Glue ear in children is a significant cause or morbidity, in which hearing loss may affect development. Because glue ear resolves spontaneously in most children, watchful waiting has been recommended for up to six months before surgery [1
], and this recommendation is now common practice. Variability in the success rates with placebo between 5% and 38% in these seven studies is likely to reflect different policies over a quarter century, as well as the random play of chance.
On average (Table ) for every 100 children given placebo in the seven trials, 17 would have avoided surgery (95% confidence interval 11 to 23). That figure is similar to rates in contemporary British settings. In Dundee in the late 1980s, 33 of 191 (17%) of children referred for surgery resolved without surgical treatment [25
]. In the placebo group of the large randomised study in Oxford in the late 1990s, the resolution rate without surgery was 13% [19
], while in a watchful waiting arm of another large randomised study in Bristol 15% did not require surgery [2
]. In terms of resolution rates, the pooled data in the meta-analysis resembled clinical practice.
Children who have not undergone a period of watchful waiting may be regarded as having a less severe form of the disease, and S-carboxymethylcysteine may be less effective where the disease was longer established. In the four trials in this meta-analysis with placebo resolution rates of 32–38% the number of children was small, at 3, 16, 19 and 20 respectively, contributing about a quarter of those given placebo. Disease severity is unlikely to confound the results, therefore.
On average (Table ) for every 100 children given S-carboxymethylcysteine in the seven trials, 35 would have avoided surgery (95% confidence interval 29 to 41). There was a consistent benefit for use of S-carboxymethylcysteine with an overall relative benefit of 2.0 (1.4 to 2.8). One child in every five or six given S-carboxymethylcysteine for one to three months would be spared an operation for grommet insertion. Only one adverse effect was reported in these seven trials.
How confident can we be of this result? Clearly there is the potential for negative studies not to have been published, though no unpublished studies were found in searching a manufacturer's database of trials. Extensive searching both here and in a previous review [6
] using a different search strategy found the same studies. We excluded one study [26
] from the previous review because it included children having myringotomy. Quality scores were 2 in three studies and 4 or 5 in four. We know that a quality score of 2 or less may be associated with over-estimation of treatment effects [27
]. Here, both lower quality scores, lower validity scores, and trials with fewer than 50 children yielded non-significantly better estimates of effects (Table ). Intention to treat analysis produced better results than per protocol analysis, in that per protocol analysis was not significantly different from placebo in three small studies.
Sensitivity analysis showed that choosing only studies of higher quality, higher validity, those that reported intention to treat results, or were larger made no appreciable difference to the overall conclusion. The proviso, though, is that for these sensitivity analyses the numbers of children or ears was small [29
], and no definitive conclusions should be drawn other than the absence of gross effects.
Outcomes chosen by investigators over 25 years differed. We chose to use endpoints that would today be reliable indicators that a child could sensibly be spared an operation for grommet insertion. Re-analysis of the Pignatoro review [6
] was necessary because it did not choose sparing operation as an outcome, because we were unable to reconcile the data in that paper to the odds ratios given, because we wanted outputs in the form of number-needed-to-treat, and to add information from our own study [19
When treatments have small effects and studies have small numbers, a clinically significant effect of treatment may be missed [29
]. To be sure of the direction and the magnitude of treatment large numbers of children have to be studied, and for S-carboxymethylcysteine studies of over 1000 children would have to be conducted to be sure of the number-needed-to-treat with a low confidence interval [29
]. This review included fewer than 300 children and 150 ears, reflecting the wide confidence interval of the number-needed-to-treat estimate.
Despite this, S-carboxymethylcysteine is likely to be a useful treatment of children during a period of watchful waiting before surgery. It spares about 18% of children an operation, and as well as being clinically effective an economic analysis [30
] showed it to be cost-effective when the number-needed-to-treat is better (lower) than 15. With a number-needed-to-treat of about 5, use of S-carboxymethylcysteine would probably result in fewer children having surgery for grommet insertion, and actually reduce overall costs as well as relieving pressure on surgery time and waiting lists.
However, if spontaneous resolution rates were much lower than 10% any conclusion about cost-effectiveness might have to be reviewed. If the relative benefit of 2 found in this meta-analysis was applied to different spontaneous resolution rates, different numbers needed to treat would be produced. With a 20% spontaneous resolution rate doubled to 40% with S-carboxymethylcysteine, the NNT would be 4. Doubling a 15% rate to 30% gives an NNT of 7. Doubling a 10% rate to 20% gives an NNT of 10, but doubling a 5% rate to 10% gives an NNT of 20. This demonstrates that statistical outputs, while crucial to any analysis, may be less important than therapeutic effort in making clinical policy decisions.
One strength of this review came from choosing trials and patients that would today be likely to have surgery for grommet insertion because of the severity of their disease, and demonstrated unequivocally using tympanography and audiometry, and after watchful waiting. It ensured clinical homogeneity. Another strength was the choice of outcome to ensure that an operation would not be carried out because of disease resolution, again using contemporary criteria. The weakness is that despite having access to data from seven studies, the total numbers are small, and the total numbers available for sensitivity analysis makes that unreliable.