[ Pobierz całość w formacie PDF ]
.Picking Cherries and Slicing SalamiOne would think that withholding negative data and publishingonly the successful trials would be sufficient to maintain the  dirtylittle secret.But the pharmaceutical companies had other tricksup their sleeves.Whereas many of the negative trials were notpublished at all, some of the positive trials were published manytimes, a practice known as  salami slicing , and this was oftendone in ways that would make it difficult for reviewers to knowthat the studies were based on the same data.22 In some cases,the authors were different, and references to previous publica-tion of the data were often missing.Sometimes there were minordifferences in the data between one publication and another, aswell as between the data as presented to regulatory agencies andthe data as published.So a reviewer trying to summarize thedata would be likely to count the positive data more than once.Another trick was to publish only some of the data from aclinical trial, a manoeuvre that researchers call cherry-pickingthe data.Some clinical trials are conducted in more than onelocation.These are called multi-centre studies.Multi-centrestudies make it easier to find sufficient patients to conduct thetrial.They also make it easier to cherry-pick the data.Forexample, one multi-centre study of Prozac was presented to theFDA as showing a drug placebo difference of three points onthe Hamilton scale.When data from this clinical trial waspublished, the difference was reported as 15 points  a five-timesincrease in effectiveness.How was this magical augmentationof the benefits of Prozac accomplished? The full study wasconducted on 245 patients.The published paper reported datafrom only 27 of these patients.In the published version, the datafrom the bulk of the patients were left out, making the drugseem much more effective than it really was.42 The Emperor s New DrugsDrug companies also publish  pooled analyses of the trialsthey have conducted.That is, they bundle together the resultsof different trials and analyse the drug placebo difference acrossthem.This is similar to the meta-analyses my colleagues and Ihave conducted, but with one important difference.Our meta-analyses, in common with most others reported in the scientificliterature, are based on all of the studies that we were able tofind.In contrast, the drug companies pick and choose whichstudies they wish to include in their pooled analyses.For example,GlaxoSmithKline submitted 15 clinical trials of Seroxat to Swedishregulators.In addition to being published individually  some-times more than once  studies with positive results were alsoincluded in six different pooled analyses.Most of the studies withnegative results were, of course, not included in the pooledanalyses.There is yet another way in which pooled analyses can hidenegative data.Rather than not publishing the negative data atall, the companies can bundle them together with data from posi-tive trials, so that the overall result is positive.By so doing, theycan truthfully claim that they have published the data from anegative trial, while hiding the fact that those data showed nodifference between drug and placebo.The article by the Swedishregulators showed that the data from about 20 per cent of clin-ical trials were not published at all.The data from another 20per cent of the trials were bundled together with data from moresuccessful trials, so that their negative results were hidden fromview.Taken together, approximately 40 per cent of the data arekept out of sight.23 Practices like this make antidepressant drugsseem much more effective than they actually are, and theyalso make it exceptionally difficult for reviewers to establish howeffective the drugs really are.Perhaps the best indication of the difficulties that are posedby selective publication is provided by the problems that NICEfaced when drawing up its 2004 guidelines for the treatment ofdepression.Having read our meta-analysis of the FDA data, NICEcontacted me in the hope of adding the unpublished data to theirThe  Dirty Little Secret 43analysis of the published data.Although they wanted to includethe unpublished as well as the published data, they did not wantto include any of the data more than once, because that wouldhave biased the results [ Pobierz całość w formacie PDF ]