Data transparency in clinical trials: Implications of Restoring Study 329

All psychotropic drugs have significant adverse event profiles, and none are curative. This does not place psychiatric prescribing on a weaker footing than many areas of medicine. But it does necessitate that prescribing decisions be based on a careful deliberate calculation of relative benefits and harms. Unfortunately, as Mulder and colleagues point out, the medical literature is pervasively misleading, overestimating the efficacy of drugs and underestimating their adverse effects. Therefore, prescribing decisions are potentially compromised. Conse-quently, to protect patients, prescribers should both downgrade the reported benefits of a drug and assume that its harms are more serious and frequent than reported. The drug should be prescribed only if deflated benefits outweigh inflated harms. For antidepressants in adolescents, this equation will rarely favour prescribing; in younger children, almost never. This now well-established conclusion presents a major challenge to the prescribing practices of many psychiatrists and general practitioners (GPs) (Jureidini, 2016).

However, there are implications well beyond antidepressant prescribing. As Mulder and colleagues point out, clinical trial registration has not solved the problem of misleading reporting. This is not only because of the poor compliance that they have identified. The sad and frightening conclusion from our reanalysis of Study 329 (Le Noury et al., 2015) and similar exercises (Jureidini et al., 2016) is that, without access to individual patient-level data (i.e. the forms filled out when patients attend study sites) and original study protocols, we can not be sure that even registered drug trials have been properly reported.

We cannot rely on researchers, drug companies, journals, universities or professional bodies to rectify this problem, as self-regulation requires acting contrary to vested interests. Notably, as Mulder and colleagues note, the Journal of the American Academy of Child and Adolescent Psychiatry, the Academy itself and Brown University have all failed to take appropriate action over Study 329. Reputational protection (an important form of conflict of interest that has not been adequately scrutinised) is likely a major factor in this failure.

Nor does peer review provide sufficient protection against misrepresentation of evidence from trials, because peer reviewers are denied access to the real data and may also be misinformed and biased towards drug use. Yet pharmaceutical companies ‘claim piously that their publications undergo peer review’ (Carroll, n.d.), and they are supported in this stance by journals and universities that overstate the effectiveness of peer review, because it is a key tenet of their claimed objectivity.

Because we cannot trust corporate analyses or the peer review process, Carroll argues that ‘Corporations and investigators should be prohibited from publishing their own in-house statistical analyses’. Instead data should be analysed by regulatory agencies strictly according to registered protocols and analysis plans, in the same way that manufacturing corporations’ scientific evaluations of safety and performance of medical devices are checked by regulators.

There is a cheaper alternative to verification by regulatory agency oversight (one that would also be less subject to financial influence, given that regulatory agencies often rely on industry fees). As well as mandatory trial registration and publication of detailed trial methodology, researchers should be obliged to provide access to their data by publishing data summary tables as soon as the study is completed. There should also be efficient mechanisms whereby legitimate researchers could apply for access to patient-level data, with the minimal level of redaction compatible with reasonable privacy requirements (Taichman et al., 2016). Both the original investigators and other researchers wishing to make use of the data could then prepare their own or collaborative analyses and submit them for publication, with research contributions and potential conflicts of interest clearly documented. To address concerns about appropriate recognition and reward for the original researchers, protocols might be needed to protect first opportunity to publish data analysis. Credit would also accrue from producing data that generated publications from other researchers. What is most important is that real authority lies with the data, not any individual interpretation.

Too much harm has been done by biased interpretation and publication. We have an obligation, both to those who volunteer to participate in medical research and to those to whom we will prescribe, to ensure data are used honestly and to their maximum potential. We must therefore give priority to data transparency over bids to hide data on grounds of claims about intellectual property or proprietary interests.