We are disappointed by the emphasis that the articles on database research in the supplement published on 15 October 1997 place on sophisticated new mathematical models to control for confounding factors and by the classic commonplace that clinical database studies are "attractive alternatives to randomized trials" [1]. Using large databases to compare therapies remains controversial [2]. By design, databases record observations made in clinical practice. Because treatment decisions are not randomly allocated, any observed therapeutic effect may be due to unrecognized factors affecting the treatment allocation rather than the treatment itself.

It is surprising that a supplement focused on the future of databases did not mention new research methods, such as cross-design synthesis [3], directed toward the generation of results with an acceptable balance between internal and external validity. Specifically, cross-design synthesis proposes the assessment, adjustment, and combination of treatment effects obtained with randomized studies and database analyses.

Our group has recently proposed the simultaneous use of both experimental and observational methods to assess the effectiveness of drugs [4]. Because randomization is essential to minimize comparison bias, we suggested the possibility of including randomization modules in new computer-based patient records. To our knowledge, this is the first time that the use of randomization with large databases has been proposed in the medical literature. These "randomized database studies" could be conducted when there is an interest in evaluating the actual effectiveness of several different options with demonstrated "equivalent" efficacy in previous research. Ideally, new computer-based clinical records should automatically suggest that patients are candidates for inclusion in a specific study. The progressive implantation of clinical practice guidelines and the new computer support systems for prescribing might contribute to better identification of the best diagnostic and therapeutic options in every clinical situation. This research strategy could be used to complete with "medicine-based evidence" the circle that started with evidence-based medicine.