The CONSORT guidelines provide a coherent statement of standards for the practice and reporting of randomized trials-a statement that is particularly welcome for its brevity and clarity. The recommendations highlight the importance of the fundamental building blocks that minimize bias and provide the scientific rigor needed to make meaningful clinical inferences from trials. These building blocks include adequately describing the hypotheses to be tested, the sample studied, and the process of randomization; accounting for all randomly assigned participants; and providing information on either primary end points for binary outcomes or estimates of mean and variance for continuous data.

If widely implemented, the CONSORT guidelines should go far in minimizing the serious deficiencies, which are now well documented, in the design, analysis, and reporting of trials [5]. The purpose of CONSORT is to highlight these deficiencies and to help ensure that the reporting of trials provides clinicians, researchers, policymakers, and other persons who interpret trial findings with the required data [6].

The effects of many modern medical treatments are likely to be small, although potentially worthwhile, thereby making it necessary to randomly assign large numbers of patients. Although large (and expensive) single trials that address questions of clinical importance have sufficient statistical power to provide precise answers about treatment effect, such large trials remain scarce in many clinical areas. Meta-analyses that synthesize data from numerous, small, similar trials provide an alternative method with which to achieve the necessary statistical precision and may do so with increased generalizability [7]. But meta-analyses are only as good as the sample of trials they include [8]. Even with the most sensitive search strategies, however, it is almost inevitable that relevant trials-especially those that have not been published in journals-will be missed. Manuscripts that inadequately report trials only compound this deficiency because they are less likely to be published, thereby contributing to publication bias. If trials are omitted from a meta-analysis in a random manner, the results will be less precise than when all well-conducted trials are included. Such incomplete meta-analyses risk providing uncertain answers when more definite ones are potentially available from existing research.

Moreover, when availability of data is affected by some systematic influence, such as the bias against publication of trials that do not show a statistically significant difference between treatment and control groups, the results of meta-analyses will reflect that bias. It is a relatively common practice to describe results as "not significant" and to provide no estimate of variance in trials in which only a small and uncertain effect is seen. On the other hand, reporting results adequately when trials detect effects that are statistically significant or of potential practical importance is also relatively common. Such inappropriate reporting practices may similarly introduce important biases into the results of meta-analyses. In addition, relevant data are frequently not available in published reports; for example, in a meta-analysis of published comparative trials of antidepressant drugs, primary outcomes were adequately reported in only 20 of 53 (38%) trials [9].

A strong case can be made to patients to participate in randomized trials [10]. However, this imperative is seriously undermined when data on the primary end points from existing trials are not made available to guide the design of future trials and the treatment of future patients. Even if the CONSORT recommendations are universally adopted, occasions will arise when reviewers will require unpublished data from trials (for example, in the analysis of rare events that are not considered to be primary outcomes in small trials and can only be handled appropriately in large overviews) or will wish to use individual patient data that may provide more precise and valid estimates of treatment effect [11, 12]. In such situations, funding bodies, including pharmaceutical companies, should make data available rather than withholding or obfuscating it on grounds of serving commercial or other interests. Given the potential importance of publication bias [5, 8, 13], this obligation should extend to unpublished trials and unpublished data from published trials. Many trials funded by the pharmaceutical industry fail to report valid data for further interpretation and analysis [9, 14]; in some instances, such trials may appropriately be considered marketing rather than science.

To make efficient decisions about the use of health care interventions, policymakers and clinicians need to know about potential costs, as well as diagnostic and therapeutic effects [15]. There is a good case for including economic analyses alongside large, "real world" randomized trials [16]. Many mistakes can be made, however, in the course of designing, implementing, and reporting such analyses. The CONSORT guidelines clearly cannot solve all of the problems surrounding the conduct and reporting of clinical trials, but if the CONSORT checklist [1] included one additional line in its protocol section that read "Planned economic analysis (if warranted)" and provided a key reference (for example, Drummond and colleagues [17]), this area would be dealt with simply and adequately.

A major challenge surrounding the CONSORT guidelines is implementation. Although these guidelines should improve the usefulness of trial reports in journals that formally adopt them, usefulness is more likely to be enhanced if editors request referees to review manuscripts with these reporting standards in mind.

Some persons may see the suggested structure for a trial report as arbitrary and intrusive. For example, it is unclear why describing planned subgroup analyses or covariate analyses in the introduction is preferable to describing them in the methods section, which currently seems to be a common practice. In light of this concern, it may not be surprising that the lead author of a "guinea pig" paper that rigorously followed the earlier SORT guidelines described "mixed feelings" on the experience [18]. To those interested in interpreting results of trials for the benefit of patients, the availability of adequate descriptions of trials is what really matters; the precise location of data in reports is less important.

If use of the CONSORT guidelines remains restricted to a few general journals of international prominence, their effect on most trials (the reports of which are primarily published in subspecialty journals) may remain limited. One potential way to augment their effect would be to convince prospective authors (as well as referees) of the value of the guidelines to influence the authors' decision about where to send their contributions. Another way would be to actively market the guidelines to journal editors through professional networks. Ultimately, randomized trials can only achieve appropriate changes in practice if their results are reported completely, systematically, and in adequate detail. The CONSORT guidelines bring this goal closer; they should be welcomed warmly.