Commentary on Berlin et al.

Abstract

The public discourse on sharing individual patient data (IPD) from clinical trials has shifted rapidly, and the article by Berlin et al. [1] in this issue exemplifies the best and worst of this transition. The benefits of sharing IPD are clear. It allows third parties to verify trialists’ own analyses of their own data, especially where flawed analytic techniques have been used in the initial published results. It permits meta-analysis of pooled IPD frommultiple trials, which can give more accurate point estimates of benefits. That same technique offers greater statistical power for network meta-analyses on relative benefits of treatments and, crucially, greater power to conduct subgroup analyses, which can in turn help predict the best or worst responders to a treatment. Where adverse event data have been well coded, an IPDmeta-analysis can permit more powerful exploratory analyses of side effects. It facilitates innovation, by allowing new hypotheses to be explored in existing data. Finally, where data are shared on abandoned products, it presents an opportunity to identify signals that suggest potentially beneficial new uses for old treatments. All these research questions could be answered by running new large trials, instead of repurposing old data, but that would be an inefficient use of resources, the most important being patients. People often tolerate both risk and inconvenience to participate in trials, expecting that the results of their own experience, pooled with others, will help improve treatment decisions for patients like them in the future. Failing to use the data is a betrayal of past participants, but may also present ethical challenges for future work, because it is arguably unacceptable to expose new patients to randomization where the clinical question being interrogated could be easily answered with data that have already been collected. Over the past year, there have been broadly positive statements on data sharing from various bodies including the European Medicines Agency [2], the European Federation of Pharmaceutical Industries and Associations and the Pharmaceutical Research and Manufacturers of America [3], GlaxoSmithKline [4], Roche [5], Leo Pharma [6], and more, with an Institute of Medicine review on best practice to come, as well as a Food and Drug Administration consultation. This represents strong progress. But we should not be so naive as to imagine that IPD sharing will happen soon, and there is a great deal we can learn from the ongoing failure to ensure that even summary results from clinical trials are reported. Iain Chalmers, the founder of the Cochrane Collaboration, described 7 years ago how a rush of positive activity on publication bias in the late 1990s soon deteriorated into failure, despite various codes of conduct and promises from major drug companies [7]. There is an active campaign at http:// www.alltrials.net seeking to address this ongoing issue, almost three decades after the phenomenon of trial results being withheld from doctors and patients was first quantitatively documented [8]. The current National Institute of Health Research (United Kingdom) review summarizes dozens of studies on publication bias: overall, the chances of a trial being published are roughly half [9]. While much of this evidence covers trials from the past it is far from historical, as these trials cover the risks and benefits of the treatments in use today. The problem also persists for current trials: the most recent study still finds that only half of all trials on clinicaltrials.gov have posted results, 3 years after completion [10], and legislation on this issue has failed. The Food and Drug Administration Amendment Act 2007 required trials to post results on clinicaltrials. gov within 12 months, but offered no routine public audit on compliance: the only published paper on compliance estimates that 78% of trials failed to post results as required [11]. At the same time, access to Clinical Study Reports is becoming actively worse. These long documents are created for industry trials and little known in the academic community, but contain a wealth of detail