Article Text
Statistics from Altmetric.com
The systematic review and meta-analysis performed by Tasche and colleagues1 appears to have been carried out with commendable thoroughness. The results seem clear cut: sodium cromoglycate is ineffective as maintenance therapy in children with asthma. However, their conclusions are not objective but depend in the final analysis on their multivariate regression model. Perhaps this is where the most important bias has crept into the analysis.
As in the case of many new drugs, clinical trials are first carried out in adults, later in schoolchildren, and finally (if ever) in infants and pre-schoolchildren. This appears to have been the case with sodium cromoglycate since, of the studies included in their review, 12 out of 14 conducted before 1981 involved schoolchildren while, of those published since 1981, nine out of 10 involved pre-schoolchildren. The authors point out that “age of the children . . . was strongly correlated with year of publication”.
They chose to interpret the positive effect size of older studies as an indication of publication bias. An equally reasonable interpretation might be that sodium cromoglycate is more effective in schoolchildren than in pre-schoolchildren.
I wonder if the authors could calculate the size of the treatment effect in their selected trials separately for those trials predominantly involving schoolchildren and those involving predominantly pre-schoolchildren.
Systematic review and meta-analysis often lend spurious objectivity to the assessment of efficacy. In the final event statistics provide guidance, and some form of subjective judgement is required as to the clinical relevance of the analysis.
References
authors' reply We appreciate the comments by Professor Silverman and have calculated the treatment effect of sodium cromoglycate (SCG) separately for trials in schoolchildren and in children of pre-school age. We excluded Silverman's trial for the reasons given by Edwards et al in a recent letter in Thorax.1-1
For both schoolchildren and pre-schoolchildren and for both outcome measures (cough and wheeze) the test of homogeneity was negative: the study results were heterogeneous in both age groups. The pooled results, using the method of Dersimonian and Laird,1-2 are shown in table 1-1.
These results seem to confirm Silverman's assumption that SCG is more effective in schoolchildren than in pre-schoolchildren, although even in schoolchildren the tolerance interval for wheeze still includes zero. However, we think the conclusion of an age specific effect is as yet unwarranted. All studies in schoolchildren were performed in the early years of SCG, at a time when the quality of design, analysis and reporting of trials was not much of an issue. All the studies were performed on small numbers of children, thus yielding estimators with low precision, and used a crossover design, a design which is apt to yield biased results in cases of incomplete follow up. Publication bias might also account for these findings, given the funnel plot results. To confirm these results, a trial of adequate size needs to be performed in schoolchildren.