Response to Yang et al.

Abstract

( 1 ) for their interest in our study ( 2 ). Th eir main concern relates to the methodology of the systematic review in general, and the use of random eff ect models to pool the results in particular. We agree that the results of systematic reviews should not be applied uncritically, and have written on this topic previously ( 3,4 ). However, systematic reviews can be of value pro-vided results are evaluated carefully, and we believe our review is an example of his t ( 2 ). Th e problem with not looking at the totality of evidence from randomized controlled trials (RCTs), and instead sim-ply selecting particular studies, is that this can be open to bias, with authors choosing data that supports their point of view. Th is is especially true of the pharmaceutical industry, which can overemphasize posi-tive data ( 5 ). Th e latter point emphasizes an impor-tant issue that underpins evidence-based medicine: we do not only need to know whether a drug works, but also how eff ec-tive it is in a given clinical situation. Th is can be diffi cult to determine when there is unexplained heterogeneity in the data. A random eff ects model does not “ artifi cially widen ” he t confi dence interval as Yang et al. suggest, it simply uses an established statistical technique to express the uncer-tainty in the data. Th ere are examples when results of two positive RCTs are pooled and the overall result is no longer statisti-cally signifi cant, and indeed we observed a similar phenomenon when data for trials of glucorticosteroids in inducing remis-sion of active Crohn ’ s disease (CD) were pooled ( 6 ). Our interpretation of the data is that adalimumab is probably eff ective in preventing relapse of quiescent CD, given the results of RCTs of other biological therapies, and as a result the conclusion of our abstract stated: “ Biological therapies were superior to placebo … . in preventing relapse of quiescent CD”. W ith regard to their comment regarding pooling of data for fi stulae healing, we per-formed a subgroup analysis including only trials with duration of therapy in excess of 4 weeks, which demonstrated that biological therapies were more eff ective than placebo in terms of inducing healing of fi stulae. We believe that the series of systematic reviews on medical therapies for infl am-may or t bowl e disasee ( 2,6 – 10 ), all ol- flowing the same methodology, which underpin the American College of Gastro-oenlogre t s y ’ monogpah r ( 11 ), help o tinform the Gastroenterology community. Th e monograph highlights both the strengths, and the weaknesses, of the data to date. An example of the latter is that we do not have a precise estimate of the effi cacy of adalimumab in preventing CD relapse. Given all other data for biological therapies in preventing relapse of quies-cent CD, it is likely that the results of the CHARM trial will be most accurate ( 12 ). As is oft en concluded, however, more studies are needed. CONFLICT OF INTEREST Th e authors declare no confl st. erein ict tof