original desired primary outcome would not be available, because the original trialists did not evaluate outcomes in the same way. As is usual in many IPD meta-analyses, rather than presenting an ‘empty’ review of little utility we sought – through consensus with clinicians, including the original trialists – to develop a composite primary outcome that allowed inclusion of all of the trials. The aim of this is, of course, to make maximum use of the available information to develop a robust answer to the clinical questions posed. We emphasise to those who may be unfamiliar with this specific form of secondary research that re-defining a primary outcome in this way is significantly different from changing outcomes in primary research.

The composite primary outcome used was ‘failure at 12 months’, defined as one or more of the following components:

1. four or more AOM episodes (including episodes of otorrhoea) per year

2. presence of effusion for ≥ 50% of time

3. need for additional surgery

4. hearing^a improved by less than 10 dB.

^aHearing was expressed as a mean air conduction hearing level measured by age-appropriate audiometry (if possible averaged over 500, 1000, 2000 and 4000 Hz). In all children the binaural average was taken. This includes trials that used unilateral grommets and randomised ears rather than children.

The choice of this primary outcome also had consequences for the secondary outcomes. Based on the availability of data, the time points for the secondary outcomes were changed from 3, 6, 12 and 24 months to 6, 12, 18 and 24 months.

Secondary outcomes

1. Each of the individual items used as part of the definition of ‘failure at 12 months’. That is:

   • four or more AOM episodes/year

   • presence of effusion for ≥ 50% of time

   • need for additional surgery

   • hearing improved by less than 10 dB.

2. Number of episodes of AOM during follow-up:

   • in the first 6, 12, 18 and 24 months.

3. Time with effusion during follow-up measured in number of weeks:

   • in the first 6, 12, 18 and 24 months.

4. Additional surgery during follow-up:

   • in the first 6, 18 and 24 months.

5. Average hearing loss measured in dBHL:

   • after 6, 12, 18 and 24 months follow-up.

6. Improvement in hearing level < 10 dB:

   • after 6, 18 and 24 months follow-up.

7. Adverse effects and events (including morbidity of surgery).

Time schedule

The research did not start in April 2011 as suggested by the timeline, but in December 2011. The formal meeting was indeed held in September 2012 in Oxford, United Kingdom and was preceded by an informal meeting during the ESPO conference in Amsterdam, the Netherlands.

Cochrane Central Register of Controlled Trials

PubMed

EMBASE (Ovid)

Cumulative Index to Nursing and Allied Health Literature (CINAHL) (EBSCOhost)

Web of Science

Bioscience Information Service previews/CAB Abstracts (Ovid)

Introduction

As requested in the brief, we have summarised the outstanding research questions based on the specific project and its position in the range of systematic reviews that it complements, covering grommets and adenoidectomy in children with OME (‘glue ear’), and recurrent upper respiratory symptoms.

Several specific recommendations are made and these follow the generic answers to the questions posed.

Briefly explain why this research or evidence is important to the NHS or for wider public health

Upper respiratory tract infections are the most common infectious diseases in children. It is one of the most common diagnoses in primary care. Although many infections are self-limiting, some require treatment.

What is the patient/population group likely to benefit from this research and why/how will they benefit?

Children with recurrent upper respiratory tract symptoms – especially nasal obstruction and discharge and symptoms of recurrent AOM and chronic OME – are often considered for surgical intervention (grommets, adenoidectomy and tonsillectomy). Alternatively they are treated non-surgically or are simply ‘actively monitored’ with a period of ‘watchful waiting’.

In what setting will the outcome of the research be delivered (e.g. primary care, the community)?

The optimal management of the population comprising children with recurrent upper respiratory tract symptoms needs to be considered on a population basis rather than in terms of distinct strategies in primary or secondary care. We know that expectations set in primary care can have profound effects on management choices in secondary care. Shared decision-making processes, based on outcomes important to patients, must be supported by relevant evidence in all health-care settings.

Important research questions