Commentary on Berlin et al.

Abstract

Openness and transparency are noble principles in the research enterprise. Trial registration is a game changer in terms of its potential for reducing redundant research, minimizing selective outcome reporting bias, and addressing publication bias. The ‘open data movement’ has set its sights on the next step – making data sets of completed trials open to the public. Access to clinical trial data sets is a logical and constructive next step. However, we concur with Jesse Berlin et al. [1] that much work, debate, and discussion are still required to establish ground rules and procedures for the secondary use of trial data and that unrestricted access to such data sets is not without serious risk. In their commentary, the authors focus on the need for explicit data-sharing principles and agreements in the context of aggregating clinical trial data sets for meta-analytical purposes. Specifically, they suggest registration of researchers, disclosure of details regarding their proposed research, signing of data-use agreements, and scientific review of proposed research plans. We propose that the need and scope for data access principles and process likely extend beyond those required for meta-analyses. While misuse of data can certainly occur in poorly conceived or conducted metaanalyses, this represents but one type of secondary analysis of trial data. In principle, having clinical trials analysis repeated and verified may be in the public interest. However, having groups from around the world perform hundreds of subgroup and secondary analyses may be neither wise nor warranted. Without proper process, justification, and analysis, open trial data can also be manipulated and misused intentionally or unintentionally. The impact of such research could contradict the primary findings of the clinical trial, degrade effective interventions, and promote ineffective or harmful interventions. Clearly, emphasis must be placed on the primary trial results and interpretation and confirmation of those results. Biased or unsound secondary analyses can also affect the integrity and reputation of academic investigators and their sponsors. Imagine a situation where a well-conducted academic trial that demonstrates a widely used profitable drug is moderately effective but is noted to increase mortality. As a consequence, the drug is removed from the market. Those with a vested interest in the drug such as long time enthusiasts, the manufacturer, and possibly regulatory authorities may be tempted to conduct a wide range of secondary analyses that question or contradict the primary findings. These results are then discussed in closed hearings or away from peer review or public scrutiny. Somehow, they are then used to discredit the primary results and reverse a prior regulatory decision to keep the drug off the market. Moreover, there is very little recourse for the academic trialists to protect their primary research findings, reputation, or integrity. Even though the provision of open and available data represents a potential public good, if misused, it may end up causing significant harms. How do we then strike the right balance? We suggest some of the following operational elements be adopted if a policy to allow open access to clinical trials information is adopted. First, we suggest that the sponsor follow a pre-established and public analytic plan. Registration of researchers and their proposed research plans as advocated by Berlin and colleagues would go a long way into ensuring the integrity and validity of secondary analyses. Current trial registration portals provide an ideal venue to register and append secondary analyses to the primary trial. It would afford a central point of access for all studies linked to the primary trial. This is appealing for consumers, decision-makers, and researchers. Second, given the many competing interests and high stakes associated with many trials, we suggest that clear processes for investigators should accompany access to open trial data sets. Indeed, once time has come to release a data set and the original statistical code, we suggest significant controls and