Medical progress is possible only because altruistic volunteers put themselves at risk in clinical trials. The results of those trials are then used to inform medical decisions. The traditional system of relying on investigators, sponsors, and journal editors to decide whether, when, and how to report trial results was based on trust. There was no way to know what trials had been conducted, what data were collected, how they were analyzed, and whether the reported data were complete and accurate. Policies mandating the registration of trials and the reporting of summary results were instituted to provide greater transparency. In turn, this greater transparency brought greater awareness that the aggregated, or summary, data used to report results were not always a valid or sufficient reflection of the underlying data. As a result, the trial-transparency movement was broadened to include a focus on the participant-level data.
Summary data are an efficient way to communicate trial results and conduct common statistical analyses. However, the transformation from participant-level data to summary data involves the loss of information. For example, a magnetic resonance imaging scan may need to be interpreted and then coded; emergency department notes may need to be used to determine whether a trial participant met the criteria for a heart attack. Inconsistent data for a given participant need to be resolved, the analysis population established, and a descriptive statistic calculated (e.g., mean or median change in response). It is not surprising that different ways of summarizing the same data may lead to substantially different conclusions. Commonly used statistical analyses are based on the presumption that these decisions were prespecified before the data were examined, though in practice the degree of specificity varies and subjective judgments are inevitable. The trial-transparency movement argues that by opening up participant-level data, along with the protocol and other documents, the underlying decisions become part of the public record, providing accountability and allowing scientific debate. Proponents also argue that the availability of such data will allow interested parties to use participant-level data for additional analyses as a preliminary test of a new idea or to combine data from multiple studies to seek previously unidentified associations.
To read full article: http://www.nejm.org/doi/full/10.1056/NEJMe1307268